Ellis Norwood series in

MARINE SCIENCE
Series Editor: Dr. TOM ALLAN,
Institute of Oceanographic Sciences, Wormley,
Surrey

SATELLITE OCEANOGRAPHY
an introduction for oceanographers and
remote-sensing scientists
I. S. ROBINSON,
Lecturer in Physical Oceanography, Department of
Oceanography, University of Southampton
This broad-ranging book bridges the void between
brief review papers on oceanographic applications
of remote sensing and detailed technical reports
about sensors and satellites. The text fills this gap
by providing marine scientists with a broad general
introduction to the subject from first principles
and a thorough explanation of the different tech¬
niques for applying satellite data in oceanography.
It will also give sensor technologists and remote
sensing specialists an oceanographer's perspective of
satellite remote sensing in ocean study. It provides
a view across all types of satellite remote sensing
of the ocean, visible, infrared and microwave
frequencies, and both active and passive sensors.
Pitching the academic level at postgraduate/
senior undergraduate level, and assuming a general
scientific training. Dr. Robinson introduces the
subject from first principles, discussing central
ideas and theories. A section on fundamental
principles follows in a manner accessible to scien¬
tists from all disciplines, with no mathematical
competence assumed beyond that of any numerate
scientist. The next section explores the areas of
application using particular remote sensing tech¬
niques. Included here are certain mechanisms
which the non-physical scientists may not find
simple to grasp: in such cases, the author offers c|ip(iip(iiociio CiiO<^
assistance to the physical understanding by referring
to suitable texts. This is a unique book which
opens an exciting new area and will inform practi¬ Occidental
sing scientists, giving an overview of all types of
College
oceanographic satellite remote sensing. There is
enough information for specialist readers to
understand oceanographic remote sensing pro¬
cesses and their applications, and sufficient breadth
to inform the non-specialist where remote sensing
makes a contribution to marine science. LIBRARY

Readership: Oceanographers and sensor technologists

in field work, research, teaching or study, and marine cs^<3g)ag)ag)<jg>cg>apcg>
scientists and engineers. Environmental scientists, marine
geologists, space scientists, photogrammetrical scientists,
and applied physicists.
 SATELLITE OCEANOGRAPHY
An introduction for oceanographers
and remote-sensing scientists
ELLIS HORWOOD SERIES IN MARINE SCIENCE
Series Editor: T.D. ALLAN, Institute of Oceanographic Sciences, Wormley, Surrey
SATELLITE MICROWAVE REMOTE SENSING
T. D. ALLAN, Institute of Oceanographic Sciences, Wormley, Surrey
PHYSICAL OCEANOGRAPHY OF COASTAL WATERS
K. F. BOWDEN, University of Liverpool
REMOTE SENSING IN METEOROLOGY, OCEANOGRAPHY AND HYDROLOGY
Edited by A. P. CRACKNELL, Carnegie Laboratory of Physics, University of Dundee
SATELLITE OCEANOGRAPHY
1. S. ROBINSON, University of Southampton
NEW PERSPECTIVES IN MARINE GEOLOGY
R. C. SEARLE and R. B. KIDD, Institute ot Oceanographic Sciences, Surrey
MARINE CORROSION IN OFFSHORE STRUCTURES
Edited by J. R. MERCER, University of Aberdeen
 ^ SATELLITE
^OCEANOGRAPHY
An i^oduction for oceanographers
^nd remote-sensing scientists

I. S. ROBINSON, M.A.,Ph.D.
Lecturer in Physical Oceanography
Department of Oceanography
University of Southampton

ELLIS HORWOOD LIMITED

Publishers • Chichester

Halsted Press; a division of

JOHN WILEY & SONS
Chichester • New York • Ontario ■ Brisbane
 OCCIDENTAL

First published in 1985 by

ELLIS HORWOOD LIMITED
Market Cross House, Cooper Street, Chichester, West Sussex, P019 lEB, England

The publisher’s colophon is reproduced from James Gillison’s drawing of the

ancient Market Cross, Chichester.

Distributors:
Australia, New Zealand, South-east Asia:
Jacaranda-Wiley Ltd., Jacaranda Press,
JOHN WILEY & SONS INC.,
G.P.O. Box 859, Brisbane, Queensland 40001, Australia
Canada:
JOHN WILEY & SONS CANADA LIMITED
22 Worcester Road, Rexdale, Ontario, Canada.
Europe, Africa:
JOHN WILEY & SONS LIMITED
Baffins Lane, Chichester, West Sussex, England.
North and South America and the rest of the world:
Halsted Press: a division of
JOHN WILEY & SONS
605 Third Avenue, New York, N.Y. 10016, U.S.A.

© 1985 I.S. Robinson/Ellis Horwood Limited

British Library Cataloguing in Publication Data

Robinson, I.S.
Satellite oceanography: an introduction for oceanographers and
remote-sensing scientists. —
(EUis Horwood series in marine science)
1. Oceanography — Remote sensing
I. Title
551.46’0028 GC10.4.R4
Library of Congress Card No. 84-25142
ISBN 0-85312-598-8 (EUis Horwood Limited)
ISBN 0-470-20148-7 (Halsted Press)
Typeset by Ellis Horwood Limited.
Printed in Great Britain by The Camelot Press, Southampton.

COPYRIGHT NOTICE -
All Rights Reserved. No part of this publication may be reproduced, stored in a retrieval
system, or transmitted, in any form or by any means, electronic, mechanical, photocopying,
recording or otherwise, without the permission of Ellis Horwood Limited, Market Cross
House, Cooper Street, Chichester, West Sussex, England.
 Table of Contents

Foreword.11

Preface . 13

Chapter 1 Oceanography from space?

1.1 A new field of marine science?.17
1.2 A quarter of a century of ocean observation from space.18
1.3 The value of satellite data to oceanography.19
1.4 The scope of the book.22

Section A FUNDAMENTALS OF REMOTE SENSING

IN OCEANOGRAPHY
Chapter 2 The possibilities in space — space hardware and data
transmission
2.1 Space platforms and their orbits.26
2.1.1 Satellite vehicles.26
2.1.2 Satellite orbital dynamics.31
2.1.3 Spatial and temporal sampling characteristics of orbits... 36
2.2 Sensors on satellites.41
2.2.1 Sensors in relation to the electromagnetic spectrum.41
2.2.2 Sensor measurements.45
2.2.3 Spatial sampling capabilities of sensors.47
2.2.4 Recent and planned satellite sensors which have
oceanographic applications.51
2.3 Data retrieval.51
2.3.1 Satellite-to-earth data transmission.54
2.3.2 Data dissemination.56

Chapter 3 The possibUities for oceanography

3.1 The oceanographic capabilities of satellite sensors.60
3.2 Ocean science in the 1980s.64
3.3 Dynamical oceanographic processes.65
3.3.1 Ocean circulation.66
3.3.2 Tides.VI
6 Table of Contents

3.3.3 Fronts.72
3.3.4 Upwelling.74
3.3.5 Mixed-layer dynamics.75
3.3.6 Coastal and estuarine dynamical and topographic
phenomena.76
3.4 Ocean waves.77
3.5 Space and time scales of oceanographic phenomena, in relation to
satellite sampling.84

Chapter 4 Principles of remote sensing of the sea

4.1 Introduction.89
4.2 Sensor calibration.90
4.3 Atmospheric correction.92
4.4 Positional registration.96
4.5 Geophysical calibration ... . . . 99
4.6 Oceanographic sampling for‘sea truth’.100

Chapter 5 Principles of image processing

5.1 Image processing and oceanography.106
5.2 Digital image data from satellites.107
5.3 Image-processing hardware.112
5.4 Geometric processing.114
5.5 Masks and overlays.119
5.6 Smoothing, filtering, and noise reduction.121
5.7 Measurement of spatial variability.124
5.8 Enhancement of single-band radiometric information.124
5.9 Presentation of multichannel image data.129
5.10 Atmospheric correction and geophysical calibration.130

Section B OCEANOGRAPHIC APPLICA TIONS OF SA TELLITE

REMOTE SENSING
Chapter 6 Visible wavelength ‘ocean-colour’ sensors ^
6.1 Introduction.134
6.1.1 Scope.134
6.1.2 Colour sensors on Landsat.135
6.1.3 The Coastal Zone Colour Scanner on Nimbus 7.139
6.2 Aspects of optical theory relevant to ocean-colour viewing
from space.141
6.2.1 Definition of some optical quantities.141
6.2.2 Measurements made by a satellite sensor.144
6.2.3 Optical pathways in the atmosphere.148
6.2.4 Practical strategies for the atmospheric correction of
visible wavelength data.152
6.2.5 Scattering and absorption of light in sea-water.155
6.2.6 Oceanographic Interpretation of ocean colour.157
6.2.7 Light-scattering at the sea surface.164
 Table of Contents 7

6.3 Calibration and application of CZCS data.166

6.3.1 Sensor calibration.166
6.3.2 Atmospheric corrections.168
6.3.3 CZCS calibration algorithms.173
6.3.4 Applications in biological oceanography.175
6.3.5 Applications in dynamical oceanography.180
6.4 Oceanographic uses of Landsat MSS data.184
6.4.1 Data processing.184
6.4.2 Calibration of Landsat data for sediment and
chlorophyll.185
6.4.3 Sediment movement studied with Landsat data.188
6.4.4 Other oceanographic processes viewed by Landsat.190
6.4.5 Bathymetry using Landsat.191
6.4.6 Coastal geomorphology.192
6.5 Oceanographic use of other visible-wavelength satellite data.192

Chapter 7 Sea-surface temperature from infrared scanning radiometers

7.1 Introduction.194
7.2 The physics of infrared radiation.199
7.2.1 Thermal emission.199
7.2.2 Atmospheric absorption.201
7.2.3 Infrared sensors.204
7.2.4 Brightness temperature calibration of AVHRR data.... 205
7.3 The near-surface thermal structure in the ocean.207
7.3.1 The skin temperature.207
7.3.2 The thermocline structure of the ocean.208
7.3.3 The sea-surface skin-temperature deviation.210
7.3.4 Surface slicks.214
7.4 Atmospheric correction and SST calibration techniques.215
7.4.1 Atmospheric correction strategies.215
7.4.2 Multispectral and multilook techniques.216
7.4.3 Cloud-removal techniques.219
7.4.4 SST operational products.221
7.5 The potential uses of SST data from satellites.224
7.5.1 Climatology.224
7.5.2 Global changes in SST.225
7.5.3 Atmospheric response to SST anomalies.225
7.5.4 Weather prediction (1 to 10 days).225
7.5.5 Gas exchange across the air—sea interface.226
7.5.6 Surface heat flux as it affects the ocean heat budget. . . . 226
7.5.7 Deep convection and water-mass formation.227
7.5.8 Dynamical oceanography.227
7.5.9 Pollution studies.228
7.6 Examples of the application of satellite SST data.232
7.6.1 Global SST, and SST-anomaly, maps.232
7.6.2 Ocean eddies.233
8 Table of Contents

7.6.3 Upwelling.237
7.6.4 Ocean fronts.237
7.6.5 The Mediterranean Sea.240
7.6.6 Shallow-sea fronts.240
7.6.7 Coastal-fringe fronts.244
7.6.8 Commerical fisheries.246

Chapter 8 Passive microwave radiometers

8.1 The physical principles of passive microwave radiometry.248
8.1.1 Overview.248
8.1.2 Thermal emission in the microwave.249
8.1.3 The radiation received by the satellite antenna.250
8.1.4 Atmospheric and solar radiation effects.252
8.1.5 Sea-surface emissivity. ..254
8.1.6 Skin depth of sea-surface emission.258
8.2 Microwave radiometer design.260
8.2.1 Radiometer design constraints..260
8.2.2 The Seasat and Nimbus 7 SMMR.261
8.2.3 Other satellite microwave radiometers.263
8.3 Oceanographic interpretation of passive microwave data
from space.264
8.3.1 Potential applications.264
8.3.2 SST and surface-wind retrieval algorithms for SMMR . . . 266
8.4 Comparison between infrared and microwave radiometers for
SST measurement.270
8.5 Applications of SMMR data.272

Chapter 9 Satellite altimetry of sea-surface topography

9.1 Introduction.278
9.1.1 An exciting new oceanographic instrument.278
9.1.2 Principles of satellite altimetry.279
9.2 Distance measurement with a radar altimeter.281
9.2.1 The instrument design.281
9.2.2 Corrections for atmospheric transmission.284
9.2.3 Surface roughness errors.285
9.2.4 Altimeter specifications.286
9.3 Establishing a datum.287
9.3.1 Orbit determination.287
9.3.2 Summary of height errors relative to the reference
ellipsoid.289
9.3.3 The geoid.289
9.4 Application of altimetry to the study of ocean currents.290
9.4.1 Geostrophic balance.290
9.4.2 The possibilites for ocean-circulation determination. . . . 292
9.4.3 Examples of ocean-circulation phenomena observed
with Seasat altimeter data.293
 Table of Contents 9

9.5 Tidal observations with the altimeter.300

9.5.1 The value of satellite altimetry to tidal studies.300
9.5.2 Extraction of tidal information from altimeter data. . . . 301
9.5.3 Examples of altimeter applications to ocean-tidal
science.303
9.6 Ocean bathymetry from satellite altimeters.306
9.7 The selection of satellite orbits for altimeters.306

Chapter 10 Active microwave sensing of sea-surface roughness

10.1 Introduction to the remote sensing of sea-surface roughness.310
10.2 Radar reflection from the sea surface.311
10.2.1 Backscatter cross-section.311
10.2.2 Specular reflection and scattering.311
10.2.3 Resonant Bragg scattering.313
10.2.4 Polarisation.315
10.3 Wind-generated surface-wave roughness.315
10.4 Surface films and slicks.317
10.5 Dynamical causes of sea-surface roughness patterns.319
10.6 Artificial causes of sea-surface roughness patterns.321

Chapter 11 The altimeter as a surface-roughness sensor

11.1 A nadir-viewing radar.324
11.2 Physical principles of operation.324
11.2.1 The effect of waves on the return-pulse shape.324
11.2.2 The effect of surface roughness on the return-pulse
strength.328
11.3 Performance of wave-height algorithms.329
11.4 Performance of wind-speed algorithms.331
11.5 Oceanographic applications of satellite-altimeter-derived
wave height.332

Chapter 12 Synthetic aperture radar

12.1 High-resolution imaging radars.344
12.2 Principles of SAR operation.345
12.2.1 SAR geometry.345
12.2.2 Range resolution.347
12.2.3 Azimuth resolution using the synthetic aperture
principle.351
12.2.4 SAR processing.356
12.2.5 SAR image errors due to moving targets.358
12.2.6 Image degradation due to speckle.362
12.3 SAR imaging of ocean waves.363
12.3.1 A controversial problem.363
12.3.2 Hydrodynamic modulation.365
12.3.3 Electromagnetic modulation.366
12.3.4 Motion effects.368
10 Table of Contents

12.3.5 High sea states.371

12.3.6 Directional wave spectra from SAR images.371
12.4 Observations of ocean waves with Seasat SAR.372
12.4.1 A description of the Seasat SAR.372
12.4.2 Wave imagery.375
12.4.3 Wave spectra from the Seasat SAR.380
12.5 Internal waves imaged by SAR.384
12.5.1 Observations of internal waves in Seasat data.384
12.5.2 Internal-wave theory.387
12.5.3 Surface-roughness imaging of internal waves.390
12.5.4 Oceanographic information content of SAR images of
internal waves.391
12.6 Other oceanographic phenomena imaged by SAR.392

Chapter 13 Microwave scatterometers

13.1 Introduction.395
13.2 Wind scatterometry.396
13.2.1 Surface-wind stress and the near-surface wind-velocity
profile.396
13.2.2 The development of wind-speed and wind-stress
algorithms for interpreting radar backscatter
measurements.398
13.3 Experience with the Seasat scatterometer.401
13.3.1 The SASS design and operation.401
13.3.2 SASS data processing.405
13.3.3 The SASS wind-vector algorithm.406
13.3.4 Removing the direction alias.410
13.3.5 Validation of the SASS algorithms.413
13.4 Applications of wind scatterometry.415
13.5 Wave scatterometry.417

Chapter 14 The way forward

14.1 Satellite sensors for oceanography in the future.419
14.1.1 Introduction.419
14.1.2 The ERS-1 payload.420
14.1.3 TOPEX.422
14.2 Developments in data analysis.423
14.3 Trends in ocean science.425
14.4 Conclusion.426

References.428

Index 444
 Foreword

The properties processes and populations of the ocean have always been of
scientific interest and are becoming of greater and greater practical importance.
The traditional uses of the ocean, as a support for ships, as a source of food, as
a sink for waste, become increasingly significant as the population of the world
increases: the concern about climatic change, and the possible effect of man’s
activities, has underlined the major part the ocean plays in the coupled air, ice,
water system.
The ocean itself is a complicated system with movement and variability
on all time and space scales — yet the oceanographers’ traditional instrument,
the ship, gives neither a time series at a fixed point nor a synoptic chart at fixed
time. This, as the author of this book points out, has steered oceanographers
towards problems that can be attacked on the basis of detailed observations
in a vertical column of the ocean. The relatively recent arrays of instrumented
moorings and drifting buoys are so expensive that they can cover only a small
area of the ocean surface.
One technique, adumbrated by some enthusiasts about twenty years ago,
is now seen to have a vast potential for observing the near-surface layer of
the ocean on a global scale - satellite observations are becoming quantitative
and increasingly accurate: they are being used to extend the scope of in situ
observations so as to measure surface variability, to study motions in relation to
space rather than to time. They have already contributed to our knowledge
of the general circulation of the ocean, to meso-scale eddies and upwelling, to
small-scale frontal systems, to internal and to surface waves.
There are two major problems in exploiting remote sensing ofthe ocean. One
is the need to deal with the vastly increased number of observations available;
we must acquire the ability to cope with large data-sets rather than individual
observations. The other is that since the remotely sensed fields are essentially
confined to the sea surface there is the need to relate them to the internal,
three-dimensional structure; there will be a continuing need for measurements
within the ocean.
12 Foreword

To make the most effective use of this rapidly evolving satellite observing
system oceanographers must learn more about the techniques of remote sensing,
so that they can interpret the observations confidently, while the specialist in
remote sensing must learn more about the ocean and its problems.
The author of this book has seen the need for a body of information
common to oceanographers and specialists in remote sensing technology — I am
confident that his account of the problems and the promise will go far to bridge
the gap between the two communities and so will speed the day when satellite
oceanography is fully integrated into marine science.

Professor H.Charnock, F.R.S.

 Preface

This book has grown out of my own developing interest in the use of data from
earth-orbiting satellites in the study of dynamical oceanography. As I sought
to understand the techniques employed in remote sensing, and the physical
processes underlying those techniques, I became frustrated at the difficulty of
tracking down information and finding it presented in a form readily accessible
to a physical oceanographer who is not a remote-sensing specialist. There
seemed to be a gap between the brief review papers which did not penetrate far
enough and the technical reports on sensors and satellites in which it was easy to
get lost in a wealth of material, much of it irrelevant to the oceanographer who
is trying to develop satellite remote sensing as a marine science research tool.
At the same time it became apparent that whilst a lot of space technology
was being offered in support of ocean applications, there was not always
evidence that the remote-sensing specialists fully understood the scientific
problem areas being studied by oceanographers.
Faced with the prospect of teaching a course in satellite oceanography
to students of oceanography, and finding no suitable course text, I decided to
attempt to write it myself, if only to encourage others to improve on it.
This book is the result, and it is written for the three kinds of people I
' have mentioned. Firstly, it is for practising oceanographers, to acquaint them
with an exciting new observational tool which is now available to them and
which is beginning to revolutionise branches of marine science. It is both a
background reader providing a broad general introduction and a primer to lead
oceanographers aspiring to use remote sensing through the first principles of
the subject and on into examples of ocean applications, and to leave them
equipped to grapple with the most recent scientific literature. It is also written
in order that sensor technologists and remote-sensing specialists can gain an
oceanographer’s view of the relevance of satellite remote sensing to the study of
the ocean. It therefore includes some background information on those aspects
of ocean science which are benefiting most from this new technique. Thirdly,
14 Preface

as well as being a book for practising scientists, it has been written with post¬
graduate and senior undergraduate students of both oceanography and remote
sensing in mind.
I have therefore tried to pitch the academic level at around that of the
senior undergraduate/postgraduate, assuming a general scientific training and
ability to handle scientific concepts,but aiming to introduce from first principles
those ideas and theories which are central to the subject of the book. The first
section on fundamental principles should be readily accessible to scientists from
all subject disciplines, and no mathematical competence is assumed beyond that
expected of any numerate scientist. The second section, which explores areas
of oceanographic application using particular remote-sensing techniques, is
also intended to be accessible to all scientists. However, such is the complexity
of some of the mechanisms (for example, synthetic aperture radar) that the
non-physical scientist will find it harder to grasp than the physicist, engineer,
or applied mathematician; but I have tried so far as practicable to make the
text self-contained even for the biological oceanographer, marine chemist, or
sedimentologist.
Where help may be needed with physical concepts I have tried to refer to
a suitable text. I have also referenced the key papers which thoroughly describe
the techniques which I am introducing, and mentioned examples of marine
applications in the oceanographic and remote-sensing literature. The reference
list is not an exhaustive bibHography, but should provide an adequate source
of material for a literature search, and a useful springboard for diving into a
detailed review of a particular subject.
I am indebted to many people who have given me help along the way
with the writing and preparation. The book took shape in my mind during a
NATO-funded study visit to North America in 1982 with encouragement from
Pahl Le Blond, Bill Emery, and Jim Simpson. It developed with some guidance
from Tom Allan whose advice as series editor I was fortunate to have at every
stage of the writing. Fellow staff and students at Southampton Department
of Oceanography have taken a lively interest in the work, and colleagues have
kindly commented on parts of the manuscript, notably Simon Boxall, Henry
Charnock, John Daniels, Trevor Guymer, Paul Mather, and Neil Wells. Report
material I had previously written with J. O. Thomas for Oxford Computer
Services was made available for use in the book. Image originals were helpfully
suppUed by K. S. Baker, P. Bayhss, H. Gordon, B. Holt, R. Legeckis, J. G. Marsh,
N. MacFarlane, R. Pingree, R. K. Raney, and R. C. Smith. Permission to copy
other illustrations is acknowledged in the figure captions. I must also thank
those whose labour contributed to the typescript and figures: Elsie Foss, Jenny
Mallinson, Graham Johnson, and especially Jean Watson whose patient days and
hours at the word processor have made the book possible. Finally I am, grateful
to my family for letting me spend time on the book outside my normal working
hours, and most of aU to my wife, Diane, whose encouragement made me believe
 Preface 15

I could start writing, and whose patient love and support has enabled me to
complete the task.
Southampton
July 1984
To DIANE
 CHAPTER 1

Oceanography from space?

1.1 A NEW FIELD OF MARINE SCIENCE?

‘Oceanography from a satellite’ — the words themselves sound incongruous
and to a generation of scientists accustomed to Nansen bottles and reversing
thermometers, the idea may seem absurd. This is all the more so because
decades of technological constraint have all but forced oceanographers
into consideration of the class of problems that derive from the vertical
distribution of properties at stations widely separated in space and time.’

So wrote Gifford Ewing in 1964, introducing the proceedings of the first

conference convened to bring oceanographers together to consider whether
the science of oceanography might benefit from man’s new found success in
exploring and occupying space outside the earth’s thin atmosphere. (Ewing
1965). It is an appropriate quotation with which to commence this book because
there is a need to establish from the outset whether ‘satellite oceanography’ is
indeed an incongruity, or, as the author believes, a new branch of marine science
which has developed sufficiently far to justify the publication of a book which
brings together the fundamental ideas and techniques of the subject.
Prior to 1957, the phrase ‘satellite oceanography’ would have had no
sensible meaning to marine scientists. However, such was the growth in space
technology in the decade following the launch of Sputnik 1 that by the
1960s satellites (that is, artificial earth-orbiting platforms lauched from earth
by rockets) were an accepted and commonplace part of man’s efforts to master
his environment. The word ‘satellite’ had rapidly gained a new meaning which
today we all take for granted. Now we use the title Satellite oceanography to
encompass all aspects of the study of the ocean which use surveillance of the
sea from platforms orbiting the earth in space. In a quarter of a century the
investment in satellite remote sensing of the earth’s environment has grown
to become a significant commitment of the space programmes of the USA,
USSR, Canada, the European Space Agency, France, and Japan. Now, whether
oceanographers are prepared to make full use of it or not, there is a flood of
information about the sea being relayed day and night from several satellites
18 Oceanography from space? [Ch. 1

carrying a variety of sensors exploiting the width of the electromagnetic

spectrum. Indeed the data rate of marine science information arriving from
space must be many orders of magnitude greater than that from all the
oceanographic research vessels deployed at any time over the face of the world’s
ocean. The question remains whether the satellite data have anything useful to
offer the oceanographer which he cannot obtain more cheaply by conventional
measurement techniques from ships at sea. It is useful, therefore, to survey
briefly the history of ocean surveillance from space, and to note the different
types of information that are available.

1.2 A QUARTER OF A CENTURY OF OCEAN OBSERVATION

FROM SPACE
The first views of the earth from space were photographs taken from the
earliest high-altitude rockets, as part of the US space programme from 1946
onwards (Lowman 1965). It was with the first manned space flights of the US
Mercury programme in 1961 that astronauts could look down and see the
ocean for themselves from heights of over 100 miles, and it was soon apparent
that there was a lot of information of potential interest to oceanographers to be
gained from an observation platform in space (Duntley 1965). With the Gemini
and Apollo spacecraft came much more spectacular colour photographs of
the sea and the first multispectral photographs on Apollo 9, but there was no
systematic ocean surveillance available with these programmes, the aim of which
was principally to develop new space technologies.
The Earth Resources Experimental Package flown on Skylab in 1973
provided the opportunity for more systematic photography and multispectral
scanning of the earth’s surface, and demonstrated the potential of a continuous
programme of earth surveillance using visible and near-visible wavelength remote
sensing. Erom this developed a series of high-resolution multispectral (visible
and near-visible wavelength) scanners carried on the Landsat satellites which
has continued to the present day and which supphes information on water
colour and turbidity in estuarine and coastal areas.
Erom the late 1960 s the TIROS series of meteorological satellites relayed
images in the infrared and visible wavelengths of considerable value to
meteorology. Under cloud-free conditions, some information about sea state
and surface roughness could be recovered from sun reflection patterns on
these images. However, it was not until the improved spatial and radiometric
resolution of the scanned images of the NOAA series of meteorological satellites
became available in the mid-1970s that oceanographers began to receive
potentially useful information concerning sea surface temperature and water
turbidity. This satellite series continues to provide an operational environmental
service.
Sec. 1.3] The value of satellite data to oceanography 19

The launch of Seasat in 1978 saw the first satellite specifically designed
for and dedicated to ocean surveillance. To overcome the cloud cover problem,
most of the sensors operated in the microwave part of the electromagnetic
spectrum, in which clouds are transparent. Information was recovered about
surface temperature, surface roughness, sea surface height, and ocean wave
parameters, for a period of three months until an untimely power failure
terminated the satellite’s useful life (see Allan 1983).
In the same year an experimental satellite in the Nimbus series carried a
visible wavelength scanner designed to observe ocean colour — the Coastal Zone
Colour Scanner — which has supplied a wealth of synoptic views of the colour
of large areas of the world’s oceans, at a spatial resolution of better than 1 km.
The same satellite has observed sea surface temperature by microwave
measurement.
Now, and in the coming few years, satellites are being built and plarmed
which will exploit and improve the observational methods already developed to
survey the colour, temperature, surface height, and surface roughness parameters
of the sea. Such is the momentum of space technology development that ocean¬
viewing sensors are likely to be produced irrespective of the degree of coopera¬
tion and collaboration between space technologists and ocean scientists. It is
therefore of importance that oceanographers not only learn how to make full
use of the ocean data reaching us from space, but take an active part in the
development of new space technologies. Indeed, it is hoped that this book will
encourage further the dialogue between the space scientists who promote the
new technologies and the oceanographers who will apply them, by introducing
the techniques of satellite remote sensing to oceanographers and by providing
the satellite technologist with a survey of oceanographic applications.

1.3 THE VALUE OF SATELLITE DATA TO OCEANOGRAPHY

It should already be apparent that the range of oceanic parameters capable of
measurement from space is very wide, including ocean colour, temperature, sea
surface height, and surface roughness on a variety of length scales from ripples,
to swell waves and larger. This suggests that the use of satellites is not merely
an extra technique of relevance only to one narrow branch of oceanography, as
for example ocean colour scanners might be considered as a branch of optical
oceanography. Rather, the applications of satellite data within oceanography are
so diverse and extensive that if satellite-collected data are useful at all they
are useful to the whole spectrum of oceanography, biological, chemical, and
physical, just as ship-collected data are.
Of course it is easy to understand the concern of the conventional
oceanographer that information remotely gathered from space cannot be as
accurate and relevant as that collected first hand by scientists on research vessels
at sea. But a moment’s reflection will remind us that ‘remote-sensing’ techniques
20 Oceanography from space? [Ch. 1

have been used for many years by conventional oceanographers using sound
waves to penetrate the ocean and view the sea bed, or to observe suspended
sediment or else biological activity in the deep scattering layer. More recently,
acoustic remote-sensing techniques have been developed to detect the thermo-
haline structure of the oceans by acoustic tomography (Flatte et al. 1979) or to
measure velocities by the doppler shift of backscattered sound pulses. There can
be no fundamental objection, therefore, to extending these remote-sensing
principles to the viewing of the sea from space by electromagnetic means through
the intervening layer of atmosphere.
A more fundamental objection to satellite oceanography lies in the fact that
the electromagnetic radiation used by the sensors barely penetrates the sea
surface. Apart from the visible wavelength sensors which may penetrate to
30 m depth, it is effectively true that satellites observe only the surface skin
of the ocean, and are limited to a two-dimensional view. This is a major limita¬
tion since the vertical structure of the ocean is of very great significance to
the oceanographer, whether he is concerned with current velocities, salinity,
temperature and density, distribution of chemical forms, distribution of primary
productivity, or fish populations. If he is on board a research vessel with a
suitable winch and cable, vertical samphng to great depths does not present
insuperable difficulties.
Furthermore, the satellite’s orbit characteristics limit its time-sampling
capabilities at a given location to discrete overpasses once or twice every day, or
every several days, except in the case of the geostationary satellites which can be
parked high above the equator. In comparison with the continuously samphng
or integrating sensor mounted on a research vessel or buoy, the sateUite sensor
begins to appear less effective, particularly if it is further Hmited by cloud cover
which in some locations may be present for nearly 100% of the time.
For both these reasons, there has tended to be a pessimistic view prevalent
amongst many oceanographers that sateUite remote sensing has little of signi¬
ficant value to offer oceanography, with the exception of the use of microwaves
to provide aU-weather sensing of surface phenomena such as height, roughness,
and wave statistics which do not have a depth dependence at all. This is
obviously the task which satelUte remote-sensing performs best, and accounts
for the choice of microwave sensors for Seasat, and the proposed ERS-1.
However, at least three reasons can be advanced to argue that it is extremely
short-sighted to suppose that satellite remote-sensing has no contribution
to make to the study of other ocean parameters which do have a strong depth
dependence or a high-frequency temporal variabiUty which cannot be detected
from a space platform. In the first place, whatever variabUity occurs at depth,
it is the surface parameters of temperature and velocity and of concentrations
of salt and dissolved gases etc., which control the exchange of energy and matter
between the ocean and atmosphere and therefore control the global ocean/
atmosphere heat engine. Thus whilst the satellite may be able to sample at only
Sec. 1.3] The value of satellite data to oceanography 21

one depth level, it does so- at the most important level — the surface. Indeed,
MacIntyre (1977), in seeking to draw attention to the importance of the surface
skin layer of the ocean, points out that on a logarithmic scale the first millimetre
below the surface represents the top half of the ocean! This ‘top half’ is what
is sampled by the satellite. In this, the last quarter of the 20th century, there is
a serious effort being made by both oceanographers and meteorologists to under¬
stand the factors controlling climate, and its variabiHty. Remote sensing of the
surface water parameters promises to contribute significantly in this important
field of scientific study.
The second reason why it is inappropriate to reject remote-sensing tech¬
niques because of hmitations in depth sampUng and high-frequency temporal
sampling capabilities is that to do so is to ignore the positive advantages of
satellite-gathered data — notably its two-dimensional synoptic view and high
spatial resolution (in the case of scanning sensors) and its ability to provide a
low-frequency time series over long periods ranging from weeks to years, even at
isolated oceanic locations. To achieve the synoptic view is impossible by other
means, and the long-period sampHng capabilities would be extremely expensive
by other means. It would require many research vessels to obtain a truly synoptic
survey of a large region, and even then the spatial sampling would be very coarse.
Moreover, since tides with a 121^-hr period are present over most of the ocean, it
is not possible to stretch ‘synoptic’ beyond a period of about 1 hour. For this
reason, the colour and infrared scanning sensors have great potential value, even
if they record only the occasional cloud-free image at irrjegular intervals, for they
supply a synoptic view of the horizontal spatial structure which is impossible
to achieve by other means, and which in fact has not been seen before. In this
novelty of finding new structures and patterns for the first time lies much of
the excitement and enthusiasm shown by other oceanographers for the use of
satelhte data. All too often this enthusiasm hardly persists beyond the adorn¬
ment of the laboratory and office wall with colourful satellite pictures! The
reasons for this may lie in the methodology of science itself. The very questions
that we ask, and the theories we formulate, are only scientifically valid if they
are capable of being tested against data. Therefore conventional, pre-satellite
oceanographic science has tended to develop theories and ask questions con¬
cerning the depth variability and high-frequency time variability of the ocean -
questions which we can answer and theories which can be tested against the sort
of data which ship and buoy sampling methods are good at collecting, that is,
time series and depth profiles. Questions about horizontal spatial variability
and synoptic structures have been kept more in the background. However, now
that we have the means to observe spatial structures, it is important that the
theoretical framework of oceanographic science should develop to start asking
more questions and to formulate more hypotheses concerning spatial structure.
Only as the theoretical base of the science develops will the full value and
significance of synoptic satellite data fields be reahsed.
22 Oceanography from space? [Ch. 1

A final advantage of satellite data in comparison with conventionally-

collected data is the automatic area-averaging which is incorporated in a
remotely-sensed data sample. This is particularly relevant when data are being
used as input to, or to test the predictions of, numerical models. In the model,
the parameter value is of necessity representative of average conditions over a
discrete area element. A ship or buoy can provide only a point sample, and there
is no way of knowing whether it is representative of the area element in which
it lies. The remotely-sensed observation is therefore much better suited for
applications in the validation of numerical modelling work.
It is the contention of the author that for these reasons satellite remote
sensing, far from being at best a peripheral activity for oceanographers, in fact
may hold the key to some of the major advances in oceanography which will
be made in the remaining years of the 20th century. In the long term, remote¬
sensing techniques will become accepted as just another weapon in the
oceanographer’s armoury for use in exploring the sea’s processes, alongside and
in conjunction with ship and buoy methods. Until that acceptance, books such
as this will aim to present the different techniques involved in satellite remote
sensing of the sea, under the collective title of ‘satelHte oceanography’.

1.4 THE SCOPE OF THE BOOK

There are many textbooks of remote-sensing now available, but they are
generally concerned with the application of data collected over land, and are
occupied as much with airborne remote-sensing techniques of air photography
and photogrammetry as with satellite surveillance. The purpose of this book is to
introduce to marine scientists the fundamentals of satellite remote-sensing over
the ocean. Its bias is that of the oceanographer, and the intention throughout
is to present satellite remote-sensing techniques in their context as tools for
oceanographic appUcation. Notwithstanding this, it should also be of interest to
the professional remote-sensing scientist and sensor technologist because remote¬
sensing methods are evaluated in relation to those tasks which oceanographers
genuinely wish to perform, as distinct from those which the space hardware
designer would like him to perform!
The subsequent chapters are divided into two main sections. The first (A) is
a broad review of the fundamentals of satellite remote-sensing, from the oceano¬
grapher s viewpoint. Within this section Chapter 2 provides an overview of
‘the possibilities in space’ - the satellites and orbits available, the different types
of sensors and their capabilities, and the way in which the data are transmitted
to earth and disseminated to applications scientists. Chapter 3 follows with an
introductory survey of ‘the possibilities for oceanography’. This seeks to place
remote sensing in the broad context of ocean science, with a review of the types
of phenomena which oceanographers are studying, a reminder of their length-
and time-scales, and the sampling strategy necessary to measure them, followed
Sec. 1.4] The scope of the book 23

by a survey of the oceanographic parameters capable of measurement by sensors

in space. Section A concludes with two chapters which present the principles of
interpreting remotely-sensed sea data and the principles of image processing in
the oceanographic applications context.
The second section (B) examines in more detail specific applications areas,
usually related to a particular class of sensor. Chapters 6 to 8 consider respec¬
tively ocean colour scanners and their application, infrared sensors used to
measure sea surface temperature, and passive microwave radiometers. Chapter 9
presents the microwave altimeter as used to measure the absolute height of the
sea surface, and introduces the exciting possibilities which follow for studying
large-scale ocean dynamics and tidal motions.
The remainder of the book contains four chapters concerned with the
measurement of waves and surface roughness by active microwave devices.
Chapter 10 is an introduction to the physics of the interaction between radar and
the ocean surface. Chapters 11 to 13 then present the principles and applica¬
tions of three types of microwave sensors — the altimeter used as a nadir surface
roughness sensor, the synthetic aperture radar, and the scatterometer.
The final chapter seeks to point towards the likely direction of future
advances in oceanographic remote sensing, arising out of present trends in the
science.
To contain the size of the book, there is no mention made of airborne
remote-sensing techniques. Although these are harder to use over the sea because
of problems in fixing the position of the resulting images, there is increasing
interest in the use of aircraft in oceanography, often in conjunction with satellite
observations. Whilst much of what is presented in a book on satellite remote¬
sensing is relevant, those problems unique to aircraft surveillance platforms
deserve more detailed consideration than there is room for here.
Furthermore, no mention is made of the oceanographic use of satellites for
navigation and data-communication purposes. This is an application of no small
significance to oceanography, for it enables the deployment of buoys and other
drifting or moored instrument packages in inaccessible locations, thus permitting
the time- and space-samphng capabilities of conventional oceanographic instru¬
ments to be stretched beyond the logistic limitations of research vessel servicing.
However, the topic is omitted because it does not offer the oceanographer the
fundamentally new vantage point in the sky from which to take a fresh look at
the ocean which is the essential subject of this text.
The book is intended to introduce the subject in its breadth and therefore
seeks to present ideas and results which are generally established and accepted.
To cover the whole field in a single volume of this size precludes an in-depth
study of all recent innovations and developments. The proceedings edited by
Gower (1981) of the 1980 ‘Oceanography from Space’ symposium provide a
broad selection of recent research results, whilst the recent monograph of
Stewart (1985) is a valuable source book for further reference.
 ' ’ #^.“1

i ■ -.' • v>l * ‘ ••-' 'P H.'" ^ ’*•

-,'' \f hi: /' » > •■
* • 1’ . .’tJ-.''^5 3^* • il^^'-^.'“'t ' J-j^' |‘>>?«'.' ■ r ^l‘♦•>^*l *1 ’ih

■ ' -r^ ■ -*.

.K^.-.rt■ •^- 1^^■'-Id*. . . .i..:*• r^ 1.J• * • *»’:
•7 i,» •ir-idt rr>('’t '4*' ■*t. '.' ~'» ' S^4 i ' "■

ii; cJ*jf, t-. .»•.■!# l>e\siMnf'-.<,••* U>fi?a|i[«H«^ »?ifi*'i;'fr** *■- •-;,/?i

»i V*; . • . -', J--,,?., ‘-II* X ..»'liF»<|r^('f.» .,V .tfc-u.

/n^'J/uhr-fei^ U AHj. enwf •u r.s -■ ' ‘V4

f' * •. *>.' ijiti 1 fT-- y. ««»<f =.-^^t«|L*4jru * j»
n»». ' rHt^•-'‘^7 fca<Sk46lL^(*r»r^^
<ay. t^i Aiwn.-v.* io*.'? it,#! >rt. -»<r-tiawk •v.-^»C''-»- .■ ,.“T .

WflK**<r.:ar* »n Jur-- . i».Y;»i-. ji-«-i'■ ^»i^iau3ci u<

■ -■'u .) rtui^itt^. f^ > i.’j

-..»■ 4*1* ^Tij i •«{e<flJcqfi^V-a;i.*i>’'} y dJirsutsI'.*>*. irst^^
c (tCirtiLHr.- ~. .•. .-r-f^^ J.* ■.- u
■ >- wr', i.’*: .. ^ ';!' riffr;^iVt

; ;V ? •‘'IfpyO -firMi Slit

sd^ tc r^4v* l4* f‘t-'i r,ti;jl*e ^f?rw»t 4. <ct5- iSfvatJii fh

‘il - 'S' S/♦•|^■..| j , I'lVjt^ **f?l4* > t,4i ^1

ir3v7j ,»rfc,. -i .i/* .. ^•»«.'4**V>r*?!4
.r':*-^U igrs'r/r: *^fi ' ;4i*.Tr ^
it«t<ittikl, _ >i ,. M.. - . v'ji si
i
• -.W> /-t- . rVi,

I rtf) >iJs -tCV’-an "i-.^,-1.' ••Ml’. • .;■ »4t' .H

‘li#!. 'icttl - .= •:-f.'/ '• -.iJlifcu*/' -«» :i4!|j^V9i*v

lU fiJi 'v;;t#Sv 4’iWli<^^Kcw I»»i:9«c»W ;fl|

■ • •■ I..A £’!■:^ ji:v t- h"^
itki loo{*,4 «•»/ 'kj ,>> ?»n:-;
-r»- '^;4<y,(?"^r^SC .Tts jtfHT-ii wil«» •*• a
dMX^4|J'r);i«Si*nf*VvV.-!^:^ 'U ’4Ulw?'fe* <,a-.
,' - • I wr*'*!' .. *.v V' i>>.* <U '•r jtJ 7iS.iJ»^34^V;r< .tU
w-.t f.arf‘ V^. -xiiL v^tiH tu lat^i v.- '.^JWirganjivO •4^,; Mi

!•*- !<: ^ijui^ ‘Wi* ^ m Usi,; oi'u4-‘iito« «?

SI'X. h&-as*. wtiw»S(si fesAmile^ju f4*r
'a'.i»4<. i-.-t'-; .-^’*••4 a;sr.-.-* '" '^f. i•
p <--s »*:sa»« 34<. .rsx>M tir*
 SECTION A

FUNDAMENTALS OF REMOTE
SENSING IN OCEANOGRAPHY
CHAPTER 2

The possibilities in space -

space hardware and data transmission

In this chapter we examine the current developments of space vehicles and

sensors to find out the possibilities which are available to the marine scientist
for ocean monitoring and applications. This is an introduction for the oceano¬
grapher, covering only oceanographically useful space systems, rather than
an exhaustive treatment for which standard remote-sensing texts should be
consulted (for example Reeves 1975, Lintz & Simonette 1976, Slater 1980).

2.1 SPACE PLATFORMS AND THEIR ORBITS

2.1.1 Satellite vehicles
Satellite engineering is unlike any other type of scientific equipment construction
because of the very different conditions encountered in space. Like oceanographic
instruments the design is matched to the prevailing environmental conditions,
but there the similarity ends. The type of factors which must be taken into
account include the lack of gravity in a free orbit, the high-vacuum conditions
in which materials behave very differently from our earthbound experience of
them, and the presence of energetic particle radiation and micrometeoritic dust
(Massey 1964). Lifetimes of several years are expected from those satellites
viewing the oceans, and since regular repair and maintenance is not possible a
very high standard of component reliability is called for. In addition, the satellite
must be able to withstand the high inertial forces and other rigours of launch,
either by unmanned rocket or by the space shuttle. Despite such stringent design
constraints, a quarter of a century of experience has achieved the reliable vehicles
which are now collecting earth environmental data. Table 2.1 lists the lifetime of
several such satellites useful to oceanography, and indicates their size and weight.
They are illustrated in Fig. 2.1.
Apart from being a frame on which to mount the earth-monitoring instru¬
ments themselves, which will be discussed later, the satellite vehicle or bus must
provide certain necessary facilities: a power supply, thermal control, aspect
control, a data-handling system, and a communication system, without any of
which the monitoring devices are useless. Power is normally supplied through
[Sec. 2.1] Space platforms and their orbits 27
ON * 0 0
c/2 0 0
Os *
w t
fI t t

o > 00
1
*n
O'
O'
00
O'
t a>
o 0 OS O' ON Os 0 00
On
a>
H z os
t-H
1-H 00
ON
r-H w> -2 s
cd
Ce:^
CO
< > t
<u <D <u
m S o CO
c/3 d d d
CL os o
O Z d
0 d d
il> <N 04 6
w 04 os SO so so CO
H 04 «-H ON
W
IS oi 04 m
S s a 0 0 0
00
00 g oo
H c^ 00 5 c
o^ ^ S o
< 00
CO Lj ^
c/3 o o c d c
< O OS O cd i<5“
s
W <N o ^ a: ^
(N
c/i 00 !S <c <=>
a>
Satellites for oceanography — general descriptions

(N GO
>

00
r- r-
c/3 Os 00
00
d
o Pi CZl
D *
M E S
pa 52 o
o t o e *3 5 N

S O os cn c/i ^
E
CN 04
o
OS Zl <L>
r- oo
Os
2® oo
■*-»
< On ON
o ic.
O ^ > ■ *
00
d
o E
r o
z o Z C c * .52
o
p<

a: ^
c^ ;3
O ' t ro 04

o CO t r-- m
oi (N <N 00
00

Z Oo 00

H Z Z Z

<N
00 00
H Os 00 d
E
< ^ *
o 2
CO
c/i CO

Q
>> * lo E "O S H
o Functional, but in standby mode, November 1983.

Z lO
E
c U-)
Table 2.1

oo (N fO
r- 00 00
OS
in Os
o 2 00 ^ 00
(U

H t--
r- ^
r- 0)
ON C 2^
00
* Still operational in November 1983.

<L)
o d E CO
H C
IT)
ON
.2 o CO
3 SO «-H ’O o
< a I e
CO ^ tI ^ (N ^ SO
Q ro (N * ro
(N OJ t lO t
Z
< (N

C
C3 op
0) g
D*
‘S .ac/5 Q,
§ a> ^ O

<U Vh ^ d d
I 2 =
(U ■3 J2 o
ft g
^ (D
c/5 00 cd
cd E o o
e .d d
X bO
.yCL 3CT" CL
'n ^
cd

o
2 -S >> D

0) CL H •-
GO CU CL
O < 1>
28 The possibilities in space — space hardware and data transmission [Ch. 2

AHITUDE CONTROL
^SUBSYSTEM PACKAGE
ORBIT DIRECTION

ORBIT ADJUST TANK

WIDEBAND
ANTENNAS (2)

DATA COLLECTION S-BAND ANTINNAS

ANTENNA MSS
RETURN BEAM ^ATTITUDE SENSOR
VIDICON
(a) CAMERAS (3)

ARRAY €QUIPMENT HIGH-ENERGY

DRIVE SUPPORT PROTON AND
ELECTRONICS MODULE ALPHA PARTICLE
SOLAR ARRAY DETECTOR MEDIUM ENERGY
DRIVE MOTOR PROTON AND ELECTRON
DETECTOR

SUN SENSOR
DETECTOR
INERTIAL
MEASUREMENT
UNIT
INSTRUMENT
MOUNTING
PLATFORM
SUNSHADE
INSTRUMENT
MOUNTING
PLATFORM
HYDRAZINE
TANK (21
ADVANCED
REACTION VERY HIGH
SYSTEM RESOLUTION
SUPPORT RADIOMETER
STRUCTURE
STRATOSPHERIC
BATTERY SOUNDING UNIT
MODULES 14)

ROCKET UHF DATA

HIGH RESOLUTION
ENGINE COLLECTION
INFRARED
ASSEMBLY (4) SYSTEM ANTENNA RADIATION
(b) SODNDFR
 SDi.An /vnuAY

Sec. 2.1]

(c)

Solar array (two)

(single axis Spacecraft
tracking) Agena
6.1 m

TRANET beacon
antenna
Telemetry,
r tracking &
; command Sensor
module
6.1 m
SAR antenna

SASS
antennas

Laser retroreflector
SAR data link VIRR \
(d) antenna '-TTAC antenna

Fig. 2.1 - Sketch of some satellites used for oceanographic purposes (not
necessarily to the same scale), (a) Landsat 1 (b) TIROS-N (c) Nimbus 7
(d) Seasat
30 The possibilities in space — space hardware and data transmission [Ch. 2

solar cells, backed up by batteries to store energy for when the satellite is on the
night-time side of the earth. Typical power requirements on operational satellites
are given in Table 2.1. In general, sensors with larger power demands must be
mounted on a larger bus which is able to support the larger solar cells which are
required.
There are relatively strict temperature ranges under which the sensors and
other electronic equipment will operate accurately, and care must be taken with
the design of the satellite to prevent overheating. Heat generated by electronic
equipment, or absorbed from incident radiation, must be matched by long¬
wave radiation emission into space, which is dependent on the choice of surface
coating. Active control of the temperature can be achieved by varying the net
long-wave radiation, through attitude control, or by the operation of shutters to
increase or decrease the area of radiation surfaces pointing towards cold space.
The satellites’ own environmental control systems therefore need to be sophis¬
ticated and reliable for a long operational life, and designs have evolved and
improved with experience. Within series of satellites the basic design is the same
although the sensors carried and the communications system may differ between
satellites. Amonst oceanographically important satellites, the Landsat and
Nimbus platforms are of the same basic design, whilst the TIROS/NOAA
meteorological satellites form a different series which includes the Defence
Meteorological Satellite Programme (DMSP). The GOES-Meteosat vehicles
have very different design criteria, being geostationary. Seasat was yet another
design, but based on the earlier Agena system. There are clearly advantages in
being able to test new sensors on a proven vehicle, but this is not always possible,
and when ERS-1 is launched later in this decade most of the satellite subsystems
as well as the sensors will be completely new designs.
One of the subsystems which differs from series to series is the Aspect
Control. A space vehicle’s orientation may need to be controlled for a number of
reasons. Polar orbiting satellites such as Landsat, Nimbus, and NOAA require a
slow spin so that the same face always points down towards the centre of the
earth. In contrast, the GOES vehicles in geostationary orbit spin once every six-
hundred milliseconds on an axis parallel to the lines of longitude at the earth’s
equator to provide a built-in east/west scan for the sensors. Precise spin rates can
be maintained by alteration of the mass distribution of the satellite and hence its
moment of inertia. Other methods of attitude control include inertial systems
which torque with respect to the earth’s magnetic field, and reaction jets, but these
have a limited endurance. Jet thrusters are required to produce orbit corrections,
or alterations if the mission demands it as in the case of Seasat. The more inher¬
ently stable the platform is, and the greater the natural damping of unwanted oscil¬
lations and rotations, the longer its useful life will be. Atmospheric drag would
reduce the operational life, and necessitate frequent course corrections for an op¬
erational satellite, but fortunately at altitudes above 300 km it is very small, and
lifetimes of several years are achieved provided no major system faults develop.
Sec. 2.1] Space platforms and their orbits 31

2.1.2 Satellite orbital dynamics

It can be shown by a simple energy argument (see, for example, Ma.ssey (1964),
Chapter 2) that a satellite moving without friction in the gravitational field
of a spherical planet has a trajectory which is either elliptical, parabolic, or
hyperbolic, depending on its starting velocity. For an earth-orbiting satellite, we
require an elliptical orbit, or the special case of a circular orbit. For the elliptical
orbit shown in Fig. 2.2, the distance r of the satellite from the centre of the
earth is given by
a (1 -e^)
(2.1)
(1 4- e COS0)
where 6 is the angle between the satellite’s present radius vector and that at
perigee (the orbit’s closest point to the earth), a is the semi-major axis of the
ellipse and ae is the displacement of the ellipse centre from the centre of the
earth, where e is the eccentricity of the ellipse.

Fig. 2.2 - Elements of a satellite elliptical orbit

From Newtonian dynamics, the period T for the satellite to travel round
the orbit is

T = (2-2)
where G is the constant of gravitation and M the mass of the earth, and
GAf= 3.98603 X 10^'* m^s“^. The instantaneous rate at which the satellite
describes its orbit is

dd/dt = [GMa(l- (2.3)

 The possibilities in space - space hardware and data tramsmission [Ch. 2

For a circular orbit centred on the earth, e = 0 a = r. In this case the horizontal
speed of the satelhte is

Vo = iGM/af\ (2.4)
In terms of the height h above the earth (Fig. 2.3) and the acceleration due to
gravity at the earth’s surface g = GM/R^, where R = 6378 km is the earth’s
mean equatorial radius, (2.4) may be written

Vo = R[g/(R + h)]^/^ , (2.5)

During launch the rocket must be fired to obtain a trajectory such that at the
desired height h of the satellite, its speed is Fq, assuming that a circular orbit is
required.

satellite orbit

Fig. 2.3 - Circular satellite orbit

If when it reaches h it is travelling horizontally at speed V, then if F< Fq

the satelhte will fall into an elliptical orbit for which a < {h+R). Alternatively
if F> Fo, the satelhte moves out into a higher ellipse and a > {h+R) If
V>2Vo then the elliptical orbit becomes parabolic and the satelhte never
returns it has reached escape velocity.
ru IS mteresting to note the effect on the orbit period of small changes to
the speed. If the satelhte is retarded by slight atmospheric drag, then although
1 now travels more slowly it falls into a lower orbit in which, from (2.2)
Its period is less. Thus from the earth it appears to be travelling faster round
the earth. Smce atmospheric drag is greatest at the lowest point of the orbit the
retardation at the perigee of an elliptical orbit tends to reduce the height of the
subsequent apogee, thus deforming the elliptical towards a circular orbit.
Sec. 2.1] Space platforms and their orbits 33

So far we have identified three of the orbital elements which characterise a

satellite’s position. These are:
6 — the angular position of the satellite in its orbit,
a — the semi-major axis of the ellipse,
e — the eccentricity of the ellipse.
There remain another two elements which are needed to define the orbital plane
relative to the fixed stars, and a sixth which fixes the ellipse orientation within
the orbital plane. These are:
i — the inclination of the orbital plane to the earth equatorial plane.
— the right ascension of the ascending node, measured eastward from the
point of Aries which is a fixed point in the heavens,
w — the angular distance of perigee around the orbit, measured from the
ascending node.
Fig. 2.4 illustrates the definition of the three orbital elements, by showing
the projection of the orbit on to the celestial shell, the sphere concentric with
the earth, but fixed relative to the stars.

Fig. 2.4 - Satellite orbital elements on the celestial shell

The apparent satellite orbit, as viewed from the surface of the earth, may
appear very different because the earth is rotating on its axis beneath the celestial
shell once every day, from west to east. For earth observation, two types of orbit
are most useful.
A geostutionaiy orbit is achieved if the satellite orbits in the same
direction as the earth with a period of one day. If it is positioned in a circular
orbit above the equator it becomes stationary relative to the earth and there¬
fore always views the same area of the earth’s surface. From (2.2) with T as
1 day = 86 400 seconds, we obtain a = 42 290 km, and therefore for a
geostationary orbit /2 = a — /? = 35910 km.
A much lower orbit has a much shorter period. Satellites at a height of
 The possibilities in space — space hardware and data transmission [Ch. 2

between 500 and 2000 km are normally placed in a nearly circular polar orbit
which carries them approximately over the poles with a period of about 1 to
2 hours. As the earth rotates under this orbit the satellite effectively scans from
north to south over one face and south to north across the other face of the
earth, several times each day, achieving much greater surface coverage than if it
were in a non-polar orbit.
The simple expressions presented above to represent orbital dymanics are
based on the assumption of spherical symmetry and zero drag. This is not the
case in practice, and perturbations therefore occur in the elliptical orbit. The
drag of the atmosphere on low orbits has been briefly mentioned and will result
in the acceleration and eventual burn-up of vehicles in low orbit. However, for
the operational life of ocean-monitoring polar orbiters and geostationary
satellites, the effect can be neglected, as can the effect of solar wind and radia¬
tion unless the vehicle has a very low density and a large surface area. The
principal deviation from a pure elliptical orbit is due to non-symmetrical gravity
forces owing to the irregular figure and mass distribution of the earth. Solar and
lunar gravitation is not important, but the equatorial bulge of the earth produces
the major effect. Orbit dynamics with these extra influences are discussed in
more detail by King-Hele (1959, 1964). It is shown that the equatorial bulge
causes a slight change to the period, and results in the perigee of an elliptical
orbit changing position with time, that is, dw/dt 0. Of most interest in our
context is the precessional mechanism which causes the orbital plane itself to
rotate, that is, df2/dr 0. In fact

da/dt cx (l~ e^)-^ cos(i).

For a given orbit height, by a suitable selection of the inclination i it is

possible to achieve a value for dS2/dt of 0.986° per day, which is equivalent to
one rotation of the orbit plane per year. In this way the orbit plane is not fixed
relative to the stars, but can be fixed relative to the sun as the earth orbits it
once a year, as shown in Fig. 2.5. This results in a sun-synchronous orbit, in
which the satelhte crosses the equator at the same local solar time on each pass
throughout the year. To achieve this, df2/dr must be positive and therefore
cos(/) must be negative, that is, / > 90°. The orbit is therefore retrograde, the
component of rotation about the earth’s axis being in the opposite direction
to the earth’s rotation. A sun-synchronous orbit cannot be polar and cannot
cover the highest latitude with a sensor which only points downwards, but for
most orbit heights the required i is usually around 100°, that is, the orbit is
nearly polar. There are many advantages to a sun-synchronous orbit, including
regularity of data sampling and uniformity of solar irradiation on the earth
surface being viewed. Most polar-orbiting satellites are therefore given a sun-
synchronous orbit, and the equator crossing time is fixed by the launch
condition, or by the subsequent post-launch course adjustment. Table 2.2 lists
some of the orbit characteristics of the oceanographically interesting satellites.
 o O o
00 00 o (N
o^ ON VO
VO d
CM
VO
CO o-

o o o
00 VO o 04
‘O
<N CN
On
d
VO
VO 04 CO
CO

CO o o o
W 00 00 o VO
DS o^ VO
(N
ON On
VO d rT o
CO
o o^
< CO

g O o o
O 00 o VO
H On 00 o-
< (N VO d o
04
H CO
IZl
O O o O
w
ro o oo
(S CO o-
o CO <N d CM
VO CM
CO

CM O o
U o o
H
< (N o
.ss CO fS VO CM
CO
0) O
'M u
o o o
cd H
W _
o
Tj-
o
o
<N VO CM
CO

H
<C 00 ^ ^ -H CO
o
< ^
■S
to 00 C--
o o 00 o
wd < 00 o o C =«
zo W
C--
CO
CO
J3
&
p lo CO
o CO o o»
CQ
IH «I CO
CO
CN
CO
»o ON
o
^ o CO
s ON ON
SO r--
o
o CN
o
■ o-
a> CO m °
u VO CM
(N <N 00
o 00 oo ON o o
< r-
>1 D

<2 O
z o
z <N (21) CM
VO
</)
O) o O
<N
00
00
oi r-
00 ON
X
a> u
-I-* CO
« z ON
t/3 §z (N
00
CM m °
t S/B = southbound overpass, N/B = northbound overpass.

CO <N <N o
I r- r- ON
fS z
ri p
CO
VO
04 (On
VO «
VO O'
05
oo
o o OO 00 C/2g
ON
CQ <
r~“
ON

H J 00
O 00
Co lo CO s®. ^
H o ON
ai < 00
CN
ON ON O
^ g CO
C/D
< Q
w CO
2:
Z < VO
00
«s
o CO
OO ON c/5 g
On On
o CO
CN
r-
00
O
00 CQ
lo CO
00
o 00 o cog
On ON CO
<N

(/3
c
4S g 0) o >»
cd
•w 0) Cd
o I § §•»- 'O
XJ c
(U
CO <u .2.
c?^
E .S
cd
3
(D >»
l-H
.
>. D ^ O B -2
cr M'S .3 )-< *s a
3 E « s
■*cd->
(1> "cd S c 4_. 0) 5i) cd ,

X)
O
C/D
S 4 .S "
z
c o
<u
II

O u- c
2 "E
o t-H
S «
S (U
>—j
C/D ^ o (X
o 3
z C/5
 36 The possibilities in space - space hardware and data transmission [Ch. 2

Fig. 2.5 - Orbit precession to achieve a sun-synchronous orbit

2.1.3 Spatial and temporal sampling characteristics of orbits

The type of orbit chosen for a satellite controls the spatial and temporal earth
coverage that can be achieved. As might be expected, there is a compromise
required between spatial and temporal sampUng rates. At one extreme is the
geostationary satelhte which with a downward-looking (nadir-viewing) sensor
can provide continuous viewing of a single point, that is, infinite temporal and
zero spatial sampling frequency. With a wide-angle sensor the spatial coverage
can include that area of the globe within the horizon as viewed by the satelhte.
For other parts of the earth another satelhte must be used. Fig. 2.6 illustrates
the coverage achieved by the existing network of five geostationary metero-
logical satelhtes. In practice, the wide-angle view is achieved by scanning
the area, so that the temporal sampling rate is typically 30 minutes, but in
principle a postationary satehite is capable of giving a continuous time series
of observations at a point, comparable with the type of measurement made
with buoy-mounted oceanographical instruments.
The polar-orbiting satelhtes have samphng capabilities towards the other
extreme of high spatial but low temporal samphng frequency The precise
capability depends on the details of the orbit, and particularly the repeat
period of the ground track, if any. Fig. 2.7 shows the dayhght ground track
of Landsats 1, 2, or 3, which were typical polar-orbiting sun-synchronous
Sec. 2.1] Space platforms and their orbits 37

GOES-W GOES-E METEOSAT GOMS GMS

USA USA ESA USSR JAPAN

Fig. 2.6 - Ground coverage (within dashed lines) of the geostationary

meteorological satelhtes, as planned for the First Garp Global Experiment
(FGGE), 1978-1979

Fig. 2.7 - Typical daylight passes of Landsat 1, 2, or 3. The night-time pass

between day-time passes 2 and 3 is shown dashed.
 The possibilities in space - space hardware and data transmission [Ch. 2

satellites. With a period of about 103.2 minutes there are almost 14 orbits a day,
and therefore 14 southward sweeps across the earth in daylight. To calculate the
distance between successive ground tracks we simply calculate the longitude
through which the sun has moved in 103.2 minutes, that is, about 25.8 degrees.
(Were the orbit not sun-synchronous, then the calculation would be more
complicated, taking into account both the precession of the orbit and the fact
that the earth rotates through nearly 361° per day relative to the stars. For the
sun-synchronous orbit these two effects cancel exactly.) This corresponds to a
spacing at the equator of about 2865 km. After 14 orbits, the ground track has
traversed 361.43 degrees of longitude, so that orbit 15, or the first on the second
day, is shifted 1.43° to the west of orbit 1 on the first day, that is, 159.4 km to
the west at the equator. After 18 days (that is, 251 orbits) the ground track
starts to repeat itself, so that day 19 will be nominally the same as dayl. In
practice, small orbit perturbations due to non-sphericity, and course correction
by the gas-reaction jet system, result in small deviations of a few kilometres
from exact repeatability, but over a year the repeat coverage over the 18-day
cycle is accurate to within about 30 km.
The sampling rates of the remotely-sensed data depend not only on the
orbit but also on the swath-width of the sensor. Thus for the multispectral
scanner on Landsats 1 to 3, with a swath of 185 km, one day’s daytime observa¬
tions will only cover 14 strips 185 km wide (Fig. 2.8). It will take 18 days to
fill in all the gaps, and the sampling period is therefore 18 days. If a particular
location falls in the small overlap zone between strips (which increases in width
with latitude) then sampling will occur on two successive days in every 18.

DAY 2. ORBIT A

Landsat 4 illustrates how a slight change in orbit pattern can alter the
samphng characteristics. It will be seen from Table 2.2 that it is rather lower
Sec. 2.1] Space platforms and their orbits 39

than the previous Landsat series, and therefore has a shorter orbital period.
Successive orbits are separated by 24.72° (that is, 2752 km at the equator), and
on the second day the tracks lie about halfway between those of the previous
day (Fig. 2.9). The repeat period is now 16 days (233 orbits), and it will be seen
from Fig. 2.9 that locations in the overlap zone will be resampled after 7 days
(on the western edge of the swath) or 9 days (on the eastern margin).

ORBITS 1,0 DAY 1

ORBIT 16 DAY 2
ORBIT 30 DAY 3
ORBIT 44 DAY 4
ORBIT 69 DAY 5
ORBIT 73 DAY 6
ORBIT 88 DAY 7
DAY 8 ORBIT 102
DAY 9 ORBIT 117
DAY 10 ORBIT 13^
DAY 1 1 ORBIT 146
DAY 12 ORBIT 181
DAY 13 ORBIT 175.
DAY 14 ORBIT 180
DAY 16 ORBIT 204
DAY 16 ORBIT 210
DAY 17 ORBITS 234,233
DAY 18 ORBIT 248

Fig. 2.9 - Orbit repeat cycle pattern of Landsat 4. Orbit numbers are shown,
counting from orbit 0 on day 1 at the eastern edge of the diagram.

Fig. 2.10 - Typical swath coverage by meteorological polar-orbiting scanners

(compare with Fig. 2.8). The overlap between two adjacent passes is shaded.
It may sometimes cover more than half the single swath
 40 The possibilities in space — space hardware and data transmission [Ch. 2

If sensors have a much greater swath-width, as is the case on meteorological

satellites such as the NOAA series and Nimbus 7, then the swaths on successive
orbits overlap, and the whole earth is covered in one day by day-time passes
(Fig. 2.10). At higher latitudes a location may be observed by two or more
successive orbits. The resulting time-sampHng period is at worst once per day,
but usually twice, about an hour apart, every day. The exact ground track of
these satellites can be determined from the specifications in Table 2.2, but the
exact repeat period is obviously less important to the applications scientist in
this case when usable data can be obtained at least once per day.
The discussion so far has considered only day-time passes. If thermal
infrared or microwave sensors are employed, which do not need solar illumina¬
tion on the earth, the night-time passes can also be included. These will be
inchned at the equator in the opposite sense to the daytime passes, and therefore
except at the equator will not be symmetrically spaced in time between the cor¬
responding daytime passes. If more than one satellite is used for an observational
programme, and the equator crossing times are arranged to be about 6 hours apart
for a two-sateUite system, then an approximate 6-hour coverage is obtained with
day and night passes. At latitudes where there is a large swath overlap, the
frequency is even greater. The NOAA series has operated intermittently in this
way for several years.
Non-scanning, non-imaging sensors may have very different requirements
in terms of pass repeat properties than those being used to generate images of
the earth’s surface. For example, in the case of a scatterometer measuring the
average surface wind stress over a large area of around 100 km square, it may be
more important to the construction of a climate data base to resample the same
area regularly, rather than cover all sea areas very infrequently. Similarly, the use
of a radar altimeter to measure the sea surface height may benefit from frequent
coverage of the same track, even though large sea areas are left undetected.
These requirements will be discussed in more detail in Chapters 9 and 13. On
certain satellites there is a conflict between the demands of frequent repeats of
a single path, and dense area coverage with no exact repeat paths. In the case
of the Seasat mission the satellite was capable of having its orbit changed to
match different applications requirements.
It is therefore apparent that the temporal sampling resolution of satellites
observing the ocean is many orders of magnitude poorer than ship- or buoy-
based instruments, although in terms of available record length the satellite’s
ability to sample regularly over a period of years is comparable with or better
than marine instrumentation. Nothing has been said yet about cloud cover which
further reduces the samphng rate for some satellite sensors, but this will be
discussed further in the applications chapters. The use of satellites has most to
offer in the global coverage which is possible, and in the dense spatial sampling
which can be achieved. Spatial resolution depends largely on the type of sensor
used, and this will be discussed in the next section.
Sec. 2.2] Sensors on satellites 41

2.2 SENSORS ON SATELLITES

2.2.1 Sensors in relation to the electromagnetic spectrum
All sensors employed on ocean-observing satellites use electromagnetic radiation
to view the sea, and therefore we first review the properties of e.m. radiation as
applied to remote sensing. Fig. 2.11 shows the electromagnetic spectrum ranging
from gamma rays at 10 pm to radio waves at 1 km wavelength. The figure also
shows the corresponding frequency of the radiations, all of which travel through
free space at the speed of light, c, almost 3 X 10®ms”\ Wave length X and
frequency / are of course related by / X — c.
Different wavelength bands are known by different names as defined in
Fig. 2.11. Some of the nomenclature is not precisely defined in the ultraviolet
to infrared range. The visible spectrum itself lies between 0.4 and 0.7 qm,
but the other range names are used rather loosely. At infrared wavelengths and
below, the radiation is usually described in terms of its wavelength, whereas at
wavelengths larger than about 15 jum it is normally referred to by its frequency.
The microwave, radar, and radio-frequency bands are defined by international
agreement in terms of their names from EHF to ELF, or in terms of numbers
where band number N indicates a frequency range 0.3 X lO'^to 3 X 10‘^Elz.
Despite this official nomenclature, remote-sensing radar and communication
jargon still refers to X, S, and L Band radars as shown in Fig. 2.11. The original
definition of these was;
L Band, 0.39 to 1.55 GHz, 193 to 769 mm
S Band, 1.55 to 5.20 GHz, 57.7 to 193 mm
X Band, 5.20 to 10.90 GHz, 27.5 to 57.7 mm,
although the terminology is now sometimes loosely used with the meaning;
L Band — 250 mm (1.2 GHz)
S Band — 100 mm (3 GHz)
C Band — 50 mm (6 GHz)
X Band — 30 mm (10 GHz)
K Band — 10 mm (30 GHz).

Fig. 2.11 also indicates which parts of the spectrum are used by remote¬
sensing sensors on satellites. The choice of bands is governed firstly by the
atmospheric transmission spectrum, and secondly by the application. Fig. 2.12
illustrates the shape of the emission spectra for blackbody radiation emitted
from a body due to thermal excitation. The source at a temperature of 6000 K
corresponds to the sun, and at 300 K to the earth’s surface. Thus if features of
the land or sea are to be observed by the reflection of incident solar radiation
in the same way as the human eye observes, then the frequency range of high-
energy solar radiation should be used, between 100 nm and 100 qm, peaking in
the visible range. Alternatively, if the self-emission of radiation by the sea is to
be the means of remote sensing, sensors should be used for the 3 to 40 qm wave-
 WAVE DESCRIPTION BAND NAME REMOTE SENSING APPLICATION
42

ZH‘AON3nD3aJ

UJ‘H10N3T3AVM
The possibilities in space — space hardware and data transmission
[Ch. 2

Fig. 2.11 — The electromagnetic spectrum, showing some band definitions and typical remote-sensing applications
Sec. 2.2] Sensors on satellites 43

m
O
z
<

2
in
>
o
o
m
o
<
_i
m

Fig. 2.12 — Emission spectra at different temperatures

length range. However, not all the parts of these ranges are useful, since the
atmosphere will not transmit them, as illustrated by the typical transmission
spectrum of the atmosphere drawn in Fig. 2.13. The atmosphere (that is, both
the air itself and water vapour and other aerosols) absorbs most radiation with
a wavelength less than 400 nm and much above about 1 fim. There is a definite
window at the visible wavelength range, and also at select bands within the
infrared. These bands have therefore been exploited for remote sensing, leading
to the visible/near infrared and thermal infrared families of sensors (Fig. 2.11).
Between around 20 /xm up to around 10 m wavelengths there is more absorption,
but above this (that is, for frequencies less than around 100 GHz) there is very
httle absorption. Although there is little reflected or emitted energy at such high
frequencies, these radar bands are exploited by the active microwave sensors
which create their own radiation with which to illuminate the target, and then
observe the nature of the reflected signal. These sensors are known as active
devices as distinct from the passive IR and visible wavelength sensors which rely
on naturally occurring radiation. According to their configuration and applica¬
tion, these active microwave devices are known as altimeters, scatterometers, or
synthetic aperture radars. In addition to active microwave devices, passive
microwave radiometers have also been developed to observe the low level of
thermally emitted microwave radiation.
The possibilities in space — space hardware and data transmission [Ch. 2

aoNviiii^SNvai
Sec. 2.2] Sensors on satellites 45

More will be said about aspects of the spectrum of naturally occurring

radiation, and the atmospheric absorption spectra, when individual sensor types
and their applications are presented in section B (Chapters 6, 7, and 8).
Definitions of the energy flux parameters used in visible wavelength radiometry
are to be found in the section on optical theory (6.2). Thermal emission
spectra and the details of atmospheric absorption are given in Chapter 7. Other
properties of e.m. radiation will be briefly explained when required later in
the book, including such concepts as doppler shift and coherency which are
important in understanding the operation of active radar sensors. For a more
general review of electromagnetic theory relevant to remote sensing the reader is
referred to Chapters 3, 4, and 5 of Slater (1980) or Chapter 3 of Reeves (1975).

2.2.2 Sensor measurements

In the above section the different families of sensors were introduced in relation
to the frequency range of the electromagnetic radiation used. Now we consider
what the different sensors actually measure, and in the subsequent section we
look at the spatial sampling strategy of different sensor types.
All sensors respond to the flux of electromagnetic energy incident upon
them within the frequency/wavelength band for which they have been designed.
However, there is a variety of other parameters which define the electromagnetic
radiation as well as simply the amplitude, and different sensors have been
developed to measure different aspects of radiation which may carry information
about earth and sea surface conditions.
The basic sensor is a radiometer which measures the flux of electromagnetic
energy reaching the sensor from a given small cone of directions.f Radiometers
are employed to measure in the visible, thermal infrared, and microwave parts of
the spectrum. Their use in remote sensing is readily grasped by analogy with the
way we use our eyes to distinguish bright and dull surfaces. In the infrared and
microwave parts of the spectrum, the interpretation of radiometric measure¬
ments depends on whether the energy is emitted by the surface itself, or merely
reflected as is the case for the visible spectrum. The science of understanding
how radiance measurements may be interpreted in terms of the environmental
parameters of the sea which control them forms the heart of the later chapters
devoted to particular sensor types.
Continuing the analogy with the human eye, we use not only intensity but
colour of hght to convey information about remote objects. Colour as perceived
by the eye depends on the relative magnitude of radiation in three overlapping
wavebands within the visible part of the spectrum. Since this is such a fruitful
source of information for distinguishing objects in everyday hfe, it is not

f A more precise definition of radiant flux quantities, units, and symbols will be found in
section 6.2 in the context of visible light, although the principles apply across the whole
of the e.m. spectrum.
46 The possibilities in space — space hardware and data transmission [Ch. 2

surprising that its electronic counterpart has been developed. This is the multi-
spectral radiometer in which the incoming radiation is measured within several
discrete spectral windows in order to describe its spectral composition or colour.
The spectral windows depend on both the spectral response of the electronic
sensing device, which may be tuned to a particular narrow frequency band, and
the spectral characteristics of any filters used. Whilst the term ‘colour’ is given
to the information obtained, it is not likely to correspond exactly to colour as
registered by the human eye, since the wavebands are likely to be different.
Multispectral sensors are also not limited to three bands, and some may have
many more. In the ocean-viewing context the colour or ‘spectral signature’ of
the radiation as measured by multispectral radiometers is used in two ways.
Firstly, it carries useful information about environmental parameters at sea, and
this is particularly so in the visible/near infrared when the energy is reflected.
Secondly, it provides a very useful way of distinguishing between variability
due to changes in the ocean parameters and variability due to changes in the
atmosphere through which the radiation travels from the sea to the sensor.
This is the principal benefit of multispectral data in the thermal infrared and
microwave regions of the spectrum.
Passive sensors are limited to the measurements of radiance and colour, and
possibly polarisation although the only ocean-monitoring technique so far
developed to exploit the latter operates in the microwave. Since passive sensors
depend on either solar illumination or emitted radiation from the sea, their
energy source generates continuous incoherent electromagnetic radiation, and
there is therefore no chance of obtaining phase or wave speed information. Since
active sensors illuminate their target (the sea) with their own pulse of electro¬
magnetic radiation, there is scope to measure not only the amplitude but also
the phase of the reflected signal, and the travel time of the pulse, both of which
can be made to yield further information about the sea surface when suitably
analysed. All the active sensors so far developed for satellites operate in the
microwave part of the spectrum (0.1 to 1000 GHz), because of the atmospheric
transparency there. The altimeter measures the travel time of a radar pulse to
estimate the height of the sea surface, whilst the difference between the shape
of the outward and return pulses can be interpreted in terms of the sea surface
wave height. The synthetic aperture radar (SAR) uses the ampHtude of the
return signal as a measure of the radar reflectance (and hence roughness) of
the sea surface, but it is able to gain a very high spatial resolution by using the
travel time of the return pulse as a measure of the distance between the point
being illuminated and the satellite subtrack, and by using the shift in frequency
of the return pulse compared with the emitted as a measure of doppler shift, and
hence of position in the along track direction (see Chapter 12).
Another property of electromagnetic radiation which is capable of being
measured is the polarisation, if any, of the electric and magnetic vector. Because
a rough sea will reflect incident radar energy in differing amounts according to
Sec. 2.2] Sensors on satellites 47

its polarisation, instruments such as the scatterometer can gain additional

information about the ocean surface conditions by transmitting and receiving in
both horizontal and vertical polarisation modes.

2.2.3 Spatial sampling capabilities of sensors

Sensors may be broadly divided into two categories — those which scan and
those which do not. The design of the sensor, that is, the arrangement of optical
lenses and mirrors or the geometry of the microwave antenna, controls the
angle of view from which a signal is received by the sensor at any one instant.
This is called the instantaneous field of view (IFOV) of the sensor. As Fig. 2.14
illustrates, in one dimension the angular response of the sensor is usually
a continuous variable so that the IFOV is not precisely demarcated by a transi¬
tion between zero and total response, but rather by some arbitrary threshold
response value. The assumption is often implicitly made that the radiant energy
measured by the sensor is that which would be uniformly integrated over the
two-dimensional IFOV, whereas in reality there will be smaller contribu¬
tions from directions outside the defined IFOV, whilst within the IFOV not
all directions will contribute with the same weight. The extent to which this
assumption is effectively valid depends on the optical design of the system. This
is a limitation on the spatial resolution of the sensor. The directional response
curve (Fig. 2.14) is related in optical systems to the point spread function which
describes the distribution over the sensor image plane of light emanating from
a single point (or direction) in the field of view. In optical systems it is possible
to achieve a very clean cut-off to the IFOV with suitable lens and mirror
configurations (see (Slater, 1980)). Moreover, it is normal with optical systems
to produce a square or rectangular IFOV which enables IFOVs to be patched
together to provide efficient areal coverage without overlap or blind spots.
Radar antennae, working with much larger wavelengths, are not able to focus the

Fig. 2.14 - Typical sensor response to uniform radiance from directions on

either side of the nominal pointing direction of the sensor
48 The possibilities in space — space hardware and data transmission [Ch. 2

radiation so easily, and the IFOV is likely to be circular and less precisely defined,
except in the case of the SAR where the phase and frequency information of the
received signal is used to effectively synthesise the small discrete field of view
and hence to achieve a fine spatial resolution.
Some sensors have a wide IFOV and simply make one downward looking
observation at periodic intervals during the overflight of the satellite,
to give an average value of radiation over a wide area of earth surface, typically
50 km X 50 km or more centred at the sateUite sub-point. The altimeter on
Seasat is an example of this type of field of view. Other sensors, particularly
those operating in the visible and infrared wavelengths, are sufficiently sensitive
to resolve the small amounts of energy to be found in much narrower fields of
view. Thus the AVHRR (Advanced Very High Resolution Radiometer) on the
NOAA satelhtes which measures in the visible and infrared, or the CZCS (Coastal
Zone Colour Scanner) on Nimbus 7, measuring in the visible and near infrared,
have IFOV of about 1 km square or 800 metres square respectively. The visible
wavelength MSS (MultiSpectral Scanner) on Landsat has an IFOV of about
80 metres sqauare, while the thematic mapper (TM) on Landsat has an IFOV
as small as 30 metres square. To produce areal coverage greater than a single
swath of 1 IFOV width around the globe on each satellite orbit, the sensors are
made to scan across the satellite tracking directions. For polar orbits the scan
timing is arranged so that subsequent scans occur after the satellite has travelled
a distance approximately equal to the IFOV projection on the satelhte subtrack
(Fig. 2.15). The radiometer is sampled at a rate so that subsequent IFOVs just
overlap along the scan direction, and there are typically 2000 to 4000 samples
along a scan line. In this way the whole area within a wide swath is efficiently

SATELLITE
track

SCAN DIRECTION -^
~~ _ . - --
~~~ —
- 1''°J1-SENS.N^SC^AN R^TURN)

—-——

1 IFOV '
1

r-- SWATH WIDTH

Fig. 2.15 Swath-filling geometry of scanner ground track

Sec. 2.2] Sensors on satellites 49

sampled once only. The scanning is usually achieved by a rotating mirror, and
may be performed in one direction only or backwards and forwards across the
swath depending on the sensor design. In the case of the proposed Along Track
Scanning Radiometer on the European ERS 1 satellite, a mirror will rotate about
an oblique axis to describe curved scan hnes, crossing beneath the satellite in one
direction and returning back across the swath at an angle 60° in front of or
behind the satellite. In the case of the geostationary satellites (for example
Meteosat) the whole satellite rotates about its axis to achieve scanning parallel
to lines of latitude, whilst the sensor’s field of view is rotated north-south to
point at different latitudes on each satelhte rotation, thus achieving a coverage
of the whole face of the globe as visible from that location in space.
The details of individual sensor geometries will be examined in the applica¬
tions chapters, but some general comments concerning spatial resolution can be
made here. It should be noted that the spatial resolution which is possible for a
given sensor is dictated by the IFOV and not by the sampling rate. It is the
latter which normally dictates the pixel (picture element) spacing, but this is
not the same as the spatial resolution. This is illustrated in Fig. 2.16. In each
case the IFOV and hence the true spatial resolution is the same. However,
whilst there is efficient matching of sampling range and IFOV size in (a), in (b)
the low sampling rates lead to much larger pixel spacing, but since the IFOV is
now smaller than the pixels there are significant areas of ground unseen by the
satellite. In (c) the sampling is more frequent than it need be, with overlapping
IFOVs. Although the pixels are much smaller this does not mean that the
spatial resolution is improved from (a), since objects in the overlap region will
contribute to more than one pixel, thus blurring their position. Careful numerical
analysis of the data, bearing in mind the point spread function of the sensor,
may be able to achieve higher resolution in case (c), but in terms of simply
reproducing an image from the data recorded for individual pixels, case (c)will
not give more spatial detail than (a).
The way in which the sensor integrates the incoming radiation over the
IFOV may be a distinct advantage in oceanographic applications, where the
average conditions over an area are often of more interest than the specific
conditions at a point as measured by a ship, particularly if there is significant
variability over short length scales. On the other hand this can be a disadvantage
if there is a single or a few discrete anomalously bright points in the field of
view, for example, a bright white object such as a ship (or even a sun reflection
from a small but suitably angled reflecting surface on a ship) in the the field of
view of a visible wavelength sensor, or a corner-cube reflector on a buoy in the
IFOV of an active radar sensor. In these cases the radiant energy recorded for
that pixel may be so anomalously high that it must be rejected as a false record
for the general sea conditions, and a data point is lost. Even worse, the data
value may be higher than appropriate for the actual sea conditions, but not
sufficiently high to be rejected, leading to interpretation error.
50 The possibilities in space — space hardware and data transmission [Ch. 2

IFOV

PIXEL
IFOV

^-

PIXEL
IFOV

PIXEL

Fig. 2.16 — Along-scan sampling frequencies

(a) Sampling interval (pixel size) equals IFOV. Efficient ground coverage
(b) Sampling interval too large. Some ground areas missed
(c) Sampling interval inefficiently small. Shaded areas sampled twice

Finally, mention must be made of the way the ground resolution varies with
position along the scan line. Fig. 2.17 illustrates this. Because resolution and the
IFOV are based on angles, the metric ground resolution varies with the distance
from the satellite and the look angle relative to the local vertical. On scanners
such as the AVHRR and CZCS which cover a swath of several thousand kilo-

SENSOR

NADIR IFOVs OBLIQUE IFOV*

Fig. 2.17 - Variation of IFOV (spatal resolution) with view angle

Sec. 2.3] Data retrieval 51

metres this is exacerbated by the curve of the earth too. Thus the resolution
length scale at the extremes of the scan may be several times longer than that
at the satelhte sub-point. Furthermore, if the IFOV is square, the projection
from the sea surface will be square at the sub-point but increasingly rectangular
towards the extremes of the scan line, the longer axis being along the scan line
direction. This degradation of resolution with obUque viewing is particularly to
be borne in mind because although the pixels are equally distorted in the scan
direction, in the satellite track direction they remain equally spaced, or on scales
where earth curvature effects are important they even become closer towards the
scan extremes. Thus there is considerable overlap of IFOVs from one scan line
to the next, which is not immediately apparent from the geometrical distortion
of the image data which will be discussed in Chapter 5.

2.2.4 Recent and planned satellite sensors which have oceanographic

applications
Table 2.3 lists most of the sensors of use to oceanographers, and categorises
them in terms of their e.m. frequency range of operation, whether they are
active or passive, their sampling strategy, and the resultant data rates issuing
from them. The appHcation of many of them will be discussed further in the
second section of the book.

2.3 DATA RETRIEVAL

No matter how long-Hved a satellite is, or how sophisticated, robust, and
sensitive a sensor it carries, neither is of any value to the oceanographer unless
the measurements observed by the sensor can be communicated to him in a
useful form. This requires the encoding of the data on board the satellite, its
transmission to earth, and its subsequent dissemination to users. All of these
are vital aspects of the total concept of satellite remote sensing, and none
of them can be regarded as peripheral to the basic design of the satellite. The
encoding and transmission of the data back to earth place demands on the power
supply, and constrain the data rate which can be generated by the sensor, which
in turn limits the spatial resolution and sampling rate. Without data transmission
the satellite is useless, and since routine maintenance of failed systems is
not yet an available option it is of the utmost importance that the on-board
data-handling systems are carefully designed for trouble-free operation.
The data reception on earth, or ‘ground segment’ as it is called, must also
be reliable. Although it can be maintained and repaired if it fails to receive
transmissions from the satellite, any data that is missed will be lost completely.
Ironically, the one link in the chain which is most likely to be treated as an
afterthought is the dissemination of data to users. From the space technician’s
point of view there is a tendency to judge the success of the satellite mission
by the data retrieval rate, just as experimental oceanographers are sometimes
52 The possibilities in space — space hardware and data transmission [Ch. 2

o
c/3 c/3 X
C3 <L) 00 00
o •+-»
C/D cd x; 3 X X
u. 00 >> 00 00
7d cd o
o -4-»
Cd
*Sh nd
>>
H

high-resolution scanner
XI I (O ' o <u
«_ 00 ix 00
tJO c/3 (U cd <D Cd c

narrow-swath
<D c p
C <D c > cd
O ’o. I 0) cd > cd o
o o o o
a C X5 c/5 o o
cd 173 cd
&0 c/3 O k- cd
<L> <u
a ’> ■4-» c/3 (1>
cd cd
cd
t-H > cd
Satellite sensors used for oceanographic applications

g o
CL.
'o
Cd O. <u
cd C
C .3^
o o
i 2 T3
C/D £
I ^

(U (U a>

passive
a> (U
? > > > > > >

o n) cd cd cd
■< Cl. n) Cl. CX a- O.

<D Pi
>
CTj

near IR
<u Pi ^ jj Pi ^
c:: a> S

visible
& 173 03 3 cd cd
o 1- P ^
c/3 cd
c
C a
X ii c
cd g
« 5> bH
(U
£ 2 §
<= x:
^ p ^ X > P «

73 00
LANDSAT

H z t
<i> c/3 <; I M3
c/3
> W c/3 c/3 D s
O) O < C/D c/3 o <
<
CQ
CJ
3 O W Pi S X
o on w w
H z
o
cd
C/D
Table 2.3

00
Multi-spectral scanner

a> 00
CX .3 oo o p .3
>> c a.
c £ a Ou
Cd "o
O <l> o <u WH a S
<u o c/5 a> ^ cd (D (U
P
s
cd a > Pi
p
oN Pcda 1
• -t
o o
a
^
C ^ J <D —H O cd
o o
00 i-i C S-J cd 2
c cd 3 o ^
CJ o > o Cd
T3 C/D O •*-»
cd
C/D < o U 0)
Pi x

O a
>L Pi
cd c H Di Pi C/3 Pi
MSS

•? o hJ C/D CJ s
(D ok-l H K N
l-H CJ
X) < U E
.£3
<d
Sec. 2.3] Data retrieval 53

(U s
00 e
o 00
'O
.2
O
>-1 <u d>
X
0)
2

d d>
T3 o
d) o
4=1
-+-»
O W) H-*
a
<30
<D •^H O l-H
c Cd cd a cd T3 4-» 4-» Cd
<D <U rj .a Cd Cd
t-H O 4-/ d>
> « s s ^ o d) .■^ o
*5i)
G d) 00
cO > 'f > 4-J 1-H cd
2
:3
Id cfl 'a
(1> 4^
3
^
s =« XI O Cd
4->
d) <-H
o o
o ct; a 00 a c/5 CO
Cd CO 3h >
O) a a c« d> 00 l-H o
C 2
on "eS
o Jn
<1> ^
-4-^ o Oh H-J O o O t-H
2 o
2h Uh d) Q o
e O o Cd o
_ 0^ C lo 3h
cd cd d>
o o o .g & “ Oh d «1>
§ “ 1 ^ 1-H
2 00
CO cd
>.
Td 00
Ch a> o d
« O > o cd CO o *cii
H at Vi
<D cd d o 0) CO /-s
o o O
• 1-H CO CO
o
w
O 4-4 4->
00
a o

<D <U a> o d>

> > >
> >
*00
00
cd o o cd cd
Oh cO cd O. Oh

<u o d) 0:5
> > > I—<
CO cd cd d) ^

3 & 3 52
o o O
'E^ 2
*5 S

H
H <
< H -t- H -t-
< 7 < 7 < g
CO CO
^ m !Z1 I t/5 I O W
C c/5 <* C/5 <« O cd
w w oi s § H O
c/5 W c/3 W c/3 g PJo
S<1>
00

o »-
c a “
o D
1-H 4/ 7 d) 4->
o t-H O
o d) S
4-* CD O <2 a
■> o O
cio ^ a ^
.a °
s
Oh
cd
i-H
cd
s •*0 cd
o
CO o •TS a d) C ^
0) cd
> cd d
-4-»
<l) a cd
c cd 2a o
_p 00
t-l Q
o .a o

<u >.
(/3
C/!)
o
c
^

-a
'3> -S
9^
oi Cd 2
o C

H 5 S 05 05 TJa>
> 05
S
(/)
CQ < c/^ CO
O
0!^: c/^ w S o
C/5 O c/5 C/D o
t-H
a Qh
o
54 The possibilities in space — space hardware and data transmission [Ch. 2

content with a successful recovery of a moored instrument. In both cases the

task is only half complete until scientists have had the opportunity to analyse
and interpret the data. In the case of satellite remote sensing the problem is
compounded by the sheer volume of data which is collected.

2.3.1 Satellite-to-earth data transmission

The earth-viewing sensor will normally generate a voltage or a frequency signal
corresponding to the measurement being made. In some cases (for example,
VHRR, RBV) this information is encoded in a video waveform for transmission
back to earth, effectively as an analogue signal. However, most satellite systems
now convert the analogue to a digital signal for transmission to earth. An
analogue signal can be corrupted by noise and interference during transmission
and storage, whereas a digitally encoded signal is either noise-free, or totally
corrupted and hence irretrievable. The digital encoding of scanned images also
facilitates their storage, their dissemination, and their analysis and enhancement
by image-processing computers as discussed in Chapter 5.
In binary coding, n bits of information are required to represent a whole
number in the range 0 to 2”—1. Scanned radiant data is often digitised in the
range 0 to 255, which requires 8 bits for every pixel value, though in some cases
(for example, AVHRR) the sensitivity of the sensor warrants a 10 bit digitisation
(that is, 0 to 1023). For non-scanning sensors which make single measurements
relatively infrequently the digitisation resolution can be made as accurate
as possible without incurring serious data transmission rate penalties, but for
scanning sensors, generating data almost continuously, a compromise must
be reached between the sampling rate, the digitisation interval, and the cost of
providing high data transmission rates. If a sensor is designed for a specific
marine use, then it may be possible to restrict the maximum and minimum
values which are digitised, thus gaining accuracy over a limited sensor output
range for a given data rate. For example, the CZCS has high resolution at the
lower radiances which are encountered over the sea, but saturates at a digital
value of 255 for the moderate radiances measured over land. It is even possible
to switch to lower or higher resolution ranges, according to the application.
As well as sensor data relating to earth viewing, many sensors generate a
small amount of self-cahbration information which must be incorporated in the
data stream for transmission. For example, an infrared scanning radiometer will
normally view cold space and a heated on-board target during its scan cycle, and
as well as these values the thermometer record of target temperature must also
be communicated to earth. The analogue-to-digital convertor is often designed
to generate a test ramp of digital numbers corresponding to a fixed scale of
input voltages, which enables the digital signal to be accurately calibrated in
terms of the sensor output. These also must be included in the data stream, and
in the case of scanning sensors, each scan line of data is accompanied by some
of this calibration data. Finally, the satellite will supply information about its
Sec. 2.3] Data retrieval 55

position and orientation, and its own environmental parameters and the state
of its power supply. This ‘housekeeping’ information must also be incorporated
into the transmission back to earth. With several sensors on board most satellites,
a fairly complex multiplexing system is required to feed the various pieces of
information into a continuous data stream.
However, it is the scanning sensors which largely dictate the capacity of the
communication downlink to earth. As sensors have been developed, so the data
transmission rates have increased, as indicated in Fig. 2.18. Higher bit rates
demand higher frequency transmission links. Thus the MSS on Landsats 1
to 3 had a data rate of 15.06 Mbits s”^ and used S-band communications, that
is, a carrier frequency of around 2 to 5 GHz, whilst the NOAA satellites carrying
the AVHRR have a lower data rate of 665.4 kbits s“^ (because the scan lines are
further apart than for Landsat), and a carrier frequency of 99 kHz can be
used. The lower frequency transmission and lower data rate make it easier to
receive the data from the NOAA satellites, and high-quaUty data can be received
at relatively moderate costs, as exempHfied by the receiving station built at the
University of Dundee, Scotland. Bayhss (1981a,b) provides a useful guide to
the size of antennae required for a given power of signal in order to achieve a
sufficiently high signal-to-noise ratio for the bit error rate (BER) to be accept¬
ably small. If the satelhte can be accurately tracked by the antenna as it passes
overhead, then the antenna size can be reduced.

Fig. 2.18 - The logarithmic growth of data transmission over two decades of
satellite remote-sensing of the earth (after R.W. H. Stewart)

Data transmission at these high frequencies must be by line-of-sight trans¬

mission, which would limit the ground coverage of the scanner to those areas
viewed by the satelhte when within hne-of-sight of a receiving station. For a
scanner with a small swath such as the Landsat MSS this would require a
56 The possibilities in space — space hardware and data transmission [Ch. 2

large number of receiving stations around the world if global coverage is to be

attempted, and even then areas of sea would be missed. However, on satellites
like Landsat a tape recorder is included in the payload which, on command
from the control station, will record an image of a specified part of the orbit for
transmission to earth when it next passes over a ground station. Other satellites
are able to store a poorer (spatial’and/or radiometric) resolution image of an
inaccessible part of the world on tape, whilst it is customary to record all
samples of low-bit-rate sensors (for example, the altimeter and the scatterometer
on Seasat) for subsequent downlinking. Of course certain sensors like the
SAR on Seasat or the thematic mapper on Landsat 4 have such an
extremely high data rate that there is no chance of tape-recording this informa¬
tion. For Seasat, line-of-sight transmission of SAR signals as they were
received by the sensor was the only possible mode of communication, severely
limiting the possible coverage of the SAR (the network of ground stations for
SAR will be mentioned in Chapter 12). However, the planned operation of
Landsat 4 was to relay the thematic mapper data from any part of the orbit
to a single ground reception station through a network of telecommunication
satellites known as the Tracking and Data Relay Satellite System (TDRSS). In
the future it is to be expected that this will become the normal method for data
retrieval, eliminating the need for on-board tape recorders which, in the very
nature of their having mechanical parts, have tended to be much less reliable
than solid-state electronic components.

2.3.2 Data dissemination

Once the remotely-sensed data has safely reached the tracking station, there is
a temptation for the space technologists to feel that their task is successfully
completed. However, if the satellite is worth launching it is so that the data it
collects can be used by applications scientists. The route by which these data
are disseminated to users varies with the sateUite or sensor, and depends on the
country of origin of the satellite. Here we also face the question of the cost of
the data.
Perhaps the most successful remotely-sensed imagery is that from meteor¬
ological satellites supplied directly to ships at sea and other users by a LIHF or
VHF picture transmission. In the case of the NOAA series, one visible and one
infrared channel of the AVHRR is selected with reduced spatial sampHng by an
on-board computer and broadcast as the automatic picture transmission (APT)
signal. In the case of the geostationary satellites the image data received at the
ground stations are rapidly processed, map grids are added, and a picture at
less than full resolution is broadcast via the satellite, using it in a secondary
telecommunications role.
Operational satellites such as the Landsat and NOAA series are those
Sec. 2.3] Data retrieval 57

for which a continuing sequence of satellites is envisaged and which are designed
to supply data for regular commercial and research appHcations. For both these
series a set of ground stations is established worldwide, and data dissemination
networks are set up. In principle it should be possible to obtain image data from
any part of the world by request if on-board recording facilities are available,
but in practice it is normally easiest to obtain data received at a local receiving
station. Even then, with notable exceptions such as the Dundee University
receiving station, it may take several weeks to obtain images, and it is to be
hoped that near-real-time data dissemination will soon become more widely
available. Image data may come in the form of black-and-white photographic
plates, or in digital form on computer-compatible magnetic tapes (CCTs),
capable of being read on standard computer tape-drives in contrast to the higher-
density digital tapes on which the data is stored at the receiving station. If
a satellite is designated operational, then it is assumed that many of the users
will prefer to have the data in as advanced a state of processing as possible.
Landsat data for example, are usually destriped (see Chapter 5), and in
future it is to be expected that atmospheric corrections and even sea surface
temperature calibrations will be applied to infrared images. This may be useful
for certain applications, but for research use it is often preferable, though not
always possible, to be supplied with the data as transmitted by the satellite,
since the ground station processing may unwittingly introduce errors as well as
corrections, depending on the use being made of the data. At the time of writing,
the cost of data from operational satellites is under review, and it is being
proposed that full economic costs be charged, although it is very difficult to
establish just what these should include (for example, the development costs, the
launch costs, the running costs of the satellite, or merely the cost of maintaining
the data archives). It is further suggested that US environmental satellites could
be sold to and operated by private companies rather than government agencies.
Such a development would undoubtedly lead to further price rises which would
inhibit research use. This would be a pity since there is still scope for the
development of further applications in the marine environment, in addition
to those already identified as justifying the commercialisation of the satellite
programme.
Other satellites such as Nimbus 7 and Seasat are looked on as research
and development programmes. For these a data dissemination network is not
necessarily set up for open access,but it is normal to establish panels of scientific
investigators who have a first right to analyse and interpret information from
the satellite. As nations other than the USA develop their space programmes
it remains to be seen what arrangements will be made to make data widely
available. In Europe, the European Space Agency has established the Earthnet
Organization to disseminate data from some of the American satellites, as
illustrated in Fig. 2.19. Some or all of the ground stations are also expected to
receive ERS-1 data.
 58
LANDSAT HCMM NIMBUS-7 SEASAT
The possibilities in space — space hardware and data transmission

£
CO

0
0

o
c
£
E
n
0

0
[Ch. 2

Fig. 2.19 - The Earthnet structure for dissemination of some satellite data in Europe
Sec. 2.3] Data retrieval 59

One final point worth discussing in this chapter is the question of the
archiving of the data. For land coverage, such as the MSS on Landsat, it is
reasonable to assume that apart from specific requests for many repeat images,
it is only necessary to archive a few cloud-free scenes of each given area, corres¬
ponding perhaps to different seasons of the year. Beyond that the indiscriminate
archiving of every image received from the satellite can be considered to be an
unnecessary expense. For oceanographic appUcation the perspective is different.
Unlike the land, the sea is constantly in motion, and jthe images received by
satellite over the ocean, from whatever sensor, are likely to be constantly
changing. Since research use of the data may not commence until some years
after the satellite s launch, particularly as oceanographers are only slowly
discovering the potential value of remote sensing, there is a much stronger case
for archiving all the data received, so that spatial and temporal variability can
both be studied. With the vast quantities of information retrieved by remote
sensing this may turn out to be very costly, and oceanographers will have to
decide whether the cost of indiscriminate archiving is justified in the light of
the potential value of the applications to be discussed later in this book.
CHAPTER 3

The possibilities for oceanography

Having seen what hardware is available in space, we take a general view in this
chapter of the variety of ways in which oceanographers are able to use satellite
remote sensing as an observational tool for their science. We first consider what
oceanographic parameters can be measured using the different types of sensor.
As a reminder of the environmental background to satellite oceanography we
then take a brief look at the range of problems and areas of research being
tackled by oceanographers which may benefit from the input of remotely-
sensed data. In particular, it is useful to examine the space and time scales of
oceanographic phenomena in order to identify the spectral windows through
which they should be observed, and to compare this with what is available from
satellite remote sensing. It is then possible to consider how remote sensing from
space not only can make observations relevant to existing research frameworks
but also promises to make possible the opening of new areas of oceanographic
study.

3.1 THE OCEANOGRAPHIC CAPABILITIES OF SATELLITE SENSORS

Different information about the sea is carried by different frequencies of electro¬
magnetic waves, with the result that particular sensors are associated with
particular oceanographic parameters.
Visible wavelength sensors are passive, and depend on solar illumination as
the initial radiation source. This is reflected from the sea back to the satellite,
and therefore the information to be retrieved from colour scanning relates to the
reflection process. If the sensor viewing angle is arranged to avoid direct sun
glitter, the upwelling radiation must come instead from backscattering processes
inside the sea, and can therefore convey data concerning suspended sediment,
plankton, or yellow substance (dissolved organic detrital material) in the water.
A multispectral scanner yields information about the ocean colour which is
required if we are to distinguish between backscatter due to plankton and that
due to sediment. Visible light can penetrate several metres below the sea surface,
and visible wavelength sensors are therefore unique amongst remote-sensing
[Sec. 3.1] The oceanographic capahilities of satellite sensors 61

devices in being able to respond directly to conditions in the upper part of

the water column. All other wavelengths can only reveal conditions at the surface
or within a depth of less than a millimetre. In shallow seas, less than about 20 m
depth, light reflected at the sea bed can be seen from space provided the water
is fairly clear. The amount of reflection depends on the material of the bed and
the depth. Thus bathymetry and bottom sediment identification are two further
possible apphcations. Visible sensors also respond to some extent to surface
materials such as oil slicks, depending on their optical transmission properties.
A final application of visible wavelength data, even from single-band sensors
which do not offer a colour signature, is to provide information about the
amount of short-wave radiant energy from the sun which is not absorbed in the
ocean. Clouds and atmospheric backscatter are not such a hindrance to this
apphcation as they are for all the others.
Moving to longer wavelengths, the near-infrared wavelength sensors are
effectively complementary to the visible wavelength sensors, although absorption
in the sea rapidly increases, and 1 nm wavelength radiation penetrates only a few
millimetres. This is in contrast to land application where the reflected infrared
radiation is a valuable remote-sensing indicator for identifying plant species or
surface substrate types. With reference to Figs 2.12 and 2.13 it can be seen that
sensors operating in the next waveband where the atmosphere is fairly trans¬
parent, at 3 to 4 fim, will record appreciable amounts of reflected solar energy
in the day-time, but in the absence of solar reflection at night-time will record
radiation emitted by the sea surface itself. At 10 to 12 /rm the emitted radiation
dominates, and thus the thermal infrared sensors, as their name implies, provide
a means for estimating the surface temperature of the sea. As we shall see in
Chapter 7, this is a valuable tool for the study of a large range of dynamical
processes in the surface layers of the ocean.
Passive radiometers measuring yet longer wavelengths, in the microwave
region, are again indicators of sea-surface temperature. The emitted radiation
depends also on the emissivity of the sea surface. If multichannel measurements
are made across the infrared and microwave part of the spectrum it should be
possible in principle to extract information not only about temperature but
also emissivity and the parameters on which it depends, including salinity and
surface films, oil spills, etc. The potential applications of the passive sensors are
summarized in Table 3.1, although it will become clear later in the book that not
all these applications are as well developed as others, and some such as salinity
detection or bottom substrate identification may turn out never to be achieved.
Active sensors have been developed with specific applications in mind, such
as those listed in the second half of Table 3.1. However, whereas it is unlikely
that future development of passive sensors will add any further possibilities to
the list in Table 3.1, the possible applications for active sensors are by no means
exhausted. If laser measurements could be made from space, then a whole
range of new possibilities would open up, but we shall restrict ourselves in this
62 The possibilities for oceanography [Ch.3

swell wave patterns, internal wave

rC G

patterns, sea-bed topography

& •G .2
CtS
t-H
u t« <u
4)
o CD
u
Cd
kH
W)
o T3 ^ cd
CD
3 2 O 4) a
c/3
0) c bi) _bp •G 0
c c/) (D 73
r~
4^ cd 'G «G CD
O ’S
T3
c/3 G -C C Td
^ c/3
<D
S
O (D 'O
C
8 o
• E ^ cd >» o
o I i CD
>
Cd
'?
CD
« ^
0) C 2 CD <D XJ cd
<U -3 P
^ £
c/5 c> -C
c/3 s: o. G >
ii -t-^
cd
4) ^
? 2
<D 03
3 ^ C
c •— 2 ^
C
q;>
cd
c/3
^ 4) O G
t-i
G ^7d
c2 cd Vi
H a. r' CD
■4—> D ^ G E .S V
^
c •-
0
o G cd CD
0)
l-H c2 O a. s
«-.+-> w ciS ^ 'G G G cd '5
> ^
(D
cd bO
d
'f X3 M
0) «-H ^_.
3 cd
G cd
0)
Vi 4).G
0 '3
3 M CU o 2 a ^
CD
o
cd
OD
cd ^ 'tl cd c/5 M-H
4) kH
o G
(3^ S
Cd ^ c/3

1/3 ^ o
rX
c CD
^ (D ;:3
.2 #v cd
4)
o
cd
b£)
G
»-i a>
+-»
cd 55
o
cd S c 3S 1-2
a ^ ^ C*-. ^ S 3 E

surface roughness
"a
a
c« ^ i::
c/5 S <D
*-* C 4> 4) CD 7d

o
l-< _ c/5 O- c/3 ag G C •S 3
CQ 2 E '■ E > ■*s ^
4= DG
c/3 0
s cd O ^ £ E o ^ c
.-M o bO bX) Vi Cl
G G 0) r-
as &£)- 4> O 0
tH G G •-
W) Vh'' g ^ “ J3 T3
CD ■4x' M
a (-4
4= —
o d W) cd " TD o CD OD bf) cd
& o ^ g- cS <D O G O CD G C
c
oj t-i O o cd .3 0 .0
&
o
O -2
Scg G >> • GJ Vh
kH
o o O 0) G
G OD CD
fl o C 3 C .G
cd c/3 CD <D
cd ><<2 cd 5^ cd c/3
0) Vi
-4-» G .JG
a; ^ -G
^ o ’G
o
<D 5
cd
3
c«
O Ss
W O C
bO
G
o CD • C/l
c
(U
.fcs
a CD

pC small areas of sea surface, along

(D Cd G
T3
strength of return pulse from

'a O CD
c CD c/3
*C/5
with doppler information
CQ <D i^\ >-> -S Vi
sc3 “3 52 'O 0)
c/3
0
o;
a< CtJ ocd 2J
c«
i s G
"g
cx
.2
3 ■§
>» C14 ^
C/5
■ •
c d> 4)
>
cd CI4 G G- OD
O 2 E^ o E kH G
>>
Cd x>
^00 c/3
1) o G kH *3
o X>
O T3 CD
O ^ P Q> CD
Cl.
QD
kH
3 ^
>> « a c w
^ R 'C cd C4-I £ S
■?
a U S
c
a>
rP .2 S Cd C SQ^ »G X
!y5
0 <3
C/D
(X
^cd ^ E OD
^ D
G CI4 c G
0
4) (D ^
o tH
<D 'E ^
cd rt)
CD O
,cd O G IH
G
"S) *5 S
7d Cd
(D
G
ID G
Wi S
E 3 C kH
as
CD M E G
t/i
as H-*
e/5 £ ^
H-J "G
o o (
cd
c/3
4>

goes C<
SCAT

a, C/D U
ALT

SAR

SAR

H
E S
cd
X ^ u <
$ K CO
1^
High-resolution
viewing

imaging radar

<D
Qh
(D
+-» oC 2
I—I <0
bO
>* CD 4> .s
■4-J
E
dar

E CD
0 cd
.2 ■E E
i3 o E 0 G
1 G' G
kH
c cd .ta •-
CD H ^ =S G
CO H 2
•G
cd G
0^
00

cd O
cu <
Sec. 3.1] The oceanographic capabilities of satellite sensors 63

text to systems which have already been developed and flown on satellites. All
of these are microwave devices which can penetrate cloud cover, and rely on the
backscattering of radar waves from the sea surface to convey information about
that surface back to the satellite.
By recording the return time for a pulse directed towards the satellite
sub-point, the altimeter measures the height of the sea surface relative to its
own position, and if this can be fixed then it is possible to determine the
absolute height of the sea surface. Oceanographic appHcations of this informa¬
tion include the study of tides and ocean circulation, and revealing details
of ocean bathymetry. In addition, the deformation of the microwave pulse on
reflection carries information about the significant wave height of the ocean
waves.
The scatterometer emits an oblique pulse of microwave radiation, and
measures the intensity of the backscattered return energy. Since this depends on
the surface roughness at length scales comparable to the radiation wavelength,
it carries information about the sea state over the illuminated area. Depend¬
ing on the particular wavelength used, this can be interpreted in terms of the
near-surface wind magnitude, the surface-wind stress, or the energy in the
surface-wave field. By viewing the same piece of sea surface from different
directions a measure of wind or wave direction can be achieved.
The synthetic aperture radar, SAR, operates in essentially the same way as
the scatterometer, but by measuring the timing and phase of the backscattered
signal as well as its amphtude sophisticated processing is able to reproduce an
image of the backscattering strength of the surface (that is, its roughness as seen
by the radar). It achieves very high spatial detail, with a resolution of order tens
of metres in contrast to the average return from an area of order 50 km X 50 km
achieved by the scatterometer. The roughness it measures is due to small waves
and ripples of a few centimetres in length. The capacity of SAR to image
anything of oceanographic interest depends on the extent to which oceano¬
graphic features are reflected in the surface roughness patterns. This has turned
out to be a most exciting aspect of satellite oceanography. Before SARs were
flown on satellites it was hoped that they would be able to image swell patterns
(that is, regular surface waves with lengths of over 100 m), ship wakes, and
perhaps slicks due to sea-surface contamination by hydrocarbons or floating
debris. In fact the Seasat SAR proved to be capable of revealing much more,
including internal waves and bottom topographical features. Because oceano¬
graphers had never had a radar bird’s-eye-view of the ocean before, it had
hardly occurred to them that dynamical or topographic phenomena centred
many tens of metres below the surface could have a surface roughness signature
capable of being imaged by a SAR. The discovery that this does happen under
certain circumstances opens up new applications possibilities for active radar
sensors, as well as raising many questions concerning the mechanisms by which
the phenomena are imaged.
64 The possibilities for oceanography [Ch. 3

We shall consider this further in Chapters 10 and 12. First we must examine
the science of oceanography as a whole to see where these remotely-sensed
observations can make a useful contribution.

3.2 OCEAN SCIENCE IN THE 1980s

The ocean covers the greater part (70%) of the earth’s surface and has a
dominant influence on both the chmate and the natural environment in which
we live. It is used in a variety of ways by mankind, and in the future it is likely
that increasingly more resources, organic and inorganic, will be sought from the
seas. At the same time demands will be made on the ocean as a convenient
depository for unwanted waste materials. Yet it is still comparatively unexplored.
As the science of oceanography has developed over the last century, the sea has
revealed more and more complexity. The challenge to understand the processes
occurring on and below the sea surface is increased rather than diminished as
advances are made in oceanographic research, although the ability to apply the
knowledge gained by research is steadily developing.
Oceanography is essentially a multidiscipHnary field of study, employing
all the physical, chemical, biological, and geological sciences and mathematics
in an attempt to understand ocean processes. Whilst individual discipHnes have
progressed a long way with isolated problems so that we may, for example, be
able to describe the biology of a particular zooplankton in some detail, or
produce a mathematical model of the tides in a particular shallow sea, we are
still strugghng to understand the ocean as a whole system with complex inter¬
locking connections. This is the preoccupation of much of the international
research effort in marine science in the 1980s. Connections are being sought
between processes occurring on different length and time scales, for example,
how does a small ocean eddy affect the ocean-wide circulation? We are also
trying to understand the links between chemical, biological, and physical
processes, for example, is the patchiness of biologically productive regions in
the sea dictated by the population dynamics (biology), the levels of nutrient
present (chemistry), the turbulent mixing and dispersion within the water itself
(fluid dynamics), the limited penetration of solar energy into turbid water
(physics), or a combination of these and other effects? By providing a new
perspective, sateUite remote sensing is beginning to generate new ideas and
insights for these types of multiscale, multidiscipHnary questions.
The motivation for oceanographic research is not merely the seeking of
understanding for its own sake, but so that the ocean can be exploited fully
to benefit mankind, and also carefuUy managed in order to preserve it in a
healthy state for future generations. As the world’s population increases, the
need to make use of the ocean increases, but it becomes clear that indiscriminate
use could lead to permanent ill-effects. Only by understanding the complex
interactions involved can we hope to propose reaHstic safeguards.
Sec. 3.2] Dynamical oceanographic processes 65

One application of oceanographic knowledge is associated with the increas¬

ing extraction of offshore minerals, particularly oil and gas. At present virtually
confined to the relatively shallow continental shelf zones, this seems likely to
move to somewhat deeper water before being gradually replaced by deep-sea
mining, mainly for the metals in manganese nodules but also to exploit exotic
areas such as the Red Sea brines. The efficient and safe conduct of such off¬
shore industries needs a knowledge of, and preferably the ability to forecast,
environmental conditions such as winds and waves, currents, etc.
Less spectacular, but commercially more important, is surface shipping
which is gradually changing to more specialised vessels with smaller crews,
vessels which need accurate cUmatic data for their optimum design and accurate
weather forecasts for their economic operation. Similar considerations apply to
the fishing industry, where continued improvements in boats and gear have
gradually increased the world’s fish catch. Here the preservation of stocks needs
an improved understanding of zoogeography and of the effect of environmental
factors on the abundance of the various species which make up the marine food
web. The next few years may see increased exploitation of the vast kriU stock in
the hitherto httle-studied Southern Ocean.
It has become increasingly obvious during recent decades that the ocean is
not a limitless sink for man’s waste. Research is therefore increasing in order
to determine the distribution and the dispersion of material from rivers and
engineered outfalls or from the atmosphere. This will go hand-in-hand with
studies of the toxicity and sublethal effects of various chemicals on marine
populations and on man. This is particularly important in the case of radio¬
active waste disposal — the proposal that high-level radioactive waste can be
safely disposed of on or under the deep-ocean floor has stimulated a lot of
interdisciphnary work on the benthic boundary layer of the deep ocean.
The ocean contains various sources of natural power which can be tapped
to assist in meeting man’s thirst for energy. The tides, OTEC (ocean thermal
energy conversion), and the waves are all being studied in this connection.
Finally, a major motivation for much of the study of dynamical oceano¬
graphy is to contribute to the study of weather and climate. The ocean
provides the greater proportion of the bottom boundary of the atmosphere,
and meteorologists and chmatologists demand better knowledge of how the
ocean-atmospheric heat engine works. Clearly there are many reasons why
oceanography is an active science. To what extent can remote sensing contribute
to the development of our understanding in all these areas?

3.3 DYNAMICAL OCEANOGRAPHIC PROCESSES

To identify the contribution which remote sensing can make in helping oceano¬
graphers tackle these important questions, it is useful to take a brief look at
some of the dynamical features which occur on a variety of different length
66 The possibilities for oceanography [Ch. 3

scales and which control many other chemical and biological oceanographic
processes. We exclude here a discussion of surface-wave phenomena, important
to the analysis of active radar sensor data, but this will be treated in greater
detail in the next section.

3.3.1 Ocean circulation

The study of the general circulation of the ocean is the central core of aU
dynamical oceanography. It is the overall movement of the ocean which trans¬
ports heat polewards from the low latitudes and contributes to the control
of the earth’s climate, and it is the rate of circulation, poleward by surface
currents and equatorward by deep currents, which controls the time scale of
sea-temperature variations. With a timescale of hundreds of years, the ocean
has an almost infinite thermal capacity in comparison with the atmosphere, and
thus tends to exert a stabilising influence upon long-term climatic fluctuations
of the atmosphere. Nonetheless, the variability of seasonally-normalised sea-
surface temperatures, on timescales of months to years, can be coupled to
similar variations of the atmospheric pressure distribution. Thus, ocean currents
are of great significance in the development of dynamical meteorology.
The ocean circulation is also important in that it transports nutrients and
populations of phytoplankton and zooplankton, and it governs the distribution
of water masses of different temperature and salinity. In these ways it controls
the overall productivity of the seas. Similarly, it controls the fate of other
chemicals whether naturally occurring, or dumped, in the sea.
In considering ocean circulation it is convenient to split the subject into two
parts - the long-period mean circulation (averaged over several years or more)
and the fluctuating part.

Mean circulation
This is the ocean circulation that oceanographers have traditionally sought to
describe. As an example. Fig. 3.1 shows the mean circulation thought to occur
in the upper layers of the world ocean in July. The more energetic parts are well
documented from observations, whereas the weaker gyres are more conjectural.
The main features are a series of gyres which extend longitudinally across the
entire ocean, and which vary with latitude in their sense of rotation. The rotation
sense is correlated with the latitudinal shear of the prevailing wind stress, and
this circulation is therefore described as the wind-driven circulation. The circula¬
tion at depth is considerably different, and depends on the distribution of mass
due to temperature and salinity variations — hence the deep-ocean circulation
is considered to be a thermohaline circulation. Since remote sensing only
penetrates the surface layer, it is the wind-driven circulation of the oceans which
is most relevant in the context of satellite oceanography.
 Sec. 3.3]
Dynamical oceanographic processes

Fig. 3.1 — The general circulation of the surface layer of the world ocean in July
67
68 The possibilities for oceanography [Ch. 3

An important feature of the wind-driven circulation is the longitudinal

asymmetry. Through most of the ocean there is a very slow southward or north¬
ward drift, sometimes with stronger flows at the eastern coasts of the ocean due
to the effect of wind forces in the vicinity of the coastline. However, the return
flow of the gyres occurs in strong western boundary currents. Thus, for example,
the water which flows southwards in the Atlantic at latitude 30 degrees N
through a width of over 2000 km returns northwards in the Gulf Stream through
a width of about 100 km (see Fig. 3.2).

Fig. 3.2 — The general circulation of the North Atlantic. Flow rates indicated in
10‘ m’s-' (from (TOPEX 1981))

There are clear dynamical reasons why this should be so, relating to the
Coriolis force due to the earth’s rotation, and its variation with latitude.
Stommel (1958) gives a general explanation in terms of the balance of vorticity
in the ocean. Because it is a boundary current, the path of the western flow
is controlled to a certain extent by topography. Thus the Florida current flows
energetically from the Gulf of Mexico through the Florida straits, with a velocity
of the order of 2 m s”\ and bends northwards to follow the American coast as
the Gulf Stream.
The ocean gyres tend to be composed of distinct water masses, with
different temperature, salinity, and chemical constituents, and often distinct
biological populations too. These aspects are capable of being studied by visible
and thermal radiation remote sensors, whilst the dynamics of the gyres may be
detected by the sea-surface slopes which accompany them, measurable using
satellite altimeters as discussed in Chapter 9.
Sec. 3.3] Dynamical oceanographic processes 69

The time-variable ocean circulation

In the last two decades, as more direct observations of ocean currents have
been made using moored and drifting current meter arrays, it has become
increasingly apparent that, apart from strong currents such as the Gulf Stream,
there is in fact far more energy in the fluctuating part of the ocean circulation
than in the steady part. Fig. 3.3 from the TOPEX Working Group Report
(TOPEX 1981) shows the distribution of general circulation kinetic energy in
frequency—wavenumber space.

Fig. 3.3 - Sketch of the frequency-wavenumber spectrum of circulation energy

at mid-latitudes, with arbitrary contour units (from (TOPEX 1981))

The high energy of variability at a period of one year, over a broad range of
length scales, indicates the importance of annual fluctuations of the general
circulation. A large amount of variability is found at length scales between
50 km and 500 km, with a peak around 100 km and 100 days period, cor-
70 The possibilities for oceanography [Ch. 3

responding to the mesoscale eddy field. Little energy is expected at length

scales less than around 50 km, that is, around the so-called Rossby radius of
deformation. This is given as

where D is the depth of the layer of less dense water overlaying the deeper
ocean, with a density contrast of Ap, and / is the Coriolis parameter. The
Rossby radius, Lr, governs the size of the smallest scale of motion which can
be maintained in geostrophic balance (see, for example, Pedlosky (1979)).
It turns out that over most of the ocean where the supposed mean circula¬
tion is weak, the eddy field is much more energetic. Thus it is very hard to
determine exactly what the mean circulation is. Moreover, there may well be a
contribution to the mean flux of mass, momentum, energy, salt, heat, and other
chemical tracers, which is not found by simple time averaging of the flux but is
due to nonhnear interactions within the eddies (Rhines 1977). However, neither
the contribution of the eddies to the mean fluxes nor the distribution of energy
within the eddy spectrum is yet well understood. Even the origin of the eddies is
not fully clear. Mesoscale eddies could result from baroclinic instability (which
is related to the density stratification), from topographic generation (by inter¬
action of the mean currents with bottom topography), and from wind forcing.
One mechanism of their generation occurs when the western boundary current
reaches higher latitudes and becomes detached from the coast. Meanders occur
in the detached jet which breaks up into rings. For example, in the Atlantic, the
Gulf Stream rings entrap warmer (Sargasso) water from the south-east side of
the Gulf Stream and transport it to the north and west, whilst cold rings do the
opposite with water from the Continental slope (Fig. 3.4). These rings may then

Fig. 3.4 - The formation of a cyclonic eddy or ring from a Gulf Stream meander.
The left sketch shows the meander developing, and the right sketch shows the ring
of cooler slope water detached from the Gulf Stream and isolated within the
warmer Sargasso water. The solid line represents the 15° C isotherm at 200 m
depth. The dashed hne is the approximate limit of the Gulf Stream on the
Sargasso side (after (Parker 1971))
Sec. 3.3] Dynamical oceanographic processes 71

wind up into the mesoscale eddies. Much of the energy probably shifts to shorter
length and time scales until eventually it is dissipated as heat in the turbulent
eddies at scales less than 1 cm. The study of the turbulent energy cascade in
the ocean circulation and the extent to which energy may also be transported
to larger length and time scales is of considerable interest to oceanographers at
present.

3.3.2 Tides
Tides are the other major dynamical process governing water movements in the
oceans besides the general circulation. Forced by the gravitational attraction
of the moon and the sun, the tides are identified by their regular periodicity
covering a spectrum of discrete frequencies (tidal harmonics) corresponding to
the frequencies in the gravitational forcing due to the orbital parameters of
the earth, moon, and sun. Principally, tidal motions are horizontal, although the
most obvious effect of the tides is the periodic rise and fall of sea level. The
dominant frequencies are the lunar semidiurnal (period 12.42 hours), the solar
semidiurnal (period 12 hour), the lunar diurnal frequency (25.8 hour period),
and the luni-solar diurnal harmonic (23.9 hour). These can account for the main
features apparent in typical time series of observed tidal heights, such as those
illustrated in Fig. 3.5. These include the normal semidiurnal tide of period 12.42
hours (a lunar period because the lunar influence is strongest), amplitude-
modulated over the fortnightly spring/neap cycle by the addition of the weaker
solar 12-hour cycle. In addition, there is a diurnal inequality of the semidiurnal
tides (one tide each day is higher than the other), due to the diurnal frequencies
present. This can be a dominant effect resulting in primarily diurnal tides if the
dynamic response of the ocean is locally stronger for diurnal than semidiurnal
forcing (see the lower curve of Fig. 3.5). The ocean as a whole appears to show a
greater response to the semidiurnal forcing than the diurnal. There are longer-

IMMINQHAM

TIME, Interval* of on* day

KUALA BARAM

Fig. 3.5 - Time series of tidal elevations at (a) Immingham, UK, showing a typical
semidiurnal tide, and (b) Kuala Baram, Malaysia, showing a combination of a
strong diurnal and weak semidiurnal tide. (Tidal data provided by I.O.S., Bidston
Observatory)
72 The possibilities for oceanography [Ch.3

period variations of monthly, six-monthly, annual, and longer periods, as well

as many other smaller modulations of the principal frequencies. In shallow seas,
non-linear dynamics give rise to further periodicities notably at shorter periods
(approximately 6 h and 4 h).
Like the ocean circulation, the study of tides on a global scale is most likely
to benefit from satellite altimetry rather than other remote-sensing techniques.
The length scale of tidal elevation variations is controlled by long-wave propaga¬
tion, so that in the deep ocean this is of the order of 1000 km, and in shallow
seas around 100 km. However, in shallow seas the tides are often the dominant
motion, and many of the small-scale gyres off headlands, and water circulation
in estuaries, are driven by the tides. Tidal motions in shallow seas may therefore
have a strong thermal or colour signature which can be viewed from space.
Local interactions between wind and tide may also generate surface roughness
patterns which can be imaged by SAR.

3.3.3 Fronts
Turning from ocean-scale phenomena to smaller-sized dynamical features, we
first consider fronts. As in the atmosphere where the terminology was first used,
fronts in the ocean represent a boundary between two distinct water mass types,
in which the temperature and/or salinity, and hence the density, have strong
horizontal gradients. Simple hydrostatic considerations indicate that such a
situation cannot occur in a stationary fluid unless the interface between the two
water masses is horizontal. If it is inchned, horizontal pressure gradients exist,
which must either produce acceleration of the fluid or be in equiUbrium with the
Coriolis force. If a front is recognisable as a quasi-stable feature, it is assumed
that it is in this condition of geostrophic balance, in which case one water mass
must be moving relative to the other in a horizontal direction parallel to the
interfacial plane. Fig. 3.6 illustrates this schematically.

surface expression

Fig. 3.6 - Schematic of ocean front between water masses of densities pj and pj
(Pi > Pj). Velocity directions are appropriate for N. Hemisphere
Sec. 3.3] Dynamical oceanographic processes 73

Fronts are found in several situations in the ocean. The Gulf Stream itself
may be considered to be a large-scale front separating the cooler continental
slope water from the warmer Sargasso water. Within the mesoscale eddy systems,
convergences may occur which will steepen the horizontal density gradients
leading to frontogenesis. Such fronts exist on length scales of a few kilometres
across the front and tens of kilometres along it. Fronts of this nature can
develop instabilities along their length which finally lead to their break-up.
In shallow seas a different type of front has been observed (see, for example
Simpson, Hughes, & Morris (1977)), controlled by tidal energy and solar heating.
In summer the solar heating of the ocean heats the surface layers of the sea
which become less dense and tend to lie stably above the deeper, cooler, and
denser water. Wind mixing of the surface layer deepens and strengthens the
thermochne, but in shallow tidal seas the turbulence associated with tidal flow
over the rough sea bed is often sufficiently vigorous to break down this stable
stratification and mix the water column to a uniform density from surface to
sea bed. The boundary between the stratified and well-mixed sea is characterised
by a front, where the thermocUne rises to the surface, shown schematically in
Fig. 3.7. Such fronts are found around the British Isles during the summer
months. Their location appears to be governed by the parameter hjU^, where U
is the surface tidal stream amplitude at Springtide and h is the sea depth. For
small values of hjU^, in shallow water with strong tidal streams, the sea is
vertically well mixed. Stratification occurs at high values of the parameter, and
the frontal region occurs where h/U^ is about 70, {U in m s“\ h in m).

SURFACE EXPRESSION

Another type of front is also possible, described by Pingree, Forster, &

Morrison (1974) as a turbulent convergent tidal front. Such a front is generated
by a convergent residual flow pattern bringing together water masses of different
types in regions with strong tidal streams. Both water masses are vertically well
mixed — their different densities may have arisen from differential heating or
74 The possibilities for oceanography [Ch. 3

cooling, in different depths of sea, or from fresh water runoff from an estuary.
It is not clear whether such fronts are maintained solely by advection or whether
they are in dynamic geostrophic balance as are the other types of front discussed
above.
Fronts are important to oceanographers in several ways. Dynamically, they
represent a step in the turbulent energy cascade from large-eddy scales down to
billow-turbulence scales of a few metres. They are of considerable importance to
the productivity of the sea since they tend to bring together cooler but nutrient-
rich water, with warmer less rich water. The combination of temperature and
nutrients arising from cross-frontal mixing leads to greatly enhanced productivity
of plankton, attended by increased fish stocks. Because they represent boundaries
between water masses, fronts may also define the limits of patches of pollutant
in the sea. There is often a surface convergence associated with fronts due to
secondary circulation in the vertical plane at right angles to the geostrophic flow
of the front. This causes an accumulation of surface debris, and may trap surface
pollutants, natural surface films, oil slicks, etc.
For a variety of reasons, therefore, it is valuable to be able to identify frontal
regions. Some fronts have a strong thermal signature and are readily located on
satellite infrared imagery as discussed in Chapter 7. Others may have a signature
in the visible wavelength spectrum, due to the enhanced productivity, the
different colours of the two water masses, and surface debris and slicks. There is
also the possibility of identifying fronts with synthetic aperture radars due to a
change of surface roughness at the front, or because the surface wave pattern
changes abruptly across the front.

3.3.4 Upwelling
Upwelling is the term used by oceanographers to describe the situation where
cool, but nutrient-rich water from the lower layers of the ocean, too dark for
plant Life to grow, is raised towards the surface. Brought into the light, such
waters become very fertile and rich feeding ground for fish. For upwelling
to occur, there must be a divergence of the surface currents, and this usually
occurs as a result of the wind field causing surface water drift in the presence of
topographic constraints. For example a longshore wind, flowing in the Northern
Hemisphere with a coastline to the left of the direction of travel, sets up
a geostrophic flow away from the coast, and upwelling tends to occur. This
upwelling is a regular occurrence off the North-West African coast, and off the
Peruvian coast, where wind conditions are suitable. Nonetheless, upwelling may
well be intermittent, depending both on the weather and, to a certain extent,
on the ocean-wide thermohaline circulation. Because of the importance to
commercial fisheries, much oceanographic research has been performed in order
to understand and predict upwelling, and remote sensing from satellites is
Sec. 3.3] Dynamical oceanographic processes 75

beginning to be used to develop this. For example, upwelling occurring in the

Gulf of Guinea has been clearly revealed by infrared imagery (Citeau & Domain
(1981)).
Upwelling is normally revealed on remotely-sensed images by the identifica¬
tion of the boundary between the cooler upwelled water and the surrounding
surface water. This often takes the form of a sharp frontal change as discussed in
the previous section.

3.3.5 Mixed-layer dynamics

Because the ocean is heated from above, and because warmer sea-water is less
dense than cooler, the ocean temperature decreases with depth. This decrease
is not uniform over depth, but in most parts of the world there is a surface layer
having a depth of between a few metres and several hundred metres, which is
well mixed by the surface winds and has a uniform temperatue. Below this the
temperature decreases rapidly in what is known as the thermocUne, resulting in
the typical temperature profiles shown in Fig. 3.8. Much oceanographic research
has been devoted to the study of this mixed layer, its depth, its stability, its
horizontal variability, and the dynamical processes which occur within it (see,
for example, Krauss (1977)). This is because it is a vital part of understanding
the processes of air—sea interaction, as evidenced by the recent Joint Air—Sea
Interaction Experiment (JASIN) in the eastern North Atlantic (Charnock &
Pollard (1983)). There are also strong military interests in being able to model
and predict the surface layer, so that acoustic detection of submarines can be
facilitated.
Temperature degC

-5 0 5 0 10 0 10 20

E 500
x:
a
O)
o 1000

1500

HIGH MID LOW

LATITUDE LATITUDE LATITUDE

Fig. 3.8 - Vertical profiles of ocean temperature at different latitudes

76 The possibilities for oceanography [Ch. 3

Remote sensing, capable of investigating the surface and near-surface layers,

can be expected to play a significant part in mixed-layer research in future.
Although it was fortuitous, it was also most appropriate that the J AS IN experi¬
ment should have occurred during the brief life of Seasat in late summer 1978.
This enabled not only the calibration of some Seasat sensors from a valuable set
of synoptic sea measurements, but also demonstrated that remote sensing can
make its own particular contribution to oceanographic exercises (as illustrated
by the papers in Allan (1983)).
Microwave sensors which determine wind stresses and wave conditions can
perhaps contribute most to the study of mixed-layer dynamics, but thermal
infrared SST measurements also have a part to play, particularly in determining
horizontal variabiHty of mixed-layer conditions. However, it may not be only
the most obvious sensor types which turn out to be useful. For example it
has recently been pointed out that some of the solar heating penetrates beneath
the mixed-layer in certain latitudes and seasons of the year. Hence the remote
sensing of ocean colour to determine turbidity and solar radiation absorption
characteristics may have a vital role to play in developing a climatology of
solar heat penetration.

3.3.6 Coastal and estuarine dynamical and topographic phenomena

The physical oceanography of estuaries and coastal seas tends to be less
amenable to scientific study than are the physical processes of the deep ocean.
The reason for this is that the dynamics of coastal waters are strongly influenced
by the local topography (that is, by the sea depth variations and the shape of the
coastline), and the water mass distribution in estuarine areas is influenced by
local river inputs of fresh water. Because of the dominance of unique local condi¬
tions it is hard to make scientific generalisations about coastal oceanography
from observations made in particular localities. In contrast, the deep-ocean
dynamical and physical processes are less influenced by the location of particular
observations and are capable of description by generalised models. Surface waves
are an exception, since they interact with coastal topography in a predictable
and clearly understood way, but other processes must be studied in the context
of their immediate environment, and only the most broad generalisations can be
made to enable the experience gained in one location to be utiUsed elsewhere.
It is significant that books written on the subject of physical processes in
estuaries and coastal seas, for example. Dyer (1973), Officer (1976), devote as
much of their attention to case studies of particular regions as to broader general
principles.
The types of phenomena which are studied in the coastal region include;
the distribution of tidal currents,
the response of semi-enclosed bodies of water to tidal forcing,
residual flows and circulation.
Sec. 3.4] Ocean waves 77

the density distribution pattern in estuaries,

sedimentation patterns,
the movement and growth of mobile sea-bed features such as sand waves,
beach erosion processes,
the distribution of suspended solids,
mixing and dispersion processes (lateral and vertical).
The apphcations of scientific knowledge in this area are many, ranging
from aspects of coastal and beach protection, through navigational applications
of hydrography, the impact of currents on offshore structures, circulation and
mixing studies relating to pollutant dispersal, plankton blooms, and water
quality conservation, to the definition of temperature and salinity distributions
because of their impact on biological Ufe. There are numerous commercial and
engineering interests which could make use of a scientific understanding of
dynamical processes in coastal waters. However, for the reasons outlined above,
there are still many gaps in our broad scientific understanding. For most applica¬
tions, scientific knowledge has to be gained from local measurements and
observations, obtained ad hoc when they are required, which can be extremely
costly.
The remote sensing of coastal and estuarine regions is beginning to be a
valuable complement to ship-board observations. For example the consulting
engineer, faced with a coastal engineering problem on the other side of the
world, can now look at satellite imagery for a general introduction to some of
the dynamical processes which occur in the region, before any expense needs
to be committed to exploratory ship surveys. Visible wavelength images are
particularly useful in coastal studies since their penetration of the water column
enables water quaUty (turbidity, suspended sediment concentration, chlorophyll,
etc.) to be measured, whilst infrared observations can be important in studying
the dispersal of outfall plumes, off-shore circulation, etc.
SAR cannot penetrate the water column, but nonetheless SAR imagery of
coastal waters promises to yield valuable information that would be difficult
or impossible to gain by direct sea observations. Indeed, prehminary results from
Seasat have already proved this, and have indicated a previously unexpected
potential to reveal complex bottom topography, as will be discussed further in
Chapter 12.

3.4 OCEAN WAVES

One branch of physical oceanography which is especially amenable to the use of
satelhte remote sensing is the study of ocean-surface waves. This is because it is
the surface of the sea which is viewed directly by satellite sensors, and micro-
wave instruments in particular are capable of measuring the roughness of the sea
surface in a variety of ways which will be discussed in detail in Chapters 10 to 13.
78 The possibilities for oceanography [Ch.3

Surface waves are found in the ocean with periods ranging from a few tenths
of a second to around twenty seconds. There are of course wavelike motions
occurring in the sea on much longer timescales of minutes, hours, or days, but
these are associated with the dynamical phenomena such as tides or mesoscale
eddies mentioned in the previous section. Within the range of surface waves it
is convenient to make the distinction between capillary waves at the short
millimetre wavelength end of the spectrum, and swell, which are waves of several
hundred metres length. The middle range of wavelengths between centimetres
and tens of metres are generally termed wind-waves. Surface-wave phenomena
have received considerable scientific study, and there are several useful texts
available on the subject. Kinsman (1965) provides a broad introduction from
first principles, Phillips (1977) presents the mathematical basis of the wave
interaction processes which are important in interpreting remotely-sensed
surface roughness data from active radar devices, whilst Le Blond & Mysak
(1978) offer a thorough survey of recent research work.
The mechanism of water-wave propagation can be modelled by a relatively
simple mathematical theory, if an assumption of small amplitude waves is made.
Being linear, this theory enables a sinusoidal wave profile to be assumed, since
any more complex profile can in principle be synthesised by the addition of
different wavelengths. Gravity is the controlling force which causes wave
propagation once the surface is displaced from its rest position. It can be shown
that the frequency o rads s“^ and the wavenumber k rads m“^ (wave oscillation
period T and wavelength X are given by
27r 2n
T = — and X = —
a k

are related by the dispersion relation

= gk, or (3.1)

for waves longer than a few centimetres, provided that the water depth is at least
half the wavelength. The phase speed c, at which troughs and crests travel, is
defined as a X

^ k T
1/2 1/2
g g\
(3.2)
_k_ _2tt_
Consequently the longer waves, with longer periods, travel faster than shorter
wavelengths, and dispersion of the waves occurs. In shallow water, where the
wavelength is 20 times the depth, the propagation is constrained by the depth
h, in which case
Sec. 3.4] Ocean waves 79

= gh (3.3)

and c = {gh)^l'^. (3.4)

In this case, all wavelengths travel at the same speed (nondispersive), but the
shallower the water the slower the waves, leading to refraction of waves over
shallow sandbars or sloping beaches. These waves, controlled by the presence of
the sea bed, are often referred to as ‘long gravity waves’.
Fig. 3.9 shows the relationship between the period and the wavelength of
surface gravity waves, and Fig. 3.10 the phase velocity variation with wavelength.
In each case the parabola represents the case of deep-water propagation, obeying
equations (3.1) and (3.2). The other curves correspond to the influence of the
sea bed at different depths. Where they become straight lines, the motion can be
considered to be that of long waves, obeying equations (3.3) and (3.4). The
phase speed, c, of (3.2) and Fig. 3.10 represents the speed at which troughs and
crests appear to travel normal to their own alignments, and as such is of some
importance to the imaging of surface waves by remote-sensing techniques.
However, it is a parameter of less fundamental significance dynamically than the
group velocity C. This is the velocity of energy or information propagation. In a
wavefield of gradually varying wavenumber or frequency it is the speed at which
a group of waves of the same frequency will travel. Within the group the troughs
and crests will normally travel at a different speed from the group itself because
in general the phase speed and the group velocity are different. It can be shown
that the group velocity is C = bo/dk. For surface gravity waves, therefore,

C =
1/2
, ,
0
1

whereas for the nondispersive long waves the phase and group velocity are the
same, C = c =
The practical significance of the group velocity is illustrated when consider¬
ing the rate at which wave energy radiates out from a storm centre. The longest
waves propagate away fastest, followed by the shorter waves, but at the group
velocity rather than the phase speed. The waves of 8 to 20 second period
are attenuated less by dissipation processes than the shorter waves, and can
propagate thousands of kilometres across the ocean as swell, characterised by
long, nearly-parallel crests of almost sinusoidal waveform. These are the waves
which are often revealed by imaging radars. The distance and time from the
storm which generated them have been shown to correspond exactly to the
group velocity of the given wavelength. An observer will note that the wave¬
length of swell from a particular storm will gradually decrease, since the later
waves must have a lower group velocity than the earlier ones.
80 The possibilities for oceanography [Ch.3

Fig. 3.9 - Variation of period with wavelength for surface gravity waves in water
of different depths

Fig. 3.10 - Variation of phase speed of gravity waves with wavelength, for
different depths, h, of water
Sec. 3.4] Ocean waves 81

The information in Figs 3.9 and 3.10 is not the complete story concerning
water waves. At short wavelengths surface tension contributes to the restoring
forces which control the wavelike propagation of a surface disturbance. At
wavelengths less than about 1 cm, surface tension dominates, and gravity effects
are negligible. The resulting wave is called a capillary wave and is characterised
by the dispersion relation = yk^ where y is the surface tension divided by
the water density. The corresponding phase and group velocities are c —
and C = 3l2(ykyi\
In this case the waves are dispersive, but the shorter waves travel faster than
the longer ones, and the group velocity is faster than the phase speed. The case
of ripples of a few centimetres wavelength is of particular interest to radar
remote sensing since this is often the scale which interacts strongly with active
radars having a similar electromagnetic wavelength. Expanding the region near
the origin of Fig. 3.10 results in a phase-speed versus wavelength curve of the
shape shown in Fig. 3.11. There is a minimum wave speed if

k = kc and thus Cml„ =

The actual values of Cmin \c~ depend on the surface tension,which

is controlled by the presence of surface impurities, thin oil films, or monolayers
of organic material, but for pure water at 20° C, 7 = 74 X 10”® m^ s”^,
Cmin = 0-23 ms”* and Xf. = 17 mm. Thus ripples with capillary wave charac¬
teristics occur at length scales of a few millimetres. Surface tension is still
important at wavelengths of 1 to 2 cm, but above a few centimetres length the
surface tension can be ignored.

T5
O
O
0.
CO
a>
0)
(0
x;
0.

I'ig. 3.11 - Variation of phase speed with wavelength for short wavelength
(capillary-gravity) waves
82 The possibilities for oceanography [Ch. 3

Although the water viscosity has little influence over water waves, other
dissipative processes, including surface tension effects and nonlinear wave
interactions, cause short wavelengths to be damped out rapidly. Capillary waves
are therefore short-lived, but being generated by wind stress, they significantly
roughen the surface as ‘felt’ by the wind, and have a considerable influence on
the transfer of momentum between the wind and the waves. The mechanisms
by which waves are generated by the wind are still not fully understood, and
it may well be that the development of radar remote-sensing techniques is able
to offer new ways of studying the problem. A gentle breeze of less than 5 m s“^
has very little effect on the sea surface except to roughen the surface with
ripples. A fresh breeze of 10 m s"^ is able to produce quite a rough sea, and
higher winds produce correspondingly rougher seas. In this situation, the waves
do not have a regular sinusoidal shape with a given wavelength and amplitude.
Instead, the fluctuations of wave-height are random, and an aroused sea must be
described by its energy spectrum.
Fig. 3.12 illustrates typical wave-energy spectra appropriate to different
wind conditions. The energy spectrum represents the amount of wave energy
present in a small bandwidth of frequencies, and it is obtained by spectral
analysis of a time series of surface-height measurements at a point. One way
of interpreting it is to imagine the waves in the aroused sea to be dispersed into
the component individual sinusoidal waves of different wavelengths, as the wave
field propagates away from the storm area. The energy of waves at a particular
wavenumber/frequency will correspond to the energy indicated in the spectrum
at that frequency. The amplitude of the sinusoidal waves is proportional to the
square root of the energy. Fig. 3.12 indicates clearly how higher winds are able
to produce more wave energy. The spectral distribution of this energy is worth
noting. At the high frequency end corresponding to the shortest ripples, the
wavefield is quickly saturated and is unable to contain more energy as the wind
rises. Instead, the energy peak moves to lower frequencies corresponding to
larger waves, and it takes a strong storm to generate the longest swell waves.
Fig. 3.12 can be used as the basis of wave-prediction algorithms. It corres¬
ponds to a fully-aroused sea, and only appHes if the given wind strength has been
blowing for a sufficiently long time (the duration), and over a sufficient length
of sea (the fetch) upwind of the observing position. If not, the wave conditions
are said to be duration-Hmited or fetch-limited. Whilst the wave-energy spectrum
contains much information about the wavefield, it has no information about the
direction of propagation of the different wave components. It is very difficult to
measure the directional wave-energy spectrum (that is, the distribution of energy
in different wavenumbers in different directions) by conventional methods.
The abihty of satellite-borne radars to sample the wavefield in two horizontal
directions will therefore make a significant contribution to studying wave
directionahty. A wave spectrum requires a certain amount of interpretation, and
a much simpler measure of surface roughness is given by the significant wave
Sec. 3.4] Ocean waves 83

Fig. 3.12 — Typical surface wave amplitude spectra at two wind speeds (adapted
from Hasselmann et al. (1973))

height. It is defined as the average of the highest one-third of the ‘wave height’
measured between adjacent lower and upper turning points on a time series
record of surface height. It is therefore given the symbol and has proved to
be a useful oceanographic parameter because it corresponds quite closely to the
visual estimates of wave height made in weather reports from ships at sea. It can
be related to the wave-energy spectrum, and it is therefore possible to construct
hij-i wave-height prediction curves such as that shown in Fig. 3.13. The construc¬
tion and improvement of such predictions which tend to be location-specific
provides significant help to the shipping and offshore engineering industries. One
problem for remote-sensing techniques is to make sure that even though the
radars may be measuring roughness at length scales corresponding to the high-
frequency end of the spectrum, they are able to accurately estimate hxj'^ which
corresponds principally to waves at frequencies near to the spectral peak. These
and many other aspects of radar remote sensing of surface waves will be dealt
with further in Chapters 10 to 13.
84 The possibilities for oceanography [Ch. 3

WIND

PERIOD. 8

Fig. 3.13 - Wave-prediction diagram showing the significant wave height as a

function of dominant wave period for different wind speeds. The curves are
traced from the right until either the fetch or duration limit is reached, or until
the curve levels out. For example, a wind of 10 ms“‘ with unlimited fetch and
duration is predicted to generate 11 s waves of about 2 m height (adapted from
Pond & Pickard (1978))

3.5 SPACE AND TIME SCALES OF OCEANOGRAPHIC PHENOMENA,

IN RELATION TO SATELLITE SAMPLING
It is obvious, when pointed out, but not always remembered, that in any observa¬
tional science, the length and time scales of the observations must match those
of the phenomenon under investigation. We must therefore consider whether
the length and time scales of ocean phenomena are capable of being sampled
adequately by the techniques of remote sensing from space.
As the setting for the dynamical phenomena. Table 3.2 reminds us of the
typical dimensions of the ocean basins, the continental margins of the ocean,
the semi-enclosed continental shelf seas, and coastal embayments/estuaries. It is
clear, but worth emphasising, that the horizontal/vertical ratio of length scales
is generally around 1000:1, similar in aspect ratio to a puddle in the road.
Therefore, whilst the sea traveller is often most conscious of the local vertical
motion of the sea due to waves, virtually aU the kinetic energy of the oceans is.
Sec. 3.5] Space and time scales of oceanographic phenomena 85

in fact, contained in horizontal rather than vertical motion, whilst because of

stable stratification of the sea, vertical length scales of variability are about
1/100 — 1/1000 those of horizontal variability.

Table 3.2 Typical topographic dimensions in oceanography

Horizontal extent Depth

Ocean basins 5000 km 5 km

Continental margins 50 km 200 m
Shelf and enclosed seas 100 — 500 km 50-100 m
Embayments/estuaries 1—20 km 5 — 30 m

The approximate length and time scales of various dynamical ocean

phenomena (excluding surface waves) are shown in Table 3.3 with an indication
of the type of remote-sensing signature which the feature is likely to display.
The vertical amplitude column indicates the variations of sea-surface height
which may accompany the feature — important in considering the potential of a
satellite altimeter to detect it. Nondynamical processes are not indicated in the
table, and it should be remembered that while passive tracers will have the same
scales as the controlling dynamics, non-conservative tracers will in addition be
governed to some extent by the processes of physics (for example, heating),
chemistry or biology (for example, growth rates, population dynamics) which
control the tracer. Where appropriate they will be considered individually in
later chapters.
Table 3.4 shows that waves on the ocean surface are characterised by a
variety of scales. As well as the vertical amplitude of waves and their individual
wavelengths and periods of oscillation, it is necessary to consider also the length
and time scales of the wavefield, that is, the length and time scales over which
the wave amplitudes and individual wave lengths and frequencies vary. There
are in fact no spectral gaps between capillary waves, wind waves, and swell, and
so the divisions in Table 3.4 are somewhat arbitrary but necessary to illustrate
the broad range of scales. It should be noted, too, that maximum amplitudes
are related to wavelength, since there is a physical stability limit on the wave
steepness (that is, height/wavelength) of around 1/7, above which the waves will
break. The length and time scales for swell are related to the order of magnitude
of travel distances across the ocean. The length and time scales given for wind
waves are related to their generation. Variation of wind field, and fetch limitation
near coastlines, will give variability of amplitude over distances of 100 km, and
the wavefield may change significantly in character within an hour, as the wind
speed changes.
86 The possibilities for oceanography [Ch. 3

Table 3.3 Length and time scales of dynamical ocean phenomena

Typical Possible
Surface horizontal Horizontal remote-sensing
displacement water length surface
amplitude velocity scale Time scale signature of the
m m s"‘ km phenomenon

Equatorial currents 0.3 0.01 5000 months to years surface slope

Large ocean gyres 0.5 0.01 3000 one to many surface slope
years colour

Western boundary 1.5 1.0 100 days to years surface slope

currents (e.g. colour
Gulf stream) temperature
surface roughness
Eastern boundary 0.3 0.1 100 days to years temperature
currents colour
surface slope
Rings (e.g. from 1.0 0.1 100 weeks to years'
surface slope
Gulf stream
colour
meanders)
temperature
Mesoscale eddies 0.25 100 100 days
Ocean fronts 0.05 10 10 days colour
temperature
surface roughness
Ocean tides 1.00 0.10 1000 12 hours-1 year surface slope
Shelf sea tides 5.0 1.0 100 4 hrs-1 year and height

Internal waves 0.10 cm 0.1-100 secs —hours surface roughness

(colour?)
Storm surges 1.0 1.0 100 hours-days surface slope
Tsunami 0.1-1.0 1-100 mins-hours 9

Table 3.4 Length and time scales of ocean surface waves

Wind waves Swell Capillary waves

Typical maximum height

from trough to crest 5 mm to 10 m 10 to 20 m a few mm
Wavelength 0.05 to 100 m 100 to 500 m 1 to 50 mm
Periods 1 to 10 s 10 to 20 s <1 s
Variability length scale 100 km 1000 km 10m
Variability time scale 1 hr 10 hr 10 s
Sec. 3.5] Space and time scales of oceanographic phenomena 87

In the case of lower-frequency waves of several seconds period, propagating

into an area after being generated elsewhere, spatial variability of the wavefield
can be caused by refraction of waves due to interaction with ocean currents,
or with shallow-sea topography. This can take place over length scales one order
of magnitude greater than the wavelength. It should be noted, too, that the
variability of the capillary wavefield is related to the scale of individual wind
waves.
Processes of interest in the ocean span a horizontal length scale range of
nine decades from 1 mm to 1000 km, and a time scale range of six decades from
seconds to years. Observation of such wide ranges must necessarily employ a
variety of experimental strategies, none of which is likely to cover more than
a couple of decades in one dimension. As Woods (1977) has pointed out, the
spectral window through which information is obtained should correspond to
the spectral window containing the phenomenon being studied. Thus, for
example, to study the oceanic gyres requires sampling in time every month or so
to obtain yearly mean values, at stations spaced 1000 km apart, whereas to study
Gulf stream meandering requires sampling at 10 km intervals every day. It is
always necessary to sample frequently enough in space and time to enable the
averaging out (without aliasing) of phenomena at smaller scales than that of
interest, whilst the samphng span must encompass the scales being studied.
Waves are particularly difficult to sample since it is necessary to resolve the
length and time scales of the wave motion, requiring fairly frequent sampling.
On the other hand, if the evolution and variation of a wavefield is being studied,
the sample span may have to be several decades larger than the sampling interval.
It is obvious from the nature of the satellite remote-sensing process that
there is no possibility at present of measuring any vertical variability in the
water column except to a very limited extent by multispectral visible wave¬
length sensors, in which different frequencies penetrate to different depths. In
this respect, there is a serious gap in the observational window of the satellite’s
samphng capability, compared with ship-sampling techniques. Nonetheless, it
would be foohsh to reject remote sensing as a technique for this reason, when
it can offer more impressive samphng capabilities in other respects.
Observations at sea generally find it easier to resolve and to span time rather
than space. Satellite observations, on the other hand, can give reasonable spatial
resolution (from 30 m upwards) and broad areal coverage. The synoptic length
span of a scanning instrument may vary from 100 km up to several thousand
kilometres, whilst a polar-orbiting satellite achieves global coverage in a few
hours. Time resolution is usually poor (no better than four passes a day, and
often much worse), but the time span is limited only by the life of the satellite
or the sensors, potentially several years. If there is a known or expected
frequency/wavelength relationship as for surface waves, it may not matter for
many applications that the time domain is not adequately resolved, since
occasional synoptic views of the wavefield can give much valuable information.
88 The possibilities for oceanography [Ch.3]

If the phenomenon is phase-locked to a known forcing frequency (as is the case

with tides), then, provided that the sampling interval is not at a tidal frequency
and does not cause aliasing, it will be possible by harmonic analysis to resolve
the phenomenon given a long enough record span, even if the sampling interval is
much longer than the fundamental period of the tides.
However, for many applications it is impossible to cover the whole spectral
window necessary to contain a particular dynamical process. In these cases
it may weU be possible to combine the time and depth sampling capabihty of
recording instruments at sea, with the occasional detailed synoptic view from a
satellite, to yield a powerful combined research tool. Such developments require
collaboration between the oceanographer and the space scientist, and may
require some rethinking by the oceanographer of the categories in which he
presents his hypotheses for testing against observations. Much oceanographic
theory is concerned more with temporal rather than spatial variability, since that
has been more easily measured. Now that synoptic views of high spatial resolu¬
tion are available, which could never be possible from sea-based measurements,
there is considerable scope for ocean science to develop as new hypotheses
concerning spatial aspects of the dynamics are put forward to be critically
examined against the new satellite data.
 CHAPTER 4

Principles of remote sensing of the sea

4.1 INTRODUCTION
In this chapter we take a brief look at the scientific methodology which is
necessary if satellite remote sensing is to be used for making valid measurements
of oceanographic parameters. Many other books on remote sensing deal with the
topic of fundamental principles in some detail (for example, Barrett & Curtiss
(1976), Lillesand & Kiefer (1979), Reeves (1975), Sabins (1978)), but they
are mostly concerned with land applications and have a particular bias towards
visible and near-visible wavelength imagery. Our concern here is to point out
the ways in which the methodology of remote sensing of the sea differs from
the quite well established approach developed for land applications. Some tasks
are easier over sea, whilst others are much more difficult.
Bearing in mind the point of view of the oceanographer who is sceptical of
remote sensing ever producing much more than a ‘pretty picture’ of the ocean,
this chapter introduces the supporting measurements and processing tasks which
must accompany satellite-derived data if useful quantitative information about
ocean parameters is to be derived. When presented with a clear and colour¬
ful computer-produced image based on satellite data, there is a tendency for
the remote-sensing enthusiast to jump immediately to unjustified conclusions,
interpreting as oceanographic features patterns which may be artefacts of the
sensing mechanism, the atmospheric interference, or the processing technique.
Conversely, the cautious experimental oceanographer may declare that any
apparently interesting features are nothing more than such artefacts. The true
‘satellite oceanographer’, if we may use the title, follows a careful path between
these extremes. On the one hand he must be sceptical and avoid reaching con¬
clusions which are not justified, but at the same time he is motivated by the
expectation that the remotely-sensed data received by the satellite contain
at least some information about the sea, even if corrupted by various sources
of error. He will also be searching particularly for the type of information which
conventional measurement techniques cannot observe. The scientific approach in
90 Principles of remote sensing of the sea [Ch.4

these circumstances is to be as objective as possible, incorporating into the data

processing corrections and checks which are capable of validation independently
of the particular data set under analysis.
The problem can be viewed in four separate stages, although very often
they overlap and interact with one another. These stages are sensor calibration,
atmospheric correction, positional registration, and geophysical calibration.
These four aspects must be included in the analysis of all satellite data, whether
from an imaging or non-imaging sensor, and for sensors operating in all parts
of the electromagnetic spectrum. However, the approach to the correction/
calibration problem varies greatly with different types of sensors, as will be
discussed in the applications chapters. Here we briefly explain the elements of
each of these stages of satellite data interpretation, particularly apphed to
oceanography.

4.2 SENSOR CALIBRATION

The sensor calibration stage of data processing is intended here to account for
all aspects of information flow from the environmental signal entering the sensor
to the receipt of an analogue or digital data signal at the ground. The applications
scientist requires a sensor calibration model which will determine the radiance,
phase, or colour etc. of the radiation entering the satellite sensor, given the
analogue or digital value transmitted by the satelHte and received by the ground
station. This includes the conversion of the received radiation into an electrical
response (normally a voltage amplitude or frequency), the conversion of that
response into a digital value, with consequent digitisation error, and the trans¬
mission of the information through a suitable carrier frequency to the ground
station, where it is electonically retrieved.
The sea-going oceanographer is famihar with the problems encountered
in placing environmental monitoring sensors in hostile enviroments and tele¬
metering observations to a base station. The equipment which operates perfectly
in the laboratory may inexplicably fail when suspended from 1000 m of wire
beneath a research ship! The equipment is therefore tested and caHbrated before
and after deployment. Whilst it is never possible to be certain that deployment
at depth does not introduce spurious calibration errors, and only independent
checking by a different type of sensor can confirm this, it is always reassuring
to be able to recahbrate after use and thus estabUsh if any serious instrumental
drift exists. This can be incorporated as necessary into any calibration algorithm.
The satellite sensor, once it has reached its orbit through the stresses of launch,
is in a relatively kind environment. It is subject to stresses of radiation, but little
mechanical stress. However, the high vacuum of outer space is not as yet a
familiar environment for which to design deUcate equipment. Moreover, the
power limitations on board the satelHte impose severe design constraints, and
there is inevitably a gradual deterioration in the power supply on the sateUite.
Sec. 4.2] Sensor calibration 91

A gradual degradation in sensor response, and a corresponding drift in the

calibration, is therefore to be expected eventually, although one hopes that it
will be very small during the design life (say one to two years) of a satellite. In
comparison with ship-deployed equipment, the problem arises that there is no
means of retrieving the instrument for periodic recalibration in the laboratory,
and therefore it is difficult to construct a model of calibration drift errors. How
then can error in the interpretation of remotely-sensed data arising from sensor
calibration inaccuracies be minimised?
The data user is very much dependent at this stage on the space technologist
to supply a well-engineered sensor and electronic package, to perform a careful
pre-flight calibration, to build in as many in-flight calibration procedures as
possible, and to publish the details of the calibration algorithms in readily
accessible documentation. Communication between the space technologist and
the user is essential, but the process is two-way, and requires a measure of trust
on the part of the environmental scientist, who must simply accept the calibra¬
tion data as given to him. This can come hard to that breed of oceanographer
who has built his reputation for careful science on a poUcy of only beUeving
those ship observations which he has personally supervised at sea! However,
as oceanographers become more involved with the concept and design stage of
a new generation of sensors, it is evident that cooperation and confidence in the
value of the data is increasing.
The voltage or frequency response of the sensor to the incoming radiation
is known from the pre-flight calibration. With certain sensors it is possible to
update the knowledge of this response by an in-flight sensor calibration. Thus on
some scanners, part of the scan will view a reference target, a lamp of known
brightness for the visible wavelength scanners, or a black body of measured
temperature for thermal IR sensors. In this way gradual drift of the sensor can
be detected and corrections made in the data analysis. Of course, this refers
the problem of accuracy back to the reUability of the on-board standard. Is the
target at the temperature indicated by the thermometer, or is there a thermal
gradient across the black body? In the case of visible sensors, is the apparent
drift due to sensor degradation or caused by the reference light wearing out?
Unfortunately it is very difficult to determine conclusively whether in-flight
sensor caUbration drift is real or not, without the opportunity of retrieving the
satellite for post-flight examination and calibration. Perhaps one day this may
be possible with the USA’s space shuttle. Of course, if the in-flight calibration
remains steady it is reasonable to assume that it genuinely is steady, particularly
if it is in agreement with the pre-flight calibration. In this case the quantitative
accuracy of the data can be relied upon. Where apparent drift occurs, quanti¬
tative accuracy depends upon the ability to compare satellite observations
with airborne or near-ground measurements of the same parameter (for example,
upwelhng radiance), but this can never be precise, because of the need to make
atmospheric corrections. In the absence of any synoptic ground truth, the best
 Principles of remote sensing of the sea [Ch.4
92

that can be done is to use the most recent calibration (in-flight or pre-flight).
Even if this is slightly in error, it will be an absolute error, applying equally to
the whole scene of a scanned image. The relative accuracy, that is, when com¬
paring the values of two points on the same scene, will not be affected by slow
sensor drift, but will depend on the resolution sensitivity of the sensor. Sensor
drift creates worse problems when comparison is made between two images of
the same sea area received several months apart.
The conversion of the sensor output into a digital value for transmitting to
earth is generally a reliable procedure with modern microelectronics, and can be
monitored by running a standard voltage ramp through the system periodically.
Once in digital form, the signal can be transmitted with high accuracy, and
effectively no further errors need be introduced into the digital data. It is
possible to lose sensor radiometric resolution in the digitisation processes. If the
transmission is hmited to 8-bit numbers, then a range of only 256 digitisation
levels is permitted. Ideally this should be arranged so that a digitisation interval
corresponds to the sensor sensitivity at each incident radiation level. If the
digitisation interval is too large, then the true sensitivity of which the sensor is
capable is not reflected in the digital data. If the digitisation interval is smaller
than the ■ sensitivity, then digital levels are being wasted in communicating an
apparent radiometric resolution which is not present in the sensor output.
Furthermore, users presented with such data may try to interpret variability
of the integer values which is purely an artefact of the digitising circuits, and
powerful image processors may be made to enhance patterns which have no
oceanographic significance whatsoever. Care should always be taken, therefore,
not only to check the calibration of the sensor but also its sensitivity in relation
to the digitisation interval present in the data.

4.3 ATMOSPHERIC CORRECTION

If the primary advantage of satellite remote sensing is the wide synoptic view
it gives us of the sea surface, the principal disadvantage is that we must look
through another medium — the atmosphere — to see the ocean. As shown in
Fig. 2.13, the atmosphere is opaque to electromagnetic radiation at many
wavelengths, and there are only certain wavelength windows through which
radiation may be fully or partially transmitted. The atmospheric gas molecules
themselves may absorb or scatter the radiation, and in addition water vapour,
aerosols, and suspended particles of dust will do the same. If water droplets are
present in the form of clouds, they may completely change the transmission
properties of the atmosphere.
Fig. 4.1 summarises the processes of interaction that can take place between
rays of electromagnetic energy and the atmosphere, as they affect what is
observed in the field of view of a satellite sensor. Fig. 4.1 is vahd irrespective of
the source and wavelength of the electromagnetic radiation. Rayl represents the
Sec. 4.3] Atmospheric correction 93

useful signal. Radiation leaving the sea surface reaches the sensor, containing
information about the sea which is of value to the oceanographer. Ray 2
represents the radiation leaving the sea which is absorbed by the atmosphere
en route, whilst ray 3 is that which is scattered by the atmosphere out of the
sensor field of vision. The sum of 1,2, and 3 represents that energy which should
be received by the sensor if a complete correlation is to be made between the
received signal and the ocean environmental parameter under investigation. The
absence of 2 and 3 reaching the satellite must therefore be allowed for in some
form of atmospheric correction to the received data. In addition, rays 4,5, and
6 reach the sensor without having left the sea surface in the field of view, and
therefore constitute extraneous ‘noise’ on top of the signal, so far as the ocean
scientist is concerned. Ray 4 is that energy emitted by the constituents of the
atmosphere. Ray 5 is energy reflected by scattering into the field of vision of
the sensor, and ray 6 is a special case of scattering into the field of vision —
energy which has previously left the sea surface but from outside the field of
view. Ray 1 may of course represent energy which has been forward scattered,
provided it originally left the sea surface within the field of view.

Fig. 4.1 - Atmospheric pathways of e.m. radiation between the sea and the
satellite sensor
94 Principles of remote sensing of the sea [Ch.4

These possible atmospheric effects cover all types of sensor, active or

passive, visible, thermal, IR, and microwave. Not all the effects are present in
each case. For example, '•ay 4 is not found in visible and near-IR wavelengths,
where there is no emission by the atmosphere, but is relevant to thermal IR
when the cool atmosphere absorbs radiation (ray 2) and re-emits it with lower
temperature characteristics (ray 4). In the case of active sensors, the noise may
be due to energy from extraneous sources being scattered into the field of view
by either the atmosphere or the sea itself, or by the emitted pulse being back-
scattered by the atmosphere instead of the sea. By categorising different rays as
signal or noise in this way, we are taking the point of view of the oceanographer
seeking to observe the sea surface. One man’s noise is another man’s signal, and,
for the meteorologist or atmospheric physicist, it is the scattering and emissions
by the atmosphere which create the signal and the sea reflections and emissions
which constitute noise. Sometimes atmospheric effects contain useful informa¬
tion for the oceanographer too, in providing a means for applying an atmospheric
correction to another sensor, as discussed later in this section.
The detailed efforts made to eliminate atmospheric effects will be discussed
in the context of different sensors and appHcations in later chapters. It is worth
mentioning briefly here the broad strategies that can be adopted in atmospheric
correction of satellite data used for earth observation. One approach used for
visible scanners is to make no separate attempt at an atmospheric correction, but
to cahbrate each scene with ground data, incorporating both sensor and atmos¬
pheric effects into a calibration valid for that image only. For land applications
this can be a most satisfactory approach, but for sea applications it is difficult to
obtain synoptic ground truth, as discussed later in this chapter, and the approach
is unsatisfactory. Because atmospheric effects vary significantly in space and
time it is not advisable to use a calibration incorporating the atmospheric
correction for a scene other than the one on which it is based.
Sea applications, on the other hand, sometimes lend themselves more
readily to a multispectrally-based atmospheric correction than do land applica¬
tions. In visible and near-lR wavebands, for example, it is often possible
to assume that certain near-lR channels are unlikely to have any upwelling
radiation from the sea, and what is recorded by the sensor must be due to
atmospheric effects. This enables an estimate of the atmospheric effect to be
made not only over the whole scene, if it is a scanned image, but individually
for each pixel. This is particularly valuable when the length scale of atmospheric
variability may be comparable with length scales of ocean parameter variability,
precluding any distinction between the two being made at the interpretation
stage. For this type of multispectral atmospheric correction to be possible, a
model of the spectral dependence of atmospheric transmission is required, and
this requires assumptions or observations concerning the optical properties of
the atmosphere.
The above two approaches make a unique atmospheric adjustment either
Sec. 4.3] Atmospheric correction 95

for each observation, or for each complete image scene. It is simpler and less
time-consuming to construct a universal atmospheric correction based on an
average model of atmospheric effects. This is the case with the routine analysis
of sea-surface temperature data from the AVHRR on the NOAA satellites. This
has the effect of removing the greater part of the atmospheric error, but cannot
allow for spatial and temporal variability in atmospheric aerosol and water
vapour composition, resulting in a level of uncertainty which can often be
tolerated provided its hmiting magnitude is known. Such an approach depends
on a broad database of meteorological observations from which to construct an
average atmosphere, and is invalidated when an exceptional atmospheric condi¬
tion is encountered. Such was the case in 1982 when the Mexican El Quichon
volcano produced unusually high levels of stratospheric dust in a low-latitude
belt around the world.
A fourth strategy, not unhke the second, multispectral, approach, is to
mount an atmospheric sounding sensor on the same satellite as an oceanographic
sensor for which an accurate atmospheric correction is required. A microwave
sounder is able to resolve separate components within the atmosphere and thus
provide a means of correcting an IR radiometer for thermal IR absorption
and emission in the atmosphere. It is also proposed that the microwave channels
of the Along Track Scanning Radiometer (ATSR) to be flown on ERS-1 will
provide a means for modelling more accurately the atmospheric corrections to
be apphed to the altimeter on ERS-1. The problem in this case differs from
those discussed in the context of Fig. 4.1, since the interpretation of altimeter
data requires a knowledge of the actual speed of microwave radiation through a
variable atmosphere.
The ATSR will be described in Chapter 14, but its concept is an example
of another possible atmospheric correction strategy. Fig. 4.2 illustrates the way
in which an oblique view of the earth’s surface necessitates looking through a
much longer path length of atmosphere than for nadir viewing. This of course
must be taken into account when making atmospheric corrections of images
scanned by large-swath-width scanners such as the CZCS or the AVHRR. This
extra problem has been turned to good effect in the ATSR design. By viewing

Fig. 4.2 — Increased atmospheric pathlength encountered by oblique viewing

96 Principles of remote sensing of the sea [Ch.4

the same piece of sea twice, through different lengths of atmosphere, an

objective estimate of the atmospheric effect can be made, provided that the
sea conditions have not changed significantly between the two observations.
The importance of having an accurate atmospheric correction depends on
the use being made of the remotely-sensed data. If a continuously variable
ocean property is to be accurately measured from space, and compared with
similar measurements made at different times and places, then accurate atmos¬
pheric correction is vital. On the other hand, if shapes or patterns, that is, spatial
measurements, are more important than parameter values (based on radiometric
data), then provided that the atmosphere does not mask or introduce ambiguities
into the shape of, say, an ocean front, or the line of the edge of sea ice, it is less
important to make atmospheric corrections, and may not even be necessary.

4.4 POSITIONAL REGISTRATION

By positional registration we mean the identification on a map of the place to
which a remotely-sensed measurement refers, that is, the location to which the
sensor was pointing when it recorded the measurement. In some texts and papers
the problem may be referred to as navigation, or location. The adjustment of an
image to bring it into conformation with a map base is sometimes called the
navigation or the rectification of the image.
Fundamentally, the problem is one of knowing where the satellite was when
a measurement was made, and in which direction the sensor was pointing, but
the accuracy with which this information is required will depend on the type of
sensor and the application. For single measurements with a downward-pointing
sensor providing an area-integrated measurement of, for example, microwave
radiation or significant wave height over a 100 km X 100 km square, precision
is not going to be very important, and a few kilometres’ error in locating
the nadir point will make no difference to the interpretation of such a large
area average value. In these cases it is sufficient to know the expected satellite
orbit and to assume the sensor was pointing downwards.
For other non-imaging sensors, such as the altimeter or scatterometer, a
more precise positional registration is required, involving ground-station tracking
of the satellite and accurate on-board monitoring and transmission of the
satelhte vehicle’s pitch, roll, and yaw. From the user’s point of view it is
valuable for this information from diverse sources to be brought together and
published with the sensor measurement. A satellite carrying an altimeter is
bound to provide this information since an accurate knowledge of the satellite’s
position and the direction in which the altimeter points is essential to the
functioning of the altimeter.
This is not always the case with other satelUtes, and imaging sensors may
have to rely on different techniques for positional registration. The usual
approach for land remote sensing is to identify recognisable points on the image
Sec. 4.4] Positional registration 97

known as ‘ground control points’. For a scanned image, it is not sufficient to

register just one point on the image, which may be rotated and distorted
as well as translated laterally from its theoretically expected position based
on the predicted satellite orbit. In land applications it is customary to use as
many as several hundred ground control points for a single Landsat image.
A generalised warping function is applied to the whole image to bring it onto
a map grid base (see Chapter 5), and the accuracy of registration of each pixel
depends on the accuracy of identification of ground control points as well as
on having a uniform distribution of them throughout the scene. Having located
some clear and unambiguous ground control points on one scene, it is possible
to register automatically other images of the same area, through the use of
pattern-recognition software which enables the computer to find the points on
further images.
Fig. 4.3 shows schematically the sort of distortion to be expected from
a wide-angle view of the curved earth. The larger the field of view the greater
the distortion. With a large image it is also necessary to define carefully the
type of map projection onto which it is to be registered. Over the sea, it is not
possible to do this by using ground control points, as over land, because there
are unlikely to be sufficient, if any, features which can be identified. Coasthnes,
prominent headlands, and small islands are most suitable for use as control
points, but there may be none on an image, or they may be unevenly spaced.
Under these circumstances the most satisfactory approach is to make use of the
orbital and sensor geometry to construct a theoretical rectification transforma¬
tion. An example of this is given in section 5.4. The resulting image should
thus have the major earth curvature and oblique viewing distortions removed,
although the expected inaccuracies in predicting the satellite orbit will lead to
small errors in locating latitude and longitude lines on the rectified image.
If there are a few ground points available on the image, these can now be used
to control a simple translational or rotational adjustment to fix the image onto
a map base. If the image is open sea containing no coastlines, then a large
positional error must be tolerated. This may be acceptable if spatial distributions
of ocean parameters are being measured on that image only. On the other hand,
if a time sequence of images is being used (for example, to plot the seasonal
change of shape and position of a front or an eddy, or to measure the speed of a
frontal instability wave), it is in the nature of the problem that each image must
be accurately located with respect to others. The same is true when images from
different sensors are being compared (for example, to examine the relative
positions of the colour and the thermal signature of a front or ocean eddy).
Sensors such as AVHRR or CZCS provide such a wide swath that it is unusual
not to find some identifiable coastal feature somewhere on the image provided
that the cloud cover is not too dense. This ground control point may be remote
from the area under study which would normally be selected from the image
and examined at high resolution. If accurate position fixing is required, it
98 Principles of remote sensing of the sea [Ch.4

is necessary to adopt a strategy of geometrically correcting the whole image

using (a) an analytical warping onto map coordinates using the nominal orbit
parameters, and then (b) adjusting it with the available ground control points.
Only then is it appropriate to select a region for more detailed study. For similar
reasons the use of Landsat MSS or TM data for oceanography is appropriate
only in coastal and estuarine areas where positional registration from coastal
landmarks is possible. After all, the most detailed image of an interesting oceano¬
graphic feature is of very little value if there is no way of locating its position
accurately, just as oceanographic data from observations at sea are of little value
if not accompanied by positional coordinates.

Fig. 4.3 - Distortion due to oblique viewing of the curved earth surface, (a) The
projection of the square IFOV on the earth’s surface, (b) The distortion of
a square on the ground as viewed from space, plotted in rectangular pixel
coordinates without allowing for oblique viewing and curvature effects
 Sec. 4.5] Geophysical calibration 99

4.5 GEOPHYSICAL CALIBRATION

The terminology ‘geophysical calibration’ is intended to include all aspects of the
process of assigning an oceanographic parameter value to a given value of the
reflected or emitted e.m. radiation as recorded by the sensor, after sensor
calibration and atmospheric correction have been performed. This is the goal
of quantitative remote sensing. Indeed, in the case of operational satellites,
the aim of the space agencies is to make the calibrated (geophysical) data the
product which is offered by the satellite management authority to the user.
Whilst this approach will make remotely-sensed data more accessible to a wider
range of users in commercial applications, there is a danger that the calibrated
data on offer will not always be of sufficiently high accuracy for use by oceano¬
graphers in scientific pursuits. However sophisticated the commercial service
offered by operational satellites, there will always be value in the oceanographic
scientist being able to receive the raw sensor data and thus make a separate
atmospheric correction and cahbration as a check on the performance of the
operational algorithms. With the present generation of satellites, separate
calibration is still necessary since most satellite data is only available in the form
of the basic sensor response.
As in the cahbration of all scientific measuring equipment, so the geo¬
physical calibration of remote sensors requires a combination of empirical and
theoretical methods. It is customary to start the construction of a calibration
model by turning to a theoretical analysis of the physical processes by which
an oceanographic parameter influences the electromagnetic radiation observed
by the sensor. In the case of emitted thermal infrared radiation, for example, the
theoretical model of the radiation emission from black bodies goes a long way
towards providing a complete calibration algorithm. Visible wavelength remote
sensing of suspended particles in water is less amenable to theoretical analysis.
Whilst quite a lot is known about the optical processes in the ocean and
atmosphere they are so complex as to make it impossible to produce from first
principles a calibration of radiance data in terms of suspended sediment load.
In the case of active radar sensors there is not yet a clear consensus concerning
the physical processes which control the radar backscatter from rough surfaces.
Thus, for example, little ‘first-principle’ theoretical input can be used for the
calibration of the scatterometer in terms of surface-wind stress.
AU calibration models require an empirical input to a lesser or greater
degree, ranging from merely fine-tuning a sound theoretical model, to construct¬
ing a crude model based not on any physics but on the regression fit of a satellite
data set to a synoptic set of ocean parameters measured at sea. The advantage of
the former approach is that a sound theoretical base gives us confidence to apply
the cahbration algorithm to circumstances and satellite data sets other than the
particular image which was calibrated against ground truth. The latter approach
gives us a workable calibration which can be used with known confidence limits
on the dataset which has been matched to ground truth, but it is likely to be less
100 Principles of remote sensing of the sea [Ch.4

accurate in universal application to other satellite data. The relevance of this

discussion will become apparent as the different approaches to the different
sensors with different applications are presented in later chapters. Here we
establish the principle that it is useful to have a theoretical basis for a calibra¬
tion algorithm, but a synoptic matching set of satellite and ocean data is essential
if any confidence is to be placed in a calibration. Because of the necessary
dependence of the calibration process on the availability of relevant ‘sea
truth’, it is important that this should be collected at sea and interpreted by
oceanographers who better understand the oceanographic processes which are
controlling it. Thus, for example, in the construction of sea-surface temperature
(SST) algorithms for infrared radiation sensors, there is a need to understand
that it is the surface skin temperature which should properly be matched to
the space data, and not the bulk sea temperature of the top metre which oceano¬
graphers traditionally measure as SST. Such problems of detail highhght the vital
need for oceanographers and space-sensor scientists to work together at this
stage of producing geophysical calibration algorithms.

4.6 OCEANOGRAPHIC SAMPLING FOR ‘SEA TRUTH’

In the previous section it was shown to be essential to measure at sea those
parameters we hope to deduce eventually from the satellite observations. In this
section some aspects of sea truth collection are considered, in particular those
which differ significantly from the obtaining of ground truth for overland
satellite data interpretation.
The major difference which must be stressed is that in general the remotely-
sensed characteristics of the sea change on a much shorter time scale than those
of the land. This means that whereas ground observations made within days,
weeks, or even years of a satellite overpass may be valid for interpretation and
calibration purposes, sea truth maybe useless unless it is collected simultaneously
with the satellite overpass. The governing factor is the time scale of the control¬
ling process. Geological appHcations on land have a time scale of years, centuries,
or more, whilst in land use appUcation the vegetation is unlikely to change much
over a week, unless a harvest intervenes. In the sea, by contrast, if individual
swell waves are being imaged, they have a time scale of the order of seconds. If
the gross statistics of a wavefield are being studied, these may take of the order
of an hour to change. The time scale of variability is also linked to the spatial
resolution. Studies of ocean colour in estuaries using Landsat with 80 m resolu¬
tion ideally require sea truth within a few minutes of the overpass, since tidal
streams are capable of advecting the spatial structure of turbidity over a 100 m
length scale in a few minutes. On the other hand, if the CZCS with resolution
around 1 km is used to study water turbidity in a shelf sea, the patterns are
unlikely to be changing significantly on this length scale in less than an hour, and
Sec. 4.6] Oceanographic sampling for ‘sea truth’ 101

sea truth collected within a few hours of the overpass may suffice for calibration
purposes, although there will always be more confidence placed in the results the
closer the sea truth can be timed to coincide with the overpass.
The provision of facilities for synoptic sea data collection presents severe
logistic difficulties not encountered by land remote-sensing workers. It is much
more costly to have to use a boat to reach a sampling point than simply driving
there on land. If the measurement demands that a scientifically-equipped
research vessel be used, then the cost escalates. If the sensor being calibrated is
an all-weather microwave device, then the only limitation on obtaining synoptic
sea and satellite data is that the weather should not be too bad to prevent the
sea observations being made. If visible or infrared sensors are being calibrated,
then cloud-free conditions are also required for there to be any useful satellite
data for comparison with the sea truth. Waiting for the suitable combination of
clear skies but not too strong a wind can be an expensive exercise with a fully-
equipped and staffed research vessel. For a satellite such as Landsat with a
16- or 18-day repeat cycle, only one or two clear weather passes a year may be
gained for some locations. An alternative is to moor buoys to collect sea data
over a period of weeks or months, in the expectation that at least some samples
will be obtained synoptic with a cloud-free overpass. An instrumented buoy
programme is itself costly, however, and it is evident that a dedicated sea-truth¬
gathering exercise for satellite calibration purposes requires a significant
commitment of resources and research effort, however it is performed.
Another means of obtaining synoptic sea truth is to make use of oceano¬
graphic data which happened to be collected within the field of view of the
satellite sensor at or near the time of the overpass, as part of an unrelated
oceanographic research programme. This might be termed a ‘serendipity
method’! It can include measurements made from ships of opportunity, and
may combine meteorological records made from commercial shipping, as well as
scientific observations from research vessels and instrumented buoys. It almost
invariably has the drawback that the oceanographic parameter has not been
measured in quite the way that would have been chosen for a ‘sea truth’ exercise.
For example, sea surface temperature may have been measured from the
engine-intake water, at a few metres below the surface, whereas ideally the skin
temperature should be measured for calibrating IR sensors. This suggests a fourth
approach to sea truth collection which is to mount instrumented packages on
ships of opportunity which make regular crossings of certain seas and oceans.
The instruments have to be simple and robust, but can be designed to measure
the particular parameters of relevance to the task of satellite data calibration and
validation.
Another related problem in the comparison of sea and satellite data is
that of spatial sampling. To obtain as broad a database as possible for calibra¬
tion algorithms, it is desirable to obtain sea data over a wide range of possible
values corresponding to a wide range of satellite data values. This requires sea
102 Principles of remote sensing of the sea [Ch.4

measurements to be made at many different locations within an image field,

simultaneously with the satellite overpass. This implies the use of many ships, or
the deployment of an array of buoys, both extremely costly for open-ocean
experiments. A compromise can be made by letting a single vessel steer a suitable
course across strong gradients of the parameter being calibrated, making con¬
tinuous measurements as the ship is underway. This compromise is Hmited by
the space—time variability of the particular parameter relative to the speed of
the ship. In other words, the ship should be capable of travelhng a distance L
in a time which is small compared with the time it takes for patterns of the
parameter to change significantly over a length scale L. This is not always
possible with ships, but may be possible if heUcopters are used for sea-sampUng
purposes. Even then it is not often evident from the perspective of sea level
where the gradients of parameter are strongest, and therefore which course to
steer with the sampling vessel or aircraft. On land it is obvious that a desert will
give a different signal from a forest, and a motorway a different signature from
a grain field. At sea the variability is more subtle, and it may require several
previous images of an area to be studied to give guidance about typical spatial
distributions of parameters before it is possible to design an efficient strategy
for gaining a broad range of parameter values from a limited number of samples
or from a single transect of continuous measurement.
There is another difficulty in matching ground-level data to remotely-
sensed data, and it must be taken into account when planning sea truth exercises.
The satellite sensor measures an integrated value of the electromagnetic radiation
over the whole of its instantaneous field of view. Now the IFOV may have length
dimensions as small as 25 m in the case of the Seasat SAR, or as large as 100 km
in the case of the SMMR on Seasat and Nimbus 7. The sea-level measurement
from a buoy, research vessel, ship of opportunity, or helicopter, is sampled at
a point which is usually less than a few cm in size. There is clearly a mismatch
in the two methods of sampling which can be important if there is significant
variability over length scales larger than the field-sampling instrument size but
smaller than the IFOV. Fortunately this is one problem which tends to be less at
sea where variations are smoother, than on land where the texture of a surface
can be composed of discrete elements which make very different contributions
to the remotely-sensed measurement (for example, visible light reflected from
a shingle beach is an average of the high reflections from lighter stones such
as chalk and the low reflections of darker stones). Even though the small-scale
variability is less at sea, the differences in parameter values which we attempt to
resolve at sea are also much smaller. Moreover, to a large extent, we do not have
a very good understanding of spatial variability at sea, because conventional
oceanographic techniques have never been able to measure it.
Care must therefore be taken when comparing a pixel value on a remotely-
sensed image with a point value measured at sea, within that pixel, because
it may not be representative of the average parameter value within the whole
Sec. 4.6] Oceanographic sampling for ‘sea truth’ 103

pixel. One simple example is the skin temperature measured in the presence of
large waves, where it has been noted that whilst the underlying sea surface
temperature may be uniform, the actual surface skin temperature may vary by
as much as a few tenths of a degree Kelvin between trough and crest, a length
scale of tens to hundreds of metres. To avoid the problem one must make
statistical allowance for small-scale variability, if it is known, in the calibra¬
tion and comparison procedure, or seek to measure an average value at sea
by subsampUng within a pixel. The only practical method for doing this is by
performing continuous sampHng along a transect and making sure that the
sampling rate is very much more spatially dense than the resolution of the
satellite sensor being cahbrated.
It is worth emphasising here that the spatial-averaging capabilities of
satellite sensors is of significant advantage to the oceanographer in many applica¬
tions. When seeking to observe the variability of large-scale processes it is a
problem if measurements of the phenomena are corrupted by local variability
due to small scale processes. The satellite sensor conveniently avoids this problem
so far as sub-pixel variabihty is concerned, whilst variability at medium-length
scales larger than the pixel size can be smoothed out because of the complete
sampling of the area achieved by imaging sensors. This enables large-scale
variability to be revealed in circumstances where ship measurements might not
have been able to distinguish it from the noise of the smaller-scale processes.
So far in this section we have considered data collection at sea in the
context of the geophysical cahbration of satellite data. There is another related
use, and that is the testing of atmospheric correction algorithms. Since these
effectively deduce the water-leaving radiation characteristics from radiances
measured at the satellite, they require sea-level observations of the radiation
for their validation. Such radiation measurements at ground level can also serve
to demonstrate whether a remote sensor is capable of detecting certain sea
parameters. For example, in coastal waters the CZCS is believed to be capable of
revealing chlorophyll and suspended sediment patterns in the sea. The imagery
is often highly contaminated by atmospheric effects, and it is therefore difficult
to estabhsh a correlation between satellite colour signature and the sea-water
parameters. If colour measurements at sea level can demonstrate such a
correlation, then it is worth improving the atmospheric correction until use¬
ful information can be extracted from the satellite data. If colour measurements
at sea level do not correlate with the water-quality parameters, then there is
clearly no point in looking for such a correlation in the satellite data.
For both these reasons, radiation measurements at sea, viewing the ocean
from above as the satellite does but without any intervening atmosphere, have
an important part to play in the quantitative interpretation of satellite data.
Unfortunately these are not measurements traditionally made by oceanographers.
Such optical measurements as have been made have usually been subsurface.
Only recently with the development of satellite oceanography has work seriously
104 Principles of remote sensing of the sea [Ch.4

begun which deploys radiometers and radiance meters from the mastheads or
the booms of research vessels or platforms. The spatial sampling and spatial-
averaging problems mentioned above present obstacles to the interpretation of
the data. Aircraft carrying analogue satellite sensors can take a wider view, but
then the atmospheric problems are introduced once more, in a less tractable
form than for the satellite since the variable height of the aircraft and its varying
tilt and pitch continuously alter the path length of the radiation. There is clearly
scope for the further development of radiation measurements at sea, not only in
the visible and infrared, but also in the microwave, seeking to understand how
radar backscatter is related to sea-surface roughness.
Finally, having mentioned in section 4.3 that the positional registration of
satellite data is essential if it is to be useful to the oceanographer, it should be
remembered that accurate positional registration of sea-sampled data is also
vital if it is to be correlated with satellite data. The errors in position-fixing at
sea away from land are comparable with those of the positional registration of
imagery, although the increasing number of navigational satellites in orbit will
improve navigation for ships and buoys equipped to use them. The question of
position fixing is the limiting factor in the use of instrument packages on ships
of opportunity, and ideally requires regular positional updates from the ship’s
navigational equipment to be logged along with the oceanographic data. The
discrepancies in sea and satellite data correlation which arise from errors in
positional registration of either or both will depend on the sensor sensitivity, the
spatial variability of the parameter in the region surrounding the measurement,
and also on the size of the IFOV. For example, if it is known that the SST is
uniform (that is, constant within the sensor’s radiometric resolution after
digitisation) for 10 km in all directions, then position fixing to within 10 km
would suffice. If, on the other hand, the sampling point is close to a front,
it is very important that the sampling point and the corresponding pixel with
which it is identified are located on the same side of the front. However, even
with a highly sensitive sensor, and strong spatial gradients, there is little value
in improving positional accuracy of sea-truth measurements to less than the
IFOV length scale.
It is difficult to generalise any further on the question of comparability
between ‘sea truth’ and satellite data. Each sensor and situation must be
considered individually, bearing in mind the points that have been made in this
chapter. Calibration errors are due to inadequacies in both the satellite remote
sensing and the sea-sampling techniques. It is worth remarking that satellite
remote sensing of the ocean has developed in some areas to the point at which
the limiting accuracy is that of the sea measurement rather than the satelhte. For
example, significant wave height could be measured with the Seasat altimeter
to a precision as good as could be achieved by measurements at sea. However,
there is certainly no question of the satellite sensor replacing the instrumented
buoy or research vessel. On the contrary, the way towards real scientific progress
Sec. 4.6] Oceanographic sampling for ‘sea truth’ 105

in improving the accuracy of remote sensing lies not only in better space instru¬
ments, but in improved oceanographic techniques to match them. Quite apart
from the temporal and depth resolution which can only be achieved by measure¬
ments at sea, the remote sensor can only achieve its potential performance in
spatial sampling of the ocean if it is backed up by a well-coordinated programme
of careful acquisition of sea truth.
CHAPTER 5

Principles of image processing

5.1 IMAGE PROCESSING AND OCEANOGRAPHY

Image processing is the term given to those operations performed upon a set of
image data in order to improve it in some way, to assist in its interpretation, or
to extract useful information from it. In remote sensing we are presented with
either photographic images, analogue TV-type images, or digital image data sets.
Processing and enhancement of the first two can be performed in a variety of
ways, but since virtually all oceanographic data from imaging satellite sensors
are now in digital form we shall look in particular at digital image processing.
There are several helpful texts which discuss image-processing methods in both
concept and in mathematical detail, for example, Gonzalez & Wintz (1977),
Castleman (1979), Reeves (1975), Rosenfeld & Kak (1982), and Colwell (1984).
The intention of this chapter is to present the principles of those processes
which are useful in the enhancement and analysis of satellite images of the
sea. Whilst the same image-processing techniques can be used for both land and
sea apphcations, the oceanographer tends to use more often certain techniques
which are used less by the land remote-sensing scientist, and vice versa. For
example, classification techniques are discussed at length in image-processing
texts with land application in mind, since they provide a powerful tool for the
delineation of land-cover types or of Hthological regions, but the method is little
used by oceanographers. This chapter should be viewed as an oceanographer’s
guide to image processing. It aims to introduce those techniques which are
worthy of further study in the context of oceanographic applications.
Since image-processing computers are now available with software ready-
written to handle most common image-processing tasks, the remote-sensing
applications scientist is generally spared the task of developing numerical
algorithms. This permits us to omit here a mathematical description of the
processes, and to concentrate on their principles of operation and the results
they can achieve.
It is obvious, but nonetheless worth stating, that image processing cannot
produce information out of nothing. If there is no information in the data set
concerning a particular application or interpretation, then no amount of sophis-
[Sec.5.2] Digital image data from satellites 107

ticated image processing can generate it! Yet such is the power of modern
computer techniques to enhance the smallest variability, smooth the result, and
enhance again to produce a clear, crisp, multicoloured pattern, that the uncritical
user of this equipment can easily convince himself that he has stumbled on
a genuine oceanographic feature. In fact he has merely used his imagination in
much the same way that we can ‘see’ all sorts of interesting pictures, shapes,
or even a map of Australia if we gaze long enough into a sky full of cumulus
cloud patterns! Whilst there is a place in satelhte oceanography for looking
particularly at pattern shapes and the length scales of variability, there are also
dangers in merely generating pretty pictures. The safeguard against uncritical
conclusions is to keep track of the numbers involved, both the raw digital values
from the sensor and their calibration in terms of oceanographic parameters.
For example, a thermal ‘front’ may appear by enhancement out of what was
previously an apparently uniform piece of sea. If we then find out that the
‘front’ has in fact a temperature step of only 0.2 deg K, when the thermal
resolution of the IR sensor is known to be 0.2 deg K and the noise another
0.2 deg K, we must conclude that the front is in our imagination and not really
there. Thus it is important to understand exactly what the image-processing
software is doing to the numbers from the original satellite data set, so that
the correct brightness value of a pixel on the screen can always be related to
a physical parameter.

5.2 DIGITAL IMAGE DATA FROM SATELLITES

Satellite data in a form in which it is ready to be used by applications scientists
is normally supphed on computer-compatible magnetic tapes (CCTs), which can
be used on standard tape-drive equipment. The way in which information is
stored on the tapes will differ from sensor to sensor and also will depend on the
standards adopted by the space agency or receiving station supplying the data.
The user must therefore first ascertain from the supplier such details as the data
density, the number of tracks (9 track 1600 bits per inch is standard), and the
way in which non-numerical ancillary data are coded (normally in ASCII or
EBCDIC) before he can provide the necessary computer hardware and software
to be able to extract a list of numbers from the tape. The format specification,
which should be supphed with the tape, can then be consulted in order to
interpret the significance of the numbers in the data stream.
Fig. 5.1 illustrates a typical way in which an image data set may be stored
in a sequential file. At the beginning is some header information which defines
the whole image by giving such information as the date and time, orbit number,
etc., and information about any processes which may already have been appUed
to the data. There then follow data from the first scan line of the image.
Interspersed with earth-view data are calibration and other houskeeping data.
Normally, if there is a voltage/digital calibration ramp supplied by the satellite it
will come at the beginning of every line, or be spread over the header information
 108
Principles of image processing

Fig. 5.1 — Typical data stream on a CCT from a satellite sensor

[Ch. 5
Sec.5.2] Digital image data from satellites 109

of several lines. Any sensor calibration information, for example, a space-view

radiance and a target-view radiance, along with any information about the
brightness or temperature of the target, will normally be supplied every scan
hne, as it is sampled by the sensor in this way. Similarly, information about
the dynamic range of the sensor being used (if it is switchable) and the status of
other sensor parameters adjustable by remote control from earth, will be included
in the data stream. Therefore software must be written either to skip this
information, or to read it and apply it in the data-processing operations.
If the data have already been extensively pre-processed by the suppliers,
much or all of the housekeeping or calibration information will be absent, and
only the earth data shown in Fig. 5.1 wiU remain. It is necessary to know the
way in which the image data is arranged. In Fig. 5.1 it is assumed that data are
being presented from a 3-channel multispectral sensor, essentially in the
sequence in which they were observed and multiplexed on the satellite. Thus for
each position (known as a sample point) along the scan line, the digitised value
of observed radiance is given in turn for the separate wavelength channels (or
bands) 1,2, and 3. If there are 2000 samples across the scan, there will be 2000
groups of three numbers each for each scan line, followed perhaps by calibration
information for the line just completed or the line about to commence, before
the 6000 numbers for the next scan Hne, and so on until the complete number of
scan lines for that image (in this case 2000) has been presented. The number of
scan lines per image is defined, if at all, purely for ease of data ordering and
distribution. Thus for the Landsat MSS each image area is clearly specified by
Earthnet, whereas AVHRR data obtained from the University of Dundee satelUte
receiving station is not spUt into predetermined image data sets but ordered in
terms of a requested number of scan Unes centred around a given latitude.
To create a viewable image from the sequential information, the whole data
set must be read into the computer store and resampled channel by channel,
first picking out all the band 1 data, and so on. To obtain an image which most
closely resembles a map (albeit distorted by earth curvature and sensor view
angle) with north towards the top, it is necessary to know the direction of travel
of the spacecraft and the direction of scan. Thus assuming the data is from a
polar-orbiting satellite which travels approximately northwards, scanning from
left to right across its track, the data of Fig. 5.1 should be arranged as shown
in Fig. 5.2(a) to create a band 1 image. On the other hand, if it overpassed
from, north to south during the data collection, but still scanning left to right.
Fig. 5.2(b) is the way in which the data should be ordered on the display to
generate a recognisable image.
The organisation of the satellite data as presented in Fig. 5.1 can be
described as Band, Samples, Lines (BSL) organisation, because the band number
changes most rapidly, then the sample number, whilst the line number changes
least frequently. It is possible for data to arrive in a different organisation,
particularly if it has been preprocessed. Thus SLB would be expected when the
110 Principles of image processing [Ch.5

Band 1 Image

Band 1 image

Fig. 5.2 - Image display organisation of data from the CCT of Fig. 5.1 to produce
a correctly-oriented image, assuming a left to right scan, for (a) satellite travelling
from south to north, and (b) satellite travelling from north to south
Sec.5.2] Digital image data from satellites 111

whole of band 1 is sequentially filed, line by line, before band 2, etc., as shown in
Fig. 5.3(a), or LSB when the first sample on each line is stored line by line, then
the second sample and so on as in Fig.5.3(b). It will be noted that the latter
case in particular is not suited to the incorporation of line by line calibration
data.

(a)
Samples, Lines, Bands
Band 1 Band 2 Band 3
Sampla 1 32 32000 31 32 31000 32000 31 32 31000 32000
Lina 1 LI LI L2 L2 L2000 L2000 LI LI L2000 L2000

(b)
Lines, Samples, Bands
Band Band 3
Band 1
*7
3ample 1 31 31 31 32 52000 32000 31 31 32000 32000
Line 1 L2 Liooe L2000 LI Liooe L2000 LI L2 Liooe L2000

Fig. 5.3 - Ordering of image data on a CCT in the case of (a) Sample, Line, Band
organisation, and (b) Line, Sample, Band organisation

The data values are normally stored as 8-bit binary codes, that is, as integer
values between 0 and 255, although sometimes (for example, for AVHRR
data) 10-bit words are used enabling an integer range between 0 and 1023. As
discussed in Chapters 2 and 4, the integer value corresponds to different levels
of electromagnetic radiation, and it is sensible to assume that a higher number
corresponds to a higher radiance level. This is true for visible and microwave
data, but for the thermal infrared the convention is the reverse of this. Thus
as the sea temperature rises, the infrared radiation at the satellite increases, but
the digital value returned to earth decreases. The reason for this is that a simple
grey-level visual reconstruction, using Ughter shading for higher numbers and
darker shading for lower, causes cloud tops to appear whiter than land or sea,
which assists in meteorological interpretation. The result is an image in which
the warmer parts are darker, which suits better our conceptual grasp of black
objects being warmer than white ones. Particular care must therefore be taken
when dealing with multispectral data from sensors which have both thermal IR
and visible/near-IR channels, since the data is a mixture of both conventions.
The calibration information accompanying the earth-view data will define
whether the calibration gradient is positive or negative, and hence define which
convention is being used.
112 Principles of image processing [Ch. 5

Assuming 8-bit data values are being recorded, the typical image dataset
described in Fig. 5.1, apart from the calibration and housekeeping data, will
have 2000 X 2000 X 3 X 8 = 96,000,000 or 96 M bits of information. In terms
of disk storage on the computer, with 1 byte = 8 bits, a 12 M byte disk capacity is
required for the three-channel image. In practice many scanners produce larger
datasets, with longer scan lines and more channels. With so many numbers to
be processed, dedicated image-processing computers are the most satisfactory
means of handing satellite image data, although it is still possible to extract a
lot of valuable information, and to perform calibration processes using ordinary
mainframe computers with sufficiently large disk storage capacity.

5.3 IMAGE-PROCESSING HARDWARE

With several commercial image-processing systems available, and given the pace
of microprocessor development, there is little value in addressing this subject in
great detail. It is useful instead to list the necessary or desirable elements of an
image processing system.
Fig. 5.4 illustrates the typical components of an image-processing system as
developed in the late 1970s. The system is built around the Central Processor
Unit of a standard computer which may or may not be used for other work, and
the magnetic tape drive and large disk storage are essential for handling the large
data quantities on CCTs discussed in the previous section. Alternative inputs
of digital data from digitising tables and video camera systems can be a useful
means of introducing data from ship measurements, map bases, etc., in a digital
form in which they are compatible with the satelHte data. The CPU and large
disk storage can be used for splitting up the CCT data into manageable subsets
of data, coping with the samples, bands, and lines organisation of the data
store, interpreting any useful calibration data, and applying the atmospheric
and calibration algorithms. This can all be done without the use of image display
at all, and does not need a dedicated image-processing system.
The image processor itself is an additional processing system designed to
enable very rapid interaction between the user, the image memory planes, and
the colour display monitor. Typically a display of 512 X 512 pixels is provided,
and several memory banks of 512X512 bytes are available for the storing of
several image datasets, either different channels of the same scene, or different
scenes. Thus each image memory bank requires over 14 M byte of storage. The
image processor can rapidly alter the way in which information in the image
memory is presented in the corresponding position on the screen, adjusting the
brightness or colour according to the interactive instruction of the user. Data for
several image memory banks can be merged or fed through different colour guns
to the display. The display processing can be driven either

(i) by commands on a standard VDU terminal keyboard.

Sec. 5.3] Image-processing hardware 113

with remotely-sensed image datasets

114 Principles of image processing [Ch. 5

(ii) by moving a cursor over a digitising tablet and selecting instructions from
a menu on the tablet,
(iii) by selecting commands from a menu on the display screen or the VDU
screen, using the digitising tablet cursor, a joystick, or a trackball, or
(iv) by a combination of these methods,
depending on the proprietary system used. It is customary to be able to enter a
mode of operation whereby the trackball or joystick can be used to zoom in
upon a detailed view of part of the image and to roam over the entire image
in this magnified perspective. Various image-enhancement processes can be
performed, with interactive intervention to produce the clearest image which
best emphasises the feature under study. However, there is a danger in these
interactive processes that quantitative and critical science can degenerate into
merely producing colourful pictures, and the user can lose track of the real data
values in the pre-written software provided with the computing system. Those
computer algorithms which operate in order to correct the image geometrically,
and those which use multispectral information to correct atmospherically
and to calibrate the data in terihs of oceanographic parameters, tend to require
more complex arithmetic processes. These operations are therefore much
slower, having to be performed on the main CPU rather than in the image
processor itself. However, as a new generation of array-processor-driven machines
becomes available, it is likely that even these tasks will be accomplished almost
instantaneously, that is, taking a few seconds rather than many minutes.
Once an image has been processed, it is useful to be able to store the final
result for future use. Floppy-disk storage offers a convenient and cheap method
at present, but whilst CCTs provide a suitable method for transferring data from
supplier to user, and from one user machine to another, floppy disks do not have
an acceptable standard format, and are not advisable for data transfer between
users or different computers, unless the floppy-disk drives and associated driving
software are identical.

5.4 GEOMETRIC PROCESSING

The simplest form of geometric processing in order to change the shape or
size of an image is achieved by selective sampUng. Thus given a CCT with
2048 X 2048 data points, before proceeding to apply geometric and atmospheric
corrections to the whole scene when only a portion of it may be of interest,
it is useful to display the whole scene by selecting only every fourth sample
on every fourth row to produce a 512 X 512 image. This enables any clouds
on the image to be located, and in a coastal region the area of interest can be
located accurately from the coastline, enabling the correct part of the image
to be isolated for further processing and study at full resolution. In the case of a
Landsat MSS image with pixels nominally of size 56 m along the scan line and
Sec. 5.4] Geometric processing 115

79 m between scan lines, by selecting every 6th or 3rd sample and every 4th
or 2nd line, an approximate correction is automatically apphed to allow for the
non-square pixels.
This approximate form of geometric processing results in loss of information
from the rejected pixels. A more satisfactory geometric correction technique
requires the resampling of the image according to a mathematical transforma¬
tion, utilising all the data values in the original image to contribute to the data
values in the final image. Fig. 5.5 illustrates two possible cases. In (a) the final
pixel size is similar to that of the original pixels, and in (b) the final pixels cover
a much larger area than the originals. The reverse of (b) could be envisaged, with
the final pixels much smaller than the originals, but in that case it is not possible
to find enough information in the original data to make every new pixel value
independent of its neighbour, and resolution cannot be improved by that means.
In Fig. 5.5 resampling can be performed most simply, but less accurately, by
‘nearest-neighbour substitution’ in which a given pixel, for example, that with
centre at A, is assigned the value of the pixel on the original grid whose centre
is closest to A, that is, a. An improved bilinear interpolation procedure is to
assume a hnear gradient of the pixel values between pixel centres on the original
grid, so that resampling for B will utilise the values at pixels b, c, d, and e,
effectively fitting the parameter value to a twisted plane through b, c, d, and e,
and locating the resampled value at B on that plane. By using higher-order
polynomials, or spline curves to fit the parameter surface on the original grid, it
is possible to obtain even better sampling on the new grid, incorporating the
values of the neighbouring 9 or 16 old grid centres. Such processing can be costly
in computer time, and introduces a smoothing or filtering effect,which increases
with the number of original pixels used to contribute to each new pixel value.
If smoothing or filtering is to be performed anyway, it may turn out to be
cheaper in computer time to use the nearest-neighbour resampling and then
apply a standard filter, rather than use a complex geometrical resampHng
polynomial which has to be re-evaluated for many blocks of pixels across
the image. Given the availability of several different resampling algorithms on
proprietary systems, the user must make a judgement based on the application
concerning which one to use.
The geometrical transformation itself, that is, determining the shape of
the new pixel grid pattern relative to the old, can be achieved by two distinct
methods, as discussed in section 4.9. In the first, ground control points on the
original image are identified and their pixel indices (row and column number)
noted. These pixel coordinates are then linked to a set of new pixel coordinates
based on the map projection of the imaged area. A general warping algorithm,
available with most proprietary image processors, then fits the old points to
the new locations. By treating the rest of the image as if it were a rubber sheet,
capable of being stretched and rotated unevenly across its surface, all the old
pixel coordinates are located relative to the new, so that resampling of the actual
116 Principles of image processing [Ch. 5

values can be performed. The more ground control points there are, the more
likely is the map to be a satisfactory one. Such a generalised warping process is
very time-consuming, when million or so data points have to be re-evaluated.
It is somewhat quicker if a simple rotation or linear stretch can be applied to
the whole scene, to bring it onto a map base, but this is only feasible if the
underlying distortions of the satellite perspective have been removed.

(a)

NEW QRID ORIGINAL QRID

Fig. 5.5 - Typical examples of geometrical resampling grids: (a) similar size,
(b) resampled grid much coarser than original

The second approach is to perform this basic correction using a mathe¬

matical transformation based on the geometry of the satellite orbit and sensor
viewing angle. This can be done without recourse to special image-processing
software, using a standard computer facdity. As an example we can examine
the correction of NOAA AVHRR imagery, which is warped in the manner
approximately sketched in Fig. 4.3.
There are two stages in the transformation. The first is to ‘stretch’ each scan
line to allow for the distortion arising from oblique viewing over a wide range of
viewing angles. The second is to map the stretched scan lines onto a given map
projection base, by identifying the latitude and longitude of each pixel.
The scan-line transformation is achieved with simple plane trigonometry,
as illustrated in Fig. 5.6. The image dataset provides a scan line of pixels each
identified by its viewing angle /3. In the uncorrected dataset, adjacent pixels
differ by a fixed angular increment S/3, and provided that a given pixel can be
Sec. 5.4] Geometric processing 117

identified as being N pixels away from the pixel corresponding to the sub-point
(normally the central pixel in a scan-line), then ^ = N8^. We need to find a
as a function of (3, where a is the distance from the satellite sub-point to the
pixel found at viewing angle /3.
Now a = i? i// = /? (tt — j3 — 7).
(h+R) sinj3
From the sine rule: siny =
R
|(i?+/z) sin|31
thus a = R TT — p— sin

Using this, the scan line can either be replotted with differential spacing
between pixels, or resampled.

Fig. 5.6 — Geometric sketch for relating along-scan distance ‘a’ to along-scan
viewing direction /3
118 Principles of image processing [Ch. 5

The identification of latitude and longitude along the scan line is rather
more complicated. We assume that the latitude and longitude of the scan-line
centre (the satellite sub-point) can be obtained from a knowledge of the satellite
orbit at a given time. The direction of the scan line is assumed to be at right
angles to the travel direction, and this can be determined from a knowledge
of the orbit inclination and the instantaneous latitude. Fig. 5.7 illustrates the
problem. It is treated analytically by considering the satellite orbit path to be
the ‘equator’ of a satellite coordinate system, denoted by a prime ('), in which
the scan lines correspond to lines of longitude. Thus point P in the scan line
sampled when the satellite was at S has true latitude (pp and longitude Xp and
in the ‘satellite co-ordinates’ 0p and X'p. X'p = X'g, and if X' is measured from
where the track crosses the true equator at X, then
sin 05
X' sm -1
sin /

0s = 0 by definition, and so 0p = a/R.

Fig. 5.7— Sketch to illustrate the relationship between geographic latitude and
longitude [(pp,
\p] of point P and the ‘satellite coordinates’ [0p, X'p] based on
the satellite orbit as ‘equator’

The transformation between the two coordinate systems is given in standard

texts on solid geometry (for example, Sommerville (1934)), from which it is
readily shown that
sin 0s
0p sin —sini cos — cos/ sin
sint
Sec.5.5] Masks and overlays 119

and Xp sin sinXo cos cos

cosisinf—I-Xq) cos/—isin^c
\2 / \rI + cosXo sin / sin
sin/
where Xq is the true longitude of X.
Thus the actual latitude and longitude of a point can be defined in terms of
its distance a from the satellite sub-point which is at latitude 0^ on an orbit
of inclination / which crossed the equator at longitude Xq. Note that Xo is
the equator-crossing latitude if the earth had not rotated whilst the satellite
travelled from X to S. A time correction must therefore be applied to the actual
equator-crossing latitude.
The above analysis offers the framework of a theoretical pixel-location
procedure, which should be developed individually for given satellite orbits. Care
must be taken in defining the directions of angles related to the travel direction
around the orbit. Positioning near the poles may be better achieved by trans¬
forming to a different set of earth coordinates which does not have a singularity
at the pole. Once the pixels have been located in terms of (0p, Xp) the whole
image can be resampled either to provide uniform sampling across 0 and X, or
to achieve square pixels on whatever map projection is being used. At this
stage, the new image geometry is still subject to errors arising from inaccurate
knowledge of the satellite orbit. However, a small linear translation and/or
rotation of the new image, based on just a few ground control points, should
be sufficient to register the image accurately.

5.5 MASKS AND OVERLAYS

Image-processing systems normally include several graphics or overlay planes,
which are arrays of one-bit binary storage with the same dimensions as the
image arrays. These are used for writing graphics overlays in order to label
features, or to locate lines of latitude and longitude. These may be drawn
over the image or not, as required. The overlay plane has a feature which is
particularly useful in oceanographic applications. This is its ability to be used
as a mask, by denoting certain areas of the image as ‘on’ and the rest as ‘off’.
If it is possible to switch those parts of the overlay plane ‘on’ which correspond
to land or cloud in the image, and the clearly visible open-sea parts as ‘off’, a
binary image distinguishing between sea pixels and the rest of the image is
created in the overlay plane. It is called a binary image because it contains spatial
pattern data, but no densitometric (pixel grey level) information.
120 Principles of image processing [Ch. 5

Such a binary image can be used in two ways. One is to ensure that certain
image-enhancement techniques operate on the sea pixels only. The mask acts as a
label attached to those pixels to be included or excluded from image-processing
commands. The detail of how this is implemented vary with different processing
systems, but in one form or another the facility is to be found in all well-
developed imaging software. The other use is in the preparation of slides or hard¬
copy records of imagery. If two masks are prepared, one of the visible land and
the other of cloud, it is visually helpful to overlay the sea-processed image with a
mask of green or black for the land and white for the cloud (if these colours
do not create ambiguity with colours already used in the sea processing). The
observer can thus readily locate the image by the coastline pattern, and quickly
appreciate the masking effects of cloud so that attention can be concentrated on
the open-sea parts of the image with which the oceanographic user is concerned.
The binary masking image can normally be created from the original image.
Cloud is usually apparent on both visible and thermal IR imagery as pixels with
much higher digital values than the sea, because on visible images the clouds
reflect more sunlight than the sea, and in IR data they are generally cooler than
the sea, emitting less radiation and hence appearing lighter because of the
thermal IR densitometric convention. Thus by assigning all pixels with values
over a certain threshold as ‘on’, the cloud mask is created. The land mask can
cause problems since in visible wavelength imagery, depending on the channel,
the land and sea may have similar brightness, and for thermal IR data the land
may be warmer or cooler than the land, depending on season, time of day or
night, etc. In general the sea will be cooler than land in summer and warmer in
winter, and whilst sea temperature varies very little at night, the land cools down
significantly in the night. There are often situations, therefore, when land and
sea have similar temperatures. One way round this is available when a satellite
delivers both a thermal IR and a visible image of the same area. It is quite likely
that when the thermal IR image cannot produce a suitable land mask, the visible
channel will, or vice versa. In practice the most reliable channel for the produc¬
tion of land masks is in the near-IR if available, where there is very little solar
reflection by the sea, but significant amounts from the land. Band 7 of Landsat
MSS, for instance, or channel 2 of the AVHRR, makes a useful coastline
delineator. Provided that the different channels have been collected by the same
sensor, the pixels of one channel coincide with the pixels of another, and the
masks made from one channel are available for use with all the other channels.
If it is not possible to create a mask from the image itself, then it might be
necessary to create one from a map, using TV input to the imaging system, or
drawing on a digitising tablet. This is rather laborious, however, and presupposes
further that the image has been very accurately position located, something
which is not so important when the coastline shape is already directly evident on
the image.
Sec. 5.6] Smoothing, filtering, and noise reduction 121

5.6 SMOOTHING, FILTERING, AND NOISE REDUCTION

One of the principal differences between image processing of land surfaces and
that over the sea is the much smaller range of variability of signal from the sea,
at least in the visibile and infrared channels which apart from SAR are the
wavebands used by the present generation of earth-imaging sensors. A sensor
such as the CZCS which is specifically designed for sea use copes with the small
variability of subsurface reflection of solar light by having a high sensitivity
within the expected range of sea-viewing radiances. As a consequence it often
saturates over land. When using other sensors such as Landsat MSS, NOAA
AVHRR, or Meteosat, it often occurs that the variability being studied is not
much larger than the sensor resolution, or the digitisation interval. Therefore
the enhancement of imagery to emphasise the patterns in the sea also reveals the
noise inherent in the image, resulting in a spotting effect.
Fig. 5.8 illustrates this. It is supposed that a contour of an ocean parameter
crosses the image centre froni top to bottom, and that the resulting digital value
of this contour is 20. To the right the values are above 20, and to the left they
are below 20. The digitisation process may be modelled in the following way.
If there are 10 pixels which have recorded a sensor radiance which should
convert to a value of 19.9, we can expect on average nine of them to return a
digital value of 20 and one of them 19. Similarly, out of ten pixels which should
be 20.2, eight will be set at 20 and two at 21. If then we assume a random
selection of which pixels return which value, within these overall average con¬
straints, the resulting image might look Idee Fig.5.8(a). The resulting image noise
creates at least two problems. Firstly, it clutters the visual impression and can
mask the genuine oceanographic features being presented. This can be avoided
by, as it were, ‘standing back’ to view a large area all at once, in which case the
digitisation noise of individual pixels is not only hidden, but can contribute to a
a visual smoothing of sharp contours which realistically images the gradual
gradients which occur in the sea. This is directly analogous to the way in which
the overall texture of a newspaper picture apparently improves as one views it
from further away, whereas viewed closely it appeares to be merely a matrix of
dots.
The second problem of noise in an image is the effect it has on other
digital-processing algorithms. If, for example, gradients are being measured, then
simply taking the difference between the individual pixels of Fig. 5.8(a) would
produce a meaningless distribution of gradient values. Similarly, if the image
dataset is used to enable contours to be plotted by computer graphics software,
the algorithm would have great difficulty in interpreting Fig. 5.8(a). There
is therefore a requirement to smooth out the noise by numerical filtering.
A wide variety of two-dimensional digital filters is available in commercial
image-processing software, and the reader should consult the software manuals
of individual systems to determine the exact nature and purpose of them all.
In general, a filter can be represented by a matrix as in Fig. 5.9. The operation
122 Principles of image processing [Ch. 5

■ pixel at value 21

KEY B pixel at value 20

□ pixel at value 10

(b)
(a)

(0
19.2

0
<D o
1 a CO
-1

ro 00

1 1 r- I 1 1 1 I 1

Fig. 5.8 — Digitisation effects in an image, (a) A noise-free linear ramp of values
between 19.0 on the left and 21.0 on the right, as it might appear when sampled
with a digitising interval of 1.0. (b) The effect on (a) of a smoothing filter and
rounding down, (c) The effect of smoothing and contour plotting of (a)

1/9 1/9 1/9 1/32 1/32 1/32 1/32 1/32

1/0 1/9 1/9

1/32 1/20 1/20 1/20 1/32

1/9 1/9 1/9

1/32 1/20 1/10 1/20 1/32

1/32 1/20 1/20 1/20 1/32

1/32 1/32 1/32 1/32 1/32

Fig. 5.9 — Typical moving average filter matrices: (a) 3 X 3 unweighted

(b) centre-weighted 5X5
Sec.5.6] Smoothing, fUtering, and noise reduction 123

of the filter is most easily conceived as moving the centre square of the matrix
over the image, pixel by pixel, and placing in the new (filtered) image dataset
for that pixel a number corresponding to the operation specified in the matrix.
Thus in the 3X3 moving-average filter of Fig.5.9(a) the pixel value is replaced
by the average value of the pixel and its eight immediate neighbours. Such a
filter would give a considerable smoothing effect to Fig.5.8(a), and might result
in Fig. 5.8(b), where the averaged value has been rounded down to the nearest
digital value. If noise still persisted, a larger-area filter, 5X5 or 7 X 7, could
be used, although the larger an area covered by the filter the more likely it
is to start smoothing out steep gradients of ocean parameters which are real.
A compromise can be achieved by using a weighted average, which allows the
pixels near the centre of the filter to influence the output value more than those
at the perimeter. A hypothetical example is illustrated in Fig. 5.9(b). It should
be noted that in performing the filtering operation the computer must use real
rather than integer arithmetic. If instead of rounding down to produce an
integer digital image, the real numbers are stored, it is possible to plot contours
as in Fig. 5.8(c). However, whilst such detailed contouring may give an illusion
of high parameter resolution, it cannot in fact contain more information than
was present in the original data. Thus, if patterns begin to emerge in such
contours they must be regarded as artefacts of the filtering process unless they
include a contour interval of at least the digitisation interval.
Images from the Landsat MSS have a characteristic 6-line stripe effect,
produced because the sensor scans six lines together with six separate, parallel
sensors with slightly different characteristics. De-striping algorithms are available
to remove this stripiness by a statistical technique, but it is found that whilst
this is adequate over land, such algorithms may introduce more stripe noise into
over-sea parts of an image. For oceanographic applications, it has been found
that a 6 X 8 moving average filter is the most satisfactory way of removing the
stripes and generally smoothing the data before further atmospheric correction
or geophysical calibration processing is performed.
Filters can have uses other than smoothing, and a variety of different types
are described in advanced image-processing texts and proprietary software
manuals. One application is edge-enhancement (that is, highlighting sudden
changes in the parameter values as would occur at a coastline in a near-IR image,
at a front on a thermal IR image, or where a surface slick, or wind shelter,
produces a much smoother patch of sea surface on a radar image). Another use
is to emphasise regions of increased parameter gradients either in a particular
direction, or in general. Other techniques such as Sobel and Pseudoplastic
filters can create some fascinating visual effects, but it is not clear how much
apphcation they have to satellite oceanography. The available image-processing
software tends to be dominated by land applications, and it should not be
assumed that it can necessarily contribute much to oceanography, although there
is room for further experimentation.
124 Principles of image processing [Ch. 5

5.7 MEASUREMENT OF SPATIAL VARIABILITY

The developments most useful to oceanography which have occurred in image
processing are those which enable quantitative measurements to be made of the
spatial features which are apparent in imagery. Simple processing to emphasise
spatial patterns on an image can be of some use, but it is scientifically more
valuable to be able to measure the spatial length scales on an image. Thus if
gyres and eddies are present in a visible or infrared image, it would be of value
to the development of dynamical theories to be able to extract the typical gyre
radius, and its variability over the image. CZCS imagery can reveal a general
turbulent structure which may be due to variability in the phytoplankton
productivity, and in this case it is not easy to pick out the dominant length
scale simply by measuring with a ruler on a hardcopy image, as is sometimes
possible when dealing with isolated ocean eddies. For these applications the use
of spatial Fourier transform algorithms can indicate the distribution of length
scales of variability over an image. By choosing a suitable algorithm, one can
obtain length scales of variation up and down or across the image, in a specified
direction or direction-independent. For turbulent eddy structure the direc¬
tion may be unimportant, whereas if plane wavelike features appear on an image,
it is most appropriate to measure the wavenumber distribution in a direction
normal to the wave crests. To obtain useful oceanographic information it may
be necessary to isolate the sea area containing the wave patterns, or the eddy
features, before applying the spatial analysis algorithms. Otherwise a feature
which is genuinely present in the image becomes insignificant in the spatial
wavenumber spectrum of the image because it is masked by the variability in all
the other features. Our brains tend automatically to filter out or ignore irrelevant
patterns when we look at the whole scene, whereas the computer must be
directed to concentrate on the relevant area. Thus although regular spatial
structure is often very obvious on satellite images of the sea, there is a need to
develop ways in which meaningful numbers relating to spatial variability can be
extracted objectively with a minimum of subjective image enhancement. Recent
advances in automatic digital extraction of directional wavenumber spectra from
SAR images point the way forward in this area of satellite oceanography, but as
yet little has been done to apply these techniques to visible and infrared data.
A few examples will be mentioned in later chapters.

5.8 ENHANCEMENT OF SINGLE-BAND RADIOMETRIC INFORMATION

We now consider image-enhancement techniques which help to produce clear
pictures of oceanographic features, and are therefore a helpful stimulus towards
creativity and the development of new scientific ideas, but are of less importance
in quantitative ocean science from satellites. It has already been mentioned
that the variability in the ocean part of an image may be small compared with
Sec. 5.8] Enhancement of single-band radiometric information 125

the fuU range of pixel values available. The result of this in a monochromatic
image is to make the sea look an almost uniform grey. If there are 256 digital
levels available, and these are distributed evenly over a grey scale from black at 0
to white at 255, a difference of one or two in the pixel value results in a grey¬
scale difference which is too small to be resolved by eye. Processing algorithms
known as contrast stretches are able to enhance the contrast in one part of the
image, at the expense of contrast in another part.
Contrast-stretch techniques are best understood in relation to the image
histograms which in themselves can carry some useful oceanographic informa¬
tion. Fig. 5.10 shows a typical histogram for a single-band visible wavelength
image. It will be seen that apart from a significant number of zero pixels, which
may be data dropout points, and a peak at 255 corresponding to cloud, most of
the pixels lie in the range 50 to 150. The cumulative histogram, showing the
number of pixels having a value less than or equal to a given level, shown in
Fig. 5.11, gives a clue to how an automatic contrast stretch can be applied. The
shape of the cumulative histogram is fitted between 0 and 255, and the resulting
curve (Fig. 5.12) is used to re-value each pixel. This is achieved on the computer
by creating a look-up table (LUT) to which each original pixel value is referred
before displaying it on the screen with its new value. It is easy to see that the
cumulative histogram for the modified image will be a straight-line ramp through
the origin (Fig. 5.13), and the actual histogram will be uniform (Fig.5.14). This
has the effect of spreading all the variabihty in image density evenly across the
range of grey levels, and it is therefore called a histogram-equalisation stretch.
It is clear from Fig. 5.12 that the range between 50 and 150 on the original
image which contained most of the pixels has now been stretched out to fill the
values between about 30 and 220, whilst the other parts of the full range have
been correspondingly compressed.

Fig. 5.10 - Typical image histogram Fig. 5.11 - Cumulative histogram of Fig. 5.10
126 Principles of image processing [Ch. 5

Fig. 5.12 - Calibration curve for histogram Fig. 5.13 — Cumulative histogram of
equalisation contrast stretch. image of Fig. 5.10, recaUbrated using the
curve in Fig. 5.12

Fig.5.14 - Histogram of recahbrated image

Such a stretch is easy to apply to an image, and if it is available on an image-

processing computer it is invariably used for quick automatic enhancement to
gain a closer look at the structure of a scene. However, it does have drawbacks in
oceanographic appUcation. Firstly, it is a nonlinear stretch, and the use of the
curve in Fig. 5.12 to provide a LUT makes it very difficult to relate the new
values to the old. The danger of losing track of the original data values was
pointed out at the beginning of this chapter. It is of course possible to work
back through the LUT to retrieve the original values, but once an image has been
enhanced and a hard copy produced, it is easier to believe the evidence of one’s
eyes and make value judgements based on the grey levels in an image than to
consider the actual values that those grey levels represent. One useful visual
reminder of the radiometric transformation which has been applied is achieved
by including on all original data sets a grey-scale wedge across one edge of the
image. The algorithm to do this is usually supplied with image-processing soft¬
ware packages. Whatever stretching is applied to the satellite data is also applied
to the grey scale. It is then possible to compare grey levels in the image with
Sec. 5.8] Enhancement of single-band radiometric information 127

their new position along the grey scale, and make an approximate estimate of
original parameter values. The other drawback of the automatic histogram-
equalisation stretch is that it may enhance a lot of variability on the image in
which the oceanographer has no interest. Thus, cloudy parts of the image are
enhanced to show different cloud shading, or land parts appear much clearer,
whilst if open sea covers only a small part of the whole image the automatic
contrast stretch may in fact compress the sea variability into a smaller range of
values. This can be avoided by masking out the land and cloud as discussed in
section 5.5 and applying the equalisation techniques to the histogram of the sea
pixels only.
In contrast to the nonlinear histogram-equalisation stretch, the linear
stretch applied to a selected part of the parameter range is slightly more time-
consuming to apply, may not result in such a good contrast being achieved over
the whole image, but is nevertheless more useful in most oceanographic applica¬
tions. Fig. 5.15 shows a typical example of the recalibration curve for a linear
stretch. In practice it is possible with proprietary software to make up a variety
of different stretches, to suit different applications, but the linear one has
the widest use in oceanography. The range between /jnin /^ax is stretched
to fill the whole 0 — 255 range. Outside and all pixels are set to 0 or
255 respectively, and therefore any variability in this range is lost. The art of
applying the linear stretch is to keep /max~-^min as small as possible, but to set
^min and to include all the parameter values for sea pixels. In practice,
Iinax

the interactive application of a binary image-masking technique can readily

reveal the appropriate values for and /^ax-

Fig. 5.15 — Recalibration curve for a linear stretch

Alternatively, the histogram of the original image can be consulted when

selecting and /max- Fen example. Fig. 5.16 shows the principal features of
the histogram which might be encountered on a summer thermal IR image of
the British Isles. Three distinct ranges are to be seen, corresponding to the warm
128 Principles of image processing [Ch.5

land (darkest pixels on a thermal IR image), the cold clouds, and the sea with an
intermediate range of temperatures. On the basis of such a histogram it would be
safe to choose and /^ax shown, to enhance the sea-surface temperature
variability patterns. Unfortunately, not all histograms are as clear-cut as Fig. 5.16.
The land and sea may be at similar temperatures, in which case masking using
another channel may be necessary. Moreover, there is usually a significant
number of pixels with values between the main groups, which can be explained
as those which contain both cloud and sea, land and sea, or land and cloud
within the IFOV. These may make it difficult to choose appropriate limits
for a hnear-contrast stretch. They also, incidentally, introduce errors in the
estimation of sea-surface temperature, as will be discussed further in Chapter 7.

Fig. 5.16 — Typical histogram of a thermal IR image of the British Isles taken in
summer daytime when the land is significantly warmer than the sea. Note the
distinctly separate land, sea, and cloud bands in the distribution

Having performed a suitable contrast stretch on the data, the details of

spatial structure should show up more clearly in the new grey-tone image. The
eye does not find it easy to compare grey tones, particularly from one side of an
image to the other across intervening areas at other grey levels. For presentation
purposes it is often helpful instead to substitute different colours for different
grey tones in what is termed a colour-density slice. With some systems, the
number of colours is limited, and the grey-level range must be divided into
discrete bands for assignment to different colours. In other systems the colour
can be varied over at least 256 different combinations of hue (the balance
between the primary colours), depth of colour (the extent to which the colour
is mixed with white), and brightness. This permits the grey range to be presented
in colour without loss of grey-level resolution. Different combinations of
colour suit different applications and are a matter of personal taste, although
certain colour conventions suggest themselves, such as selecting red for warmer
sea-surface temperatures ranging through to blue for cool water.
Sec.5.9] Presentation of multichannel image data 129

5.9 PRESENTATION OF MULTICHANNEL IMAGE DATA

The use of colour-slicing techniques to present single-channel radiance-density
information should not be confused with colour images which contain multi¬
channel data. Multispectral imaging sensors generate several images with
essentially the same pixel positions, but containing information on radiation
in different wavebands. The simplest way of analysing them is to treat each
channel separately, employing appropriate enhancement techniques, and then to
compare the final images. Such an approach gives at best a qualitative analysis,
and may miss a lot of the information about the ocean parameters which can be
extracted from multispectral data.
Because the pixels of different channels correspond geometrically it is
possible to utilise image-processing techniques to present the multispectral
information. We have already seen how land-masking for one channel may be
defined from another because land and sea are revealed differently in different
channels. The same may be true of phenomena in or on the sea. For example,
in visible and near-visible channels the ratio of sea reflectance between two
channels may differ according to whether phytoplankton or inorganic sediments
are the dominant scatterers. Therefore the calculation of ratios of pixel values
between different channels can be oceanographically meaningful. The ratio of
two channels, pixel by pixel, produces another single-channel image which can
be enhanced and presented in a grey-tone or colour-sliced image.
When three channels are being examined together, a single ratio is not so
helpful, and in this case it is customary to produce a false colour composite. The
data from one channel is placed on the image-display screen as a monochromatic
‘grey-level’ image in one of the primary colours, and the data from the other
two channels is overlaid in monochromatic images, each in one of the other two
primary colours. It is customary with near-visible waveband imagery to display
the lowest wavelength band in blue, the next in green and the longest wavelength
data (normally a band in the near infrared) in red. The colours are therefore not
true colours unless the wavebands used happen to centre at green, blue, and
red in the visible spectrum. The changes of hue, colour depth, and brightness
in the resulting image convey information about the different spectral responses
of different areas of sea. The technique is a powerful one for land studies.
Algorithms are available to classify pixels in terms of ther spectral characteristics,
and thereby it is possible to identify different regions having the same land-cover
or geological properties. In the oceans, classification techniques are not so useful.
The value of the false-colour composite is that it shows up spectral as well as
spatial variability in the ocean which is not so readily seen by comparing three
grey-tone images of different bands. If quantitative use is to be made of spectral
information in oceanography, it is usually necessary to study in more detail
those areas which look interesting on the false-colour composite, and examine
the actual water-leaving radiances by returning to the original satellite data
for certain pixels and applying atmospheric corrections. Because atmospheric
130 Principles of image processing [Ch.5

effects, as well as any preprocessing contrast stretching of individual bands, will

alter the colour ratios in a complex way, it is not usually very fruitful to seek
a precise interpretation of the actual colours which emerge on a false-colour
composite. Rather the false-colour composite can be used to provide an overall
spatial perspective of the colour changes (and hence an approximate measure of
water-parameter changes) occurring across a sea area.
One way of manipulating multispectral data in order to extract colour
(that is, spectral-ratio) information rather than absolute brightness levels is to
apply a chromaticity transform. Thus, if for a given pixel f, and /s are the
values for channels 1, 2, and 3, the data are transformed into image datasets of
X\, A"2,.and where

h
-
Ii +12+ h
h
/j + /2 4- It,

h
^3 =
I1 + I2+ I3

This approach has only limited value in oceanography, having been applied only
to Landsat images of coastal and estuarine waters.
Another technique which is used extensively in land application to
emphasise spatial structure in images is principal-components analysis. This
applies a linear radiometric transformation to N spectral bands to produce N
transformed orthogonal image datasets in such a way that the variability in the
image of the first transformed band (the first principal component) is maximised.
The second transformed band maximises the remaining variability, and so on.
This sometimes has the effect in ocean images of enhancing the land —sea —cloud
contrast in the principal component, since these features give the greatest vari¬
ability in radiance across the image, whilst the oceanographic features of interest
appear in the second or third components. Because it is a computer-controlled
transformation, which separates the user from the original satellite data values
in a complex manner, the quantitative interpretation of an image treated by
principal-components analysis is obscured, and its use in satellite oceanography
appears to be limited. Indiscriminate application of the technique is certainly
not recommended. A more thorough description of the principal-components
analysis technique can be found in the standard works on image-processing
mentioned at the beginning of the chapter.

5.10 ATMOSPHERIC CORRECTION AND GEOPHYSICAL

CALIBRATION

In the oceanographic interpretation of satellite data, although the spatial struc¬

tures are important, and indeed are the unique contribution of satellite remote
Sec. 5.10] Atmospheric correction and geophysical calibration 131

sensing to oceanography, it is still important to interpret the data in terms

of ocean parameters. This usually requires the application of an atmospheric
correction and a calibration algorithm as was discussed in Chapter 4. The multi-
spectral nature of the data contributes significantly to the eventual extraction
of ocean variables. At present the correction and calibration algorithms require
more mathematical manipulation than is available on image processors, and
to some extent the use of image-processing computers to analyse and display
raw satellite oceanographic data can be a diversion away from useful scientific
activity. If their result is a colourful picture of doubtful oceanographic interpre¬
tation, this is certainly the case. On the other hand, if the image analyst is able
to enhance the image and keep track of the original satellite data values, it
may be possible to apply an approximate calibration to the whole of an image.
For example, if it is possible to assign a sea-surface temperature to a particular
data value of an IR image, or a near-surface suspended sediment concentration
to a particular single-channel visible-wavelength image-data value, then a colour
density-sliced image is capable of being interpreted as a contour map, by defining
the colours in terms of ranges of the ocean parameter.
Furthermore, if a multispectral data set has undergone a full atmospheric
correction and calibration to produce a spatial dataset of the oceanographic
parameter, image-analysis techniques can be used with confidence to display the
oceanographic information and to extract spatial measurements from it. The
attractive promise of a new generation of image-analysis computers is that they
may be able to perform the atmospheric correction and calibration operations
with the same convenience and speed as they can at present perform the simpler
processes described in this chapter.
The validity of such calibrations depends on our scientific understanding of
the mechanisms for the remote sensing of different oceanographic features. It is
to these that we turn in more detail in the remaining chapters of this book.
. M«'.-ivf«‘>; .'■» '^ir.v<^''-« J?'-'4ft->((|l6i *; Wi ^en-t^^KbgCrfWId'. •■'< -.tA'*:•-'<■
w'vv:'.-;.-<.ifvr; ick - u^
**(';& 1 ««iii. yS'? “ /ih'??i»<&#aii' Aif-^TiflthS • :-
W'i'^ '»?'»> *-*Hi 1*»'- ■'•-
'f?**’• ^. .r.f ••.

'■' '*• 'irf- •Vf^W9:VfTfo-<,i#-

rf>. ^> ' ^;<t S Jvtr?i»A

’« Ji- ;4>*l* -^IrtiJi ’-'ll •■ •yiwrt.'*^ til ^

■•j^u'4 /-.») 1o y-.inib. ^ii Cl jitiiitn4JifiJ .-i r.<.}nr, \rf ».,«ii^..-*-ct •/

p. *wt»<ri --^ fc.q^it 0’ ^ ,*^mJV5 •'"'H
a^iin^'fi'w ^s»«i« H< la'ui
i r <!?*>,‘ii-ij . •■rfiffitf ' i«r’/iftrti-^t)3'-.i. •' 01
'i;i^^tth l^ aidr.fjK: >i ’ <',4-v k4j\«?

-frrtfc iV -i .’n'-^aiw^sui •I'S* 4i

.v(fl^iitj;..u>» c- «7 ttsifVit iisTfjJi* j v.yl'/<A<|;jfr» -i .*» <ifJ<4 ■ 1* teo.r-jr-^i?-

/'.MTrase’"

, jmit^ , - ’ ■ r».^ »iw ’ •• *•••*» I r-V

*• '1. ■' ’• ^•' • • wcv . ' '»■' m:Mpr^ w -^^wm r'V:.i>

tK^sl •*,<'.tj.< .j-»y'tisi^ 'l: 4^l<^S}iu , TW-ft'ij pj f*refc} Wi.Ti'rt

.»v - V i--'-- >-,. -, ^ - ,x

•■'■ :- ^ i' -' ’"r*--.^^ _£3C ■? *
^•. w^V %i I
. Iff?-! • f» •■ i{.;b- ■i-.n/'-'t* •
►' * ' 4 rC ».* :\i fc jr» »

' «■■ ‘.i '• <i ^r-T-t.4 ‘tf. * 'a -O 'i»- *•.. ‘

* >./»;. tivf<yA> J*-

>'??MV^ ''^.tJ''4;'-t t' V ■■.^r i» II tie- % !«•%>
—*lit> lit -. ^ . ^ - ■_-,

'.r. .^ ."ilftii • i. «> 4p . n .■; at-.* wr'o&ICLVA

.. •i,X '■ ’ ';^^
f
•j '♦ M, ^-y4i-4>.v i'H. ij-''?tv•^•o.'•. *1^"*“’; <»,4v, >|f
; i»4» .) ■' .:,i^^i“^ .-1: *?- ; ‘»4^rf4! "* ' c»^
 SECTION B

OCEANOGRAPHIC APPLICATIONS
OF SA TELLITE REMOTE SENSING
CHAPTER 6

Visible wavelength ‘ocean-colour’ sensors

6.1 INTRODUCTION
6.1.1 Scope
Of all the techniques used in ocean remote sensing, the observation of ocean
colour from satellites is perhaps the most easily understood in concept. For
as long as men have sailed the sea they have noted its characteristic colour
and sought to navigate, to fish, to chart its depth, or to seek out its riches by this
means. In the twentieth century the science of optical oceanography has grown
up in order to assist the study of water quality, and in particular the measure¬
ment of the primary productivity of planktonic organisms, which can be detected
by underwater optical observations. Most optical oceanography has concentrated
on subsurface measurements of scattering, absorption, attenuation, and their
variation with spectral wavelength. Apart from such easily-observed phenomena
as ‘red tides’ associated with certain plankton blooms, shipboard observations
of water colour from above the sea surface have not contributed very much to
marine science. However, since the use of colour photographs or scanned images
from aircraft and satellites has provided a wider spatial perspective to define
larger-scale structures in ocean colour, oceanographers have begun to see the
scientific, quantitative, possibilities in studying the colour of the ocean as viewed
from high above the surface.
In this, the first of the chapters devoted to individual techniques of
oceanographic remote sensing, we look in some detail at those sensors which
operate in the visible part of the electromagnetic spectrum. The principles of
atmospheric and underwater optics are introduced before considering the way
in which scanned ocean-colour data are processed, atmospherically corrected,
and analysed for oceanographic interpretation. Now that satellite ocean-colour
data have been received for more than a decade, there is a variety of application
areas to describe, making contributions to both the commercial and research
interests of marine science.
It is appropriate to consider visible-wave length remote sensing first, not
only because it was historically the first method developed to study the ocean
[Sec. 6.1] Introduction 135

from space but also because it is the only technique which penetrates beneath
the sub-millimetre surface skin and ‘sees’ directly into the surface layers of the
sea down to a depth of several metres.
For further reading on satellite observations of ocean colour, the 26 papers
on the subject in Gower (1981) give a broad coverage of recent research, and
reviews have recently been written by the author (Robinson 1983) and by
Gordon & Morel (1983).

6.1.2 Colour sensors on Landsat

Originally called the Earth Resources Technology Satellite, and renamed Landsat
in 1975 to distinguish it from the Seasat programme, this series of satellites has
been designed primarily for the monitoring of earth resources on land. Tables 2.1
and 2.2 list the principal orbital characteristics of the first four satellites
launched in the series. The first three had similar orbits with an 18-day repeat
cycle, and during the overlap of their lifetimes a 9-day repeat was achieved.
Landsat 4 and the most recently launched Landsat 5 have a lower orbit with
a 16-day repeat cycle. They are all in a near-polar orbit extending to 81° latitude
(see explanation of satellite orbits in Chapter 2).
The Landsat sensor which is of most interest to oceanographers is the
Multi-Spectral Scanner (MSS) whose characteristics are defined in Table 6.1.
It has four broadband radiometers measuring the radiance in the visible and
near-infrared wavelengths. The bands are determined by absorption filters
whose spectral characteristics are illustrated in Fig. 6.1, and were not designed
with any oceanographic properties in mind as were more recent ocean-colour

Table 6.1 Characteristics of the Landsat Multi-Spectral Scaimer (MSS)

Band 4 Band 5 Band 6 Band 7

Nominal wavebands,/am 0.50 -*0.60 0.60-* 0.70 0.70-*0.80 0.80-*1.10

Saturation radiance, Wm ’ sr ' 24.8 20.0 17.6 46.0
No. of digitisation levels
(after processing) 128 128 128 64

Radiometric sensitivity
(noise equivalent reflectance %) 0.57 0.57 0.65 0.70

Angular field of view 0.086 m rad

Ground IFOV at nadir 76 m X 76 m
Blur circle 30 m
0
00
CA
1+

Maximum scan angle from nadir

Length of scan line (swath width) 185 km
No. of samples in scan line (approx.) 3240
Approximate pixel size at sub-point 81 m X 57 m

Data rate 15 X 10‘ bps

136 Visible wavelength ‘ocean-colour’ sensors [Ch.6

Fig. 6.1 — Nominal spectral responses of the Landsat MSS visible and near-infrared
channels (from NASA Landsat User's Handbook)

monitors. The scanner operates in the way described in Chapter 2, but because
the total field of view scanned is so small (less than 12 degrees) the scan mirror
oscillates rather than spinning right round. (On Landsats 1 to 3 the oscillation
period was 33 ms.) Another feature of the MSS is that in order to gain the high
resolution of 79 m scan-line width, without leaving large gaps between successive
scan lines as the satellite speeds on its way, it is necessary to scan 6 hnes simul¬
taneously, with 6 separate parallel sensors. Of course it is impossible for the
radiometers and filters to be exactly identical, resulting in a tendency for 6-line
stripiness to appear on the images reconstructed from MSS data. The other
pecuharity of the Landsat MSS is the way in which the scan samples are
obtained. Although the IFOV is nominally a square of 80 m sides at the satellite
sub-point, the radiometer is sampled while the mirror scans along the line at a
rate equivalent to a distance of 56 m at the sub-point. Fig. 6.2 illustrates the
overlap which consequently occurs, and the result is that the nominal pixel size
when reconstructing the image should be 56 X 79 m. This rectangular rather
than square pixel shape must be allowed for by appropriate image-processing
techniques. A display system which is always used for Landsat images can be
adjusted to generate rectangular pixels itself. If square pixels are displayed on
the screen a simple way of approximately removing the distortion is by sampling
every 3rd scan sample, but every 2nd line. If a full-resolution image is required
without loss of data, using a square pixel display, the whole image dataset must
be resampled using the techniques described in Chapter 5.
Sec. 6.1] Introduction 137

PIXEL
IFOV WIDTH

79m

iiI 1-- 56m [ I6 6m- ■5em

k-— 7 9m-►!

Fig. 6.2 - Nominal pixel size of Landsat MSS compared with its IFOV

Since the swath width is 185 km, the data stream is framed into ‘scenes’
which are also 185 km long in the satellite track direction, but with 10% overlap.
Each scene therefore has 2340 scan Hnes, and 3240 pixels per hne, that is,
7 581600 pixels per channel, or over 30-miUion separate radiance observations.
Such high resolution, resulting in so much data, can be an embarassment to the
oceanographer who may be interested in much larger length scales of ocean
variability than the 80 m which is still often too coarse for land appHcations. For
this reason, and also since positional registration of a scene is difficult without
landmarks, Landsat is in fact of much less value over the open ocean than in
coastal and estuarine regions where the 80 m resolution can be exploited and
where coastal topography facilitates position fixing. The archives of Landsat
imagery have mostly been compiled with respect to the requirements of land-
use applications, and therefore there are few scenes available which do not
have some land on them, that is, there are few open-ocean scenes available.
A further serious drawback for oceanographic applications is that because of
the narrow swath width of 185 km and only a small overlap with the next day’s
adjacent overpass, the repeat period for imaging the same point is in general
18 days. Given the problems of cloud cover in many coastal and sea areas,
the frequency of useful scenes may be reduced to one or two per year — of
apparently little value in the study of continuously-variable dynamical processes.
Landsat MSS data therefore tend to be viewed by oceanographers as
‘second best’ compared with the observations gained from satellites and sensors
designed primarily for ocean applications. Nonetheless, nearly a decade of
archived Landsat data covering estuaries and coastal seas is now available, much
of it as yet unviewed and very little of it having been processed for atmos¬
pheric and geometric corrections. Despite the drawbacks mentioned above there
is considerable scope for extracting oceanographically-useful data from this
still-growing archive. Some of the marine application areas which have begun to
be exploited are discussed in section 6.4.
Landsat 4 marked a new generation of Landsat technology, and although it
carries an MSS essentially the same as the earlier ones it also carries the first of
138 Visible wavelength ‘ocean-colour’ sensors [Ch. 6

O
to o o
<N fN
(N o
NO lO X X
11 to
(N o
£ E
O o o
1-H fN rs

to
CO
CN NO o
1 to
cs (N fS
00
o
<N
Table 6.2 Characteristics of the Landsat Thematic Mapper (TM)

to

VO
1 to
to cs
to

o
ON
o NO
1 to
NO <N

83 X 10* bps
ON
NO O
m O
o NO
E o m
o
1 to
o
X lo <N X
CO
NO
tN
E +1 00 VO E
o o
o ro m
o
VO
o NO
1 to
(N <N
to
d

(S
to
d V£)
1 to
to CN

OCI
•o o
.p I-t
D.
T3 Oh
o 2
C
cd C w
i-t
*->
o 0) E
O
Oh o cd
G
P3
E
:t G(U E(U .in
Ocd
CA T3cQ
Ij) ot/i o
c a c c ccd o
o ao a> ■S O)
I

ccd
II
C/3 1 > G
cd p
U 3 o o o o
60 *s D a* cd
o D a> (D 6 E
Data rate

(l>

E c/}
3 5 .ac/)
•a .B o .2 ’o c <«-■
c c p 60 o
*3
E
o o o
I-t
X
ed
Ca> 6 .S
n Z 2 O s H-] 2 Oh
Sec. 6.1] Introduction 139

a new design of earth-monitoring sensors, the Thematic Mapper. The principal

design details of this are presented in Table 6.2. It has improved radiometric
and spectral resolution with slightly narrower bandwidths and a pixel size of
only 30 m. Its oceanographic potential has yet to be explored, but to some
extent it is closer in concept to the CZCS which was designed specifically for
marine science appHcations.

6.1.3 The Coastal Zone Colour Scanner on Nimbus 7

The Nimbus series of satellites has principally been used to develop new sensors
and remote-sensing techniques rather than to provide an operational service such
as the TIROS and NOAA meteorological satellites. Thus Nimbus 7 was used to
carry the first visible/near-visible-wavelength scanning radiometer designed
primarily to observe ocean colour, known as the Coastal Zone Colour Scanner.
The principal orbital characteristics are given in Tables 2.1 and 2.2, and sensor
details are presented in Table 6.3. Like Landsat and the NOAA satellites.
Nimbus 7 is in a near-polar orbit. The much greater total scan view, or swath
width, of 1636 km ensures that successive orbits overlap, providing for at least
one daytime pass per day over every point on the earth. Being an experimental
rather than an operational sensor, the CZCS has not always been switched on,
and there has been no commitment to a continuous archiving of data, nor of
full global coverage. Nonetheless, the satelhte has long outlived its design life,
and in June 1984 is still supplying CZCS data, over five years after its launch,
so that an impressive archive of data is available for analysis and oceanographic
interpretation.
As displayed in Table 6.3, the bandwidths (using interference filters) are
much narrower than the Landsat MSS, and the detectors are more sensitive,
being designed for the relatively low radiance of sea surfaces and therefore often
saturating over the land. To further improve the resolution in low-illumination
conditions the gain of the amphfier in the analogue-to-digital convertor has
switchable values. The wavebands were chosen with certain properties of
chlorophyll absorption in mind, as will be discussed later. In addition, the look-
angle of the sensor (that is, the angle between the vertical and the plane of the
scan line) was designed to be tilted forwards or backwards 20 degrees as well as
straight down, to avoid any problem of sun glitter from the sea surface which
could arise with the sun-synchronous orbit having a daytime overpass at local
noon. In many respects, therefore, as a sensor designed especially for oceano¬
graphy the CZCS contrasts sharply with the Landsat MSS. A further description
of the sensor and early results from it will be found in Hovis et al. (1980) and
Gordon et al. (1980).
Although no operational colour-monitoring satelhte is yet promised by
the various space agencies, it is expected that the Americans will fly a CZCS
successor within a few years, whhst the European Space Agency, the French, and
the Japanese ah plan to launch ocean-colour-monitoring sensors within a decade.
140 Visible wavelength ‘ocean-colour’ sensors [Ch.6

s
3.
lO
cvi
VO
T
IT)

o
o
00 On ON
lO cn ro
o (N (N
t/5 o
u
N
QJ
u
00
C +1 ro 00

g o (N
u
VO
<-i
3 'O
cd
(N

1636 km
O
u
+1
VO NO B 00 0ON 00
a> 00 (N in m X
m VO
fi o a^
o c4 NO CN (N
S +1
iri
lO
00
N
O (N
"a
-w 00
(/)
n
o O
u +1 NO
(N
a> O m

.a NO
i-i +1
V
■<-»
ro
(3 ■vt
JS
u

so
(when not tilted in track direction)

.S .s
cd
n CU!) tiO
H
><
cd a>
s c
33
cd
o C C
.q cd
o
<D cd S c/5
O 3^ C o Ui
c c *> th!
M-H O
o 4-. Oh
a.2 o
cd
C/5
C cd cd
c/5 > 3)
o c C4 43
I
b
cd c
T3 c:
o s ]3) C q TJ
O C o cd cd
Z nJ V-> u*
cd o c/5 '$
Ol Cd c 00
T3
c
4) Vh
3
-<-»
cd
I o
d
3 :3
oc o
q t-M
X
cd
o
d
43
"S
CQ C/5 Z < o s Z V3
Sec. 6.2] Optical theory relevant to ocean-colour viewing from space 141

6.2 ASPECTS OF OPTICAL THEORY RELEVANT TO OCEAN-COLOUR

VIEWING FROM SPACE
6.2.1 Definition of some optical quantities
For the oceanographer and the space scientist alike, the literature of visible
wavelength remote sensing presents an array of new terminology. Some of
the terms are defined here with an explanation of their significance in remote
sensing. A more thorough treatment of the whole subject of optical oceano¬
graphy will be found in Jerlov (1976), whilst Slater (1980) treats remote-sensing
optics in detail.

Table 6.4 Summary of quantities, units, and conventional symbols

used in optical remote sensing of the sea

Customary
Quantity Definition symbol SI unit

Radiant energy Q J
(joule)
Radiant flux or power </) W
dr
(watt)
Radiant energy density w J m'^
dV
(V = volume)
d0
Radiant flux density at a surface
dA
Irradiance (landing on surface) E Wm-2

Exitance (leaving surface) M Wm-^

d(j)
Radiant intensity I Wsr-^
dco
(co = solid angle)
dl
Radiance L W sr"* m"^
d(^cos0)
1 dE -1
Diffuse attenuation coefficient K m
E{z) dz

Table 6.4 lists the quantities, their units, and their normal symbols, which
are used in the measurement of visible-wavelength electromagnetic radiation. As
presented they are total quantities integrated across a broad spectral range whose
limits are assumed to be implicit in the context of their use. Spcctful (juuntitics
can be represented by restricting each to a narrow wavelength band. This is
normally denoted by the use of subscript X, and the units must be altered by the
142 Visible wavelength ‘ocean-colour’ sensors [Ch.6

addition of or nm"^ to convert the quantity into a spectral concentration,

that is, per unit spectral bandwidth. Thus, for example, the spectral radiant
energy in a nanometer-wide spectral bandwidth centred at wavelength X is
denoted by J nm”h
Radiant energy, and radiant flux or radiant power are total quantities
whose precise limits will be obvious from the context. Thus the radiant power
of a light source will be the flux of all light energy radiated from the source, in
all directions, and irrespective of the spatial size of the source.
Radiant energy density is the amount of energy to be found in a unit
volume.
Radiant flux density is a quantity commonly used in remote sensing, and it
is the energy flux intercepted per unit area of a given plane surface (Fig. 6.3).
Thus for a total flux <E> intercepted by a surface of area A, the average radiant
flux density is £" = ^/A. The symbol E is used to denote flux arriving at a
surface, or the irradiance. For energy flux emitted by a surface, the symbol M
is used, and this is called the exitance, or in some literature the emittance. E and
M do not specify the direction of the radiation, since they integrate the total
flux over a hemisphere. If the energy flux is stronger from a certain direction,
the irradiance will of course vary according to orientation of the plane, and is
therefore to be understood only in connection with a given plane surface. In the
sea, it is helpful to discuss the radiant flux density passing through a horizontal
plane at a particular depth below the sea surface. Flux passing down into the
lower part of the water column is described as the downwelling irradiance E^,
whilst that which has been backscattered in the lower water and is travelling
towards the surface is the upwelling irradiance E^J (see Fig. 6.4).

AREA, A

Fig. 6.3 - Definition sketch; radiant flux density, E or M

Sec. 6.2] Optical theory relevant to ocean-colour viewing from space 143

Sea Surface

area A

T T \ ' ''
<t>u
Fig. 6.4 — Definition sketch; downwelling and upwelling iiradiance, and

Radiant intensity, I, introduces the concept of the direction in which light

energy is travelling. It is the radiant flux per unit solid angle leaving a point
source in a given direction, and hence has the units of watts per steradian.
A steradian (sr) is a dimensionless unit of solid angle, analogous to the radian
measure of plane angles. The solid angle of a cone in steradians is defined as
the area subtended by the cone on the surface of a sphere whose centre is the
apex of the cone, divided by the square of the radius of the sphere. In Fig. 6.5,
the solid angle is AjR'^. The total sohd angle of a sphere is therefore 47t, and
an isotropic (uniform in all directions) source of radiant intensity / therefore
emits a total flux $ = 47r/. Radiant intensity can only have a meaning in the
context of point sources, or in calculations in which the source can be effectively
considered as a point source.

A
y 7! flux 6

Fig.6.5 - Definition sketch: radiant intensity, /

The radiance, L, is the optical property appropriate to light energy leaving

an extended source. It is obviously a parameter of great relevance to remote
sensing, since a satellite measuring the radiation back-scattered in the sea views
144 Visible wavelength ‘ocean-colour’ sensors [Ch.6

the ocean surface as an extended light source. Fig. 6.6 illustrates the definition
of L, which is the radiant flux per unit solid angle in a given direction, per unit
projected source area in that direction. The units are therefore W sr“\ By
introducing the source area into the definition of radiance, the concept of the
brightness of an object is represented. Thus two sources of different size which
emit the same total flux isotropically will, at great distances, appear to have the
same radiant intensity, but viewed from nearby the larger source will have the
smaller radiance, corresponding to the fact that it is not as bright as the smaller
source. A Lambertian source is one in which the radiance L is independent
of the angle from which it is viewed. The assumption is normally made in the
viewing of ocean colour from space that the sea is a Lambertian source in
terms of the upwelling of backscattered light from below the surface. For a
Lambertian plane surface, L = Fu/rr.

Fig. 6.6 — Definition sketch; radiance, L

The Diffuse attenuation coefficient, Kfk) at a point in a medium is

defined as _2

K{\) --^
E{\,z) dz
where z is the depth within the medium. K is specified as downwelling or
upwelling according to the use of or E^ in the expression.

6.2.2 Measurements made by a satellite sensor

A satellite radiometer is designed to measure the radiant flux <I> in a restricted
wavelength range, which passes through the aperture of the sensor from a narrow
cone of directions. Considering first the spectral characteristics of a sensor
designed with a bandwidth between Xi and Xj, the aim is to measure
Sec. 6.2] Optical theory relevant to ocean-colour viewing from space 145

«>(Xlt0X2) = $;,dX. (6.1)

In practice, the voltage output signal, S, of the sensor is dependent on the

spectral response of the sensor, which is related to the spectral characteristics of
any filters used as well as of the photoelectric conversion process in the optical
transducer. If the radiometer responsivity is R(X) then

5= /“$;,i?(X)dX. (6.2)
0

Fig. 6.7 shows a typical sensor response function. The stated bandwidth of the
sensor, (X —Xi), is normally calculated such that
2

(X2-Xi)/?peak = /”i?(X)dX. (6.3)

0

Fig. 6.7 - Typical sensor spectral response function

Given a signal S, it is then assumed that

<I)(Xi to X ) =
2 (6.4)
^peak

but consideration of (6.1) to (6.3) reveals that assumption (6.4) is not con¬
sistent. Fig. 6.8(a) illustrates that when is smoothly varying with X, the
estimation of $ in the stated sensor range Xi, to X using (6.4) is entirely
2

satisfactory. It is only when a spectrum such as that in Fig. 6.8(b) is encountered

that significant errors could arise. If phenomena are being studied which have
significant spectral variability, such as phytoplankton blooms in the ocean, then
the sensor must be designed with bandwidths sufficiently narrow to be com¬
parable with or smaller than the variability wavelength scale. Similarly, if spectral
information for atmospheric correction purposes is to be extracted from the
data, a broad bandwidth can lead to errors and ambiguities, particularly if, as
with band 7 of Landsat, the response function has a non-rectangular shape.
146 Visible wavelength ‘ocean-colour’ sensors [Ch. 6

assumed response

Fig. 6.8 — Effect of sensor spectral response in interpretation of sensor output,

(a) Smooth spectral variation of input. Incoming flux in nominal wavelength
range is estimated correctly by the assumed response, (b) Irregular spectral
variation of input. Incoming flux within the nominal waveband is seriously
overestimated by assuming a square response shape

Turning now to the geometric properties of the sensor, Fig. 6.9 illustrates
the limiting ray paths from a small sea-area element 6.A within the IFOV of
the sensor. The sensor has an aperture ^4^ and is distant from the sea surface.
It is viewing at an angle d to the surface normal. If the surface radiance is L in
direction 6, then given the definition of L, the radiant flux from 6.A which is
intercepted by the sensor aperture is

d$s L diA cosd — (6.5)

where is the solid angle subtended by the sensor aperture as viewed from

the sea surface. Although not necessary, it is perhaps easier to understand (6.5)
by considering dA to be very much smaller than Ag. The total flux intercepted
Sec. 6.2] Optical theory relevant to ocean-colour viewing from space 147

by the sensor from the whole of its IFOV, which has area A, is obtained by
integrating the contributions of all d^l. The angle d will change very slightly for
each of these, but for a sensor with a narrow angular field of view the variation
of 6 will be negligible. The solid angle subtended by the sensor will be constant,
and L may change very slightly with 6, but on the assumption that the surface
is Lambertian, L will be constant. Hence the total flux is

T>s = Lv4cos0 — . (6.6)

Rl

Fig. 6.9 — Viewing geometry of a satellite sensor

Now A, being the IFOV, will depend on the solid angular field of view of the
sensor, a^, the distance from the satellite, and the surface inclination to the
viewing direction, that is.

A COS0
(6.7)

From (6.6) and 6.7) we conclude

(6.8)
and with (6.4) we can evaluate the radiance at the sea surface (neglecting
atmospheric interference) in terms of the sensor voltage output signal S:
 Visible wavelength ‘ocean-colour’ sensors [Ch. 6
148

(6.9)

We note that S is therefore a direct measure of L, independent of the

distance between the satellite and the ground and of the angle subtended by
the ground to the direction of view (except insofar as L itself may vary with d
if the surface is non-Lambertian). This confirms the importance of the quantity
L in remote sensing. For a given L, the flux reaching the satellite is dependent
on Os and As. These must be designed to be as large as possible from the point
of view of improving the sensor sensitivity, but Oj is governed by the desired
spatial resolution on the ground, and is constrained by the satellite’s size.

6.2.3 Optical pathways in the atmosphere

The above discussion of how the satellite sensor measures the radiance of
the earth or sea surface neglects the effect of the atmosphere. In fact, the
atmosphere interacts with the light which eventually reaches the sensor, to such
an extent that only a small proportion of the radiance measured by the satellite
may be due to light coming from below the sea surface. Fig. 6.10 illustrates the
variety of possible pathways for light rays which eventually reach the sensor. The
labels on Fig. 6.10 represent the following rays:
(a) are light rays which upwell from below the sea surface and after refraction
at the surface point in the direction of the sensor. For this to happen they
must leave the water within the IFOV, but otherwise their previous history is
irrelevant. They may have travelled directly from the sun and been reflected
once in the sea, or they may have scattered in the atmosphere before penetrating
the sea and been scattered many times before emerging, but what matters is that
having left the sea they contribute to the water-leaving radiance;
(b) only a proportion of the rays (a) which contribute to Z-w reach the sensor;
(c) are those rays from which are scattered by the atmosphere out of the
field of view of the sensor;
(d) are those rays from the sun which reflect at the sea surface directly into
the field of view of the sensor. This is the sun ghtter;
(e) are rays which are scattered in the astmosphere before reflecting at the
surface into the sensor. This is the sky glitter;
(d) and (e) together contribute to Lj, the radiance just above the surface due to
all surface reflection effects within the IFOV;
(f) are the rays of which are scattered out of the field of view of the sensor;
(g) are the rays from which reach the sensor;
(h) are rays from the sun crossing through the field of view of the sensor which
are scattered towards it by the atmosphere;
Sec. 6.2] Optical theory relevant to ocean-colour viewing from space 149

Fig. 6.10 - Optical pathways to sensor (from Robinson (1983))

(i) are rays which are scattered towards the sensor by the atmosphere after
previous atmospheric scattering;
(j) are rays which have emerged from the water outside the IFOV and sub¬
sequently been scattered into the field of view. They do not contribute to Z-vv>
which by definition is the brightness of a particular area of sea;
(k) are rays scattered by the astmosphere towards the sensor, which have been
reflected from the sea surface outside the IFOV and therefore do not contribute
to Lp
(h), (i), (j), and (k) all contribute to L^, the atmospheric path radiance.
Thus if Ls is the radiance received by the sensor, it is made up of

Ls - Z,p + TL^ + TL, . (6.10)

Here T is that proportion of the surface-leaving radiance which is not scattered

out of the field of view, and known as the direct or beam transmittance of the
atmosphere. Table 6.5 gives a typical example of the relative percentages of
made up from Z,p, TL^ and TL^ at different wavelengths, in the case of clear
150 Visible wavelength ‘ocean-colour’ sensors [Ch.6

water and turbid water. The reflected signal is small in this example because the
sun elevation was low at the time of observation (36°) and therefore only sky
glitter contributed. It can be seen that even in turbid water the maximum
percentage of the satellite signal which contains information about the sea water
is 35%, at 550 nm wavelength (yellow light). It decreases gradually towards the
blue end and rapidly towards the red end of the spectrum.

Table 6.5 Typical % contributions to the signal received by a visible wavelength

sensor from below the sea surface {TLy^), from the atmospheric scattering (Z,p)
and from surface reflections (TL^) (after Sturm (1981a))

Wavelength Contribution to signal, %

nm clear water turbid water
TL^ Lp TL, TL^ Lp TLr

440 14.4 84.4 1.2 18.1 80.8 1.1

520 17.5 81.2 1.3 32.3 66.6 1.1
550 14.5 84.2 1.3 34.9 64.1 1.0
670 2.2 96.3 1.5 16.4 82.4 1.2
750 1.1 97.0 1.9 1.1 91A 1.5

Z-p and T both depend on the scattering and absorption properties of the
atmosphere. Sturm (1981a) discusses these in some detail. Here we note that T
at a given wavelength of light, and over a path length / in the direction denoted
by distance coordinate s (see Fig. 6.11) is given by

Tx = exp[-/o/f(A,s)ds] = exp ///f(X, z) dz]

where K{\) is the attenuation coefficient and =r(A,z) is a

dimensionless quantity called the optical thickness which describes the
absorption and scattering properties of an atmosphere at different heights.

Fig. 6.11 - Definition sketch: path length I

Sec. 6.2] Optical theory relevant to ocean-colour viewing from space 151

K and hence r and T are due to the collision of photons of light energy
with air molecules and with aerosols. The latter are small solid or liquid particles
which may have been Hfted by wind from the earth’s surface, or sublimated
and condensed gases from the atmosphere. Since the molecular composition of
the atmosphere is generally uniform and well known, it is convenient to treat
molecular effects separately from aerosol effects which are variable in space and
time. The optical thickness of molecular effects alone, TmCX), is due largely to
scattering by air molecules, and depends very little on absorption except for
ozone which absorbs light around 600 nm wavelength. Since molecular scatterers
are small compared with the wavelengths of visible light, Rayleigh scattering
theory applies, and an expression for rm(X), (sometimes known as the Rayleigh
optical thickness) which is sufficiently accurate for remote-sensing applications,
has been derived:

p(0)-p(z)
rm(\,z)= 0.00879 (6.11)
Pn(0)
where the wavelength X is given in pm and p(z) is the atmospheric pressure
at height (z). Pn(0) is the pressure at sea level of a standard atmosphere of
temperature 15° C.
The ozone optical thickness, ro(X), due to ozone absorption, can also be
calculated on the basis of observations of ozone distribution in the atmosphere
(see Sturm (1981a) or McClatchey et al. (1972)).
It is not possible to calculate the aerosol optical thickness so readily. Mie
scattering theory applies in this case, since the particles are much larger than
molecular sizes. Sturm (1981a, 1983) discusses methods of calculating which
require a knowledge of the local meteorological visibility as a measure of the
amount of aerosol haze present, but this does not promise a reliable method
for atmospheric correction of satellite data. Some research indicates that the
wavelength dependence of is given by

rA(X,z) = A{z)\~^

B
rA(Xi) X2
(6.12)
rA(A2) Xi

where B is known as the Angstrom exponent. Unfortunately, whilst this has

proved useful over land with 0.8 < 5 < 1.5, it does not seem to hold generally
in marine atmospheres, and there is even doubt as to the sign of the Angstrom
exponent.
Similarly, while the path radiance due to Rayleigh scattering can be satis¬
factorily predicted, that due to aerosol scattering is less easily modelled, and the
atmospheric correction of visible wavelength data from fundamental physical
models does not appear to be possible. A different strategy is therefore required.
152 Visible wavelength ‘ocean-colour’ sensors [Ch.6

6.2,4 Practical strategies for the atmospheric correction

of visible wavelength data
The problem which prevents the analytical modelling of atmospheric effects
is the complexity of the multiple-scattering processes which occur in the
atmosphere. To calculate the path radiance requires a knowledge of the actual
radiance in all directions at all heights, a knowledge of the differential scattering
coefficient (which describes the amount and the directional properties of scatter¬
ing at all heights), and the transmittance discussed in the previous section.
The differential scattering coefficient includes a phase function P(a, P, a', P'),
dependent on the aerosol content, which expresses how light is scattered from
directions a, P into directions a',P'. A multiple integration over the 4;r solid
angle at each depth of the atmosphere is required to model scattering from aU
directions. Given a lack of knowledge of the aerosol properties of the atmosphere
it is just not possible to perform this integration accurately. However, by simu¬
lating typical atmospheres and running large Monte-Carlo modelUng programmes
it has been possible to gain sufficient understanding of the processes to suggest
simplified approximate approaches for a semi-empirical atmospheric correction
strategy.
Gordon (1978, 1981) concluded from such studies that a more useful
grouping of different ray paths is that illustrated in Fig. 6.12. On the assumption
that sun glitter could be neglected, which is appropriate for satellites designed to
avoid direct specular reflection of sunlight, the effects of reflected sky glitter are
incorporated into the atmospheric-scattering radiance. This is divided into two
parts, that due to Rayleigh (molecular) scattering, Lr, and that due to aerosol
particle scattering, These two parts are assumed from modelling experience
to be capable of completely independent treatment.
Thus (6.10) becomes

-^R + • (6.13)

T'L^ now includes all photons which have penetrated the sea surface,
even though they may have emerged outside the field of view and been scattered
into the sensor by the atmosphere. It is appropriate for them to be counted as
‘signal’ rather than ‘noise’ because they do contain ocean information. However,
in order to include them in this way, it is necessary to use the diffuse trans¬
mittance T' rather than the direct beam transmittance T. This tacitly assumes a
Lambertian surface, and also that does not vary significantly between the
IFOV and the adjacent pixels which make a small contribution to T'L^.
The strategy of atmospheric correction is therefore to obtain Za and Zr,
whence Z'Z^ can be determined and then Z^ if Z'can be estimated. Now Zr,
the Rayleigh path radiance, can be calculated quite accurately, and Sturm
(1981a,b) gives the rather complex expressions necessary to achieve this. He also
presents an approximate expression for T', but the accuracy of this is not as
Sec. 6.2] Optical theory relevant to ocean-colour viewing from space 153

important as the correct solution of the path radiance, given the relative magni¬
tude of the terms indicated in Table 6.5. It should be noted that the important
oceanographic aim is to measure the variation of across the sea surface as
accurately as possible, and therefore whilst Lr and must also be measured
accurately, particularly the latter which is Hkely to be spatially variable, an error
in T' will not alter the estimated spatial variability of very much.

Fig. 6.12 - Optical pathways to sensor, grouped differently from Fig. 6.10 (from
Robinson (1983))

To estimate Z-a, a multispectral approach is used, with the assumption that

~ *^(^2, ^l)•
(6.14)

This appears from the modelling studies to be a valid approximation,

provided that the aerosol phase function (controlling the scattering direction)
and the ratio of the aerosol optical thicknesses, Ta, at the two wavelengths, does
not vary within the scene (that is, on a length scale of several hundred pixels),
even though the aerosol concentration may itself vary significantly. Furthermore,
if the aerosol phase function is independent of wavelength, then S may be
represented by
. s ... ^ ■^0,\(^2)7’oz(^2)
. (6.15a)
154 Visible wavelength ‘ocean-colour’ sensors [Ch.6

^a(^2)'^a(^2)
where (6.15 b)
<*^a(^i)’'a(^i)

^oz(^i) is the ozone transmittance at wavelength Xj, co is the probability that a

photon will backscatter on interaction with the aerosol, and Fq \ is the solar
irradiance at the top of the atmosphere. Thus if S is known or can be estimated
as a function of X, the wavelength dependence of Lj^ is determined.
The usefulness of this is illustrated in Fig. 6.13 representing the variation
of the three separate contributions to in the different channels of a multi-
spectral sensor. i-R is assumed to be known by calculation at all wavelengths.
If at one or more of the longest wavelength channels the water-leaving radiance
is known to be zero or small (see Table 6.5) then can be obtained at that
wavelength. Given S, can be estimated for all the other wavebands, and
hence Z-w can be obtained. The success of this strategy depends upon whether
there is a waveband at which can reasonably be assumed to be zero, and also
upon there being a suitable model for S. These will be discussed in more detail
later in the chapter, in the context of specific implementations of this strategy
in atmospheric algorithms for the CZCS and the Landsat MSS.

Wavelength

Fig. 6.13 - Decomposition of received radiance (Zj) in four waveband channels

into the aerosol (Z^) and Rayleigh (Zr) atmospheric-scattering contributions
and the fraction (Z') of the water-leaving radiance (Zw) which reaches the sensor
Sec. 6.2] Optical theory relevant to ocean-colour viewing from space 155

6.2.5 Scattering and absorption of light in sea-water

We turn now to consider underwater optical processes. These tend to be even
more complex than atmospheric scattering and absorption because of the
wide variety of optically interfering materials found in the sea. It is of course,
the optical effects of such materials as suspended sediment, phytoplankton, and
organic detritus that makes these water-quality components potential candidates
as properties to be studied by satellite remote sensing. With this in mind optical
oceanographers have only recently begun to consider seriously what they can
learn about the sea by looking through the surface from above, rather than
making their optical measurements beneath the water surface. The achievements
so far have been reviewed by Sathyendranath & Morel (1983).
The water-leaving radiance which is the optical parameter measured
by the satellite after atmospheric corrections have been made, is not the best
indicator of oceanographic properties, since it is influenced as much by the
incident radiation as by the water quahty. In fact a more useful parameter is
the subsurface reflectance ratio:

where is the upwelling irradiance and F^ the downwelling irradiance just

below the surface (Fig. 6.14).

R=E,/E,

Fig. 6.14 - Definition sketch; subsurface reflectance ratio

In the case of a flat surface, optical theory shows that R can be related to
Iw by

i (6.16)

where n is the refractive index of water and p is the Fresnel reflectivity of the
water—air interface which causes some of F^ to be reflected back down into
the water. 6' is the subsurface direction of a ray which emerges in the direction B
(the viewing angle of the satellite), and therefore n sin0 = sin0 (see Fig. 6.15).
If the surface is rough, then (6.16) will not be strictly valid since the rays of
hght leaving in the direction of 6 will make a different angle with the surface
156 Visible wavelength ‘ocean-colour’ sensors [Ch.6

normal. Provided that 6 is not close to the critical angle of about 48°, however,
a roughened surface will not introduce much error. If the surface is very rough,
of course, whitecaps and foam appear, and the whole approach is invalidated.
Assuming the ocean is a perfect Lambertian surface, Q should be direction-
independent and equal to a value of n. However, Austin (1980) has suggested
from observations at sea that Q should in fact be approximately 5.0 for near¬
nadir observations. Given Ly, from the satellite sensor, R can then be estimated
using (6.16), by calculating from the known sun illumination at the top of
the atmosphere, and estimating the atmospheric transmittance. Strictly, account
should be taken of the dependence of the refractive index on sahnity and
temperature, but given the much larger errors of estimating atmospheric inter¬
ference this is probably not worth considering. Since Q and are only weakly
wavelength-dependent, it often more satisfactory, and easier, to cancel them out
in the reflectance spectral ratio:

^(>-i) ^ Lw(Ai)«^(Xi) [l-p(X2)]

Ri\2) ~ LU^2)n\X2) [l-p(Ai)] ■
Thus the reflectance spectral ratio is obtained from the remotely-observed
radiance ratio (that is, the ‘colour’) of the water.

Fig. 6.15 — Definition sketch; direction d' in relation to viewing angle 6

R itself depends on the balance of absorption and backscattering of light by

the water and its suspended contents. Since particle sizes are somewhat larger
than the wavelengths of hght, Mie scattering theory applies. Preisendorfer (1961)
has shown that the total absorption coefficient a of the water and the total
backscattering coefficient are the linear sums of the individual coefficients
due to all the different absorbers and scatterers which are present. Therefore it is
useful to express R in terms of a and by^. According to the theoretical modelling
studies of Morel & Prieur (1977),
Sec. 6.2] Optical theory relevant to ocean-colour viewing from space 157

R = 0.33 (Va) (6.18)

is a satisfactory approximation for most marine situations where by^/a is small

(<0.3). Alternatively, Gordon et al. (1975) have proposed that R should be a
function of the form

Both expressions illustrate clearly how the absorption and backscattering

properties of a water component can affect the reflectance. The greater the
backscattering the greater the reflectance, while increased absorption decreases
the reflectance. We note, however, that a very low absorption does not in itself
lead to a high reflectance, since there must be scatterers present. Thus both
clear water and turbid water containing strongly-absorbent material have low
reflectances. Equation (6.18) lends itself to the study of the effect of absorption
and backscattering properties on the spectral ratio of reflectance. From (6.18)
we obtain
/?(Al) ^ hb(Xl)fl(X2)
(6.19)
7?(X2) hb(X2)fl(Xi)'
Thus an absorption peak at Xi relative to X2 will decrease the reflectance ratio
of Xi to X2 just as effectively as a backscattering peak at X2 relative to Xi.
If the absorption and backscattering properties of sea-water and its
dissolved and suspended constituents are known, it should in principle be

possible to relate observations of R and -- to the amount of the consti-

i?(X2)
tuents present. This is the theoretical basis of water-quality algorithms using
ocean-colour measurements from space. Unfortunately, it has not proved yet
to be possible in practice to invert the problem in this way and to generate
a calibration algorithm from first principles, because of the complexity of
the observed relationship between the concentration of sediment or chlorophyll
etc. and the resulting backscatter and absorption properties. Nevertheless,
empirical calibration algorithms have been developed which are based on this
underlying approach.

6.2.6 Oceanographic interpretation of ocean colour

Following the reasoning presented in the previous section, if ocean-colour
observations are to be interpreted in terms of the water-quality parameters, we
need to know how these parameters affect the optical properties of water, and
to consider separately the spectral characteristics of the different constituents
of sea-water.
The spectral signature of pure sea-water provides the baseline upon which
the spectral characteristics of other types of water are built by the addition of
158 Visible wavelength ‘ocean-colour’ sensors [Ch. 6

other scatterers and absorbers. Typical observations of the spectral variation

of a and in pure sea-water are summarised in Fig. 6.16, based on the work
of Morel & Prieur (1977) and Smith & Baker (1981). Absorption decreases and
backscatter increases with decreasing wavelength, leading to the characteristic
blue colour of pure sea-water.

b,10^m’

Fig. 6.16 - Absorption (a) and scattering (b) of pure sea-water (after Robinson
(1983))

Water containing phytoplankton has much more complex spectral charac¬

teristics, because the hving cells of the small plant organisms and algae contain
chlorophyll used for photosynthesis. Since photosynthesis uses solar power as
its energy source, it is not surprising that chlorophyll absorbs sunlight strongly
in certain parts of the spectrum. As well as displaying the chlorophyll absorption
characteristics, the phytoplankton have a structure which optically is equivalent
to particulate material, acting to scatter Hght. They are also surrounded by
dissolved organic compounds which contain phaeophytin a, which has another
characteristic absorption spectrum. Within a phytoplankton population, the
detritus of dead organisms may contain both skeletal material which contributes
to the scattering, and hght-absorbing organic matter, even though the chloro-
Sec. 6.2] Optical theory relevant to ocean-colour viewing from space 159

phyU is no longer present in the dead cells. Consequently water containing

phytoplankton has different proportions of absorbing and scattering elements
according to the species and the age of the population. A variety of absorption
and scattering spectra have been observed, and a review of the results of many
workers has been presented by Sathyendranath & Morel (1983). Certain general
characteristics emerge, as sketched in Fig. 6.17. There is a strong absorption
peak at 440 nm, and a lesser one at around 675 nm, whilst over the rest of the
spectrum in general the absorption decreases with wavelength. The backscatter
is fairly uniform apart from decreases occurring at wavelengths approximately
corresponding to the absorption peaks.

400 700
X,nm

Fig. 6.17 — Typical spectral variation of (a) absorption due to chlorophyll,

normalised at 440 nm (after Prieur & Sathyendranath (1981)) and (b) specific
backscatter coefficient. Values of (a) at 440 nm vary between 0.01 and
0.1 m‘‘ (mg m"’)"', depending on the age and species of phytoplankton. Units of
(b) are typically of order 10"’ m"' (mg m"’)"’ (from Robinson (1983))

Where the dissolved organic matter associated with decayed vegetation is

found separately from a phytoplankton population it is known as 'yellow
substance', or 'gelbstoff. This influences water colour through its absorption
properties, shown in Fig. 6.18. In marked contrast to pure sea-water it has
strong absorption in the blue and shows a monotonic decrease of absorption
160 Visible wavelength ‘ocean-colour’ sensors [Ch. 6

with wavelength. Adding Fig. 6.18 to the absorption curve of Fig. 6.16 explains
the name given to yellow substance, since the least absorption is in the yellow,
middle part of the spectrum. Yellow substance not associated with local
phytoplankton productivity is thought to derive from land drainage (Hojerslev
1980), and is found in strong concentrations in enclosed seas such as the Baltic.
If yellow substance can be detected from space, it might therefore serve as an
indicator for the plumes of river discharge and the dispersal into a shallow sea
of fresh-water runoff.

Fig. 6.18 Approximate absorption spectrum of yellow substance (from

Robinson (1983))

The other, and most diverse, contribution to ocean colour is the presence
of suspended particulate material not related to phytoplankton. This may be
resuspended bottom sediment, riverborne sediment, eroded coastal and beach
material, or due to the dumping of sewage-sludge waste or dredging spoil. Its
spectral characteristics are as diverse as its composition in terms of natural
material colour and size distribution, and there appears to be no consensus in
the literature regarding a standard spectrum shape. Sathyendranath & Morel
(1983), in their review of the theoretical and experimental studies in the litera¬
ture of the optical properties of suspended sediment, are able to point to
Sec. 6.2] Optical theory relevant to ocean-colour viewing from space 161

evidence for neutral (flat) absorption spectra, absorption spectra similar to

yellow substance (Fig. 6.18), and U-shaped absorption spectra with increasing
absorption at both the red and blue ends. As well as these diverse observations,
there is no certainty that the amount of absorption or backscatter is linearly
related to the amount of sediment present. Consequently there is little hope
that a universal theoretical model of scattering and absorption by suspended
sediment can be achieved before a great deal more experimental research is
performed in this area.
With so much uncertainty concerning the values of a and appropriate
for many constituents of sea-water, coupled with the optical complexity of
waters where phytoplankton, yellow substance, and suspended sediment are
all present, there is little hope in the near future that reflectance spectra or
reflectance ratios can be predicted for a given water quality using (6.18) or (6.19).
Instead, if we are to begin to interpret satellite observations of ocean colour, we
must rely on empirical observations of the reflectance spectra for different
types of water.
To make progress in this complex problem, a fundamental step is to divide
all waters into two categories for optical purposes. Case 1 are seas whose optical
properties are dominated by phytoplankton and their degradation products only,
whilst case 2 waters have non-chlorophyll-related sediments or yeUow substance
instead of, or in addition to, phytoplankton. Fig. 6.19 shows typical reflectance
spectra for case 1 water. The arrow indicates the way the spectrum changes with

Fig. 6.19 - Typical reflectance spectra for case 1 water. The arrow shows which
curves correspond to increasing chlorophyll (and related suspended particulate)
concentration. The dashed line is the dear-water spectrum (from Robinson
(1983))
162 Visible wavelength ‘ocean-colour’ sensors [Ch. 6

increasing chlorophyll concentration, and the dashed line is the dear-water

baseline spectrum. The combination of absorption and backscatter from the
phytoplankton population tends to decrease the reflectance below the dear-
water spectrum at wavelengths below around 540 nm, and slightly increases
it at higher X. Increasing chlorophyll content enhances this effect, with a
definite minimum appearing at 440 nm caused by the chlorophyll absorption.
Another minimum appears at 660 nm, although the red absorption maximum
(see Fig. 6.17) is masked by an apparent reflectance peak at 685 nm due to
fluorescence. Although the absolute reflectance values differ greatly with
different chlorophyll species, the spectral shape and its variation with chloro¬
phyll concentration remain similar. This suggests that a chlorophyll algorithm to
interpret visible-wavelength radiance measurements from space should be based
on spectral ratios (that is, colour information) rather than the actual magnitude
of the reflectance. It will be noticed that the reflectance does not vary very
much with chlorophyll content in the 550—600 nm range of the spectrum.
Indeed, there is a tendency for the spectrum to rotate counterclockwise about
a hinge point’ in this region, as chlorophyll content increases, suggesting that an
appropriate spectral ratio for calibration purposes might be between a waveband
near the absorption minimum and a waveband in this hinge region.
Fig. 6.20 illustrates typical reflectance spectra for case 2 waters in which
suspended sediment is the optically-dominant constituent. Backscattering by
the sediment produces higher reflectance ratios at all wavelengths, although the
increase is smaller at low wavelengths owing to small amounts of chlorophyll

Fig. 6.20 Typical reflectance spectra for suspended-sediment-dominated case 2

water. The arrow indicates increasing sediment load. The dashed line is the dear-
water spectrum (from Robinson (1983))
Sec. 6.2] Optical theory relevant to ocean-colour viewing from space 163

absorption. Because the spectral shape does not change much with sediment
load, the water colour does not change, and a sediment algorithm based on
spectral ratios appears less hopeful. Instead, an algorithm relating suspended load
to the absolute reflectance in a single waveband is more appropriate. One excep¬
tion to this occurs in those situations where the sediment is highly coloured, and
the strength of the colour is likely to be correlated with the amount of sediment.
In this case the reflectance spectra would be rather different from Fig. 6.20,
with a peak corresponding to the colour of the sediment.
Fig. 6.21 shows reflectance spectra for case 2 waters in which yellow
substance dominates. Like the chlorophyll spectra in Fig. 6.19, the measured
concentration leads to increased absorption and a decrease in blue reflectance.
Consequently a calibration algorithm based on spectral ratios should be more
promising in this case, although there is less of a well-defined hinge point for
normahsing the spectral ratios.

400 700
X,nm

Fig. 6.21 —Typical reflectance spectra for yellow-substance-dominated case 2

water. The arrow indicates increasing concentration. The dashed line is the dear-
water spectrum (from Robinson (1983))

From these generalised spectra, it appears that there is sufficient change

in optical properties correlated with the concentration of the dissolved or
suspended material to permit calibration of colour images in terms of water
quahty, but only if the water has a single component. This fundamental limita¬
tion provides the proper perspective for the development of interpretive
algorithms. Optimistic attempts at producing universal algorithms appear to be
ill conceived. There is no justification for taking a sediment or chlorophyll
164 Visible wavelength ‘ocean-colour’ sensors [Ch. 6

calibration algorithm based on synoptic ship and satellite measurements of

one particular place at one particular season, and applying it to other locations
or seasons without further sea observations. It may turn out to be the best that
can be achieved under Umiting circumstances if the satellite data are from the
archives and cannot be retrospectively caUbrated, or if the location being studied
is inaccessible for in situ sampling. In this case the conclusions must be treated
with extreme caution.
On the other hand, undue pessimism about obtaining reliable calibration
algorithms is also out of place. It is true that theoretical modelling studies such
as those by Bukata et al. (1981) demonstrate quite clearly that such is the
complex interaction of absorbing and scattering effects when a mixture of
different constituents are found in sea-water that it is possible for different
combinations of chlorophyll and suspended minerals to give exactly the same
reflectance ratio. In other words, there may be no unique relationship between
water-quality parameters and their optical expression, suggesting that an un¬
ambiguous calibration algorithm based on simple reflectance ratios is impossible.
However, the real world beyond the confines of computer modelling of first-
principle physics is not only more complex, but sometimes more amenable than
our theoretical predictions imply. It is often possible from general oceanographic
knowledge, or from field experience of an area, to predict with confidence the
type of water to be expected. Thus in the open ocean it is reasonable to assume
that yellow substance and suspended sediments will be negligible and we are
dealing with case 1 waters, in which event a unique calibration of satellite data
may be possible. In coastal waters it is more difficult, but the more field
experience that can be gained of typical conditions and their correlation with
location, season, and tidal state, the more likely it will be that archived imagery
of that area can be interpreted in the absence of synoptic sea data. This is
the challenge of ocean colour interpretation, to derive as much oceanographic
information from the images as is possible, but to know the limitations of
the method and not to extend the interpretation into the realms of unjustified,
unscientific speculation.

6.2.7 Light-scattering at the sea surface

Much of the early interest in visible-wavelength observations from space was
directed at remotely detecting the surface roughness, and hence the surface wave
properties. Research effort into the remote sensing of sea surface roughness has
now largely shifted to microwave sensors, to be described in Chapters 10 and 13.
Specular reflection of the sun from the sea surface is thus normally considered
as a problem to be avoided in the remote sensing of ocean colour.
Cox (1974) has reviewed the subject in some detail. In order that sunlight
is directly reflected into the sensor field of view, there must be a portion of the
sea surface at just the right angle. Fig. 6.22 illustrates this. The upper curve is the
Sec. 6.2] Optical theory relevant to ocean-colour viewing from space 165

sea-surface displacement f, and the middle diagram represents the corresponding

surface slope f'. If the surface slope lies within a narrow band 8^' centred at fo)
then sun glint occurs as shown in the lower diagram. If the areas over which
the sun glint occurs can be spatially resolved by the sensor, these appear as
highlights on the remotely-sensed image. Otherwise, the sun glint contributes
to an enhanced brightness in the pixel in which it is found, although in this case
it may not be so evident that sun gUnt is occurring.

SENSOR SUN

I contribution to sensor radiance

Fig. 6.22 - Specular reflection from a wavy surface. For a given solar and sensor
direction as shown in the upper diagram, direct reflection only occurs for a small
range of surface slopes, illustrated in the centre diagram. The result is a highlight
on the image corresponding to the location of such slopes, as illustrated in the
lower diagram

The contribution due to sun glint depends on the size of 6f', and the extent
to which fo is achieved within the sea area being viewed. 5f' depends on the
solid angle subtended by both the sun’s disk and the satellite sensor aperture
at the sea surface, fo depends on the zenith angle of both the satellite and the
sun. Whether fo is achieved depends on the sea state. If the sun and satellite
are both overhead, fo = 0 and a calm sea will give total specular reflection. If it
is roughened, the sun glint will be decreased because much of the sunlight will
166 Visible wavelength ‘ocean-colour’ sensors [Ch.6

soon be reflected away from the sensor, but the area of sea over which the sun
ghtter pattern occurs will be increased because off-nadir parts of the image
will now have facets producing direct reflection of the sun. Thus for all view¬
ing directions except the one which produces direct specular reflection in calm
conditions, increased surface roughness will tend to increase the chance of some
specular reflection occurring. Most satellite orbits are planned to avoid the
occurrence of sun glitter, by ensuring that fo for the given satellite/sun geometry
is larger than is likely to be encountered at sea. If fo is in a range which may be
encountered in a sufficiently rough sea, then it is necessary to apply corrections
based on the sea-surface roughness, or the wind speed, in order to remove the
extra radiance due to specular reflectance which is measured by the satellite.
Bernstein (1982a) illustrates the use of such a correction in the context of
reflected infrared radiation.

6.3 CALIBRATION AND APPLICATION OF CZCS DATA

6.3.1 Sensor calibration
Fig. 6.23 represents the physical inputs and digital outputs of the CZCS instru¬
ment package on the Nimbus-7 satellite. The sensor views the earth radiance L^,
and once every sixteen scans it views the calibration lamp whose known radiance
is Lf., defined in units of radiance per waveband (mWcm”^ sr"^ /im“^). These
result in voltages V and V^. being output by the sensor and digitised as count
values Q and Cq. In addition, after every scan line a voltage ramp of sixteen
values, Vi, is input to the a/d converter to produce ramp counts Q.

RADIOMETER A MPLIF IE R / DIG ITIS E R

ramp

Fig. 6.23 - Schematic of data flow in the Coastal Zone Colour Scanner
Sec. 6.3] Calibration and application of CZCS data 167

Os VO C^
Tt VO VO OS
<N CO <N
t-H
a oq <N Tj-
4> a>
O o o O

cn ^ d d d d
CO '2
'4-^ lO s
•S *o
so
CO
Os
<N
CO VO
VO <N r- »-H
CO CN
o o o o
d d d d

o VO CO CN
v ^^ o 00 VO CO
o. c CO o VO
o Q oq
o o
«—1
o
o
o

e
o
^ s d d d d

.w VO ^
es
IM <
CO
r'
rs
o
r--
CO
00

CO VO 1-H
W CO (N 1-H
o o O o
o d d d d
(U
c VO VO 00
.3 CO
Os
CO
CO
(N

Qq CO (N o
2 o o o o
1 00 d d d
lo d
O ON
ON CO
lo •<
VO ;2: r- r-- 00
s C os CO CO Tf
c
o fa VO
CO CNJ 1-H
« 5 o o o o
Vi ^ d d d d
U b.
N .2d
1-H Os Tj-
W 13 o C-- VO 00
<u '—' CO VO Os
Si ^ oq (N
o
(S o
o
O o
<*- .S Tj* "*3^ d d d d
O C3 00 CO
.—V ®® ^ <
(N 2
2> « VO
00 CO CO
VO

VO 1—:
e3 X
o
CO
o
(N
o
1—1
O
C d d d d

JO
ll CO (N VO VO
o VO
OS
VO
CO
os
OS
CO
1-H
cs
CO
oq CO VO 1-H CN
■*-* o o o O o
XI
d d d d d
2f
S 1
J3 0) <N CO r-- VO VO
•M Cu U-) o VO CO 1-H
1-H 1-H
O CO <N 1-H OS
C o o O o O
_o d d d d d

.3
c3 t-.
> X) (U CN CO VO
C CO (U <D <u 0) CD
3 q> c c G C c
NO
C ^ c c c c c
o 03 C3 CT3 d ca
NO X X X X X
X ^
<U o o u U u

JO
«
H /
168 Visible wavelength ‘ocean-colour’ sensors [Ch. 6

Cg, Cc, and Q are transmitted in the data stream. V} is also independently
coded in the extra calibration data inserted in each line. Hence it is possible to
monitor the amphfier/digitiser characteristics, Oy and where

Cj — fly 1^- + by (6.20)

can be fitted by a least squares method (fly and by depend on the particular gain
setting in operation, which is also specified in the calibration data).
Hence F and can be determined as

(6.21)
and

(6.22)

Since Lc should be known from pre-flight measurements, it is then possible

to evaluate the sensor response from

fc (6.23)
(Cc-by)
that IS, fls = -.
Qy Lq

Then the earth-scan digital values can be calibrated as

^ _ V Q by

~ ~~ (6.24)
Cly Gy G^

or Ls = ^cQ + ^c
1
where Ac and (6.25)
fly flj Uy Ug

It has been found that and Be do not change significantly from one scan line
to the next, and therefore it is customary to evaluate A^. and once only for a
scene and apply (6.25) to all pixels in the image. There are slow fluctuations in
Ac and B^., as Table 6.6 illustrates. A general sensitivity loss of the sensor with
time has also been noted and possible corrections for it are discussed by Sturm
(1983). This may be due to degradation of the on-board calibration lamps, in
which case the above calibration technique cannot be regarded as absolute, and
it becomes difficult to compare images which have been observed several months
or years apart. Absolute comparisons then require assumptions to be made
about the nonvariability of Lyy in certain clear-water areas.

6.3.2 Atmospheric corrections

Assuming that the above calibration procedure is functioning properly, the
sensor-received radiances, Lg, must be atmospherically corrected to retrieve the
Sec. 6.3] Calibration and application of CZCS data 169

water-leaving radiance L^. Several practical approaches to this problem have

been adopted, depending on the particular sea-water conditions encountered.
Robinson (1983) has reviewed these, and Fig. 6.24 sets out the principal options
available. It is customary to use the first four channels only (Table 6.3), since
the algorithms depend on quite precise sensor-waveband data being available.
The basic strategy is that presented in section 6.2.4, and the ideal application is
one where there is zero water-leaving radiance in channel 4 (that is, Z,w,670 ~ 0)
and the aerosol path radiance variability with wavelength S{\), is known. This
situation provides the central pathway through the flow diagram of Fig. 6.24.
Expressions for evaluating the terms are given in Table 6.7, based mainly on
those presented by Sturm (1983).
Knowledge of 5(X) requires a knowledge of e(X) (see (6.15)). If this is
known confidently from independent measurements, then the left-hand side
of the diagram can be ignored. If it is not known, then the safest assumption is
probably that e = 1, implying that in (6.15 b)

^a(^2) _ ^
COa(^i)

and in (6.12) the Angstrom exponent, B is zero. Fig. 6.24 offers a possible
way of iterating for a better estimate of S, e, and hence r a which is required to
improve the estimate of T'. This iterative procedure works if there are several
areas of dark-water pixels, which must be well away from cloud and land, and
should also be fairly close to the satellite subtrack (small viewing zenith angle 0).
It must also be oceanographically reasonable to assume that these are clear water,
with very low turbidity because the concentrations of sediment, chlorophyll,
and yellow substances are all low. In this case the water-leaving radiances
in channels 2 and 3 can also be assumed to obey the formulae stated in
Table 6.7(D). The estimates of S and e made from these groups of pixels are
then applied to the rest of the scene. This allows a pixel-by-pixel atmospheric
correction, but assumes that the aerosol does not change its optical properties
significantly over the scene, other than its optical thickness.
Another route to S is to measure Z-vv,\ for all four channels at various
locations synoptic with the overpass. The correction algorithm scheme in
Fig. 6.24 is then entered both from the top and through the box at the bottom
left which leads to an estimate of e for the ground-observation pixels which is
assumed to hold over the whole image.
Pixel-by-pixel correction is only possible if one of the bands can be assumed
to carry information about Z-a without anyLw contribution. In relatively clear,
deep-sea water Z-w,670 0 rnay be a reasonable assumption, but in a plankton
bloom, or in turbid coastal waters, this is not the case. If Z,w,670 is small com¬
pared with Za,670 fh® iterative loop on the right can be used. If case 1 waters
are being studied then a reliable algorithm is available to predict Z-w,670 froiT^
a knowledge of Z-vv ii^ channels 1 to 3. In turbid coastal waters dominated by
170 Visible wavelength ‘ocean-colour’ sensors [Ch. 6

s
o
u

oO

a>
a

a>

•si
c/5
00
5U
1 N
•S
O- o
2 E

o. o
o ^
T3 ""
C

3
o!U
oVh
CL,
I ,
CN

.s?
Sec.6.3] Calibration and application of CZCS data 171

sediment, such an algorithm is unlikely to be valid, although with extensive sea

observations it may be possible to develop a site-specific one. The danger in
this approach is that it is difficult not to make a priori assumptions about the
water quality in order to perform the atmospheric correction, when the ultimate
aim of the exercise is to obtain an independent measure of water quality!
Alternatively, it may be possible to estimate Zw,670 or assume it to be zero for
only a few of the deeper, clearer water pixels in the scene, in which case the
value of 7^(670) deduced from for those pixels is assumed to be uniform
over the rest of the scene, 7-a,670 being allowed to vary only with geometry. This
prevents the removal of differential haze structures varying across the image, but
is better than nothing.

Table 6.7 Expressions used in the atmospheric correction of CZCS data

(after Sturm (1983)) (see Fig. 6.24)

(A) = —!-!— where X is a complex expression defined in

cosd
(11.68) of Sturm (1981a)
+ ^oz,X
(B) Tx(B) = exp---
cosa
(C) Initial estimate of 5(7., 670) can be obtained from (6.15) by assuming that
/670\
e(X,670) =
X
(D) For clear water in which the chlorophyll concentration is less than
0.3 pg r\ it is possible to assume (after Gordon & Clark (1981)) that

■^'W,520 = 0.90 COS0O 7^520(^0) [1 P(^)]/^520

T-W.SSO ~ 0.52 COSPq T’sso (^0) P(^)]/^550

(E) Use 7<a,x 7^0,670 T’oz(670,0,00)

e(X,670)= -
LA,67oFo,).To^(X,d,do)
that is, inverting (6.14) and (6.15)
Tq2 is a function of the viewing angle 0 and the sun angle 0o. At this
stage it is possible to evaluate the Angstrom coefficient v defined as

e(X„X.) = (^j .

Ideally v will be wavelength-invariant, but we may define V520 and V550 as

^670\’^”» '670\''5=o
e(520,670) and e(550,670) =
^520, .550y
172 Visible wavelength ‘ocean-colour’ sensors [Ch. 6

Table 6.7 — continued

If it has not been possible to evaluate e(443,670) because an estimate of
Z,A,443 is not available, we may assume V443 = 14(v52o + V550) and use

e(443,670) = [-)

Sturm (1983) points out that the iteration should only proceed from this
point provided V443 ^ V520 - V550
and 0.3 < V443, V520, V550 < 3.0.
(F) Use an average value of /-a,670 from clear-water pixels to calculate
_ 0.26 COSgLA,670
7’o,(67O,0,0o)
and then use e(X, 670) from (E) to obtain
'^A,x — '^A,610^(\^70)

(G) For case 1 waters, Smith & Wilson (1981) have shown that,
. 1.661
^ss,443 ^ss,443
= 12.06
•^ss,670 \-^ss,550/

where x is the subsurface radiance, related to //w,\ by ,

1-P(0) ,
'ss, X-
IJ V 1.661
I •T.'ss.SSO
Hence I33 ^70 = 0.0829 Zss,443 and Z-w,670 can be evaluated
' .^ss,443)
from estimates of //w,443 and Z-w.sso-
(H) Z,A,\ can be assumed to vary across the image in proportion to roz(X) which
depends on 9 and Oq.
Toz is the ozone transmittance, and is defined by

Toz(Kd,dQ) = exp-jroz,x(—- + ——))•

( \cos0 COS0o/J

In the above expressions:

9 is the satellite-viewing zenith angle
9q is the solar zenith angle
ris the Rayleigh optical thickness, available from tabulations
ta,\ is the aerosol optical thickness, which can be measured, estimated, or
calculated in the iterative loop
Toz,\ is the ozone optical thickness available from tabulations
p(0) is the Fresnel reflectivity of the sea surface at viewing angle 9
n is the refractive index of the air-sea interface
Fo x is the solar irradiance at the top of the atmosphere
/Sa is an empirical coefficient
Sec. 6.3] Calibration and application of CZCS data 173

The state of the art is therefore promising for the open ocean, but still
unclear for turbid coastal waters. The principal difficulty lies in the fact that
channel 4 is at 670 nm and for many waters Z-w,670 0- W^re it 150 nm greater,
then only the most turbid fluid mud suspensions would produce a water-leaving
radiance at that wavelength, and it would serve as an independent channel
for atmospheric effects. It would be harder to estimate S(X) over the wider
wavelength interval, but that would be a lesser problem than is presently encoun¬
tered in atmospherically correcting CZCS data in coastal areas. In principle,
channel 5 of the CZCS (700—800 nm) should be useful for atmospheric correc¬
tion, but its poor radiometric sensitivity and its broad bandwidth limit its value
to approximate but fast atmospheric filtering algorithms.

6.3.3 CZCS calibration algorithms

Most of the algorithms that have been developed to interpret the atmospherically-
corrected fvv iri terms of water parameters are of the form
B B
-^W,443 •^W,S20
C or 5 or K{\) = A , OX = A (6.26)
L^W,550_ L^W,S5oJ

where C is the total pigment concentration (chlorophyll a plus pheophytin d)

in /ig r\ S is the dry weight concentration of total suspended solids in mg r\
and K{\) is the diffuse attenuation coefficient of the near-surface water for a
given A.
A^(A) is itself an optical property rather than an oceanographic parameter,
but because it has been extensively measured at sea and used for bio-optical
classification of sea-water (Smith & Baker (1978a,b). Baker & Smith (1982)), it
can be used as an intermediate indicator of the water-quality parameters.
Equation (6.26) is simply the result of regression analyses of the logarithms
of the satellite radiance ratios and the water properties measured synoptically at
sea. The only physical understanding entering (6.26) is the choice of a spectral
ratio as a variable. It should be clear from the above section that the best fit to
such a form is likely to be found in case 1 water.
NASA have produced a pigment algorithm which is now used in producing
contoured images of pigment concentration. This is:

-1.71
■^ss,443
C- 1.13 used for C <l .5 idg \ ^
_-^ss,550_
(6.27)
-2.439
■^ss,520
C = 3.326 used for C> 1.5 //g 1“*
_-^SS, 550_

where the subsurface radiance 4s rather than the water-leaving radiance Lw has
been specified. As in (6.16) we have
174 Visible wavelength ‘ocean-colour’ sensors [Ch. 6

L ss (6.28)
1-P(0')

where the refractive index n and surface reflectivity p are weakly wavelength-
dependent, and p is also dependent on the view angle, and must therefore be
evaluated separately for each point along a scan line. It will be noted that
^ss,443 is used only for low-chlorophyll concentrations because at higher values
the absorption becomes so great and Z,w,443 correspondingly so small that its
accuracy in a ratio algorithm is not adequate.
Equation (6.27) is not the only algorithm to be developed, and Sturm
(1983) lists a variety of others. There is significant variability, and in general the
coefficient A (in (6.26)) is lower for chlorophyll algorithms in coastal waters,
because of yellow-substance absorption which reinforces the chlorophyll effects
on the spectrum shape. In case 1 water, an accuracy of log C within is
claimed with the NASA algorithm (Clark 1981) appUed universally.
Algorithms for total suspended sediment in case 1 waters are quoted by
Sturm as:
-0.88
■^ss,443
S = 0.4 mg 1-1
-■^ss,550_
-1.09
•^ss,443
5 = 0.33 mg 1-1 (6.29)
-^ss,520.
-4.38
-^ss,520
S = 0.76 mg 1-1
j^ss.SSO-

with accuracy comparable to the chlorophyll algorithms.

Austin & Petzold (1981) present algorithms for the diffuse-attenuation
coefficient within the upper layers (that is, the first optical depth) which are
believed to be as accurate as could be measured from a ship, given an accurate
atmospheric correction. They are;
-1.491
•^ss,443
K^{A9Q) = 0.0883 + 0.022
_^ss,550_
-1.398
(6.30)
-^ss,443
/:u(520) = 0.0663 + 0.044
-^ss,550-

The apparent success achieved with CZCS data over case 1 waters can be
misleading if taken out of context. The only reason why a good sediment
algorithm can be achieved with spectral reflectance ratios is that the particulate
load is controlled by the phytoplankton population which has a chlorophyll
colour signature. Hence the variables C, S, and A:(X) are not independent in
case 1 waters. In case 2 waters, no such correlation can be found between
Sec. 6.3] Calibration and application of CZCS data 175

sediment load and reflectance ratios (Hojerslev (1974)) and this is to be

expected for the reasons discussed in section 6.2.6 above. The only useful
algorithms in these circumstances are likely to be limited to a particular location
and seasonal condition, and heavily dependent on synoptic sea data. In those
case 2 waters where yellow substance is not dominant, there remains hope that
calibration for pigment is still possible (Gordon & Morel (1981)). Recent observa¬
tions in the Irish Sea suggest that sediment is related to the wideband reflectance
magnitude, whilst the spectral ratio continues to indicate the chlorophyll.
In general, for case 2 water and places where the water type is not known, it
makes most sense to calibrate the imagery in terms of the attenuation coefficient
K(\) since the algorithms of (6.30) are based on a wide variety of water types
and are apparently universal. C and S can then be obtained from K(X) using
site-specific caUbrations based on ground measurements without the need for
data collection simultaneous with a clear-weather satellite overpass.
Other types of calibration algorithm, involving sums and differences of
different bands, or the ratio of differences, have been tried. Most show some
limited success in accounting for a synoptic dataset, but they have little value in
predicting the calibrations of scenes other than the one from which they were
constructed, having no physical basis for their form.
Related to the calibration in terms of K(X) is the suggestion by Hojerslev
(1980) that the spectral ratio can be accurately correlated with the depth to
which light penetrates. This does not teU us about what is causing the water
turbidity, but it is nevertheless a very useful piece of information for certain
applications.

6.3.4 Applications in biological oceanography

The application for which the CZCS was designed is the study of the spatial
and temporal variability of primary production, that is, phytoplankton, in the
world’s ocean. This provides the beginning of the food chain and is fundamental
to the study of the rest of biological oceanography. Using the NASA or similar
chlorophyll algorithms (6.27), studies of CZCS data such as those by Smith &
Wilson (1981) and Yentsch & Garfield (1981) have demonstrated the feasibility
of mapping synoptic chlorophyll concentrations to a good degree of accuracy in
case 1 waters (see Fig. 6.25). These satelhte surveys cover areas the size of the
Southern California Bight or the Gulf of Maine/Georges Bank region - up to
1000 km square. This would have been impossible to achieve using research
vessels, since to cover such a wide area with even a coarse sampling interval
would have taken several days, during which time the plankton populations
could be expected to have grown or decayed, and been advected through the
region by ocean currents. The accuracy of .chlorophyll measurement using
individually calibrated scenes of satellite data also appears to be comparable
with that achieved using continuous-sampling fluorometers on research vessels.
These calibrated images represent the beginning of a new type of genuinely
 Fig. 6.25 — Satellite image from 19 October 1980 of the ocean off the Southern
California coast where land is black and clouds are white. The optical data from
the Nimbus-7 CZCS has been atmospherically corrected and absolutely calibrated
to produce an image of chlorophyll-Uke pigments (see Smith & Baker (1982)).
The grey scale to the right gives the calibration from 0.01 mg pigment/m* (dark)
to 10.0 mg pigment/m’ (light) in logarithmic steps. (Image provided by R. C.
Smith of the University of California Marine Bio-Optics Group)

synoptic oceanography. They reveal new aspects of the spatial and temporal
distribution of chlorophyll and sediment in detail previously unobtainable from
ship-sampling methods. If oceanographers, in their use of such images, are to
move beyond the descriptive phase, they need to formulate the questions that
will lead to new theoretical understanding of the factors controUing the observed
distributions. One promising field of study is the way physical processes -
advection by residual flows and upweUing, mixing across fronts, and stirring by
tides or storms etc. - control the biological processes. An example of this is the
study of Gulf Stream rings by Gordon et al. (1982) illustrated in Fig. 6.26.
 ■.ir'Y

Fig. 6.26 - Image of chlorophyll pigment concentration, with a grey-scale range

between 0 (white) and 1 mg m"’ (black) produced from CZCS data of Orbit 3226
(14 June 1979). The area covered is in the north-west Atlantic, to the south of
the Gulf of Maine. The black speckles and patches are cloud detected in the pro¬
cessing, and the white irregular line is the track of the Research Vessel Atlantis II,
from which samples* were collected to calibrate the image. At the south of
the image is the low concentration of chlorophyll in the warm Gulf Stream.
Centred at around 39°20'N, 69° W can be seen a warm core ring which has
detached from a Gulf Stream meander. The ring is identified by its low produc¬
tivity (~0.1 mg m“*) in comparison with the surrounding slope water
(~0.4mgm-*). Further north is the highly productive shelf water (~0.7mgm'*).
Gordon et al. (1982) show a sequence of images between 9 and 14 June, revealing
that the eddy is progressing westwards at a speed of about 5 km per day. Note
that images such as these are very suitable for atmospheric algorithms requiring
a ‘dear-water pixel’ assumption, because such regions can usually be found in
the Gulf Stream water to the south. (Image provided by and reproduced with
permission of H. Gordon)

As well as viewing synoptic spatial detail it is now possible to estimate the

total primary productivity of a sea area, and its seasonal variations (Smith
178 Visible wavelength ‘ocean-colour’ sensors [Ch. 6

& Baker (1982), Smith et al. (1982)). Total productivity must be integrated
throughout the whole water column. The major but not necessarily the entire
contribution to the total comes from the upper well-lit few metres which control
the satellite colour data. It is therefore necessary to construct a model of the
vertical structure of productivity in order to extrapolate from satellite observa¬
tions to total productivity. This is an area requiring further research along the
hnes of Platt & Herman (1982). Previous estimates of global and regional
productivity have been based on ship observations which, given the spatial
variability revealed by the CZCS, may turn out to be unrepresentative.
Another area of study which has already benefited significantly from the
CZCS imagery is the identification of productive regions in the vicinity of
oceanic or shallow-sea fronts. With a CZCS image it is possible readily to deter¬
mine the location of productive patches relative to the position of a thermal
front or other physical feature. Whilst ship sampling can offer a view of the
vertical structure of such patches, it cannot provide the same perspective as the
satellite in both horizontal directions. Fig. 6.27 is a good example. In this image
of the English Channel regions of enhanced productivity are revealed by multi¬
channel processing after atmospheric correction. A productive patch, probably
a dinoflagellate bloom, stands out where it would be expected along the Ushant
Front at the transition between tidally mixed water to the south and east and
stratified water to the north and west. The front allows the nutrients from the
lower layer on the stratified side, where it is too deep and dark to be used up by
plant growth, to be mixed towards the surface on both sides of the front, where
it is able to support primary production. A second productive region in Fig. 6.27
marks the line of the continental shelf edge. Although the water is stratified
here, and in general the nutrients are found in the lower, unproductive layer, the
presence of the shelf break gives rise to mixing processes which locally transport
nutrients into the upper layer where they can support growth. The most prob¬
able cause of the mixing is the generation of internal waves by tidal flow onto
and off the shelf. The shelf-break patch appears unusually strong because it is
probably due to the highly reflective species of Coccolithophores.
In the biological oceanographic applications of the CZCS so far reported,
satellite observations have been used to extend the usefulness of, but not to
replace, shipbome measurements. One near-real-time apphcation which requires
good communication from the receiving station to the oceanographer at sea is
to use the satellite image to locate the most active areas of productivity (for
example, an algal bloom). A research vessel can then be directed to those
places where physical and biological parameters require detailed vertical
profile sampling for a period of days to weeks. During this time the wider area
in which the particular process is set can be regularly surveyed by the satellite.
Holligan et al. (1983) have demonstrated the effectiveness of combined ship and
satellite surveying of a dinoflagellate bloom in the English Channel such as that
illustrated in Fig. 6.27.
 Fig. 6.27 - Image of the Celtic Sea and N. Bay of Biscay processed to enhance
CZCS data recorded on 22 June 1981 to reveal areas of high productivity. The
areas where productivity is greatest are at the shelf break (A) and along the
Ushant Front (B). Both of these are regions where mixing processes cause
nutrients to be brought into the weU-lit surface layer so that productivity can be
supported. Note the streamer S indicating a southward transport of the shelf
break bloom, probably caused by a pair of eddies in the Bay of Biscay (see Pingree
(1984)). The white straight lines across the image are due to data dropout in the
transmission or reception. (Image provided by R. Pingree, lOS)

In areas where cloud cover is less of a problem the daily overpasses of

Nimbus-7 have been used semi-operationally. Thus off the California coast,
charts showing the position and boundaries of water masses of different ocean
colour as detected by CZCS, have been issued to fishing vessels every few days
by NASA as part of a wider experimental programme to use satellites in support
of commercial fisheries (Fig. 6.28). The fishermen are able to locate particular
180 Visible wavelength ‘ocean-colour’ sensors [Ch.6

species of fish in particular colours of water, since the colour is strongly cor¬
related with the origin, temperature, and nutrient levels of the water masses in
this sea area.

Fig. 6.28 - A satellite-derived ocean-colour chart produced to assist commercial

fisheries off California (from Montgomery (1981))

6.3.5 Applications in dynamical oceanography

Besides the biological science which must develop from them, satelhte ocean-
colour data also promise advances in dynamical oceanography. Treating the
plankton or suspended sediment as a passive tracer, the satellite images reveal
patterns related to residual flows, eddies, frontal instabilities, or mixing processes.
Many such features may have a colour signature even when they do not appear
on thermal IR imagery. Attempts to obtain dynamical information from colour
images have mostly been qualitative, for example, using CZCS images to display
sediment movement in headland eddies as in Fig. 6.29. The clockwise gyre to
the south west of Portland Bill is clearly evident in the sediment pattern. Other
features are less readily explained. The streak-like fine structure in the central

Fig. 6.29 — CZCS channel 2 Image of the English Channel recorded on 26 March
1981. The more turbid water along the English coast is swept further south by
headland eddies such as that off Portland BiU (P). Note also the fine structure
lined up east-west with the tidal streams (e.g. at a) which is not yet fully
explained, and the banded feature (b) to the west of Jersey referred to in the
text. (Image supplied by and reproduced with permission of Dundee University
Satellite Receiving Station)
 182 Visible wavelength ‘ocean-colour’ sensors [Ch. 6

channel between England and the Cherbourg peninsular is apparently lined

up with the tidal streams, but remains unexplained. The cause of the clearer
(darker) water a few kilometres off the French coast near the mouth of the
Seine at Le Havre is also not clear. The feature west of Jersey, consisting of
banded areas of hght and dark water, might be thought to be atmospheric in
origin, but Pingree (1984) has shown it to be due to the west south-westward
residual transport of sediment-laden patches which are generated periodically
each cycle by the turbulent stirring of the tidal flow past Jersey.
Fig. 6.30 shows some unexpected spatial variability structures revealed
in a CZCS image of the Enghsh Channel. Since these are more likely to be a
chlorophyll rather than a sediment signature their size and persistence times
may be governed partly by biological growth dynamics, but it is to be expected
that this image also contains information about the turbulent eddy dynamics
of this area of sea, which is dominated by strong tidal currents of order 1 ms"h
We see here the advantage of a visible-wavelength image over an IR tempera¬
ture image. Both depend on there being gradients of a parameter which can
be deformed by the flow field in order to reveal dynamical processes. Because
the temperature in this area of the Channel is generally uniform, there is no
corresponding evidence for these eddy structures on IR images, whereas the
spatial patcloiness of biological productivity is able to reveal dynamical features
which were hitherto unknown.
Faced with evidence about a new-found dynamical feature, the physical
oceanographer’s reaction is to visit the area by ship and measure the velocity
field in more detail. However, in this case such an approach may not reveal very
much, since the eddy structures being viewed may have small velocities com¬
pared with the tidal velocities and therefore be lost in the current meter noise.
Indeed, it may be that the best and perhaps the only way to study features like
this is directly from the image, by measuring the length scales of variability,
and applying spatial Fourier analysis to the digital image to obtain a spectrum
of spatial frequencies. These ideas are still speculative and have yet to be fully
tested, but they represent the type of innovative approach which is required to
exploit the spatial coverage offered by satellite image data. In parallel with new
analysis techniques must come a new conceptual approach by dynamical oceano¬
graphers. On the whole, dynamical theories have hitherto been constructed
with time as the principal independent variable. We tend to study evolution
and variability in the time domain because oceanographers have become good at
measuring time series of dynamical variables using moored instrumentation. Now
that we can also thoroughly sample the spatial domain with sateUite-imaging

Too" CZCS channel 3 Image of the English Channel recorded on 4 April

1980. Note the turbulent structure in region (a) and the feature (b) corresponding
to the Hurd Deep. (Reproduced Avith permission of Dundee University Satellite
Receiving Station)
184 Visible wavelength ‘ocean-colour’ sensors [Ch.6

sensors, there is every reason to expand our theoretical framework to include all
aspects of spatial variability. So far, the testing of models of spatial variability
has largely been limited to advective systems in which the length scales could be
inferred from a point-measured time series, but the satellite scanner has removed
that limitation. It is worth noting also that the measurement of the length scales
of dynamical processes from images is not dependent on the development of an
accurate calibration algorithm. Provided that spatial variability of atmospheric
effects is removed, accurate length scales can be obtained even when there is
only a crude calibration available, in just the same way as frequency peaks can
be readily obtained from a time series whose amplitude calibration is suspect.
As more CZCS imagery becomes available for study, it is to be expected
that many new dynamical features such as those on Figs 6.29 and 6.30 will be
found, and as a result we can expect exciting developments in physical oceano¬
graphy in the next decade. Indeed, further close scrutiny of Fig. 6.30 will reveal
a long narrow feature over the centre of the English Channel which corresponds
exactly with the presence of the Hurd Deep — a long narrow trench where
the depth is 110 m compared with the 70 m surrounding it. How it comes to be
revealed in a CZCS image is at present an open question whose answer may lead
us into more areas of new oceanographic understanding.

6.4 OCEANOGRAPHIC USES OF LANDSAT MSS DATA

6.4.1 Data processing
Ideally, one would like to apply the same sensor calibration and atmospheric
correction techniques to Landsat MSS data as have been developed for the
CZCS. This is not possible, for several reasons, mostly related to the design of
the sensor for land rather than ocean use.
Firstly, it is difficult to obtain in-flight calibration information, and the user
must normally depend upon pre-flight calibrations of the sensor and digitiser.
The six parallel sensors which sweep out six lines together each have slightly
different radiometric and spectral characteristics, and whilst they may be
normalised to give the same output from a target with a spectrally uniform
radiance, observations over sea often give rise to a marked striping. This may
become worse over the sea after ground-station ‘de-striping algorithms’ have
improved the signal over the land, and it means that for most appUcations it
is necessary to smooth or filter the data with an 8 X 6 filter (8 samples, 6 lines)
to remove the noise. The stripes are even more apparent over the sea because the
sea radiance varies much less, and over longer length scales, than does the signal
from the land which is often strong enough over short length scales to make the
stripiness unimportant over land. The smoothing which is necessary over the
sea reduces the effective spatial resolution of the data, but unless very small
estuaries and embayments are being studied, this does not usually cause problems
in marine applications.
Sec. 6.4] Oceanographic uses of Landsat MSS data 185

In attempting to apply an atmospheric correction, the problems with the

Landsat MSS are its poor spectral and radiometric resolutions which one hopes
will be much less with the Thematic Mapper. The sensor bands are nominally
100 nm wide, and the radiation entering the sensor is filtered through a spectral
window which is not uniformly flat (Fig. 6.1). If the strategies presented
in section 6.2.4 are to be adopted, each of the wavelength-dependent variables
such as Lr should be calculated by convolution with the Landsat spectral
window and integration over the bandwidth — a tedious and time-consuming
process. The alternative approximation is to evaluate variables at the central
wavenumber of each band, and then multiply by the bandwidth to obtain
the total radiance recived by the sensor. In this case the use of (6.14) becomes
suspect, because S probably has a nonlinear wavelength dependence, and it is
not clear which wavelength to select as representative of a particular broad
band.
The MSS has the advantage over the CZCS in that band 7 (800—1100 nm) is
in that part of the spectrum for which, in all but the most turbid of fluid mud
suspensions, the water-leaving radiance will be zero. The advantage is lost
because the band is so wide, and the radiometric resolution so poor, that it is
difficult to extrapolate atmospheric aerosol information from band 7 into bands
4, 5, and 6.
Therefore apart from some recent work by MacFarlane & Robinson (1984)
to adapt the CZCS strategy, attempts to provide atmospheric corrections for
Landsat MSS have tended to be much cruder. One easily-implemented approach
has been to subtract from a whole scene the radiance level of the darkest pixel
(Johnson (1975)) on the assumption that it corresponds to clear water for which
the water-leaving radiance is zero and the signal is entirely due to path radiance.
Unless there is some open-ocean water in the scene, this is a doubtful assump¬
tion for bands 4 and 5, but is better than nothing. No attempt is usually made
to apply a differential atmospheric correction across a scene, and most attempts
at quantitative interpretation of Landsat data over the sea have incorporated
the atmospheric effects in empirically-derived regression constants as part of the
calibration process (for example, Klemas et al. (1974)).

6.4.2 Calibration of Landsat data for sediment and chlorophyll

Given the broader bandwidths, and the poorer atmospheric corrections, the
attempts to calibrate Landsat data in terms of water quality have been crude
compared with the CZCS work. It is much harder to pick out the spectral peaks
and troughs with a 100 nm band, and therefore chlorophyll has been very hard
to detect, except in case 1 waters where it is related to particulate scatterers
(Le Fevre et al. (1982)). Because of its high spatial resolution, Landsat data
has been applied mostly in coastal and estuarine regions where case 2 water
predominates. Calibration algorithms are therefore only locally applicable.
186 Visible wavelength ‘ocean-colour’ sensors [Ch. 6

Because of the infrequent repeat cycle, synoptic sea data have also been difficult
to obtain, and it is much harder to test the accuracy of calibrations.
Most attempts have been made to relate the satellite radiances to the
suspended sohds concentration. Algorithms fall into two broad categories,
those which use colour (multispectral information) and those which use
intensity (single-band information). Since, in general, non-chlorophyll-related
suspended solids do not have a strong colour signature, it is more useful to
seek simple relations between S and the reflectance in a single band. Often
band 5 has been the most useful since it is less contaminated by atmospheric
effects. For low concentrations (<60 mg T^) a linear relation is satisfactory,
whilst over a wider range of sediment loads the reflectance is better correlated
with log S (Munday & Alfoldi (1979)).
The most successful use of colour correlation with S has come from the
use of chromaticity analysis, as described in section 5.9. A good example is
presented by Alfoldi & Munday (1978), and is illustrated in Fig. 6.31. Radiance
values for Landsat scenes of the Bay of Fundy, Canada, were mapped on to
a two-colour chromaticity transform plane, and S (measured simultaneously
with the overpass) was correlated with the position along the resulting locus.
Atmospheric contamination was treated by shifting the whole locus to a
normalised position. The success of this technique in studying the Bay of Fundy
is probably due to the highly-coloured red sediments which are input by local
rivers, resulting in a strong colour signature to the concentration. This causes

Fig. 6.31 Radiance values from Landsat MSS channels 4 and 5 viewing Minas
Basin in the Bay of Fundy, Nova Scotia, plotted on a Chromaticity Transform
plane. There is a correlation between the suspended solid concentration measured
synopticaUy in situ and the position along the chromaticity locus of the
corresponding pixels (adapted from Alfoldi & Munday (1978))
Sec. 6.4] Oceanographic uses of Landsat MSS data 187

the locus in the transform plane to lie at an angle to both axes, so that better
resolution of sediment values is obtained by correlation of sediment with the
position along the locus than by using the position along either axis.
MacFarlane’s haze-corrected data for the Solent estuarine area in the south
of England was also calibrated against synoptic sediment data (MacFarlane &,
Robinson (1984)). As well as correcting for atmospheric effects, the radiance
values which came from four separate overpasses have been normalised by

eo-i

0 i 1-1-1-1 ■■ 1
20 40 60

Fig. 6.32 - The correlation between gravimetrically-measured suspended sediment

concentration, S mg h', and synoptic data from Landsat MSS, for the west Solent,
UK. * = 2 Nov. 1979, -i- = 18 Feb. 1980, 0 =19 May 1980, X = 24 June 1980.
(a)S versus 7^, the digital count in channel 4
(b) S versus /j, the digital count in channel 5
(c) S versus R^, the subsurface reflectance ratio in channel 5, after atmospheric
correction based on channel? has been applied. The line is S = 167 J?; — 5.51
(d) S versus X^, the chromaticity-transformed channel 4 data, after atmospheric
correction. The line is 5 = —354 + 252.
 188 Visible wavelength ‘ocean-colour’ sensors [Ch. 6

scaling with the factor l/cos0o to allow for varying sun illumination from scene
to scene (where do is the sun zenith angle). Fig. 6.32 shows the resulting cor¬
relation that can be achieved between a satellite measurement and the suspended
sediment measured gravimetrically in samples collected within the top metre of
the water column. Figs 6.32 (a) and (b) show the poor correlation between the
sediment and the raw satellite data (the MSS digital counts in channels 4 and 5);
(c) and (d) show the good correlation possible after atmospheric correction and
normalisation for solar illumination; (c) is a correlation with the channel 5
reflectance; and (d) is for the chromaticity-transformed channel 4 data. In both
cases, the atmospheric correction and illumination normaUsation has brought
the groups of data points from different overpass dates onto or close to a single
straight line. The conclusion may therefore be drawn that given sufficient field-
measured sediment data synoptic with an overpass, and by applying fairly
lengthy processing operations to the data, it is possible to achieve a multidate,
site-specific sediment calibration algorithm for Landsat MSS. Because we are
usually deahng with case 2 waters, and the sediment optical properties can
be expected to vary from estuary, to estuary, a universal algorithm in terms
of sediment seems unlikely. However, it may be possible to develop a universal
algorithm which relates atmospherically-corrected, illumination-normalised
Landsat data to an optical field measurement of turbidity. It would then
be possible to achieve caHbration of previously-received Landsat data in terms
of the backscattering properties of the water. This could be interpreted in
terms of the local type of sediment by field measurements made completely
independently of further satellite overpasses. If the appUcation merely called for
optical turbidity information there would be no need to visit the estuary at all.
Although such a universal algorithm seems promising, it is yet to be developed.

6.4.3 Sediment movement studied with Landsat data

Various workers have used Landsat MSS imagery to define sediment patterns
in estuaries, some of them caUbrated in terms of sediment concentration, others
merely quahtative. Relatively large enclosed areas of water such as the Baltic
(Horstman & Hardtke (1981)), the Bay of Fundy (Amos & Alfoldi (1979)),
and the Louisiana Bight (Rouse & Coleman (1976)) have been studied as
well as smaller estuaries Hke the Solent (Robinson & Srisaengthong (1981)).
Fig. 6.33 from the Solent illustrates the sort of patterns obtainable from a single¬
band image after smoothing and atmospheric haze removal. In this particular
image there is a strong tidal flow from east to west through the Solent, and
the turbidity streaks are lined up parallel to the tide. Sea observations suggest
that there are indeed streaks of water with higher suspended sediment concen¬
trations at this stage of the tide, and that the features are not therefore merely
haze streaks or patches of different surface roughness giving different surface
reflection effects.
Sec. 6.4] Oceanographic uses of Landsat MSS data 189

Fig. 6.33 - Landsat MSS image of R^, the subsurface reflectance ratio in
channel 5, for the west Solent, UK. Atmospheric correction based on channel 7
data and an 8 X 6 smoothing filter have been applied. Note the banded structure
between Cowes and the mouth of Southampton Water, which is ahgned with
strong tidal streams flowing westwards through the Solent at this state of the tide,
VA hours after high water. The image was recorded on 23 Feb. 1979. (Image
processed by N. MacFarlane at Space Dept., RAE Farnborough, UK)

It is important when analysing images such as these to make sure that the
patterns observed are genuinely of suspended sediment distribution. If it is not
possible to make an inspection at sea, then it can often be helpful to look at
several images. Those features which are permanent, or are consistent with tidal
conditions, are likely to have a marine rather than an atmospheric explana¬
tion. It is harder to be certain that it is not a surface-roughness pattern which is
being observed. Like the sediment, this could line up with tidal flows, revealing
streaks of faster-flowing water which against a headwind could produce a region
190 Visible wavelength ‘ocean-colour’ sensors [Ch.6

of whitecapping. Since the sediment patterns being sought give rise to very small
differences in radiance, it is not unreasonable to expect that whitecaps could
produce a comparable radiance change. The availability of sea data, even if not
synoptic, will help to resolve the issue. Great care is required in interpreting
images of remote areas when no first-hand knowledge of the local oceanography
is available.
If the interpreter is reasonably certain that sediment patterns are being
imaged, there remain two problem areas to be overcome before the satellite data
can be used for studying sediment dynamics. The first is that the satellite only
responds to sediment loads in the one to five metres or so near the surface,
and less for very turbid water. Yet the potential appHcations of the data in civil
engineering studies of sedimentation, deposition and scouring, and sediment
transport, are all much more concerned with either bedload transport or the
concentration of near-bottom suspended sediment. There is a need to gain .much
more knowledge and understanding of the vertical distribution of suspended
sediment throughout the water column before the use of Landsat data can be
fully integrated into offshore and estuarine sedimentation research programmes.
The other problem is that with such infrequent cloud-free Landsat imagery,
it is not usually possible to obtain time sequences to detect even seasonal pattern
changes. Of course, given a knowledge of the tidal flow patterns in an estuary, a
single image revealing a synoptic sediment distribution can be readily interpreted
in terms of sources and sinks of sediment, direction of sediment movement,
etc., bearing in mind the limitations of a near-surface view as discussed in the
previous paragraph. There are ways, however, of gaining a measure of time-
sequence information. One is by using the overlap region to obtain passes on
adjacent days (possible with Landsats 1-3 but not with the orbit of Landsat 4).
Indeed, at higher latitude with a greater overlap it is occasionally possible to
obtain images on three successive days, as demonstrated by Horstman & Hardtke
(1981). Another approach in a tidaUy-dominated estuary is to arrange the avail¬
able archived Landsat images in sequence according to the stage of the tide
within the 1214-hour semidiurnal cycle at the time of the overpass, irrespective
of the actual time or date of the image. This has been shown to make a fairly
consistent interpretation possible in the Solent (Robinson & Srisaengthong
(1981)). For example, the patterns shown in Fig. 6.33 are repeated almost
exactly on other images viewed at the same time of between 1 and 2 hours
after local high water, but months or years apart in real time. In contrast,
other images for other states of the tide show a marked difference in pattern^
consistent with different tidal flow conditions.

6.4.4 Other oceanographic processes viewed by Landsat

As well as studying sediment processes, Landsat MSS data can be used by
oceanographers in the coastal zone for several other applications. Very often
sediment is the tracer in these studies, but it is not being measured for its own
Sec. 6.4] Oceanographic uses of Landsat MSS data 191

sake, nor is accurate calibration essential, the spatial patterns being the important
information. On many Landsat images of the coastal zone it is possible to see
the plumes of discharging rivers and large sewage outfalls, and on occasions
the plumes of sludge dumped by barges at sea. Whilst satellite surveillance does
not have such a major role to play in the pollution monitoring of estuaries as
does airborne remote sensing, it is still possible to learn a lot about the dispersal
of discharged matter by studying several Landsat images from different states of
the tide.
Sometimes the contribution to pollution control can be more subtle,
as in the work of Klemas (1980). This examined the position of small frontal
structures in Delaware Bay which had a definite colour signature on the Landsat
imagery, and which also were influential in pollutant dispersal. It is in its ability
to offer a synoptic view of estuaries that Landsat has considerable potential
in coastal oceanography. Coastal areas can have such a complex combination of
eddies, fronts, and plumes that it is sometimes difficult to interpret observations
from ships, whereas the satellite image can place these in context.
As with the CZCS, Landsat MSS can also reveal interesting biological
features such as the plankton bloom viewed in the English Channel on Landsat
images (Le Fevre et al. (1982)). Further study of why this should have such
a strong sediment-hke signature revealed that it was due to a species of
Coccolithophores with calcite scales which outlasted the life of the organism and
produced a dense suspension of particulate material following a bloom (Holligan
et a/. (1983)).
Another interesting use of Landsat imagery apparently viewing another
plankton bloom, this time in the North Atlantic, is the work of Gower et al.
(1980). They analysed the spatial variability of the radiance in the Landsat data
and used this to comment on the turbulent eddy length scales of the sea area.
This is an example of the innovative approach to exploiting satellite data in
dynamical oceanography which was discussed in section 6.3.5.

6.4.5 Bathymetry using Landsat

In the study of sediment dynamics with Landsat one problem which occurs is
the influence of bottom reflection being mistaken for water turbidity in shallow
areas. If the water is very clear and has very little backscatter from suspended
material, it is possible to relate the upwelling subsurface irradiance to the water
depth, provided it is no deeper than around 20 m and at most 30 m. Channel 4
of Landsat MSS penetrates furthest, but is more contaminated by haze.
However, if it is atmospherically corrected, and it is known that the water is
very clear, it is possible to construct bathymetric charts directly from radiance
contours. If atmospheric correction cannot be performed, channel 5 provides a
less contaminated alternative which will not penetrate so far. Bullard (1983 a,b)
describes the method from the surveyor’s point of view, and concludes that it
has a role to play in hydrography, particularly for giving an overview of areas
192 Visible wavelength ‘ocean-colour’ sensors [Ch.6

where a chart is not available, or for checking the vaUdity of existing but old
charts. The fact that Landsat will not penetrate much below 20 m in clear water
is not too great a drawback, since most vessels that will use the charts for naviga¬
tional purposes have a draught less than 20 m. Fleming and Le Lievre (1977)
were able to chart features off the Labrador coast which could be subsequently
verified by a survey vessel.
In many locations, the bathymetric application is hindered by the presence
of sediment. By using both bands 4 and 5, and even 6, it ought to be possible
to begin to distinguish between the backscatter due to sediment and that due to
the sea bed, provided that suspended sediment is uniform with depth and has
a uniform reflectance over the spectrum. Then in deep water only the sediment
will show up, uniformly in all bands. But as shallow water is approached the
reflectance in band 4 will increase more rapidly than in band 5. Cracknell et al.
(1982) attempted a quantitative study of bathymetry in the presence of
sediment for the Tay estuary, but with inconclusive results.

6.4.6 Coastal geomorphology

The study of coastlines is straying across the bounds of oceanography into the
land application of remote sensing, but it would be incomplete in this brief
survey of coastal and marine applications of Landsat not to mention its valuable
role in mapping the sea coast. Because channel 7 is such a reliable indicator of
an open water surface, which in this waveband has a very low or zero reflectance
compared with the adjacent land, it is very easy to delineate the land—sea
boundary. This can be used in a well-mapped area for studying the detail of
coastal erosion, the migration of sand bars, etc. In a less-well-mapped area it
can be used to provide the map base of the coastline of inaccessible islands and
headlands. It can even be used to study beach profiles, since a suitable collection
of images over the tidal cycle will reveal the movement of the land—sea boundary
over a tidal cycle, at least in those cases where the movement is significantly
further than 80 m and thus resolvable on the images.

6.5 OCEANOGRAPHIC USE OF OTHER VISIBLE-WAVELENGTH

SATELLITE DATA
As well as the CZCS and Landsat MSS, several other satellites and sensors have
collected visible-wavelength data. The NOAA AVHRR has two visible channels,
and the geostationary meterological satellites, GOES and Meteosat, have one
visible-wavelength channel. These are all very broadband sensors, but they are on
operational satellites and therefore a reliable continuing source of data. Their
intended purpose is meteorological, the detection of cloud patterns, but one
example of an oceanographic application illustrates the imaginative use to
which this freely-available information can be put in marine science. Gautier
(1981) reports how visible-channel data from GOES was used to estimate the
Sec. 6.5] Oceanographic use of other visible-wavelength satellite data 193

daily shortwave energy budget over the ocean. Essentially, the method simply
detects whether clouds are present or not within each pixel, and uses an appro¬
priate model to calculate from known atmospheric absorption effects how much
solar shortwave energy is reaching the sea. By measuring the brightness of cloud
tops with the satellite sensor, it is possible to make a crude estimate of the
absorbing properties of the cloud and improve the estimate of how much short¬
wave energy penetrates through the cloud. The insolation is modelled every
hour with the hourly updates of the scanned image, enabling a daily total to be
calculated. Therefore the satellite data is used in a novel way towards producing
a chmatology of shortwave energy into the sea.
A further development of this will be to use spectral information from the
CZCS to estimate the depth of energy penetration into the sea, and thence
to model more accurately the thermal structure of the upper ocean and its
horizontal variability. Simpson & Dickey (1981) and Dickey & Simpson (1983)
have already shown how such information could be put to valuable oceano¬
graphic and climatological use. As oceanographers become more familiar with
ocean-colour data from satellites it is likely that many other applications will
come to light.
 CHAPTER 7

Sea-surface temperature from

infra-red scanning radiometers

7.1 INTRODUCTION
The monitoring of sea-surface temperature (SST) from earth-orbiting infrared
radiometers is the technique of marine remote sensing which has had the widest
impact on oceanographic science. For example, the displaying of an IR image
of a frontal or eddy structure is quite commonplace within a scientific paper
describing observational or theoretical studies of a dynamical feature. Amongst
the reasons for this must be included:
(a) The very good correlation, even without complex atmospheric correction or
calibration algorithms, between light and dark shades of grey on an IR image
and cooler and warmer water, giving a valid qualitative view of a thermal feature
in the ocean.
(b) The operational nature of the meteorological polar-orbiting satellites in the
tiros/noAA series and the Geostationary Meteorological Satellites, which has
ensured a regular and continuous supply of data from the infrared radiometers
which are carried on them, primarily for meteorological purposes but with a
ready capability for ocean surveillance given cloud-free conditions.
(c) The wide dissemination of the IR data, due to the Automatic Picture
Transmission (APT) facility on both series of satellites carrying IR radiometers,
and also the comparative ease of receiving the full resolution data at receiving
stations around the world. For example, British oceanography has benefited
greatly from quick and easy access to IR data through the Dundee University
Satellite Receiving Station, an example of how high-quality data reception
can be achieved at low cost (Bayliss 1981a,b). This is only possible with the
relatively low data rates required by the current generation of IR sensors, and
high-resolution imaging will require higher-frequency data transmission which
demands more sophisticated and expensive ground stations.
The principal source of infrared data for sea-surface temperature measure¬
ment is the TIROS/NOAA series of polar-orbiting meteorological satellites.
Tables 2.2 and 2.3 list the general characteristics of these satellites in com¬
parison with others, and details of the sensor operating wavelengths and image
[Sec. 7.1] Introduction 195

size are presented in Table 7.1. The Advanced Very High Resolution Radiometer
(AVHRR) was first flown on TIROS-N in 1978. Its four spectral channels, two
visible and two infrared, and its digital data transmission with 10-bit binary
word size (1024 levels) were a distinct improvement on the VHRR flown on
the previous NOAA 5 satellite which had just one visible and one IR channel and
an analogue data transmission with about 100 radiance levels. An AVHRR was
flown on NOAA 6, whilst NOAA7 carried an AVHRR/2, which has an extra
channel in the infrared. Being an operational meteorological satellite programme,
there is always at least one functioning satellite in orbit. The original operational
system had two satellites working, in sun-synchronous orbits separated by about
90° (approximately 6 hours) of longitude, so that one would give a morning and
the other an afternoon daytime overpass at each location. The trimming of space
programme budgets may mean that satellites will operate only singly in future,
with the sensor possibly alternating between AVHRR and AVHRR/2.

Table 7.1 Characteristics of the Advanced Very High Resolution Radiometer

(AVHRR)

Channel AVHRR AVHRR/2

Wavebands 1 0.58t - 0.68 0.58 - 0.68

(Mm) 2 0.725 - 1.10 0.725 - 1.10
3 3.55 - 3.93 3.55 - 3.93
4 10.5 - 11.5 10.3 - 11.3
5 channel 4 repeat 11.5 - 12.5

t 0.55 jum on TIROS-N

Sensitivity of thermal IR channels NEAT 0.12 deg K at 300 K

(3,4,5)
No. of digitisation levels 1024
Angular field of view 1 .3 ± 0.1 mrads
Ground IFOV at nadir approx. 1.1 X 1.1 km
(4 X 4 km in low resolution mode)
Cross-track scan ±55.4°
Swath width 2580 km
(4000 km in low resolution mode)
Data rate (includes all spacecraft
instrument data) 665.4 X 10^ bps
196 Sea-surface temperature from infrared scanning radiometers [Ch. 7

The Geostationary Meteorological Satellite programme also provides infrared

imagery of the oceans. Although its 5 km X 5 km sub-point resolution is much
poorer than the AVHRR’s 1 km X 1 km resolution, and at mid to high latitudes
it views the earth surface very obliquely, it does have a 30-minute repeat sampling
capability. Tables 2.2, 2.3, and Fig. 2.6 illustrate some of the characteristics of
this series, and Table 7.2 lists the sensor characteristics of the American GOES 4
and 5 and of Meteosat, the European Space Agency’s contribution to the
international GMS programme. The Meteosat radiometer and the Visible and
Infrared Spin Scan Radiometer (VISSR) on the GOES satellites each have a
single IR and a single visible channel for earth viewing.

Table 7.2(a) Characteristics of the Meteosat radiometer

Infrared watei■ Infrared earth

Spectral bands Visible (VIS) vapour (WV) viewing (IR)

Wavelength range, nm 0.4-1.1 5.7-7.1 10.5-12.5

No. of channels 2 1 2 identical
Operation of both
produces two simul- (1 is redundant)
taneous non-overlapping
adjacent lines
Scan options (a) both channels off one channel
(b) one channel on one channel
No. of samples per line 5000 2500 2500
No. of lines per image 5000 2500 2500
of full earth’s disk 2500t
Resolution at nadir, km 2.5 X 2.5 5X5 5X5
2.5 X 5.Of

Spin rate
100 rpm
Time to sample one line 30 ms
Time interval between adjacent lines 600 ms
Radiometer telescope rotation between lines 0.125 mrad
Total north-south scan angle 18°
Total time to produce one image 25 min
Image repeat time 30 min

t Scan option b
Sec. 7.1] Introduction 197

Table 7.2(b) Spectral characteristics of the GOES visible and infrared spin scan
radiometer VISSR

Spectral range. Ground resolution.

BAND jUm km

VISSR (GOES 1^5) VIS 0.55 - 0.70 14

IR 10.50 - 12.60 8

V-1 0.55 - 0.70 14

VISSR atmospheric sounder IR-1 14.60 - 14.81
(GOES 4 & 5 only) IR-2 14.29 - 14.62
IR-3 14.06 - 14.39
IR-4 13.79 - 14.18
IR-5 13.12 - 13.48
IR-6 4.496- 4.537 16
IR-7 12.50 - 12.82
IR-8 10.36 - 12.12
IR-9 7.143- 7.353
IR-10 6.390- 7.067
IR-11 4.386- 4.484
IR-12 3.623- 4.310

As weU as the visible and the IR radiometers, both series of satellites carry
atmospheric sounders which measure the radiance at wavelengths where there
is no clear atmospheric window through which to view the earth. These provide
information about the atmospheric composition which can help in the atmos¬
pheric correction of the SST data. On the NOAA series this is the 20-channel
High-resolution IR Spectrometer (HIRS), with a ground resolution of 17.4 km.
On Meteosat there is a single-channel water-vapour (WV) radiometer, whilst
GOES 4 and 5 carry the VISSR Atmospheric Sounder (VAS).
Other satellites which have carried infrared radiometers capable of yielding
sea-surface temperature data are listed in Table 7.3. The HCMR is particularly
useful for measuring diurnal temperature fluctuations, and Deschamps & Frouin
(1984) present an oceanographic appHcation. However, some of the instruments
in Table 7.3 have been beset with problems, or have not provided such high-
resolution or drift-free data as was expected, and in the rest of this chapter we
shall draw most examples from the highly successful AVHRR programme.
198 Sea-surface temperature from infrared scanning radiometers [Ch. 7
G
O
T3ci> &
eg
Wj

U-)
cd S _2G X
£ a. G
o s
-G V3 ^ 00 4-* -4-i
O ’-H CN G cn
<U tn ^ G .G
wc^ OT3
cd o^ ^ C 2 o CN
s (U
C/5 r- £ i ON
t t
^
O cd
>>
X*
I •r*
^>7 cd
o ^
d e e
(N
tT)^
a> ^
o
l-l s s
a o 5
D.
oc .S
o
s x od
s
cd QJ
cd
G

O c/3
C/3
c/3
s
a
X o PC
’O .2 ^ X
C3 T3 IZ)
(N X g >
it) G
00
3 cd
W-) X X
m 2 c <
X) X G cd 5/3

S On c/5 •o
X
.2 l-l X
o z
U
N
CN cd .2 o
z U
00
^ o
oJ &o
(U G* V3
cd

u 6D
s
o X »o
s .s
cd
o
TO r4 CN *cd >
IZ)
G OD
(D in > ?
cd cd t<_,
T3 o X t T3 X CO ^ O
G r-- • in o
cd • CN
X
H
d o
Si
s cd ^
Q ca
•4-*

00 X G X
'G (N CO O o
G 00 CN o
in c/3 G
cd cd
T3 o X t X 00 •G
cd G On
cn
(D l-l
CA cd
X cn
d
CO
G
o
Xcd
<N
u ^4 X

a OX) 00 X
o .G -G
o
cd
a G Cd cd (U
G
cd
G >-i u O
G
cd
CO
G CO
E c
s >.
4-> <H
o
/-s in
• <N
O
00
(N
O O
O O
I ^
’
5 ^ cd o ^ 1—( 4-* <n X
>.
620

g.2 G G Cd
cd cd 60 t t X X
o .52 S.o
o o E g
(O.f
10.5

X* "cd ^
00
(D o o S3 S
0) lo X

i X
S 2
X 2 c0)
73 JrJ
a>
0^ cd
G C
G/
u u .SP-'S
e/3 a>
X X (U
Q
>
o
t-’
pS G
o
z
Cd E *•4 X
H <D 4. X
o g x"
G C/3 CO 4-»
c/3
G cd ^G a> •G
X M x"
O G G T3 c
4-'
i-H cd 13 G
G cd
a X O H G X E
0) OX) c/3 < G
G 1 O E
cd ‘(U 0) W IH cd o
C/5 ffi C/5 1 Z a CO U
Sec.7.2] The physics of infrared radiation 199

7.2 THE PHYSICS OF INFRARED RADIATION

7.2.1 Thermal emission
It was mentioned briefly in Chapter 2 and illustrated in Fig. 2.12 that the sun
emits radiation at shorter wavelengths (peaking in the visible) compared with
the thermal emission of the earth (peaking around 10 urn) which is at a much
lower temperature. SST measurement using IR radiometry makes use of this by
measuring the radiation emitted in the 10—12ium waveband as an indicator of
the sea-surface temperature.
The spectral characteristics of thermal emission from a body at temperature
T degK are described by Planck’s radiation law:

Cl
(7.1)
\Mexp(C2/Xr) - 1]

is the spectral exitance (sometimes called emittance), that is, the radiant flux
density of radiation per unit bandwidth centred at leaving unit area of surface,
irrespective of direction (the equivalent of irradiance but for energy leaving
rather than falling on a surface).
X is the wavelength in metres.
Cl and C are constants with the values;
2

Cl = 3.74 X 10-^®Wm^
C = 1.44 X 10-2 nr degK
2

giving a value for in W m'^ m" . To obtain in W m-‘ fjLm~^ as is more

customary (7.1) should be multiplied by 10“®.
Equation (7.1) is based on ideal thermodynamical principles which will only
apply if the surface is a perfect emitter (sometimes termed a black body). The
emitting properties of a real surface are described by its spectral emissivity, e(X);
Mx (real surface at temperature T)
e(X) = - (7.2)
Mx (perfect emitter at temperature T)
which is only weakly temperature-dependent, but may vary significantly with
wavelength.
Integration of (7.1) over aU wavelengths gives the total exitance of a
black body:

M^oT^ (7.3)

where o = 5.669 X 10“® Wm'^ is Stefan’s constant.

Fig. 7.1 illustrates perfect or black-body exitance at various temperatures
according to (7.1), and the area under each curve is given by (7.3). The spectral
peak in the exitance curve is found at wavelength Xmax given by Wien’s
displacement law;
200 Sea-surface temperature from infrared scanning radiometers [Ch. 7

r = Cj (7.4)
where C = 2897 fxm K“^
3

Fig. 7.1 - Infrared emission spectra of black bodies at different temperatures

In the context of remote sensing of the sea and land surface, we note that
the peak of the emission spectrum for a perfect emitter at 300 K is around
10 /im, and sensors operating near this wavelength will be most suited for
measuring the temperature of the surface. From (7.1) and Fig.7.1 it is apparent
that even at the wavelength of the sea-surface emission peak, the direct solar
radiation will still be substantially greater than the earth radiation. At night this
is not a problem because there is no solar radiation, but in the daytime reflected
solar radiation must be allowed for. Experiments have shown that a surface
will absorb a proportion of incident IR radiation equal to its emissivity, e. For
the sea surface, e is about 0.98, and varies very little with wavelength, tempera¬
ture, or surface roughness (in the wavelength range of interest between and3

14 ^lm), although it can be affected by slicks of surface material. It does decrease

markedly with the viewing zenith angle, when this is less than about 50°, and
at such zenith angles surface slopes due to waves will have a noticeable effect.
At around 10-12 /xm, for small zenith angles, the thermal emission by the sea
surface fortunately turns out to be substantially greater than the reflected solar
radiation, whereas at 3-4 jum the surface exitance is seriously masked by the
reflected solar irradiance, so that sensors using this waveband are of little use for
SST measurements except at night-time.
Sec.7.2] The physics of infrared radiation 201

As discussed in section 6.2.2 a satellite sensor measures the radiance, that is,
the radiant flux per unit solid angle per unit projected area per unit bandwidth.
For infrared emission, the surface can be considered to be Lambertian with
uniform radiance in all directions, so that:

Lx = Mjn. (7.5)

Given L^, it is therefore possible to determine T by using (7.5) and inverting

(7.1).
Since (7.1) applies to a perfect emitter, T in this case is the apparent or
brightness temperature at that wavelength. This is defined as the temperature
of the perfect emitter required to produce the same spectral exitance at the
given wavelength. If e is known, (7.2) can be used to obtain the exitance of a
perfect emitter at the true temperature of the actual surface, using:
(measured)
M^(perfect) =-—-,
e(\)
whence (7.1) can be used with Af;y^(perfect) to retrieve the true temperature.

7.2.2 Atmospheric absorption

In the infrared waveband of interest, between 3 and 14/zm, the atmosphere
interacts with incident radiation by absorption, and by re-emitting the radiation
at a different wavelength, rather than by molecular or even aerosol scattering
as in the visible wavelengths discussed in the previous chapter. Fig. 2.13 shows
the transmittance of the atmosphere across the spectrum, and Fig. 7.2 indicates
the atmospheric molecules responsible for absorption at different wavelengths
within the infrared region. Such spectra can only be approximate, because the
actual transmittance of the atmosphere varies from place to place and from day
to day as the constituents in the atmosphere, and their distribution through the
depth of the atmosphere, vary.
The principal absorbers are water vapour, ozone, and carbon dioxide, and
only the last of these is appreciably constant and uniformly distributed through
the atmosphere. The ozone layer is found between 20 km and 30 km height and
varies diurnally, being greater during the day when sunlight generates ozone by
ultraviolet interaction with oxygen. The water vapour is found in the lower part
of the atmosphere, the troposphere, mostly below 10 km height. The water
vapour content can vary significantly in horizontal distances of order 1000 km,
and is generally much greater in warmer air, thus varying latitudinally and
seasonally.
The atmosphere being cooler than the ground or sea, the radiation absorbed
by the constituents is re-emitted as from a lower temperature, with a consequent
shift of the spectral peak to longer wavelengths. Thus the atmospheric effect
will be to reduce the radiation reaching the sensor, and to reduce the apparent
surface temperature.
 202 Sea-surface temperature from infrared scanning radiometers [Ch. 7

WAVELENGTH^m

Fig. 7.2 - Absorption by different atmospheric constituents across the infrared

part of the spectrum
TEMPERATURE DIFFERENCE (K)

Fig. 7.3 Spectral variations ot the atmospheric correction to sea-surface

temperature retrieval from infrared radiance, computed for several model atmos¬
pheres. Note the small corrections necessary in the transmission windows. (From
Deschamps & Phulpin (1980))
Sec.7.2] The physics of infrared radiation 203

Absorption is located spectrally within very narrow bandwidths, but these

are spread by broadening processes to almost fill certain parts of the spectrum.
As well as the line absorption spectrum of O3, CO2, and HjO, there is a further
absorption due to water vapour which is continuous over the spectrum, so that
even where there are no absorption lines the transmittance is not 100%. The
resulting absorption spectrum has three ‘windows’ between 3 and 5 /im, 7 and
8 /zm, and 9.5 and 13 /am, none of which is completely transparent, so that even
in these regions of the spectrum the ground can only be seen by an infrared
sensor at a brightness temperature lower than its true temperature. The task of
an atmospheric correction algorithm is to estimate what effect the atmosphere
has on the infrared radiation, and by how much to raise the apparent brightness
temperature to reach the true sea-surface temperature.
Further details of the physics of atmospheric interference are discussed by
Singh & Warren (1983) and in the several papers on infrared radiometry of the
sea surface in the collections edited by Gower (1980,1981). As an indication of
the magnitude of the atmospheric correction in the different windows, Fig. 7.3
shows the values calculated by Deschamps & Phulpin (1980) for different model
atmospheres. The 3.7 /am window is the most transparent overall, although
the 9.5 to 13 /am band performs equally well in the mid-latitude winter atmos¬
phere, with a temperature difference of half a degree or less, but through a
moist tropical atmosphere a 5 degK error occurs. The 7 to 8 /am window is
least satisfactory. Fig. 7.3 shows the rationale behind the choice of infrared

Fig. 7.4 — Relative spectral response for (a) the 3.1 channel, and (b) the
11 pm channel, of the AVHRR on TIROS N. (From Deschamps & Phulpin
(1980))
204 Sea-surface temperature from infrared scanning radiometers [Ch. 7

sensor bands centred at 3.7 /rm, 11 /Jin, and 12 nm. On balance, the higher wave¬
length window is the most useful because it corresponds to the peak of the
thermal emission spectrum, whilst the 3.7 /rm window is able to provide a useful
correction channel, particularly for night-time imagery not troubled by solar
reflection.

7.2.3 Infrared sensors

To understand the limitations of SST measurement from space, and to decide
how accurately it is worth trying to correct for atmospheric and other errors, it
is important to understand the errors that occur within the satellite instrument
itself. Unlike the visible wavelength sensors where the goal is often a colour ratio
rather than an absolute measurement, and where a few percent error is accept¬
able in most applications, infrared radiometers must be extremely accurate,
well calibrated, and stable, since a very small error in absolute radiance leads to
a large temperature error from Planck’s law (7.1).
To achieve any long-term calibration capability such as is required in the
compilation of SST databases, it is therefore necessary to have a calibration
black-body cavity at a known temperature carried on board the satellite, to be
viewed periodically by the sensor. If a reference black-body cavity is not
included in the sensor package, calibration of an image can still be performed,
since relative temperature differences across the image may be quite accurate,
but in that case absolute calibration will depend on a ground knowledge of sea
temperatures over part of the image. Whilst absolute calibration depends on
the accuracy of the on-board referencing, the sensitivity of the sensor (that is,
the difference in brightness temperature which it is able to resolve between one
part of an image and another) depends more on the resolution of the radiometer
itself.
To illustrate the problems, we consider the NOAA/AVHRR which is the
sensor with the best-proven operational performance at the time of writing.
Since temperature is usually the end result of IR data processing, errors are
always expressed in terms of equivalent temperature errors. Because of the
highly nonlinear nature of (7.1), as illustrated by the curves of Fig. 7.1, the
same radiance uncertainty will give very different temperature inconsistencies
according to the temperature being viewed. Therefore temperature errors are
often presented in ranges corresponding to the worst and best cases within the
typical temperature range of the sea (that is, about 270 K to 310 K).
The radiometer resolution is controlled by the sensor noise and the digitisa¬
tion interval in the analogue-to-digital converter. The inherent sensor noise is a
function of the target temperature and the integration time over which is summed
the electrical output of the transducer, which converts the radiant energy into
an electrical signal. The longer the integration time, the easier it is to eliminate
random thermal noise. In fact the noise equivalent difference temperature
(NEAT) is inversely proportional to the square root of the integration time, for
Sec.7.2] The physics of infrared radiation 205

a given level of noise. The cooler the sensor, the smaller the Brownian thermal
noise. With a sensor cooled to 105 K (requiring a carefully designed heatsink
radiating out to cold space), the AVHRR achieves sensor noise levels of around
0.05 degK in the very short integration time of less than 25 fxs permitted during
the scanning process.
In the analogue/digital converter, the least count is an increment in energy
rather than temperature, and therefore the digitisation resolution in equivalent
temperatures varies both with the temperature of the sea surface being viewed,
and with the wavelength channel. For typical sea temperatures between 270
and 300 K, the resolution is 0.13 — 0.10 degK for channels 4 and 5, and
0.17—0.05 degK for channel 3. The sum of the sensor noise and the digitising
error is specified for channels 4 and 5 of the AVHRR at around 0.12 degK. In
practice, Bernstein (1982a) has shown that a NEAT of around 0.05—0.07 degK
appears to be achieved. Similar noise levels are specified for channel 3.
The black-body cavity for calibration of the AVHRR is heated to around
290 K, as measured by four platinum-resistance thermometers. Errors arise from
the assumed emissivityof the black-body, the existence of temperature gradients
across it (a problem for the AVHRR), and the accuracy and resolution of the
thermometers. The total temperature uncertainty in the radiometer estimation
of sea-surface temperature, arising from errors in the in-flight calibration using
the black-body, is quoted by the makers as 0.31 degK.
The significance of these errors depends on the application of the data. The
study of spatial temperature structure in the sea requires accurate measurement
of temperature difference, whilst absolute calibration is less important. NEAT
is therefore the limiting parameter in this case. On the other hand, a measure
of average sea-surface temperature over an area of many pixels (for example, for
a climate database) can reduce NEAT over many pixels, but in this case the
requirement of comparability with the data from other orbits, other satellite
sensors, and other methods of measuring the sea temperature, requires the
absolute calibration to be accurate. Averaging over several scan lines (and hence
several black-body calibrations) will reduce some of the calibration error, but
any bias in the calibration cannot be removed without recourse to sea-truth
comparisons.

7.2.4 Brightness temperature calibration of AVHRR data

The data dehvered by the AVHRR to the ground station and obtained by the
user on a CCT includes the digitised sensor values, DNx in each waveband for
each pixel across a scan line, and end-of-scan data encoded in a predetermined
format. In this is contained the digital values obtained from viewing the heated
black-body cavity, Xj, and from viewing cold, deep space, X^, as well as four
temperatures corresponding to the four resistance thermometers. The average of
206 Sea-surface temperature from infrared scanning radiometers [Ch. 7

these gives T^, the black-body reference temperature. For a given image, X^, X^,
and are averaged over many scan lines to reduce error before performing the
following calibration to estimate the apparent temperature of each pixel.
A separate calibration is performed for each waveband. Within each wave¬
band, the actual sensor response is governed by a normalised wavelength response
curve. This is measured before the flight and is available from the satellite manu¬
facturers (NOAA/NESS 1978, 1979) in the form of n discrete values of
for / = 1 to n. The X; are separated by AX so that X] and X„ encompass the total
response of the sensor in a given waveband, and

.2 ^(X,-)AX = 1
I = 1

(see Fig. 7.4). The emitted radiance from the black-body reference, as detected
by the sensor at waveband X, is then calculated as

n Ci^\(Xr)AX
1
(7.6)
’ ( =
X^[exp[C2/X/rr] -IJtt'
Theoretically, the radiance of the deep-space view should be similarly calculated
from the assumption that the temperature of deep space is 4 K. However, to
bypass a small nonlinearity in the digitising process, a correction is incorporated
into the deep-space radiance which is supplied along with the other calibration
constants as
With two radiance values, and two corresponding digital numbers in the
data stream, a linear, two-point, calibration can be constructed, with gradient

2s,A. ^T,X
(7.7)

and intercept

A — x X. (7.8)
Then the radiance measured in each pixel within waveband X can be obtained
from
L'x= GxiDN)x+Ix (7.9)
where the significance of L'x is that it is the radiance as recorded by the satellite
in the given spectral band defined by $x. Thus if the spectral radiance actually
reaching the sensor is L(X),

L'x= /\(X)0(X)dX. (7.10)

Each value of L'x for each pixel can be interpreted in terms of a brightness
temperature T^, the temperature of the perfect emitter which would produce
that radiance if there were no intervening atmosphere. This is related to by
Sec. 7.3] The near-surface thermal structure in the ocean 207

,, « Ci0;,^(X,)AX
L\ — 2- (7.11)

which is not readily inverted to obtain from L'x- Instead, (7.11) is used to
construct a curve of the form

Ty, = A + B\n{L'^, (7.12)

A and B being obtained by a least squares fit. Thence look-up-tables to relate

every value of {DN)-/^ to a value of Tb can be constructed from (7.9) and (7.12),
and the image dataset for waveband X is now calibrated in terms of Tb.
Computationally this is far more efficient than seeking to invert (7.11), since
(7.12) need be evaluated over only a limited range of temperatures correspond¬
ing to a limited range of DN appropriate to sea-surface temperatures as cor¬
rupted by the atmosphere. The Tb image can then be used as a rough indication
of temperature gradients and differences across the area of sea being viewed, or
can form the input to an atmospheric-correction algorithm.

7.3 THE NEAR-SURFACE THERMAL STRUCTURE IN THE OCEAN

7.3.1 The skin temperature
Once infrared radiometer data has been atmospherically corrected, there is no
doubt about interpreting it in terms of an ocean parameter. As we have seen, the
radiance is directly related to the sea-surface temperature. But is this an oceano¬
graphically useful parameter? To what extent can SST measurements from space
contribute to oceanography? The question must be asked because of the nature
of the remote-sensing observation, which is truly of sea surface temperature.
Since the IR radiation is emitted only by the layers of water molecules close to
the surface, it is their temperature which characterises the radiation and controls
the observed brightness temperature. The actual thickness of the layer whose
temperature is remotely sensed varies according to the wavelength of the radia¬
tion, but for the wavelengths of interest between 3 and 14/zm, it is less than
0.1 mm. Indeed, attempts have been made to measure the temperature gradient
in the near-surface layer using airborne radiometry, with a 3.4—4.1 jum wave¬
band to sense temperature at a depth of about 0.075 mm and a 4.5 — 5Anm
waveband to sense temperature at a depth of about 0.025 mm (McAlister &
McCleish (1970), McAlister et al. (1971)).
The temperature measured from space is therefore a skin temperature, and
its usefulness in contributing to oceanography in a wider perspective depends
on being able to interpret it in terms of the underlying temperature of the top
metre or so of the ocean. This is what oceanographers normally mean when
referring to sea-surface temperature, because this is typically what they have
 208 Sea-surface temperature from infrared scanning radiometers [Ch. 7

measured as SST. Clearly it may be different from what is measured by a

radiometer. Care must be taken when interpreting any data record purporting to
be sea-surface temperature, since what it actually means will depend upon how
it was measured. Ship measurements may be made by towing a thermometer
within the top metre of the water column, by measuring the temperature of
the engine cooling water intake, which may be several metres below the surface,
or by measuring the temperature of water pumped on board from a known
depth within the top few metres. The extent to which it is valid to compare this
type of measurement with the radiometer observation depends on the physical
coiiditions controlling the temperature structure in the top few metres of the
ocean. We shall look at these in the remainder of this section. It is also useful
for the reader who is a remote-sensing scientist rather than an oceanographer
to be aware of the typical vertical profiles of temperature in the ocean when
considering what we may learn about the ocean as a whole, including the non¬
surface waters, from observations of the surface temperature. How deep a layer
is characterised by the surface temperature? To what extent do dynamical
processes, fronts and eddies, etc. in the deeper layers of the ocean have a surface
temperature signature which can be observed from a satellite?

7.3.2 The thermocline structure of the ocean

Fig. 3.8 illustrates typical profiles of temperature in the upper 1500 m of the
world s ocean. In general, the temperature decreases monotonically with depth,
except in the polar seas where a layer of cold water between 50 and 100 m may
overlie warmer saltier water in the deeper layer. Apart from high latitudes,
the ocean is characterised by a thermocline, that is, a depth zone in which the
temperature changes rapidly with depth. Below this from 1000 m downwards the
temperature decreases very little with depth and is the same at most latitudes.
Above it is a layer of nearly uniform temperature, typically 50 to 200 m depth,
called the mixed layer. The mixed layer extends to the surface in low latitudes
so that sea-surface temperature measurements from a satellite, if they are typical
of the temperature of the upper two or three metres, are probably representative
of the whole of the mixed layer.
At mid-latitudes the same structure occurs, although the thermocline is
not so steep and the surface temperature not so high as in low latitudes. The
temperature achieved by the mixed layer is a balance between the solar heating
of the surface tending to raise the temperature and the stirring of the upper
layers by wind which tends to mix the heat downwards, reducing the surface
temperature and deepening the mixed layer. Since solar heating and wind stirring
show strong seasonal variation in mid-latitudes, the mid-latitude temperature
structure is seasonally variable. In winter, the low heat input and strong mix¬
ing by winds results in a deep mixed layer at lower temperature. In summer, the
surface layer heats up with increased insolation, wind stirring is generally less,
and a shallow surface layer heats up, becoming thermally isolated from the
 Sec.7.3] The near-surface thermal structure in the ocean 209

deeper part of what was the winter mixed layer by a strong temperature gradient
called the seasonal thermocline. The summer mixed layer may be only about
20 m deep after a calm spell of sunny weather. The actual depth of the seasonal
thermocline and the temperature of the upper mixed layer are dependent on
the recent weather history of a location, and vary considerably in time and
space. Indeed, the temperature profile may have a step-like structure down to
the depth of the main thermocline, the deeper steps corresponding to strong
wind-mixing events a few weeks previously on which have developed a new
seasonal thermocline, broken up by subsequent but weaker storms. Under these
circumstances, without temperature profiles available from ships or buoys it is
not possible to deduce from a satellite-measured sea-surface temperature map
how deep the surface layer extends. In winter it can be expected to reach to
the main thermocline, but in summer the upper mixed layer depth cannot be
specified.
These structures found between 20 and 1000 m limit the depth of sea
where temperature is the same as that of the near-surface water, and therefore
limit the general usefulness of satellite IR sensors. The development of a diurnal
thermocline is even more troublesome for the interpretation of the sea-surface
temperature within the top few centimetres as representative of temperature
in the upper mixed layer as a whole. The process is similar in principle to the
development of the seasonal thermocline. As the day progresses past noon, solar
heating causes the surface temperature to increase. Because the short-wave
solar radiation is progressively reduced by absorption as it penetrates the water
column, the water temperature rises most near the surface, and a strong gradient
of temperature may occur in the top metre or so of the water column, raising the
surface temperature locally above the upper mixed layer temperature (Fig. 7.5).
If there is significant wind-mixing, the effect is very weak because all the solar
energy is mixed well down throughout the upper mixed layer or contributes
to the slow growth of the seasonal thermocline. On a calm day, however, the
surface may be up to a degree K or more warmer than the mixed layer, whose
upper limit is now not found at the surface but at a depth of a metre or so.
Under these conditions, the satellite measuring the temperature of the top of
the diurnal thermocline cannot detect the true mixed-layer temperature. Yet it
is this rather than the temperature of the top few centimetres which is important
oceanographically in studies of the heat flux of the upper ocean. Moreover, it
is quite possible that the diurnal thermocline can mask a horizontal structure
present in the mixed-layer temperature. It is very hard to predict the diurnal
thermocline strength, and it can be very patchy over length scales as short as
tens to hundreds of metres.
The problems are greatest on sunny afternoons on relatively calm days.
Radiometric measurements of sea-surface temperature under these conditions
may differ significantly, by up to a degree, from the‘bulk’sea-surface tempera¬
ture representative of the top one or two metres, as measured from a ship
210 Sea-surface temperature from infrared scarming radiometers [Ch. 7

or buoy. Furthermore, whilst the patchiness may be detected from airborne

radiometers, it may be on a sub-pixel length scale which is averaged out by
the satellite sensor, misleading the interpreter into assuming that the diurnal
thermocline is not a problem on that image. IR images obtained during the after¬
noon, even in winter, are therefore to be treated with caution unless it is known
that a wind strong enough to destroy the diurnal thermocline was blowing at the
time.
Fortunately the diurnal thermocline is completely destroyed at night, not
only because of wind-stirring but also because the surface heat losses to the
atmosphere in the absence of any incoming solar radiation cool the surface layer.
This becomes heavier than its surroundings, sinks, and promotes gravitational
mixing which destroys the diurnal thermocline quite rapidly after sunset. Night¬
time IR imagery should therefore be completely free of the diurnal thermocline
problem.

Typical temperature, degC

15.0 16.0

Fig. 7.5 - Typical near-surface temperature profiles illustrating the diurnal

thermocline in calm, slightly-mixed and night-time conditions

7.3.3 The sea-surface skin-temperature deviation

Quite apart from the diurnal thermocline effects which occur through the top
metre or two, the temperature of the sea surface is typically a few tenths of a
degree Kelvin cooler than the temperature a few centimetres below. This is
due to the vertical heat flux, through the air —sea interface. Because the sea
surface inhibits turbulent motion if it is intact and not destroyed by wave¬
breaking and spray, the heat transport near the surface is restricted to molecular
conduction processes. The heat transport is able to support a large temperature
gradient where the conduction coefficient is small. Thus most of the tempera¬
ture difference between the surface and the bulk near-surface water temperature
Sec.7.3] The near-surface thermal structure in the ocean 211

(that is, at about 10 cm depth) is found in the top millimetre. At night the
solar radiation is zero, and the net upward heat transfer through the surface is
given by
2n = Qir + Gg + 2h + Ql

where
Gir = infrared radiation emitted by the sea surface (positive)
Gg ~ downwelling infrared radiation from the sky, cloud, and atmosphere
(negative)
Gh = sensible heat flux
Ql — latent heat flux (normally positive).
Under most circumstance is positive, and the temperature deviation
Sr is negative (that is, the surface is cooler than the water a few millimetres
below). During the day the solar radiation is largely at or near the visible wave¬
lengths. This penetrates below the sea surface as electromagnetic energy and,
apart from a small amount of absorption in the skin layer, contributes little
negative heat flux to offset 67 is therefore normally negative during both
day and night, and observations suggest a typical value is between —0.1 degK
and —0.5 degK.
Under special conditions where Qn is negative, 5T has been observed to be
positive, but this is exceptional. The physical processes controlling the surface
skin temperature are illustrated schematically in Fig.7.6.
The subject has attracted a small but steady trickle of interest from physical
oceanographers, and has been reviewed in some detail by Katsaros (1980), and
previously by Saunders (1973). It is clearly a problem for the measurement of
sea-surface temperature by infrared radiometry from above, but has hitherto
been considered too small to worry about in comparison with other errors
encountered in IR radiometry. However, a recent review of the skin-temperature
deviation in relation to present and projected accuracies of sea-surface tempera¬
ture retrieval from satellite sensors (Robinson, Wells, & Charnock (1984)) points
out that developments in remote-sensing technology have caught up with the
effect. It now constitutes a potentially significant source of error in SST
algorithms.
Table 7.4 lists some of the observations made of the skin effect. It is very
difficult to measure accurately, and some of the laboratory or sheltered experi¬
mental conditions which give rise to large temperature deviations may not be
typical of open-sea conditions. As shown in Fig. 7.6, the temperature deviation
6r is defined as the difference between the temperature of the sea at the sea —air
interface, and the so-called ‘bucket temperature’ or bulk SST which effectively
is the temperature of water taken from the top metre of the sea and mixed well.
In a well-stirred sea, there is unlikely to be a significant temperature gradient
between about 1 cm and 1 m depth, and ST will be concentrated in the top
millimetre. If a diurnal thermocline develops there is likely to be a strong
212 Sea-surface temperature from infrared scanning radiometers [Ch. 7

Table 7.4 Values of ST reported by various authors

Author 6T Location Conditions

NichoUs (1979) --0.5 ± 0.3°C at sea low to moderate wind and waves
Paulson & Parker (1972) -1.14 to laboratory no waves
-1.81°C
Ewing & McAlister (1962) -0.6°C at sea low winds, night-time
HiU(1972) (i) ~-1.0°C laboratory low wind, no surface film
(ii) -2.0°C laboratory low wind, surface film present
(iii) ~-0.1°C laboratory higher wind conditions
o

Katsaros (1977) (i) laboratory

1 1

no wind

(ii) up to -3.4°C reservoir no waves

Kropotkin et al. (1978) (i) + 5.0°C at sea calm, 1 mm oil slick
(ii) -1.0°C at sea sea state 3,1 pm oil slick
Schooley (1977) (i) -0.2°C at sea 2.5 m s"' wind, no cloud
(u) ~0.0°C at sea under cloud shadow
Simpson & Paulson (1980) -0.15 to at sea winds 5 to 9 m s"'
-0.30°C
Grassl (1976) -0.17 to at sea winds 1 to 10 m S'*
-0.21'’C
1 o

Woodcock & Stommel (1947)

O

reservoir calm, night-time

r
O
Sec.7.3] The near-surface thermal structure in the ocean 213

thermal gradient within the first metre of water, the temperature decreasing
with depth, and the temperature changes across this thermocline must be dis¬
tinguished from the skin effect 8T when observations of 57’ are being made, and
if corrections for it are eventually to be applied.
Until now, no attempt has been made to predict the size of the skin-
temperature deviation in order to convert radiometric observations of SST into
the equivalent bulk SST. However, with 8T values of up to 1 degK observed
in the open sea, there is clearly a need to develop a better understanding of
the effect. Theoretical analysis (Saunders (1967b)) suggests that a possible
approach to prediction is to relate the effect to 2n> the surface heat flux,
and U* the friction velocity of the wind at the surface. Assuming that the
convective processes occurring just below the skin layer (Fig. 7.6) are due to
wind-driven turbulence rather than convective instability (which occurs in flat
calm conditions) it is suggested that

S-p — (7.13)
kU*
where k is the molecular conductivity of the sea water, v is the kinematic
viscosity, and X* is a constant to be determined. Estimates of X*. by various
workers are listed in Table 7.5. A fair scatter of values is noted, so that even if
further research confirms the basic form of (7.13) as a prediction equation, it is
clear that other environmental parameters contribute to the skin effect and
must be incorporated in the value of X* These include the size and wavelength
of surface waves, the presence of slicks and the effect of surface tension, and
the existence of whitecaps or other evidence of the complete destruction of the
thermal skin by the physical breaking of the surface. (It is interesting to note
that the thermal skin deviation seems to reappear witin 10 seconds of the broken
surface reforming.)

Table 7.5 Observational estimates of X*

Hasse (1971) 6 8 field

Grassl(1976) 2.2^5.5 field
Paulson & Parker (1972) 15 ± 1 laboratory
Simpson & Paulson (1980) 5^9 field
Weseley (1979) 6 field
Paulson & Simpson (1981) 6.5 ±0.16 field

Saunders (1976 b) 5 ^ 10 based on available field data

Equation (7.13) requires a knowledge of Qn and U*, which presents a

problem for adjusting remotely-sensed SST for the skin effect when no in situ
214 Sea-surface temperature from infrared scanning radiometers [Ch. 7

ship observations are available, f/* is capable of being measured remotely by

active microwave sensors, such as the scatterometer (Chapter 13), but 2n is not
so readily obtained. The sensible and latent heat fluxes depend on a variety of
factors such as the air and sea temperature difference, the atmospheric humidity,
the wind speed, etc., and are usually computed from bulk aerodynamic empirical
formulae. The only hint of a possible remote-sensing technique for direct
measurement of is the use of multichannel IR radiometers which directly
detect the temperature gradient in the top 100 )Um of the surface, from which
could be directly calculated. Such a technology has not yet been developed
satisfactorily for ground-level use, and certainly is not available for deployment
from satellites. Of course, if Qn could be directly measured, it is in its own
right a parameter of particular interest to oceanographers and meteorologists.
Indeed, one of the reasons for making SST observations is to use them as an
indirect way of estimating the ocean/atmosphere heat flux.

7.3.4 Surface slicks

From place to place the ocean surface is surmounted by a thin layer of lighter
material floating upon it. This may be naturally-produced organic material,
or oil from shipping. Often it forms into distinct patches, or slicks, in regions
where there is a surface convergence of the sea-water. The presence of slicks
can significantly change the surface roughness of the sea, and reveal patterns
of surface-water convergence as will be discussed in Chapter 10. In the context
of thermal IR imaging of the sea surface, slicks may have a significant, yet
difficult-to-predict, effect. If the material is in a monolayer, one molecule thick,
the effect will be small because radiation from the sea-water below will penetrate
it without difficulty. If the slick is much thicker, the reduced emissivity will be
expected to lower the brightness-temperature significantly. However, because
the slick increases the reflectance, solar radiation will counter this effect in the
day-time to an extent depending on the relative solar and satellite zenith angles.
The surface slick also affects the thermal structure of the near-surface. Even
a very thin layer inhibits the transfer of material and momentum across the
atmosphere-ocean interface. Thus wind-mixing of the surface layer is inhibited,
leading to an increased diurnal thermocline effect during the afternooir
(increasing near-surface temperature), but also permitting a larger conduction
layer to form in the surface skin with an increased temperature deviation 5 T
(reducing surface skin temperature). The slick also inhibits evaporation, reduc¬
ing Gl> and with reduced emissivity Qi^ is very slightly reduced. The overall
reduction of in the presence of a slick may therefore mean ST is smaller
(increasing surface-skin temperature).
With such a variety of contrary effects, it is not possible to predict whether
the observed radiation temperature will be reduced or increased by a slick.
Studies of oil spills suggest that the brightness-temperature change varies either
positively or negatively, depending on the thickness of the slick, and the ambient
Sec. 7.4] Atmospheric correction and SST calibration techniques 215

atmospheric and surface-wave conditions. Since slicks are likely to occur on

sub-pixel length scales for the polar-orbiting scanning sensors with resolution
around 1km, their presence may have the effect of slightly lowering or raising
the average brightness-temperature for the pixel. The size of the contribution
of this effect to the errors of SST measurement from space is a subject requiring
further research.

7.4 ATMOSPHERIC CORRECTION AND SST CALIBRATION

TECHNIQUES
7.4.1 Atmospheric correction strategies
If the constituents of the atmosphere are known, it is possible to calculate the
various contributions to the satellite-observed IR radiance (Singh & Warren
(1983)) as follows. If L!\{d) is the radiance recorded at the satellite in the
waveband centred at X, viewing in the direction with zenith angle 6, and is
related to the actual radiance reaching the sensor L(X,0) by (7.10), then
L'^(d) = + r^,;,(0) + (7.14)

where subscript e refers to the contribution from earth- or sea-emitted radiation

to the satellite measured radiance, a is from upward emission by the atmosphere,
rs is from the reflection at the sea surface of solar radiation, and ra is from the
reflection at the sea surface of downward atmospheric emission. The four cons¬
tituents of (7.14) can be expressed in terms of the temperature of emission,
and the atmospheric transmittance r(X, 0; pi, P2) which is dependent on the
wavelength and zenith angle 6 of the ray being considered. pi and p2 represent
the atmospheric pressure of the end-points of the path whose transmittance is
specified, thus allowing for transmittance to change through the height of the
atmosphere as the constituents change.
We let Bx(X) represent the spectral radiance emitted (assumed uniformly in
all directions, 0) by an emitter at temperature T, viewed through the spectral
window of the sensor, that is, after (7.11)

(7.15)

Then Ze,\(0) = e5;^(rs) r(X,0;Ps,O) (7.16)

where 7^ is the sea-surface temperature and p^ is the sea-level atmospheric

pressure.

(7 17)
216 Sea-surface temperature from infrared scanning radiometers [Ch. 7

where TfJ^p) is the air temperature at pressurep.

, 1
= (l-e)^xCos0sr(X,0s;O,Ps) - r(X,0;Ps,O) (7.18)
TT

where E\A% the solar irradiance at the top of the atmosphere on a plane normal
to the solar direction which is at zenith angle Sg. The downwelling solar irradiance
at the sea is therefore

£'xCos0sr(X,0s; O.Ps),

of which a factor (1 —e)/7r is reflected to produce a radiance in the direction^;

if the surface is assumed to be a diffuse, Lambertian, reflector. Similarly

^(X, 0; p,Ps)
^ra,\(^) = (1 - e) t(X,0; Ps,0) ^a(p) dp.
dp
(7.19)

Here, rather than calculate the diffuse reflectance of all the downwelling atmos¬
pheric radiation, it has been assumed that the plane reflection of the emission
along a column of atmosphere with zenith angle d will give a similar and more
readily calculated result.
In principle, knowing the atmospheric temperature profile and the distribu¬
tion of r with pressure, which depends on a given distribution of water vapour,
carbon dioxide, and ozone in the atmosphere, equations (7.14)-(7.19) can be
solved to yield Such a strategy is unpractical because it is not possible to
know /^(p) and t accurately enough, because of their temporal and spatial
variability. Ideally, it would be convenient if they could be measured with a
multi-channel IR atmospheric sounder such as the HIRS, but such an instru¬
ment does not necessarily yield a unique temperature profile, and high-accuracy
atmospheric sounding requires a prior knowledge of sea-surface temperature,
the very parameter we are trying to determine after atmospheric correction!
Whilst they are useful for meteorological purposes, such sounders do not yet
promise to provide an accurate means of direct atmospheric correction for SST
retrieval algorithms.

7.4.2 Multispectral and multUook techniques

Instead of pursuing a theoretically-based correction, recourse is made to an
empirical approach which makes use of the multispectral data from IR radio¬
meters. Since the atmosphere will affect different spectral channels in different
ways, the difference between the brightness temperatures in the separate
channels can be used to provide an empirical atmospheric correction. If a sensor
can be made to look at the same sea area through different thicknesses of
atmosphere, as in the proposed Along Track Scanning Radiometer (ATSR), once
Sec. 7.4] Atmospheric correction and SST calibration techniques 217

again the difference between the two brightness temperatures can yield a
measure of the atmospheric effect.
Where equations (7.14) to (7.19) are useful is in the simulation of typical
satellite-observed radiances given a variety of different, but typical, atmospheric
profiles. By modelling the different responses of the separate sensor channels
to different sea-surface temperatures, a simulated dataset of brightness tempera¬
tures corresponding to true sea-surface temperatures can be generated, from
which empirical algorithms can be extracted. This enables the accuracy of SST
retrieval from a proposed sensor to be statistically explored before the satellite is
launched or even built.
For a satellite in orbit, the most satisfactory SST atmospheric-correction
algorithm is derived from a real dataset of satellite and synoptic sea measure¬
ments. The important consideration is that the points chosen for calibration
purposes should represent a wide range of latitudes, seasons, and weather
conditions, so that all typical atmospheric conditions are allowed to influence
the dataset. If only one type of atmospheric profile were encountered, the
calibration algorithm would probably be more accurate for that particular
profile, but of less value under other conditions. Since a universal algorithm is
sought for all locations, sensors, and atmospheric conditions, in order to provide
an operational product for wide dissemination, the operational algorithms
now available are based on as wide a dataset as possible. The dataset should also
consistently represent a range of different viewing angles, if the algorithm is to
be applicable to all parts of a scanned image, at least within certain limiting
angles. Beyond about 9 = 50° the accuracy of atmospheric correction becomes
much poorer.
Table 7.6 lists the algorithms which were introduced by NOAA/NESS for
the correction of NOAA 7 AVHRR data, effective from 14 September 1982
(McClain (1982)). Since radiance values can be uniquely interpreted as bright¬
ness temperatures, there is nothing to be gained from correcting the radiances,
which conceptually might seem to be the right approach. Instead, the correction
algorithm employs the brightness temperatures for each channel, which can be
calculated from the satellite data using the method described in section7.2.4,
and delivers an atmospherically-corrected estimate of SST. In Table7.6, r^fcis
the corrected SST in degrees C, and T^.j, Tn, and are the satellite observed
brightness temperatures in degrees K from the three different channels. The
algorithms were based on a dataset of nearly one-hundred SST observations from
drifting buoys. These continuously monitor the temperature within the top
metre of the ocean, and when a cloud-free satellite overpass occurs the relevant
sea temperature can be selected from the buoy record.
Because the atmospheric contributions are different at night from the day¬
time (since there is no reflected solar radiation at night), it is necessary to use
different algorithms for night and day. The former tend to be more accurate,
since the day-time corrections should strictly include a dependence on the sun
218 Sea-surface temperature from infrared scaiming radiometers [Ch. 7

zenith angle. Table 7.6 also lists the accuracies achieved with the night-time
algorithms, as tested against a set of 94 drifting-buoy temperatures which were
independent of the dataset from which the algorithms were constructed.

Table 7.6(a) NOAA-7 MCSST calibration/atmospheric correction equations

operational from 14 September 1982
(based on a set of nearly 100 drifting-buoy observations)

DAY-TIME
Split-window rsfc(4/5) = 1.0351 Tn + 3.046 (Tn - 7^12) - 283.93
NIGHT-TIME
Split-window rsfc(4/5) = 1.0527 Tn + 2.6272 (Th-T^) - 288.23
Triple-window (3/4/5) = 1.0239 Tn + 0.9936 - T^) - 21%Ae
Dual-window rsfc(3/4) = 1.0063 Tn + 1.4544 (73.7- Tu) - 272.47

Table 7.6 (b) Accuracy test of above equations, using an independent set of
N drifting-buoy observations

Equation Split-window Triple-window Dual-window

N 94 94 94
Bias 0.06 deg C 0.14 degC 0.16 deg (
Scatter 0.60 0.64 0.72
Root-mean-square 0.61 0.66 0.73
deviation

In the daytime, 73.7 cannot be used without careful extra corrections

for sun reflection, and the split-window algorithm (using Tn and T12) is the
only algorithm available. For night-time viewing all three channels can be used
in different combinations. Perhaps surprisingly the split-window algorithm still
appears to give the best results, with the lowest bias of the mean values, and
the smallest scatter and root-mean-square deviation in comparison with the
triple-window and the dual-window (using 7^3.7 and Tn) algorithms. Deschamps
& Phulpin (1980) in simulation studies predicted that the use of the 3.1 pm
window could significantly improve the accuracy of SST retrieval, provided
that the sensor noise was sufficiently low. Bernstein (1982a) draws attention
to a noise problem on the 3.7 channel of the AVHRR and AVHRR/2
sensors which is probably the cause of the reduced performance of algorithms
incorporating this channel.
Sec. 7.4] Atmospheric correction and SST calibration techniques 219

It should be noted that these operational algorithms are based on a match¬

up between a 1 km X 1 km average observation and a point measurement from a
buoy, which may not necessarily be representative of the whole pixel. It is neces¬
sary to select the data points carefully, in order to exclude those which appear
to be near or within regions of horizontal temperature structure, although if
this is of sub-pixel size there is no way of detecting it. Another limitation of the
dataset is that the buoy records bulk SST within the top metre, no account
being taken of the diurnal thermocline or the skin-layer temperature deviation.
The former should be eliminated for the night-time algorithms, but the latter is
bound to be present. This means that the temperature product generated by the
algorithm truly represents the bulk SST rather than the skin SST, and this makes
it in general more useful to oceanographers who are accustomed to interpreting
this parameter. At the same time it means that some, perhaps much, of the
scatter and the bias noted in the testing of the algorithms using the independent
dataset, could be due to temporal and spatial variability of the skin effect
rather than atmospheric variability. As such, the scatter reflects the algorithm’s
inability to cope with a correction for which it was not designed (that is, the
skin effect), which implies that the algorithms may be performing even better at
atmospheric correction than the statistics suggest. Now that the algorithms can
produce predictions with a bias of only 0.06 degK and scatter around 0.6 degK,
any further improvements in the algorithms will need to consider the skin effect
as a separate factor.
Further improvements in accuracy are promised by taking into account the
viewing angle 0. The extra path-length of atmosphere encountered at high 6 is
already accounted for to a large extent because of an increased differential drop
in brightness temperature between the different channels. However, it appears
that the increased path-length should be treated in a nonlinear way, and thus the
correction term should be slightly 0-dependent. The other major improvement
in accuracy which is promised will come with the use of a multilook sensor
which views the same piece of sea through different atmospheric path-lengths.

7.4.3 Cloud-removal techniques

One of the principal uses of satellite-measured SST is as a database for climatic
records. If automatic processing is to be achieved so that all overpasses can be
calibrated for inclusion in a database of SST, some way of excluding cloud-
covered pixels is required. When dealing with individual pixels it is relatively
easy to establish a low-temperature threshhold below which a pixel is deemed
to be partly or wholly covered by cloud which generates the low brightness-
temperature. The problem arises when it is not clear if a rather low temperature
area is due to genuinely low sea temperatures, or due to clouds within the field
of view, too small to resolve but nonetheless biasing the retrieved temperature.
These may be apparent in the visible channels of the AVHRR, and day-time
imagery can be screened by this means. If a pattern on the IR imagery could be
220 Sea-surface temperature from infrared scanning radiometers [Ch. 7

RADIANCE COUNT VALUE

Fig. 7.7 - Histograms showing the frequency at which different radiance-count

values were registered by channel 4 of the AVHRR in a 50 X 50 pixel array for
(a) a cloud-free field, and (b) a partially cloud-obscured field. (Based on a report
in the Rutherford Appleton Laboratory’s ATSR Proposal.)
Sec. 7.4] Atmospheric correction and SST calibration techniques 221

due to oceanographic or cloud/haze effects, then if it also appears on the visible

image it is likely to be the latter. At night — the best time for obtaining accurate
SST information — the visible channels cannot be used. In this case, a pixel-by-
pbcel cloud-removal process can be developed which compares the different
SST predictions from the three different algorithms: split-, dual-, and triple¬
window. The presence of sub-pixel or larger clouds has the effect of causing
these three different estimates to diverge, so that if the three algorithms do not
agree within specified limits the value of SST for that pixel can be rejected from
inclusion in a database because of possible cloud contamination. McClain et al.
(1982) express this by showing that for a pixel to be considered cloud-free, a
linear relation of the form

{T^.i - Tn) + h,{Tn - Tn) = 0,

or ^2 ^2 ^3.7 “f ^2 ^11 ■“ 0
must be satisfied within certain limits.
In the end it will never be possible completely to eliminate cloud contamina¬
tion by very small clouds, but any resulting errors will be contained within
acceptable bounds.
A different approach can be used when high accuracy of SST measurement
is being achieved by averaging the satellite observations over areas of, say,
50 X 50 pbcels. Averaging can be performed when it is believed that there are no
significant horizontal sea-temperature gradients within the area. It leads to high
accuracy because sensor noise errors are greatly reduced by averaging. However,
a few pixels contaminated by cloud can seriously distort the average. One way
of avoiding this problem is to examine the histograms of the radiance counts,
the brightness temperature, or the atmospherically-corrected temperature distri¬
bution within the 2500 pixels. Fig. 7.7(a) illustrates a histogram in cloud free
conditions. There is a Gaussian spike centered at what is taken to be the true
radiance count appropriate to the average SST and atmospheric transmission
conditions throughout the 50 X 50 km area. Fig.7.7(b) shows the case in which
cloud is partially obscuring some pixels and completely obscuring a few. The
Gaussian spike now has a long cool tail due to the cloud-contaminated pixels,
and the true SST value can be obtained not by averaging the whole histogram,
but by fitting a Gaussian spike of 2500 values to the warm end and determin¬
ing where the centre is located. With appropriate computer algorithms this can
be done automatically, although care must be taken to exclude cases of 100%
cloud contamination. It is claimed that this approach can cope with 80% cloud
cover, provided that 50 X 50 pixels are used. For smaller sample areas, the
technique becomes progressively less reliable.

7.4.4 SST operational products

As summarised in Table 7.7, the errors of the present measurement and cor-
222 Sea-surface temperature from infrared scanning radiometers [Ch. 7

rection techniques based on the AVHRR/2 amount to an uncertainty of at least

0.6 degK. Maps of sea-surface temperature based on multichannel data from the
AVHRR only, and without any other satellite or sea-based input to the atmos¬
pheric correction, are issued operationally by NOAA/NESS under the name
MCSST (Multi-channel Sea Surface Temperature). This product superseded the
formerly-issued GOSSTCOMP (Global Operational SST Composite) data, which
used single-channel AVHRR values and coarse-grid atmospheric information
from the HIRS atmospheric sounder as well as some surface observations to
generate SST maps. The buoy evaluation of GOSSTCOMP values indicated a
bias of —0.4 degK and a standard deviation of 1.4 degK.

Table 7.7 Typical magnitude of errors in SST measurements

ARHRR/2 ATSR/M
(observed) (predicted)
degC degC

Radiometer sensitivity 0.05-0.12 0.02-0.1

Absolute calibration of radiometer 0.3 0.1

(black-body target error)

Uncertainty after cloud removal 0.2 -0.5 0.2 -0.4

Theoretically-predicted error of best atmospheric 0.5 0.2

correction algorithm for cloud-free conditions (night-time) (day & night)
and 50 km square average

rms error in direct comparisons between satellite 0.61 untested

prediction and buoy observations

The MCSST is a clear improvement, although, as will be discussed later, the

accuracies are still not sufficient for many applications. Nonetheless the spatial-
sampling benefits of the satellite surveillance technique outweigh the calibration
uncertainties, and it has been demonstrated that maps of SST constructed from
the AVHRR data are preferable to maps constructed from the sparser, if more
reliable, ship observations (Bernstein (1982a)). NOAA now treat MCSST as an
operational rather than experimental product, even though from time to time
alterations are made to update and improve the algorithms.
It would be premature to depend entirely on the satellite measurements as
the only SST database, since the MCSST is subject to further testing against
Sec. 7.4] Atmospheric correction and SST calibration techniques 223

buoy and ship observations. Following the eruption of the Mexican El Chichon
volcano in 1982, large quantities of dust were injected into the stratosphere in
equatorial latitudes, rendering the atmospheric-correction algorithm invalid and
enforcing a temporary discontinuance of SST information at low latitudes.
Now that the use of satellites to measure SST is becoming regarded as
established operational practice, it is worthwhile to consider exactly what
temperature is required by oceanographers and meteorologists. The true
sea-surface temperature is the skin temperature which is measured by the radio¬
meter, but hitherto oceanographers have used either the bucket temperature, or
even the engine-intake temperature to provide an SST database. In practice
the requested SST parameter will vary in definition according to the application.
Thus in studies of the mixed-layer and the modelling of its depth, the bulk SST
at about 1 m depth is probably most valuable, since it tends to eliminate the
very shallow diurnal thermocline which has little effect on the total heat content
of the mixed-layer. On the other hand air —sea interaction is concerned with the
sea-air temperature difference which drives the heat flow by sensible and latent
heat transfer between ocean and atmosphere. Thus the actual skin tempera¬
ture might be considered appropriate in this context. However, recalling the
discussion of section 7.3.3, it appears that the thermal skin is itself a response
to the heat flux, and that the temperature-difference driving the heat flow
is that between the sea temperature a few centimetres below the surface, and
the atmospheric temperature just above the surface. 2ir and will be slightly
modified by the temperature deviation across the surface skin, but it seems
conceptually correct to use the temperature below the skin as the most
appropriate definition of SST in the context of air-sea interaction studies. In
addition it should be remembered that the empirical formulae which have been
developed for parameterising sea —air fluxes are based on the bulk sea and
air properties and ignore the skin.
It can therefore be concluded that the provision of an SST dataset based on
the surface-skin temperature measured from satellites is less useful for climate
modelling and air —sea interaction studies than if a correction has been made
to the radiometric temperature to allow for the presence of a skin-temperature
deviation. As we have seen, the MCSST is such a database. Although it does not
set out to allow for the skin effect, it does so implicity by matching satellite
observations to buoy measurements. It is also important for the long-term
continuity of climate datasets that the satellite-based SST data are essentially
related to the same parameter as the previous data from ships, particularly if the
satellite records come to be thought of as superseding and replacing ship records.
It is important for climatologists of future generations that along with the
SST data is archived the details of the calibration, so that it is possible to deter¬
mine whether shifts of mean SST or changes in the variability of SST which
may emerge from the data for the decade of the 1980 s are a real occurrence
or merely an artefact of the changing methods of collection of the data. The
224 Sea-surface temperature from infrared scanning radiometers [Ch. 7

problem is analogous to that of today’s historical climatologists who can only

make use of very old temperature data if it is known to have been measured by
a properly-calibrated thermometer.

7.5 THE POTENTIAL USES OF SST DATA FROM SATELLITES

Having already considered some of the factors constraining the establishment
of operational SST data-archiving from satellite measurements, we look now at
those areas of oceanography in which SST measurement from space is actually
or potentially capable of making a contribution. It is also necessary to con¬
sider the accuracy required from the SST measurements in each context, the
horizontal spatial length scales to be resolved, and the temporal sampling rate
which ought to be achieved. In certain applications a distinction is made between
the absolute accuracy required (which may be improved by temporal and/or
spatial averaging) and the sensitivity (that is, the ability to detect temperature
differences within an image or from one image to another).

7.5.1 Climatology
Studies of the earth’s climate require an understanding of the complex
interactions between the ocean and the atmosphere. Fundamental to this is
the temperature of the ocean surface through which energy is interchanged.
Small changes in the sea-surface temperatures on the one hand may represent a
significant change in heat energy storage within the ocean, and on the other
hand may have significant impact on the atmospheric flow and consequent
weather patterns.
Present SST databases used in climatology have a horizontal resolution
from 1° X 1° to 5° X 5° and are averaged over periods of 5 to 30 days. The major
difficulty with the present database is that it is dependent on the monitoring of
the ocean by merchant ships, which produce measurements with rms errors of
0.5 deg K and with a spatial coverage limited by the major trade routes. Satellites
therefore offer a good opportunity to produce measurements comparable to
and eventually more accurate than merchant ships but with less temporal and
spatial aliasing. However, it is important that the satellite database does not
have a systematic bias in its errors as a result of atmospheric contamination, or
the surface skin effect, nor episodic deviations due to anomalous atmospheric
contamination as occurred following the El Chichon volcanic eruption referred
to above.
Whilst it may not prove too difficult in the near future to improve on the
accuracy of present ship-of-opportunity sampling, it is worth considering what
the ideal requirements are in different applications of a climatological database
of SST. The following three application areas are examples of the use to which
a climatological SST database may be put.
Sec.7.5] The potential use of SST data from satellites 225

7.5.2 Global changes in SST

The measurement of long-term changes in global SST will be necessary if
changes in the global heat balance are to be detected. Such changes may occur as
a result of anthropogenic changes in the composition of the atmosphere (for
example, carbon dioxide, ozone) and as a result of natural variations in the
climate system (for example volcanic aerosols, ice extent, planetary albedo). The
shortest timescale global event is most probably associated with major volcanic
eruptions, which it is estimated can change the global temperature by a few
tenths of a degree on a timescale of 1 season to 2 years. This will require global
measurements of SST accurate to 0.1 degK, and therefore it is important
that the measurement technique does not have a systematic bias of more than
0.1 degK, though a much higher random error in SST can be tolerated, since
ocean-wide area averaging can be performed.
The monitoring of the thermal effects of carbon dioxide will require a high
accuracy of similar order. Estimates of global SST warming due to a doubling of
atmospheric CO2 are about 2 degK over a period of 50 to 100 years, depending
on the rate of fossil fuel consumption. Hence a change of 0.2 to 0.4 degK over
a decade will need to be measured if this trend is to be observed.

7.5.3 Atmospheric response to SST anomalies

SST anomalies are deviations between the observed SST and the long-period
mean value expected at a particular location at a given time of the year. Thus
an annual anomaly is the deviation of the average over a particular year from the
average over many years, whilst the monthly anomaly for a particular month (for
example, June 1982) is the difference between the mean SST measured during
June 1982, and the average of all the mean June SSTs for many previous years.
Both observational and modelling studies have shown that large-scale (for
example, 20° longitude X 10° latitude) SST anomalies can produce significant
variations in the circulation of the atmosphere (for example. Wells (1979)). The
most sensitive regions appear to be located in the Eastern Tropical Pacific and
Tropical Atlantic Oceans, where temperature anomalies up to 3 degK on scales
of 1000 km can persist for 1 to 8 consecutive seasons. The monitoring of these
anomalies requires a reliable global SST climatology having an accuracy of at
least 0.5 degK over an area of 500 X 500 km. Because of the sensitivity of the
equatorial atmosphere to SST anomalies, a higher accuracy may be required in
these regions, with probably a resolution of 200 X 200 km. A sampling-rate of
once every 10 days is probably sufficient to monitor these anomalies.

7.5.4 Weather prediction (1 to 10 days)

Sea-surface temperature strongly influences the rate of evaporation into weather
systems, and therefore anomalies in SST may have a significant effect on the
local development of the system. In particular, SST has a very significant
226 Sea-surface temperature from infrared scarming radiometers [Ch. 7

influence on the location of the early development of tropical cyclones and may
affect the subsequent track of the cyclone. Modelling studies in mid-latitudes
have shown that warm anomalies may have a significant influence on the develop¬
ment of depressions in the N.E. Pacific Ocean and in the Tasman Sea. For these
phenomena an absolute SST accuracy of 0.5 degK should be sufficient, with
spatial information on scales of 50 to 100 km.
Information regarding the influence of SST on subsynoptic scales of motion
is small, though it is known that the distribution of sea fog is very sensitive to
SST, and mesoscale convective systems may be influenced by SST. To study
this further, SST would be required at a very high spatial resolution (1 km), with
accuracy of better than 1 degK.

7.5.5 Gas exchange across the air—sea interface

SST may influence the exchange of gases between ocean and atmosphere in the
following ways:
(i) The solubilities of dissolved gases in sea-water are temperature-dependent.
For instance, the buffering factor for carbon dioxide in sea-water decreases from
14 to 8 for a change in temperature from 0°C to 30°C.
(ii) The vertical stability at the air —sea interface is a function of the air —sea
temperature difference, and this may influence the transfer velocity for some
gases.
(hi) The saturated vapour pressure of gas may be temperature-dependent (for
example, water vapour).
Though little is known about the sensitivity of gas exchange to SST, it has
been shown that inter-annual variations in atmospheric COj are correlated with
the Southern oscillation and variations in the Eastern Equatorial Pacific Ocean.
It is not certain whether this indicates the sensitivity of the COj transfer rate
to SST, or to wind stress, which is also known to affect the transfer rate.
Until more is understood about these processes it is difficult to define an
accuracy and sensitivity requirement for this application. It may be only as the
consistent, global coverage data from satellites becomes available that the
influence of SST on gas exchange can be explored.

7.5.6 Surface heat flux as it affects the ocean heat budget

We turn now from those applications which are essentially meteorological to
those which are primarily oceanographic, although such a distinction may be
considered artificial, given the strong interactions which occur between the
ocean and atmosphere.
Sea-surface temperature is very important in the estimation of the heat
flux through the ocean surface. If it could be measured directly by a specially-
designed satellite sensor (perhaps by measuring the temperature gradient in the
thermal skin layer) that would be a major advance. Meanwhile, it is calculated
Sec.7.5] The potential use of SST data from satellites 227

from bulk aerodynamic formulae which require a knowledge of sea-surface

temperature. An example of the way this technique is applied is to be found
in the heat-budget calculations performed for the Mediterranean by Bunker etal.
(1982).
The demands made on an SST database are best illustrated by the speci¬
fications of the CAGE experiment. This experiment is intended to obtain an
estimate of the poleward flux of heat in the N. Atlantic Ocean between 20°N
and 60°N to an accuracy of ±20% (±0.2 X 10*^ W). Direct and indirect methods
will be used to calculate this heat flux. One of these indirect methods will
require an estimate of the mean surface heat flux to an accuracy of ± 10 W m“^.
The CAGE planning group have suggested that to obtain this accuracy the
SST will be required to an absolute accuracy of 0.25 degK. The determination
of the surface heat flux will be constrained by synoptic scales of motion in the
atmosphere, and therefore sampling at 1 — 2-day intervals over scales of 200 km
will be required.

7. 5.7 Deep convection and water-mass formation

Deep convection is the essential process which produces bottom-water in the
ocean. Recent studies have shown that the rate of bottom-water production
in the N. Atlantic has varied over periods of decades, and may be an important
influence on climatic variability. Eiowever, the process of deep convection has
been rarely observed by conventional research ships as it occurs in late winter in
localised regions. The main regions where deep convection has been observed
are in the Mediterranean, the Gulf of Lions; in the North Atlantic off Greenland
and the Norwegian Sea; and in the Weddell Sea. The process is associated with
intense coohng of the surface for a period of a few days in late winter. The
temperature changes are relatively small (0.1 degK to 0.3 degK), because the
cooling is spread through depths of 1 km to 4 km of water. SST observations on
space scales of 10 km, with accuracy of 0.05 degK would be needed to delineate
these convection zones.

7.5.8 Dynamical oceanography

In regions where the seasonal thermocline is weak, the SST distribution contains
information about the underlying dynamical processes occurring in the upper
part of the ocean. These include:
(i) Large-scale dynamical features, such as those in the equatorial ocean,
which cause large vertical excursions in the equatorial thermocline and lead to a
surface-SST signature. By studying consecutive SST patterns, inforihation on the
life-time, propagation speeds, and length scales of equatorial waves can be
obtained.
(ii) Mesoscale eddies, such as those shed by meanders of the Gulf Stream and
the western boundary currents of other oceans.
(iii) Small-scale phenomena such as frontal regions, and coastal upwelling.
228 Sea-surface temperature from infrared scanning radiometers [Ch. 7

In all these studies emphasis will be on the spatial variations of SST over the
timescale of the process or feature, rather than absolute accuracy. For instance,
continental-shelf tidal fronts have temperature variations of a few tenths of a
degree over length scales of 10 km. Mesoscale eddies, however, will often have
temperature anomalies of ±2 degK on space scales of 50 to 200 km. Coastal up-
welling regions will have temperature variations of up to 5 degK over distances
of 50 km. Large-scale dynamical motions in extratropical latitudes may be more
difficult to detect since the vertical velocity is relatively small and only small
displacements of the main thermocline occur. However, in equatorial latitudes,
because of the increase in speed of the long waves and the shallowness of the
thermocline, a strong influence on the SST may result.

7.5.9 Pollution studies

Thermal pollution from large power-stations may be monitored by satellite.
Here, requirements are similar to the monitoring of small-scale dynamical
phenomena, with the emphasis on high horizontal and temporal resolution (at
least 1 km). Surface oil films alter the skin temperature and affect the emissivity
of the surface skin (as discussed in section 7.3.4). They therefore influence the
observed brightness temperature. Regions of oil pollution are potentially capable
of being monitored by satellite. With a desired time resolution of between 1 hour
and 1 day, a spatial resolution at least as good as 1 km but preferably better, and
a sensitivity of 0.1 degK to detect the subtle effect of slicks, such an application
is unlikely to make serious use of satellite data in the near future, and airborne
radiometry is more appropriate. However, when the thermal infrared channels
on the Landsat satellites are functioning satisfactorily they will offer a spatial
resolution of around 120 m, and should make an exciting contribution to the
study of the synoptic patterns of coastal discharge plumes etc., even though
there will be no facility for time-sequence imaging, given the long repeat cycle of
Landsat.

The above applications, and their sensitivity, accuracy, and sampling

frequency requirements are summarised in Table 7.8. The uses of SST have
so far been presented in relation to research work in oceanography. Much of
this research has an applied commercial significance, and as the uses of SST from
satellites are developed it is to be expected that more direct commercial use will
be made of SST data. This is particularly true in connection with the last two
application areas. For example, it is easy to envisage a situation where the onset
of an upwelling event, or its disappearance, may herald a significant change
in the fishing potential of a region, in which case the rapid dissemination of SST
data from the satellite to fishing vessels is of considerable value. Similarly,
oceanographic research vessels are already making use of satellite IR images
to guide them to fronts or coastal upwelling regions in order to study them more
closely. The APT facility of the GMS and NOAA satellites is particularly useful
since it can be received directly by ships at sea.
Sec.7.5] The potential use of SST data from satellites 229

H OD
c/3 O
00
•o lo CO
<N O JD lO IT)
o
cO o o o d O o
t-< W)
3
o
o
cO

CO
u Cd)
0)
E
•3
T3
a> tn <N fS ^
.id o
I I dod
3
Summary of SST requirements for different application areas

e/3 c/3 e/3 V3 CM >%

>> >> >* >, >v >% cO >* >»
CO CO CO cO CO CO ^3 CO CO
•3 *3 •3 *3 -3 *3 *3 3
*3 in O
*o 0^ O in <N •3
1

*'/
o a> >. B >»

C ‘S U<
cO
bi
CO
bH
CO
b^
CO
.2 c M t-H JS
CO CO *3
i>i.
^
CO
a- s.
^ 03 03 03 03
o'^
DO —
>. >>§ ^3
5?
I
ft>
t-i
>» >> >, >» VO di
O in <N •n
I S I
1 B
.S CO
cn
A A O I
2 3

— O
O
E E E o E E E E E E
.2-B S o phd
^.3 3 o o o o in o o rH
c/5D. O
JO
0^
O'
aj
bN
o
cs
o
in
o
<N B
1 o
(N
o •—(

a>
CO S _
M CM 52 CO CO
a> 13 lo
CO
fO
2 c c
> x> x> X) ,3 •“ .2 .2
o .2
o 3
cO
*Sb ^
43 O
S '5b '5b
2 u <u
CO O a a o 03
a> O o OC OS
Table 7.8

O O

O
CO
Sri 3
<l>
a> E ^ 03
X «3 oj
5 c/3
C/3 &<
bt
Q>
Vi C^ ^ 3
•3 -H O &
^
CO
CO
*3 E 3
“ c :3 ii
C
O
o .^5 O
X3 O
“ .2
M w) c o .a
S
8 .2 3
"cO
o o *3 o ^ cO -
o O
;3 ^ O «> 2 2 8s
a .S 3 1o
a lu bo. >
cr CO 3
cj 3 "5
O 3
5
—
3
< I •§ S*s
^
_ 03. u
o 3I w
'^■2 ^ O
S fc -S
5 2 <4-.
CJ
■S § a
H 3 H co 03 S, =3
00 o o 2' o-o
CO 03
oo CO ^ o Q Q
230 Sea-surface temperature from infrared scanning radiometers [Ch. 7
eo^N^ MARCH 1982

Fig.7.8 — Global maps of (a) SST and (b) SST-anomaly, for March 1982, derived from the NOAA-NESDIS satellite data archives. The contours
in (a) have been drawn at 2 deg C intervals for clarity at this small reproduction size, but are supplied at 1 deg C intervals on the original. (Data
supplied by R. Legeckis.)
232 Sea-surface temperature from infrared scanning radiometers [Ch. 7

7.6 EXAMPLES OF THE APPLICATION OF SATELLITE SST DATA

Because of its wide availability for several years, and its ready interpretation
in terms of SST, thermal IR imagery from the NOAA orbiting satellites has
already been applied in a variety of situations. Initially, the photographic imagery
was used, since only analogue signals were available form the earlier TIROS
satellites, but latterly digital data from the AVHRR, calibrated for SST, has also
been employed in scientific research. Having discussed in the previous section
the broad areas of potential application, we conclude this chapter with some
specific examples of how the satellite data has been used. This is not intended
to be an exhaustive or comprehensive review of all the applications in the
literature, which could probably by now fill a book by itself, but serves to
demonstrate that oceanographers are no longer merely talking about the
potential of IR data, or working on its calibration, but are actually performing
useful oceanographic science with it.

7.6.1 Global SST, and SST-anomaly, maps

Fig. 7.8 illustrates global SST and SST-anomaly maps produced from the
satellite database of NOAAt-NESDIS$. By its very nature, a climatological
database is used as such only when a reasonably long time-span of data has been
collected, and therefore there is little published work yet relating to the applica¬
tion of satellite-derived as distinct from general climatic databases. Most of the
research which has been published is concerned with relating satellite-derived to
ship-derived databases, and noting any contrasts. In this context Fig. 7.9, based
on the work of Miyakoda & Rosati (1982), is illuminating. It compares the ship
and satellite SST data available for the construction of an SST monthly average
for January 1977. The ship coverage is restricted to the shipping lanes, and is
almost nonexistent elsewhere, whilst the satellite data is sparse in cloudy regions
such as the Inter-Tropical Convergence Zone, and in mid-latitude regions. It is
interesting to note that the two datasets are surprisingly complementary, at least
so far as covering the North Atlantic and Pacific Oceans, suggesting that for this
case the combined dataset would be most valuable, provided that proper inter¬
calibration of the data-collection methods could be achieved. On the other hand
a comparison between SST maps produced independently form the two datasets
would not be meaningful.
Since 1977, the NOAA series of polar-orbiting satellites has been providing
a somewhat better coverage than is illustrated in Fig. 7.9(b). It should dso be
borne in mind that for a useful monthly mean value, several samples should be
available during the month, and Fig. 7.9 does not make clear how many samples
are available at each point. In general, the satellite is likely to give the most
homogeneous coverage, apart from persistently cloudy regions.

t NOAA-National Oceanographic and Atmospheric Administration (USA).

fNESDIS-National Environmental Satellite, Data and Information Service (formerly
Sec.7.6] Examples of the application of satellite SST data 233

Fig. 7.9 - SST archived data available for January 1977 from (a) ship and buoy
measurements and (b) satellite infrared radiometer observations (from Miyakoda
& Rosati (1982), published by the American Geophysical Union)

7.6.2 Ocean eddies

Clear-weather images of the mid-latitudes of the oceans usually reveal the
presence of warm- or cold-core eddies, particularly towards the western coasts
where eddies are shed by the western boundary currents. There are in the litera¬
ture a number of cases where such eddies have been studied using the satellite
data along with ship observations. Fig. 7.10 shows an AVHRR image in which
the Gulf Stream is clearly seen as the front between the cooler inshore and
234

< -H S
< 5 ^

^ c o
■3
c .ti >»
_g o -O
^ J=
x> o
•o C
(U
•o q> 0/5
o
o o
• • ^ on

«j TS 2^
^ TJ *5
^ “ 2:
-go
• I C9

R. Legeckis of NOAA/NESDIS.)
2 SI
o cd •♦■^
« ^ O
CQ
SC f3 ^
!>
«

8.3£
u-3-

O
C/3 (l>
a> is S
60 3 n>

go's
.is 60

^■§1
I c o
C cQ
S J> J2
•c
^ E -3
'S
- “ E
S^ “
So; 5
r- 3
^00

' ££
03"^
»-i >—> o •w
r^ »o
c ?
pH O U
Sec.7.6] Examples of the application of satellite SST data 235

warmer offshore waters. The Gulf Stream front meanders, and to the north of
it is a clearly defined warm-core eddy in the cooler water. There is also evidence
of a cold-core eddy which has broken away and moved into the warmer
Mid-Atlantic water.
Using earlier VHRR images of a similar cold-core eddy, Spence & Legeckis
(1981) were able to extract dynamical information without needing to correct
the data atmospherically. They analysed a sequence of images, two between 9
and 12 March 1977 and five between 8 and 16 April, which were sufficiently
cloud-free for a cold-core eddy to be identified and followed. After digitisation
of the analogue data, and geometrical rectification, the eddy could be identified
on computer printout by tracing the locus of the steepest temperature gradient.
From this, the orientation of the eddy, which was elliptical, could be seen to
rotate in a counterclockwise direction at a rate of about 20° per day (Fig. 7.11),
equivalent to 5% of the local Corolis frequency f. This could be compared with
analytical theory of the dynamics of baroclinic eddies. Using satellite data, this
is a relatively simple measurement to make, and is achieved without the need for
careful temperature calibration, whereas to make the same synoptic observations
from ship or buoy sampling in order to explore the dynamical theory of eddies
would be much more time-consuming. Of course the acquisition of ship data
to provide vertical soundings through the eddy would enable the analysis to
be taken further than if no ship data were obtained at all. The conclusion to be
drawn is that when ship and satellite data can be combined to complement
each other, with the satellite data supplying horizontal synoptic spatial informa¬
tion, there is considerable scope for improving our understanding of dynamical
processes in the ocean. On a cautionary note, it should be pointed out that
eddies may exist below the sea surface without having a thermal surface signa¬
ture, and the absence of eddy structures on an IR sea-surface image need not
imply that there are no dynamical features present in the water below the
surface layer.

•37°N

36°N

Fig. 7.11— The location of the steepest temperature gradient of a cold-core ring
in the Atlantic Ocean similar to that illustrated in Fig. 7.10. This was determined
from infrared images observed by the VHRR on NOAA-5. A sequence of posi¬
tions between 8 and 16 April is shown, revealing an apparent counterclockwise
rotation of the ellipse shape at a rate of about 20° per day. (From Spence &
Legeckis (1981), copyrighted by the American Geophysical Union.)
Fig.7.12 - Meteosat infrared channel 2 image at 11.55 UT on 17 February 1978.
This image has been enhanced to illustrate the range of sea-surface temperatures
off the coast of northwest Africa. UpweEing of cooler subsurface water along the
coast is marked by a hght grey tone at A. (Image supplied by ESA.)
Sec.7.6] Examples of the application of satellite SST data 237

7.6.3 Upwelling
Fig. 7.12 is a Meteosat IR image from February 1978 showing the west coast of
Africa off Mauritania and Senegal. It reveals very clearly the upwelling that
occurs along this coastline as the trade winds blow to produce offshore surface
drift. It is most marked immediately south of Cape Verde and northwards
towards C. Blanc. Domain et al. (1980) used a sequence of Meteosat data during
the spring and summer of 1980 to chart the progress of the thermal evolution.
Warmer surface water gradually moves northward, and the upwelling area
recedes north of C. Blanc by July or August. This area has been well studied
by numerous ship experiments, and the satellite observations are in accord with
the expected patterns of sea-surface temperature. The satellite, however, is able
to identify more specifically the location of the upwelling zone and the timing
of its movement, and this has proved to be of some benefit to local fishing
operations.
Being close to the equator this area lends itself to study by Meteosat which
views Cape Verde with a zenith angle of no more than about 15°, achieving
nearly maximum resolution. With data-sampling every half hour, a technique
of data-processing was adopted which dealt with all the data from one day and
selected for each pixel the most persistent value. In this way, transient clouds
passing over the region could be eliminated to leave the permanent land and
sea temperature features — a useful benefit of the frequent sampling-rate
of Meteosat which offsets to some extent the disadvantage of the large pixel
size (4 km X 5 km). Meteosat data is not so readily calibrated or atmos¬
pherically corrected, and reliance has to be placed on in situ measurements
to calibrate individual scenes, but this does not prevent the upwelling patterns
being determined.

7.6.4 Ocean fronts

Fig. 7.13 is an IR image of the North-East Atlantic Ocean, with clear weather
between Iceland and Scotland. In this image, enhanced by the Dundee University
Satellite Receiving Station to emphasise sea-surface temperature differences, the
white streaks and patchy areas across the south-west corner are cloud, but the
white area east of Iceland is in fact cold sea-surface water. The sharp contrast
between the warmer North Atlantic water and the cold water of more polar
origin marks very clearly the instantaneous position of the Iceland—Faroes
front. Indeed, the position of the front as defined by the steepest temperature
gradient corresponds well with the 35 surface salinity contour measured in
various observational experiments in this area (for example, Hansen & Meincke
(1979)). Since the North Atlantic water has a salinity of around 35.3 and the
colder northern water a salinity of 34.9, the 35 contour would be expected
to mark the frontal zone quite well.
Fig. 7.13 - Infrared image recorded by channel 4 of the AVHRR on NOAA-6
at 08.27 GMT on 18 May 1980. The area shown is the North-East Atlantic
between Iceland, Faroe I., and Scotland. The Iceland - Faroes front lies east of
Iceland, the colder (Ughter-shaded) water to the north, and eddies and intrusions,
are clearly present on it. The more gradual thermal gradient towards warmer
water off the Scottish Western Isles also reveals eddy structures throughout the
North Atlantic. (Image supplied by the Dundee University Satellite Receiving
Station.)
Sec.7.6] Examples of the application of satellite SST data 239

Given a year’s images of the area between January and December 1979
from the AVHRR archives at the Dundee Satellite Receiving Station, between 30
and 40 images were obtained with a part of the front visible through gaps in the
clouds, and about half this number had essentially clear views of the front which
enabled its position to be accurately located. Several of these images came in
pairs an hour apart on the same day, given the overlap between successive swaths
of the TIROS/NOAA satellites. There were also three or four periods when two
or three images within two or three days were cloud-free. Thus it was possible
to study the long-period seasonal movement of the front (which appeared to
be seasonally stationary) and the short-period meanders illustrated in Fig. 7.14
which shows the line of the front for several different images during the year.
In this study it was possible to match the evidence of the satellite data to the
archives of ship data. What the satellite imagery is able to add to previous
understanding is a synoptic view of the front at given times, and its variability.
Thus, whilst averaged ship data might indicate that the front weakens into a

Fig. 7.14 - Composite of lines drawn to represent the steepest thermal gradients
marking the edge of the Iceland—Faroes front on about 40 clear-weather IR
images recorded between November 1978 and July 1980. The front position is
subjectively determined and depends to some extent on the photographic contrast
stretch employed on images such as Fig. 7.13. On most of the images only a
portion of the front was visible through the clouds.
240 Sea-surface temperature from infrared scanning radiometers [Ch. 7

more gradual gradient towards the east, the satellite data show that in fact the
front remains relatively strong, but meanders and moves its location more at
the eastern end than the western. A hydrographic station at the mean frontal
position will sometimes lie to the north and sometimes to the south of it, so
that if hydrographic data are time-averaged it will tend to smooth out the front.
Again we see the synoptic spatial view of the satellite being the particular
contribution which is able to bring new information into what was already quite
a well-understood oceanographic feature.
Careful analysis of the images has also revealed that the warm-water
intrusion seen on Fig. 7.13 about 180 km east of the Icelandic coast, where it
extends northwards into the cold area, is a semi-permanent feature. Although
it is masked by turbulent meanders which appear to travel along the front
and break off into eddies at this point, there is usually a tongue of warm water
reaching 50 to 100 km northwards at this location, which corresponds closely
with a deep notch in the Iceland—Faroe ridge over which the cold water
is known to overflow southwards. Further research is needed to examine the
hypothesis that the existence of the warm northward intrusion may be related
to the occurrence of subsurface southward overflow. If this was the case the
disappearance from an image of a clear warm intrusion might signify that the
intermittent overflow had temporarily ceased. Such an idea is speculative, but
illustrates the type of research possibilities opened up by satellite IR data.
It would certainly be interesting if it turned out that a sensor limited very
definitely to measuring a surface parameter was able to indicate the occurrence
of a process 300 m deep.
The Iceland—Faroes front is one example of many such features visible on
IR imagery from around the world as reviewed by Legeckis (1979).

7.6.5 The Mediterranean Sea

Fig. 7.15 from Gallagher et al. (1981) is a simplified graphic description of the
mean distribution of dynamical features in the Mediterranean Sea which can
be reviewed on IR imagery. In this case for the 1977-79 period the data were
principally from NOAA 5 VHRR. Monthly and annual charts, from which
Fig. 7.15 is taken, are produced to denote the position of elongated fronts,
eddies, and the directions of surface flow inferred from the patterns. Being
less cloudy than the seas of N. Western Europe, the Mediterranean lends itself
to study by satellite, and in the future it is to be expected that the combina¬
tion of ship sampling programmes, modelling, and satellite surveillance will
lead to progress in understanding further the oceanographic processes in the
Mediterranean.

7.6.6 ShalloW-sea fronts

A type of front which has attracted a lot of scientific interest since the 1970 s,
and which has been quite extensively viewed using satellite IR imagery, is
 Sec.7.6]

HiUON
<

h
(/)
«

Ui
k
Examples of the application of satellite SST data

Fig. 7.15 — Mean distribution of the more persistent surface thermal features in the Mediterranean Sea observed on satellite infrared image datasets
241

for the period 1977—1979, showing the inferred circulation. (From Gallagher etal. (1981).)
Fig. 7.16 — NOAA-6 AVHRR channel 4 image of the British Isles recorded at
1900 GMT on 16 May 1980, showing clearly the presence of shelf-sea tidal-
mixing fronts between the weE-mixed water column (cooler, lighter shading) and
stratified water (warm surface, darker shading). (Image supplied by the Dundee
University Satellite Receiving Station.)
Sec.7.6] Examples of the application of satellite SST data 243

the tidal-mixing front found in shallow tidal seas (Simpson & Pingree (1978)).
Fig. 7.16 shows the existence of these fronts around the UK, where they are
found between about May and September. They are prominent in the English
Channel south-west of Cornwall, in the Celtic Sea south-east of Waterford, and
stretching across to South Wales in the Irish Sea south-west of the Isle of Man,
and running north-north-east from the coast of N. Ireland.
The way in which these fronts occur at the boundary between stratified
water (having a warmer surface temperature) and well-mixed water which is
stirred by tidally driven turbulence, has been described in section 3.3.3. Fig. 3.7
illustrates the temperature profile schematically and shows why the front has
an SST signature. The surface is cooler in the weU-mixed region, essentially
because the solar heating has to heat the whole water column and not just down

Fig. 7.17 - Map of contours of the parameter log,,,(/z/u’) in the Irish Sea,
English Channel, and Celtic Sea, based on tidal stream observations. The contour,
log,o(/t/«®) = 1.85, is found in approximately the same position as the shelf sea
fronts in Fig. 7.16. (From Simpson (1981).)
244 Sea-surface temperature from infrared scanning radiometers [Ch.7

to the thermocline as on the stratified side. The frontal region has a certain
amount of upwelling associated with it, and is usually characterised by enhanced
productivity. Its location is therefore of much interest to biological oceano¬
graphers who have used IR images to guide the location of their cruise-sampling
tracks for sampling of productive patches.
The physical processes of tidal fronts are now understood, and it can be
shown that the occurrence of a front depends on the magnitude of the tidal
energy dissipation by bottom friction, distributed throughout the depth. Thus
/z/u^, where h is the local depth and u is the tidal-stream amplitude, is an appro¬
priate parameter controlling whether stratification is present or not. A certain
value of hlu^ is seen to mark the transition, and hence a corresponding contour
on a map of should define the position of a front. Fig. 7.17 shows such a
contour map, and the agreement between the contour h/u^ = 70 m“^ s^ (that is,
\ogio hj= 1.85), and the actual frontal positions in Fig. 7.16 demonstrates
how well the theory is able to model the actual process in the sea.
Satellite imagery has been used to study the meandering and spring—neap
excursions of these fronts, and to examine whether seasonal movement occurs.
Clearly, if only horizontal movement is being studied, the satellite data provides
a much cheaper and easier way of proceeding than field surveys at sea. Satellite
IR imagery from all over the world is now revealing similar fronts to be present
in continental shelf seas where the depth is sufficiently small, and the tidal
streams are strong enough, to break down the seasonal thermocline.

7.6.7 Coastal-fringe fronts

Another type of shallow-sea frontal structure which has come to light in satellite
IR imagery is that found in winter around the coasts of shallow seas. It is
particularly evident off the south coast of England as seen in Fig.7.18. Hitherto,
temperature charts of the English Channel in winter have looked like Fig, 7.19,
since ship sampling grids were usually too coarse, or smoothing of ship data was
applied too strongly, to reveal the sharp discontinuity of up to 1 degK over a few
km which appears on the images. This has now been verified by sea measure¬
ments, but only after the satellite data had pointed them out, illustrating
another aspect of the satellite data’s usefulness.
The reason for the sharp discontinuity is not fuUy understood, but it may
simply be a turbulent convergent front, in which a steady residual circulation
brings two water masses together, and removes the intermediate mixed water.
This could explain why the front is sharpest off headlands, where the typical
headland eddies could help to maintain it. The inshore water is cooler in winter
because surface heat loss is more effective in lowering the temperature of the
shallow rather than the deeper offshore water. The reverse is true in summer,
when some evidence of a warm front can be found, but tends to be masked by
diurnal thermocline noise, which is not present in the winter image.
Fig. 7.18 — TIROS-N AVHRR channel 4 infrared image of the English Channel,
13.47 GMT, 4 March 1979. The fringe of cooler water along the English coast is
evident, and in certain places the boundary with the warmer water in the central
Channel sharpens into a steep front, particularly where residual currents off
headlands force the cooler water away from the coast. (Image supplied by the
Dundee University Satellite Receiving Station.)

Fig. 7.19 - Mean distribution of sea-surface temperature (°C) for February,

based on data collected by ships between 1905 and 1954. (From Pingree (1978).)
246 Sea-surface temperature from infrared scanning radiometers [Ch.7

It is worth noting too that the front is equally visible on certain CZCS
images, indicating that it has a colour signature which persists when the thermal
signature disappears in spring. This is a reminder that thermal imagery reveal¬
ing SST patterns can only be an indication of dynamical processes so long as a
temperature gradient exists to act as a tracer. It also points out the difference
between two classes of thermally-imaged phenomena, those such as the coastal-
fringe front in which physical processes are occurring which happen to be
revealed by temperature gradients, but which could exist without having a
thermal signature, and those like the tidal-mixing front in which the tempera¬
ture structure and heat balance is an essential part of the physical process. It is
intriguing to pose the question whether some of the turbulent eddies which can
be seen at fronts on images such as Fig.7.13 are dynamically related to the front
and therefore fall into the second category, or whether they are evidence of a
turbulence structure which occurs over the whole of the ocean and just happens
to be imaged in this case because the strong temperature gradient provides a
tracer capability.

7.6.8 Commercial fisheries

We conclude this section of different examples of the use of thermal IR imagery
with a commercial application, albeit experimental. This is the chart to fisheries
issued by NASA/JPL in the Californian Bight area — Fig. 7.20. Whilst ocean
colour in this area (Fig. 6.28) was able to act as an indicator for certain fish

Fig. 7.20 - A fisheries-aid chart produced by NASA-JPL as an experimental

applications product of satellite infrared (thermal) data and other microwave-
derived information. (From Montgomery (1981).)
Sec.7.6] Examples of the application of satellite SST data 247

species, for others the temperature range is vital. These charts, which can be
communicated to vessels at sea, can be of genuine assistance to commercial
fishing fleets in helping them to find the shoals of fish they are seeking.
CHAPTER 8

Passive microwave radiometers

8.1 THE PHYSICAL PRINCIPLES OF PASSIVE MICROWAVE

RADIOMETRY
8.1.1 Overview
Passive microwave radiometers viewing the sea surface from space function in
essentially the same way as do the infrared radiometers discussed in the previous
chapter. Normally operating at electromagnetic wavelengths between 1.5 mm
and 300 mm, that is, 1 GHz to 200 GHz frequency, they observe the thermal
radiation emitted in the microwave part of the spectrum by the sea surface,
the microwave radiation emitted by the atmosphere, and that reflected by the
sea surface from downward atmospheric and solar emission and deep-space
background radiation. At these comparatively long wavelengths there is no
scattering by the atmosphere or aerosols, haze, dust, or small water particles in
clouds, so that microwave sensors are effectively all-weather devices, although
liquid water in the form of precipitation does scatter the radiation and can
render the atmosphere opaque at microwave frequencies.
This principal advantage of the microwave sensor is countered by the fact
that thermal emission is very weak at these longer wavelengths, and the signal
received at the sensor is therefore weak. To overcome noise levels a large field
of view must be received, with relatively low spatial resolution achieved by
spaceborne radiometers up to the present date. The emissivity of the sea at
microwave frequencies is also low (typically 0.3) and varies with the dielectrical
properties of sea-water, and the surface roughness. Since the dielectric constant
varies with temperature, salinity, and e.m. frequency, the observations of a
multichannel microwave radiometer must contain information not only about
the sea-surface temperature, and the atmospheric absorption and emission, as
with an IR radiometer, but also about the ocean salinity and the sea state. The
development of passive microwave radiometers for oceanographic purposes has
sought to exploit these properties of microwave surface emission in order to
produce an instrument and data-analysis system capable of yielding measure¬
ments of sea-surface temperature, salinity, and sea-surface winds as indicated by
the surface roughness.
[Sec. 8.1] The physical principles of passive microwave radiometry 249

Helpful introductions to the principles of passive microwave radiometry can

be found in Swift (1980), Reeves (1975) (pp. 499-534), and Thomas (1981).
Wilheit (1978) has reviewed the range of oceanographic applications of passive
microwave radiometry.

8.1.2 Thermal emission in the microwave

Thermal emission in the microwave is controlled by Planck’s radiation law (7.1)
which can be written for a non-perfect emitter as:

2hc'^€(9,4>)
X® [e\p(hc/XkT) — 1]

where
Mx= the spectral radiance at wavelength X, in units of power per unit solid
angle per unit projected area per unit wavelength interval
/z = Planck’s constant = 6.626 X 10”^ J s
c = speed of light = 2.9979 X10® m s"^
k = Boltzman’s Constant = 1.38 X 10“^ J
T = the temperature of the emitter in degrees K
and e(9, 0) is the emissivity of the surface in the direction (6, 0). 9, 0 and the
elemental solid angle dl2 are defined in Fig. 8.1.

Fig. 8.1 - Definition sketch of angles used in microwave radiometry expressions

250 Passive microwave radiometers [Ch.8

For microwave radiometry, the spectral radiance or brightness is often

expressed as B per unit frequency, rather than M per unit wavelength. Since
/= c/X Hz, then:

2hc e(d, (t>)

[exp(/jc/XA:r) — 1]

_ 2^ e(d,<t>)
[exp(hf/kT) — 1]

At microwave frequencies, hfjkT^l and

hf hf
exp 1+
kT kT
2kf^Te{e,(j>) 2kTe{d, (p)
Thus 5/(0, 0) = (8.1)

known as the Rayleigh—Jeans approximation.

Thus the brightness, or radiance, is directly proportional to the emitter
temperature, a much simpler result than for infrared radiometry where the full
Planck function must be used. It follows that there should be a linear corres¬
pondence between the temperature and the power emitted, although there will
be a strong directional variability depending on e(0, (p).

8.1.3 The radiation received by the satellite antenna

Microwave antennae respond to incident radiation in a different way from the
optically focused visible or IR wavelength sensors which can view a narrow
angular spread of rays. Instead, the antenna has a directional response with a
gain of AqG(6, 0) where G(d, (p) is a normalised power pattern having a typical
shape as shown in Fig. 8.2, with a main lobe and side lobes. Aq is the effective
area of the antenna, which for radiation of wavelength Xis defined as

X^
Ae=- (8.2)
f G(e,(P)d9.
HIT
(A helpful introduction to the theory of microwave antennae in the
remote-sensing context will be found in Chapter 9 of Reeves (1975).)
Sec. 8.1] The physical principles of passive microwave radiometry 251

B(ei)

Fig. 8.2 - Typical power pattern shape of a microwave radiometer antenna

The antenna also receives energy from one polarisation state only, that
is, intercepting only half the possible energy flux. Thus the total power
received over a bandwidth between / and /+ A/; at an antenna illuminated by
a brightness Bf (6, (p) of unpolarised radiation, is

1 f
B = -A, Bj-(0,<p)G(e,<P)dndf. (8.3)
2 j/ Jatt
Now from (8.1), if the emitter in direction (0,<p) relative to the antenna
(Fig. 8.2) has a temperature T(0, <p),we can write (8.3) as
kA^Af
P = -^i^Jid,<P)e{d,<t>)G{d,<l>)da (8.4)

where a bandwidth A/ has been assumed to be narrow enough for X, T, e, and

G to vary linearly over it, average values being taken in (8.4).
The brightness temperature of the emitter is

T^{e,<i>) = e{d,<t>) T(d,<p).

If Fb is invariant with direction then the power received at the antenna is

kTb AfAeX-^ r G(d, <l))dn = kTs A/, using (8.2)

47r
252 Passive microwave radiometers [Ch.8

However, T^{d, (/>), will normally vary with direction, owing to the variability of
temperature in the atmosphere and across the sea surface being viewed. We
therefore express the power received by the antenna as

P^kT^Af (8.5)

where is called the antenna temperature, by analogy with Nyquist’s theory

of thermal agitation of charge in electrical conductors (which shows that the
radiation within a microwave bandwidth A/ from a resistor at temperature Tr
is kT^Af). In reality the antenna does not have the physical temperature
If it were an ideal (loss-less) antenna its own temperature would be irrelevant to
the discussion, since it would convert all the received thermal radiation power
into electrical power without re-emitting any thermal noise power of its own.
Rather, is to be understood as the temperature of the medium being viewed
by the antenna, that is, the effective temperature of the radiation intercepted by
the antenna. The power P received by the antenna within the narrow bandwidth
Af is measured electrically by the microwave radiometer electronic systems,
with due allowance for any losses, and can then be converted by (8.5) into the
temperature T^.
If the ambient radiation is uniform from all directions with brightness
temperature Tg, then 7X = Tg. Otherwise, is related to rg(from (8.2), (8.4),
(8.5)) by
/
rg(0,(/>)G(0,0)dJ2
Ta = --■ (8.6)
/^^G(0,0)dn
The aim of antenna design is to achieve a power pattern having a strong
narrow main beam and low side lobes, so that 7)^will approximate closely to
the average value of Tg in as narrow a field of view as possible in a given viewing
direction.

8.1.4 Atmospheric and solar radiation effects

The brightness temperature T^(d, 0) as viewed by the antenna contains contri¬
butions from sources other than the earth or sea surface being viewed. In fact
the contribution of the atmosphere is similar to that encountered in IR radio-
metry (section 7.4.1). It can be simplified in the microwave case to be expressed
directly in terms of temperature, rather than the Planck function, in view of the
Rayleigh—Jeans approximation (8.1).
The radiative transfer equation gives

^b(0) = [(1 -R)Ty, + ^T’incl exp[-sec0 / ^a(z)dz]

-b + c “-

z
+ sec0 / rair(z)0£(z)exp[—sec0 / a(z)dz]dz (8.7)
-d-
Sec. 8.1] The physical principles of passive microwave radiometry 253

where 7\v is the sea-water surface temperature, R is the power reflection

coefficient at the sea surface, Tine is the temperature of radiation incident upon
the sea surface, (z) is the atmospheric temperature at height z, 6 is the nadir
angle of the viewing direction, and 0!(z) is the atmospheric absorption
coefficient. a(z) is measured in Nepers m“S and the integral inside the exponen¬
tial term can be expressed as the optical thickness r(0, z) between heights 0 and
z, that is,
t(0,z)=/ Q;(z')dz'
0

Tine is defined by:

RTinc = R Text exp [- t(0 , «>) sece]

-— C -

+ Rsecd /J°TairCz) a(z) exp[—r(O,z)sec0] dZ .

-b-
Text is the incoming radiation from space at around 3K, including the galactic
contribution and the cosmic residual radiation. The separate terms (a) to (d)
are respectively the contribution of the water surface (the oceanographically
useful signal), the downwelling atmospheric emission reflected at the sea surface,
the external radiation after absorption through the atmosphere and reflection at
the sea surface, and the upward radiated atmospheric contribution evaluated
at the satellite height h. They are illustrated in Fig. 8.3.

Fig. 8.3 - Contributions to the microwave radiation received at the sensor.

(a) The signal emitted by the sea surface, (b) The downward atmospheric emis¬
sion, reflected at the sea surface, (c) External radiation, reflected at the sea
surface, (d) Direct upward atmospheric emission

Since atmospheric scattering is not a problem in the microwave, and the

radiometers are not pointed in the sun-glint direction, solar contributions to
the radiation can be ignored.
254 Passive microwave radiometers [Ch.8

As with infrared radiometry, it is not possible to predict the value of a{z)

which depends on such atmospherically variable quantities as water vapour,
cloud liquid water content, and rainfall, as well as more constant quantities
such as oxygen concentration, a is also highly frequency-dependent. Thus (8.7)
does not offer the possibility of atmospheric correction by direct calculation,
although it enables the influence of different atmospheric composition types
to be modelled. This can assist in the formulation of empirical algorithms to
retrieve both ocean and atmospheric variables from multispectral microwave
measurements. Fig. 8.4 illustrates the contributions to antenna temperature
from different sources, from which it is apparent that the range 2 GHz to around
10 GHz is the best for surface viewing. Below this the galactic contribution
is strong, and above it the correction required for water vapour and oxygen
becomes increasingly large. Rain remains a problem within the 2 — 10 GHz range.

FREQUENCY (GHz)
Fig. 8.4 - Contributions to antenna temperature from absorption by oxygen,
water vapour, and the galactic background (After Swift (1980))

8.1.5 Sea-surface emissivity

It is the surface emissivity at microwave frequencies which carries information
regarding ocean parameters as well as sea-surface temperature, and which is
therefore of interest to the remote-sensing oceanographer.
Sec. 8.1] The physical principles of passive microwave radiometry 255

The emissivity e and the power reflection coefficient R (as used in (8.7))
are simply related by

^H.v = 1 ~ -^H,V

where the subscripts H and V denote horizontal and vertical polarisation states
which need to be considered, since the antenna measurement is of one polarisa¬
tion state only. For a plane surface, R is dependent on the incident (and
reflected) radiation zenith angle 6, and the complex relative dielectric constant
of water, e:
cos9 — v^— sin’'^0 ^
Rh =-;-■; (8.8)
[cosd + y/e — sin^_

e cosd — y/e — sin^O ^

Rv = -\
e cosd + y/e — sm^d_
. (8.9)

Fig. 8.5 - Typical shape of the variation of horizontally-polarised emissivity ej.j

and vertically polarised emissivity ey, with the viewing angle 9 (From Swift
(1980))
256 Passive microwave radiometers [Ch.8

Fig. 8.5 illustrates the general shape of the variation of and eywith
viewing angle, although the exact position of the Brewster angle at which ey
peaks will vary with the value of e.
The complex dielectric constant for sea-water has been expressed by Lane
«fe Saxton (1952) as
, (^s-O . o
e = +-i —— (8.10)
Infeo

in which i is the imaginary number \/~l, ^oo is the electrical permittivity at

very high frequencies, t.^ is the relaxation time, o is the ionic conductivity, and
Cq = 8.854 X 10“^^ farads m~^ is the permittivity of free space, e^, tj, and a are
functions of the temperature and salinity of sea-water, and have been evaluated
by Klein & Swift (1977).

Salinity (%o) 0 10

Fig. 8.6 - Variation of brightness temperature at normal incidence with sea-

surface temperature, for different water salinities, at a microwave frequency of
1.43 GHz (From Swift (1980))
Sec. 8.1] The physical principles of passive microwave radiometry 257

120 r—

12 18 24 30
SEA SURFACE TEMPERATURE Tg (®C )

Fig. 8.7 - Variation of brightness temperature at normal incidence with sea-

surface temperature, for different water salinities, at a microwave frequency of
2.65 GHz (From Swift (1980))

Figs 8.6 and 8.7 illustrate the resulting variation of brightness temperature
for different SST and salinity conditions at two separate microwave frequencies.
It is apparent that the brightness temperature is very much lower than the actual
SST. At the lower frequency (Fig. 8.6), is much more responsive to salinity
than to SST, and at high salinities the brightness temperature actually decreases
as SST increases. At higher frequencies the salinity-dependence becomes less
(Fig. 8.7), and Tg is a much better measure of SST.
The surface emissivity can also be significantly changed by the presence of
a thin layer of another material on the surface, such as an oil slick, or a film
of organic material, a raft of foam generated by breaking waves and wind action,
or fresh-water ice. Generally, the dielectric constant of the layer will be less
than for sea-water, and the emissivity consequently greater. However, because of
multiple reflections at the air-film and film-sea interfaces, standing waves can
258 Passive microwave radiometers [Ch. 8

be set up, resulting in e having a strong dependence on the layer thickness. For
a given frequency, as the layer thickness changes, e goes through alternately
high and low turning-points, which may differ by a factor of 2. This provides
for the possibility of oil-slick detection from aircraft using microwave
radiometers, but many different frequencies are required to identify the slick
thickness given this complicated e variability with thickness.
The other factor affecting surface emissivity of the sea is the surface
corrugation and roughness due to wind-driven waves. If the surface roughness
length scale can be assumed large compared with the electromagnetic wave¬
lengths, then the resultant emissivity due to an ensemble of different plane
reflecting facets at different angles can be calculated by (8.8) and (8.9), pro¬
vided that the sea-surface slope statistics are known (see Chapter 10). Such
calculations (for example, Stogryn (1967)), using Cox & Munk’s (1954) relation¬
ship between surface-wind speed and sea-surface slope statistics, show the
interesting result that whereas the emissivity of horizontally-polarised radia¬
tion is variable with wind speed at all viewing angles, the vertically-polarised
emissivity appears to be constant at all wind speeds and surface roughness
when viewed at an angle of 50°. Although the actual emissivity is affected also
by backscattering and diffraction from surface-roughness length scales com¬
parable with the microwave wavelengths, so that the model of reflection used
by Stogryn is not entirely valid, in practice there remains a view angle around
50° at which €v is surface-roughness invariant.
Since the wind speed also controls the production of foam, there is a more
complicated relationship between wind speed and emissivity than is suggested
purely by the calculations of surface-roughness statistics.

8.1.6 Skin depth of sea-surface emission

In an electrically-conducting medium, a high-frequency electrical signal will
only penetrate a limited depth into the material. The penetration depth is
related to the wavelength of the radiation in the medium, and to the conduc¬
tivity. Thus the penetration depth of microwave radiation into the sea depends
on the salinity and temperature of the sea, as well as the frequency of the
radiation. The skin depth 5^ in this context is defined as the distance into the
medium at which the power of the electromagnetic radiation is reduced by a
factor exp(— 2). Given the reversibility of electromagnetic emission and absorp¬
tion, the skin depth is an approximate guide to the depth of water contributing
to the emitted radiation, and whose temperature therefore controls the strength
of the emission.
Fig. 8.8 illustrates how S^varies with frequency in typical sea-water condi¬
tions of 20°C temperature and 36 ppt salinity. Up to about 20 GHz, 8^ is greater
than 1 mm, and at frequencies below 2 GHz it approaches 1 cm. As the salinity
is reduced (Fig. 8.9) the skin depth increases further, to about 10 cm for fresh
water, at a frequency of 1.43 GHz. It is interesting to note that even in the range
Sec. 8.1] The physical principles of passive microwave radiometry 259

Fig. 8.8 — Variation of the electromagnetic penetration depth with microwave

frequency for a sea-surface temperature of 20°C and a salinity of 36 ppt (From
Swift (1980))

Fig. 8.9 - Variation of electromagnetic penetration depth with sea-water salinity

at a microwave frequency of 1.4 GHz (From Swift (1980))

6—10 GHz which is used by satellite microwave radiometers to measure surface

temperature, the penetration depth of 2 to 3 mm is almost certainly greater
than the thermal skin layer discussed in section 7.3.3. Although the near¬
surface layers contribute more strongly to the signal, the ability of the warmer
layer below the cool skin to contribute to the radiation will tend to reduce
260 Passive microwave radiometers [Ch.8

the thermal skin temperature deviation effect. This is rather academic at present,
when microwave radiometers are not capable of resolving SST accurately enough
to make the thermal skin worth worrying about, but as improvements are made
in future, it could well be another useful feature of microwave radiometry.

8.2 MICROWAVE RADIOMETER DESIGN

8.2.1 Radiometer design constraints
From the foregoing, it appears that radiometers operating in the 1 — 10 GHz
range have the most to offer oceanography. In the low end of that range,
particularly, they are potentially capable of measuring salinity and penetrating
the thermal skin. However, there are other constraints which require satellite
radiometers at present to have higher frequencies. The principal one is the
ground resolution. The angular resolution j3 of a microwave radiometer is
normally defined as the half-power width of the antenna main beam. Whilst
this depends on the detailed antenna design and is related to the power pattern
G(0,(/>), it is always approximately related to the wavelength of radiation and
the antenna aperture diameter/) by j3 = X//) radians.
For a satellite at altitude h, this results in a ground resolution cell length (or
footprint) d — ^h= \h/D. Thus the required antenna size for a given footprint
d is D = Xh/d. Thus for a frequency of 1 GHz, X = 300 mm, a radiometer
at altitude 1000 km will need to be 6 m in diameter to have a 50 km footprint.
At 10 GHz, only 60 cm diameter is needed for the same footprint. A design
compromise must be reached between ground resolution, size of antenna, and
minimum frequency. The current limits of practicality are represented by the
design of the Nimbus and Seasat SMMR described in the following section,
which has an antenna diameter of 70 cm, and a lowest frequency of 6.6 GHz
for which the footprint is over 100 km.
The ultimate goal in the design of a microwave radiometer for oceano¬
graphic application is probably that it should have a ground resolution of 1km
for coastal sea applications, and operate down to 1.5 GHz to facilitate salinity
discrimination. If it flew at a height of 500 km, then an antenna size of 100 m
is required.
Given the very weak signal that has to be measured by a microwave radio¬
meter, thermal noise from the antenna and the other parts of the electrical signal
pathway to the preamplifier is a problem, as is the stability of the receiver gain.
The first problem of high-frequency noise is overcome by integrating over
many samples. To reduce the noise contribution, the noise must be uncorrelated
from sample to sample, requiring the samples to be spaced an interval of time,
(^/) ^ s, apart, where A/ is the antenna bandwidth. Thus within a sampling
interval N = t^Af independent observations can be averaged, reducing the
noise by a factor of (ij A/)'^^ This can be related to a radiometric (temperature)
resolution for the radiometer:
Sec. 8.2] Microwave radiometer design 261

where M is a figure of merit depending on the design of the receiver circuits.

It is interesting to note that, for a scanning radiometer, AT and the foot¬
print size d are not independent. If footprints are to be viewed across the
swath of the sensor, in the time taken for the satellite to move at speed U over
the ground a distance d, then the maximum value of ignoring the need for
calibration views, is ? = d/UN^. Thus

or, since for a given swath width S, N^— S/d,

AT o:\ld.
Once again, the SMMR represents the best compromise between tempera¬
ture resolution and footprint size which has been achieved to date. If a 100 m
antenna could be flown, then to achieve a reasonable swath width by a sequential
scanning mechanism could result in far too small a and instead a parallel
array of sensors using the same reflector dish would have to be used, being
sampled in what is called a ‘push-broom’ mode.
The problem of gain variability on a timescale of or longer is catered for
by switching the preamplifier input between its antenna and a source of known
temperature, either a heated or cooled source on board the satellite or a receiver
horn pointing skywards. Comparison between the two sources enables absolute
calibration to be achieved, provided that a reliable known temperature source
is used. The time spent switching between the sources reduces the sampling time
and hence worsens the temperature resolution. Another compromise must be
reached, between absolute accuracy and resolution, depending on the application.

8.2.2 The Seasat and Nimbus-7 SMMR

The Scanning Multifrequency Microwave Radiometer which was flown on both
Seasat and Nimbus-7 represents the most recent in a developing succession of
microwave radiometers. Fig. 8.10 is a sketch of the instrument. The receiver is
a multifrequency feed horn at the focus of a reflector which is offset so that as
the reflector oscillates about a vertical axis through the feed horn, the boresight
(the central axis of the main beam) is at 42° to the vertical. Because of the
spherical surface of the earth, this results in the footprint being viewed at an
angle of incidence close to 50°. The scan swept out is therefore a curved rather
than a straight line on the earth. On Nimbus the view is forward-looking, and
scanned symmetrically from +25° to —25° on either side of the sub-satellite
track. On Seasat the view was rearward-looking and biased 22° to the right of the
flight path (see Figs 8.11, 8.12).
262 Passive microwave radiometers [Ch.8

The geometry of the offset reflector is such that the footprint is oval rather
than circular in shape. The various frequency bands and their corresponding
footprints are listed in Table 8.1. The scan period is 4.096 s, which is sufficiently
fast to achieve a continuous coverage of the ground within the swath even for
the smallest footprint at 37 GHz, which has to be sampled more often than the
higher frequencies with larger footprints. The five frequencies are sampled
for horizontal and vertical polarisation, resulting in ten separate data channels
containing geophysical information. A considerable amount of data processing
is required to present the channels of data in terms of antenna temperature on a
grid pattern covering the swath. Further details of the procedures are presented
by Njoku (1980). Assessements of the early performance of the SMMR on
Seasat and Nimbus-7 are to be found respectively in Lipes (1982) and Gloersen
(1981).

Axi$ of Rotation

Off*at
Reflactor

Fig. 8.10 - The Scanning Multichannel Microwave Radiometer (SMMR) flown on

Seasat and Nimbus-7
Sec. 8.2] Microwave radiometer design 263

Table 8.1 Characteristics of the Scanning Multichannel Microwave Radiometer

(SMMR)

Satellite vehicle Seasat Nimbus-7

Altitude, km 800 950
Period of operation 8 July-10 Oct 1978 Oct 1978^ 1983
Scanning direction 50° scan, aft-viewing but 50° scan centred on the
offset 22° on starboard side forward track direction
Swath width, km 600 800

Frequency, GHz 6.63 10.69 18.0 21.0 37.0

Polarisation H and V H and V Hand V H and V Hand V
Radar wavelength, cm 4.6 2.8 1.7 1.4 0.8
Antenna 3 dB beam-width, deg. 4.2 2.6 1.6 1.4 0.8
Integration time, ms 126 62 62 62 30
Seasat footprint size, km 136 X 89 87 X 58 54 X 35 44 X 29 28 X 18
Seasat retrieval grid size
(effective resolution), km 150X 150 85 X 85 54X 54 54 X 54 27 X 27

Brightness temperature resolution

of sensor for 300 K target, degK 0.9 0.9 1.2 1.5 1.5

Bandwidth MHz 250

Absolute temperature accuracy,
degK 2
Brightness temperature range, K 10-330

8.2.3 Other satellite microwave radiometers

Table 8.2 lists the important parameters of several previous microwave radio¬
meters flown during the 1970 s. The general trend of development has been to
lower the frequency of operation, and also to reduce the footprint size without
greatly improving the radiometric resolution. The inverse relation between
spatial and thermal resolution is apparent if the Electrically Scanning Microwave
Radiometers (ESMRs) on Nimbus-5 and 6 are compared with the other instru¬
ments carried on these satellites, the Nimbus E Microwave Spectrometer (NEMS)
and the Scanning Microwave Spectrometer (SCAMS) respectively. The former
have a smaller footprint at the expense of a poorer temperature resolution.
The ESMR scanned electrically, using a phased-array antenna, rather than
mechanically as does the SMMR.
Although the performance of these earlier instruments may seem inferior
to the SMMR, it is worth noting that some oceanographically-usable data was
obtained a decade ago. The ESMR on Nimbus-5 was used to identify different
ice types in the Beaufort Sea (Campbell et al. (1974)), whilst the NEMS and
SCAMS, designed as atmospheric sounders, were able to provide maps of rainfall
264 Passive microwave radiometers [Ch.8

distribution over the oceans (for example, Staelin et al. (1976)). Without more
than one channel in the 1—20 GHz range the radiometers could not properly
separate sea-surface temperature from other contributions to the antenna
temperature, and applications were limited to identifying features with a
strong emissivity contrast, such as ice boundaries, or with a strong effect on
atmospheric transmission properties, such as rain and snow.

Table 8.2 Details of some radiometers other than the SMMR

Sensor name ESMR NEMS ESMR SCAMS

Satellite Nimbus-5 Nimbus-6

Height 1100 km 1100 km
Launch date 11 Dec 1972 12 June 1975

Frequency bands GHz 19.35 22.24 37.00 22.24

31.40 31.65
53.65 52.85
54.90 53.85
58.80 55.45

Bandwidth MHz 5-^ 125 250 250 220

Spatial resolution km 27-175 180 20 X 50 145
Temperature resolution degK 1.5 0.25 1.0
Integration time 47 ms 2s 1 s
Additional information electrical single H and V across-track
scan. vertical polarisation scanning
3300 km view electrical
swath scanning

8.3 OCEANOGRAPHIC INTERPRETATION OF PASSIVE MICROWAVE

DATA FROM SPACE
8.3.1 Potential applications
In principle, the physics of microwave radiometry and of microwave emission
from surfaces contains the potential for a variety of applications in oceano¬
graphy. Most of these have been shown to be possible using aircraft-borne
sensors, but not all of them can yet be achieved with satellite radiometers.
Sec. 8.3] Oceanographic interpretation of passive microwave data 265

The obvious application is the measurement of sea-surface temperature, and

this can now be achieved to within about 1.5 K. Because the brightness tempera¬
ture may be influenced as much by temperature effects on the emissivity as
by the variation of black-body radiation with temperature, the relationship
between SST and brightness temperature is nonlinear, and frequency-dependent.
For surface roughness and atmospheric effects to be removed, multichannel
sensing is essential for SST retrieval.
The other ocean application which has been developed sufficiently to be
useful is the determination of surface-wind speed from the effect it has on
the surface roughness. Accuracies of the order of 2 m s”' are possible. Again,
a multichannel radiometer must be used, and the retrieval algorithms are
more effective if the view angle is near 50° at which the vertically-polarised
sea radiation is invariant with surface roughness.
The determination of salinity from its effect on emissivity requires radio¬
meters in the frequency range 1—2 GHz, and these have not yet been flown on
spacecraft. The principle has been proved by airborne radiometer flights over
Chesapeake Bay (Blume et al. (1978)) when salinity distributions were mapped
with 2 ppt contours, and comparison with ship samples indicated agreement
within about 1 ppt. Such resolution could be useful in estuarine survey work,
but would be of no value to open-ocean studies where a salinity resolution
between one and two orders of magnitude better is required to discriminate
between different water masses. Given the high spatial resolution required for
coastal and estuarine studies, and the large antennae necessary to achieve this at
low frequencies, it is unlikely that satellites will contribute to the measurement
of ocean salinity distributions in the near future.
The same applies to the observation of pollutants, oil spills, slicks, and foam
from space. Low microwave frequencies are not demanded in this case, and the
surface slicks can have a strong surface-emissivity signature, but by the nature
of the problem, spatial resolutions of tens to hundreds of metres are desirable
for the determination of oil slicks. If slick thicknesses are to be determined,
many channels of information are required. This application is one which is at
present most appropriately pursued by airborne sensors.
Another application of some interest to oceanographers is the use of the
SMMR in Arctic and Antarctic waters to delineate sea ice. Research programmes
on Arctic sea ice using aircraft and ground radiometry have demonstrated that
first-year ice typically 1.5 m thick has a brightness temperature of 235 to 240 K,
multi-year ice of around 4 m thickness has a brightness temperature between
209 and 233 K, whereas the brightness temperature of open-sea water is between
130 and 160 K. Thus the edge of the ice sheet is readily apparent on microwave
imagery, and the proportion of ice cover can be estimated from the brightness
temperatures when the ice is broken up with open water in between. Since
fresh-water ice acts as a layer of lower dielectric constant over the sea-water, the
emission from it shows the oscillatory signature characteristic of layered media
266 Passive microwave radiometers [Ch. 8

as the thickness increases (see section 8.1.5). Since the thickness/emission

variability changes with frequency, multichannel scanning is valuable in this
work.
Although the relatively poor spatial resolution of the SMMR means that
aircraft observations are necessary for detailed studies of sea-ice distribution,
the Nimbus SMMR’s ability to record the movement and change in character
of sea-ice continuously over several years, despite cloud cover and the long polar
night, have yielded a great deal of new information about the ice cover (Zwally
& Gloersen (1977)). Amongst other things this is useful in the estimation of the
ice-cap albedo and its variability, important in the study of the global energy
budget. It was found from the ESMR satellite data for example (Campbell et al.
(1980)) that there are sizeable regions of semi-open water within the heart of
the polar ice pack, and their slow movement can be detected by time sequences
of microwave imagery. In this respect the long life of the Nimbus SMMR has
made it much more valuable to ice oceanography than the short-lived Seasat
instrument.
Finally, there are a whole range of meteorological and atmospheric para¬
meters which can be obtained from passive microwave-sounding radiometers.
Strictly, these fall outside the scope of this book, but they are often of use to
oceanographers. Atmospheric water content is a parameter measured by SMMR
(for example, Katsaros et al. (1981)) which cannot readily be measured by any
other means. It is important in air—sea interaction studies, and Chelton et al.
(1981) have demonstrated the possible contribution of SMMR water data to
climate research.

8.3.2 SST and surface-wind retrieval algorithms for SMMR

The data from the SMMR is gathered from footprints of different sizes for each
of the frequency channels. The footprints are oval, and lie along circular zig-zag
paths filling the swath, as illustrated in Fig. 8.11. These are resampled in terms
of rectangular grid boxes (Fig. 8.12), but the higher frequencies are capable of
being sampled in smaller grid boxes nested in the larger ones (see Wilheit (1978),
Njoku (1980)). There are ten separate antenna temperatures, corresponding
to the two polarisation directions at five different frequencies, from which to
construct an algorithm for the various geophysical parameters which contribute
to the signal. The spatial resolution of the geophysical parameters is limited by
the spatial resolution of the lowest frequency used. Table 8.3 shows which
channels are used for SST, surface wind, water vapour, cloud liquid water, and
rain rate, the five variables retrieved from the raw data. The table applies to
the algorithms used by NASA/JPL for routine geophysical data production,
and shows the different ground resolutions achieved for different parameters.
The two ocean parameters, requiring the lowest frequencies as input to their
algorithms, have the poorest spatial resolution.
Sec. 8.3] Oceanographic interpretation of passive microwave data 267

I DI V 1 S t G = I 0O KM

Fig. 8.11 - Instantaneous antenna 3 dB footprints for the Seasat SMMR at

6.6 GHz (large ovals displayed every 128 ms) and 37 GHz (small ovals every
32 ms) (From Njoku (1980))

Fig. 8.12 — Grid cell configuration for the Seasat SMMR (four cells in the swath
width W) (From Njoku (1980))
268 Passive microwave radiometers [Ch.8

Table 8.3 Channels used for geophysical parameter algorithms

(after Taylor (1983))

Retrieval grid size

Geophysical parameter Channels used (GHz) (km)

(1) Sea surface temperature 6.6V, 6.6H plus corrections 150 X 150
(SST) for water vapour (3) and
cloud (4)
(2) Surface wind 10.7H plus SST (1), water 85 X 85
vapour (3) and cloud (4)
corrections
(3) Water vapour 18V, 18H,21V, 21H 54 X 54
(4) Cloud liquid water 18V, 18H, 21V, 21H, 37V, 37H 54 X 54
(5) Rain rate 18H,37H 54 X 54

The antenna temperatures also have to be corrected for the contribution

by the side lobes, and for various polarisation rotations (Njoku (1980)) in
order to achieve a dataset of brightness temperatures as they would appear
at the altitude of the satellite if the polarisation vector did not rotate. The
calibration algorithm due to Chester simply applies a linear matrix operation
to these brightness temperatures in order to generate geophysical parameters.
Fig. 8.13 illustrates the sequence schematically, and the dependence of certain
parameters on others. The switchable gates in Fig. 8.13 represent coarse grid
corrections for SST and wind speed which could be developed to improve
further the atmospheric water algorithm.
The accuracy of such an algorithm depends on the choice of matrix
coefficients, determined from multiple linear regression on a large dataset of
simulated brightness temperatures resulting from a variety of atmospheric and
ground-emission conditions. The simulation is based on well-proved models of
atmospheric transmission and sea-emission characteristics (Wilheit etal. (1980)).
The algorithms have been improved since launch, by fitting the predictions based
on actual satellite measurements to synoptic sea and atmospheric data, where
this is available.
The accuracy of the updated Chester algorithm has been tested against later
synoptic sea truth, and it appears that SST can be retrieved with this algorithm
to within —0.14 ±1.1 C, the first figure being the mean bias and the second
the scatter about that mean. Since the sea-truth measurements must contain a
certain amount of noise and error, it is reasonable to suggest that the SMMR
is capable of retrieving SST to an accuracy of better than TC. It should be
Sec. 8.3] Oceanographic interpretation of passive microwave data 269

noted, however, that these comparisons are made with satellite data obtained in
the absence of sunglint, significant rain, or radio-frequency interference, all of
which seriously degrade the retrieval algorithms. Another limitation is caused
by the side-lobe correction which assumes that the side lobe views the sea at
a similar temperature to the main beam. If it views the land, serious errors
occur, and this restricts the use of the SMMR for SST retrieval at this accuracy
to locations at least 600 km away from a land mass.

Fig. 8.13 — Data-processing scheme for the Seasat and Nimbus-7 SMMRs
(Adapted from Wilheit et al. (1980))

Sea-surface wind-speed accuracy of 1.4 ± 2.1 m s“^ is also achieved, subject

to the same constraints, except that degradation occurs only within 500 km of
land.
Taylor (1983) has reviewed these algorithms, and points out that a different
algorithm due to Wentz is capable of producing SST accuracies of 0.07 ± 1.0°C
and surface-wind accuracies of 0.8 ± 1.9 m s'* from Seasat SMMR data. In each
case there is a significant reduction in the bias and a small reduction in scatter.
The algorithm uses a simplified nonlinear model to predict brightness tempera¬
tures from assumed geophysical parameters. This model (Wentz (1983)) was
originally derived from the full radiation transfer equations and sea-emission
models. Francis et al. (1983) have proposed a similar approach to the retrieval
of Nimbus SMMR data. Fig. 8.14 shows the computational scheme required to
fit geophysical parameters to the observed brightness temperatures. An initial
estimate of parameters is made, normally based on the values obtained for the
270 Passive microwave radiometers [Ch. 8

adjacent grid square. These are used in the nonlinear environmental model to
predict the brightness temperatures which they would be expected to produce at
the satellite. Comparison with the actually observed values shows if the predic¬
tion was valid. If not, a differential correction is applied to alter the original
estimates of the geophysical parameters, and the iteration procedure continues
until the satellite-observed radiation in all ten channels can be predicted
sufficiently accurately.

Fig. 8.14 — Nonlinear iterative algorithm scheme for recovering the environmental
parameters from the SMMR (From Thomas (1981))

8.4 COMPARISON BETWEEN INFRARED AND MICROWAVE

RADIOMETERS FOR SST MEASUREMENT
There are now two different sensor types available for the measurement of SST
from space. There is some debate regarding which is the most useful to oceano¬
graphers. The microwave radiometer achieves an all-weather coverage which
makes it very attractive for the construction of systematic databases, but its
overall performance is poorer, both for spatial and radiometric resolution, and
for accuracy. To make a complete comparison between the two types of sensor,
their relative merits in several different areas are listed in Table 8.4, with an
indication of which factor is a positive advantage or a disadvantage for that
sensor.
The best performance so far achieved from a satellite is also compared.
There is room for improvement in the development of both types of sensor, and
which is most suitable will inevitably depend on the application. Except in very
 Sec. 8.4] Comparison between infrared and microwave radiometers 271

c o T3
o a <u hs
<U

e <u
<u 3
> cr 1> 5d cr ’T)
o o a> a (L> Cd
<D
'4-»
c Cd o
CO S S cd
C o
C
o s g C .2 C
o
.2 g g
.2 o a>
c o *■4-* <D
o <u C 00
Table 8.4 Comparison of IR and microwave measurements of sea-surface temperature

!U o- - ^ 4Z
cd
'o
s o cd
Sn -
o - ^ ca £? !U 00
c C
O w ^ Cd o
cd «« ii H-1 cd ■$
(U
•4-)
cd
(-1 nU
cq a> O ^ Cu cd ’>
> cd ^ § o 5 o :2
a "S > a> x:
s M c C 4:5 "cd
o

achieved
^ »-i fl)
c/3
ca C.UJ t-. ^ o <u 00 CD

1.5 degK
o o c
1 10 Ui c/3 o o -4->

2 degK
c/3 Cd <D
oq d T3
Ui
>% ^ 2 W S 'o
4-1 rz3 cd
z< ro
d
0 Cd Cd
>
c/3 cd
>
c/3
Cd
oq 0 vu W Q
+ +

C Q 0 0
.2
.2
X> 3 V-*
0 u
u 0 0 a>
0-1 c/3 CL X 'O
e <U (U
3 Ui
-Td E -4->
C/3
o -4-^
-S
Cd Cd
0)
c cd
o -£ ^ .2
X
So o O
6 ta
<D
C 00

c: .2
X) & e
0
13
00 .2
SZ
c Ui u ’-M
t; S § ^ cd 0
on-board target

cd 00
s ca -o 'TU o (i>
<u
Ui n:3
CO a>
.9 C
<L» o
(U
4-J s
cd
*2 00
0
c >
<D
00 Ui o o cd 00
cd Ui
Cd 0)
Cd
CL ^
^ I
Cd
U4
Ui X> •2 > s
CO
.2
'O
0 2
O
00 o cd
Ui ■$
0.6 degK

C/3 0 .2i
0.1 degK

N Ui
Cd c -2 o o cd <1>
Cd

cd g & a> O
Ui Xl
0
0 *>
CCl
«—(
Ui
-4—• a- E X5 Ui G ca -4-J

/< ro
1
1
^
fTD ?l
(I 0
0
D <L>
c
cd
S .S L—1
Ui
:3
0 cd
<D
«j Z
c cd
u CO
1 oq

1
-- + -I- + + +

C H
Presently achievable sensitivity

tZ)
_o on
Presently achievable absolute

(D
-4—>
a> .2
O
z^d
e a>
o
Cd 0
kH C/D C
<D 0) cU
c/3
0
cd
b-i o f2
c
-4-*
Um c
c
C4—1 c^ 0
oc Ui

.s 0
C/3
S'
'u
u
D
u->
0
0
0
.2
cd
-4-^ X C u
u Cd
0 Cum
U-*
t3 X
00 0 0
0 Ui cd
accuracy

<u Ui '0
>> a> 00
XI >v Q./
-4-* ■4-M <L> X U CD
ZJ
-4-> s [> c/3
u-»
Cd
CL
c/3
00
C
u-<
o ’S00 U-> ■4-» cd
Cloud

*C/3 0
cd 0 00 00 U-4 *0
• '?
Un cd
cd c
s Cd
«D
E
■4-M
Cd
cx
0
CO
X
0
S c/5 w C/5 <C 00 > <
272 Passive microwave radiometers [Ch.8

cloudy areas, it would seem that on balance the superior resolution of the IR
devices at present outweighs their cloud limitation. Very little can be done,
however, about the IR cloud problem, whereas there is room for significant
improvement in the microwave radiometer performance through technological
development, larger antennae, array imaging, etc. Furthermore, given the micro-
wave radiometer’s diversity in measuring other parameters besides SST and
its value for ice studies, it is possible that in a decade or so it will have been
developed to a design which can compete with IR radiometers.

8.5 APPLICATIONS OF SMMR DATA

Perhaps because of the poor comparison between the SMMR and the AVHRR in
SST retrieval accuracy and resolution, despite the cloud limitations of the IR,
there appears to have been little use made of the SMMR data in oceanographic
research applications other than that required for assessing the accuracy of the
sensor. The same is true of its surface-wind measurements, which tend to be
overshadowed by the success of the Seasat scatterometer, although the
Nimbus SMMR has produced an archive of data several years long. Most use
appears to have been made of the SMMR in ice studies, and in measurements
of atmospheric and meteorological parameters which are beyond the scope of
this book.
Bernstein (1982b) and Bernstein & Morris (1983), have used Seasat SMMR
data to map the surface-temperature distribution of the tropical and mid¬
latitude North Pacific. It was found necessary to correct the SST supplied by the
production algorithms for some systematic errors. These included the existence
of artificial cross-track gradients of SST from one side of the swath to the other,
owing to incomplete allowance for polarisation rotation of the radiation passing
through the atmosphere. There was also a systematic difference between SST
measured on ascending and descending satellite passes, ascribed to the effect
of the diurnal thermocline cycle, since the ascending and descending passes
occurred regularly at different times of the day. Bernstein’s careful analysis
and correction of the data supplied by the production algorithms provides an
example of how important it is that oceanographic users should examine and
validate satellite data carefully before drawing oceanographic conclusions from
them. If systematic errors in the algorithms are found to be small, it is possible
to apply an empirical adjustment, but if they are significantly large it suggests
that the routine processing algorithms require modification.
In the small area of the Pacific for which detailed SST data were available
from routine marine weather observations on ships, comparison was made
between SST maps based separately on the ship and on the satellite data.
It is not easy under these circumstances to decide whether discrepancies in
direct comparison between the two are due to error in the SMMR data, in the
marine observations (which may not be measuring exactly the same tempera-
Sec. 8.5] Applications of SMMR data 273

ture because of diurnal thermocline and surface skin effects), or both. The maps
must be constructed by interpolating between many point observations, whilst
applying smoothing criteria so that one single measurement which is seriously
in error will not grossly distort the map. This is more of a problem with ship
data from many vessels, than with the satellite data which should have self-
consistency. The satellite data are also regular and have a fuller coverage than
the ship data, and although the ground resolution (150 km X 150 km cells)
seems rather poor, it is at least as good as can be achieved by routine reports
from ships on passage. Consequently, when the database from which they were
constructed was compared with the resulting maps of SST, the SMMR data were
found to have a scatter of 1.5°C relative to the map, whereas the ship data
scatter was 2.1 C. This suggests that the regular, uniform coverage of a single
sensor, although its inherent accuracy is only around 1°C, is able to produce a
better map than the haphazard sampling from ships.
Fig. 8.15 illustrates the monthly SST-anomaly maps produced by satellite
and ship observation for July to September 1978. They show a general
agreement, but the satellite maps are more complete than the ship maps. By
constructing 8-day coverage SST maps from the SMMR data (which would
be almost impossible with the less regular ship data) and presenting them in a
time sequence (Fig. 8.16), it has become possible to observe fluctuations in
the intensity of the cold equatorial tongue which stretches across the Pacific,
and meanders in the strong meridional SST gradient just north of the equator.
The wavelength and speed of the meanders can be estimated by presenting the
time-variation of a particular isotherm, as shown in Fig. 8.17, and individual
meanders can be tracked over 3000 km. In these interesting observations,
Bernstein & Morris have selected the type of information which the SMMR is
best able to supply. Comparison with Fig. 7.9 reveals that IR radiometers would
be prevented by clouds from obtaining equatorial SST maps with an 8-day
sampling frequency, and it appears that microwave radiometry from space is the
only means of directly examining the dynamics of the equatorial ocean at length
scales of around 2° latitude and above, and timescales of about 10 days.
 Jul 7 - Aug 4,1978 7-Aug 4.1978
274
Passive microwave radiometers

Fig. 8.15 - Monthly maps of sea-surface temperature anomaly in the North Pacific Ocean for July, August, and September 1978. (a) Based on
[Ch.8

Seasat SMMR data, (b) Based on ship-observed data. (From Bernstein & Morris (1983) copyrighted by the American Geophysical Union)
Sec.8.5] Applications of SMMR data 275

(a) (e)

(b) (f)

20*N

10°N-

Fig. 8.16 - Continued next page.

276 Passive microwave radiometers [Ch.8

Fig. 8.16 - Time sequence of 8-day composited Seasat SMMR SST maps for the
Eastern Tropical Pacific. There are 6-day steps between the maps. (All periods in
1978):
(a) 6 13 July (b) 12 ^19 July (c) 18-25 July
(d) 24-^ 31 July (e) 30 July ^ 6 Aug (f) 5-12 Aug
(g) 11 ->18 Aug (h) 17-24 Aug (i) 23 — 30 Aug
(j) 29 Aug ->■ 5 Sep (k) 4-11 Sep (1) 10-17 Sep
(m) 16 ^ 23 Sep (n) 22 - 29 Sep (o) 29 Sep — 5 Oct
(Adapted from Bernstein & Morris (1983))
Sec. 8.5] Applications of SMMR data 277

•so* (20* no* 100*

..I I I I I I I ■ I I I I I I I I I I I I I . I I I I.
1978 ^
6-13 July

12-19 July

18-25 July
24-31 July

30 July-6 Aug

5-12 Aug
11-18 Aug
17-24 Aug
23-30 Aug
29 Aug-5 SapI

4-11 Sapi
10-17 Sapi
16-23 Sapi
22-29 Sapi

28 Sapi - 5 Oel

1*0* no* «00«

LONGITUDE CW)

Fig. 8.17 — Composite plot of the 24°C sea-surface isotherm from the fifteen
maps of Fig. 8.16, illustrating the westward progression of wave-like features
along the dashed-line trajectories. (From Bernstein & Morris (1983) copyrighted
by the American Geophysical Union)
CHAPTER 9

Satellite altimetry of
sea-surface topography

9.1 INTRODUCTION
9.1.1 An exciting new oceanographic instrument
Judging by the number of scientific papers which have appeared since the three-
month flight of Seasat in 1978, the observations made by the radar altimeter
on board have attracted the attention of many ocean scientists. This is not
surprising when the achievement of the Seasat altimeter is considered. It was
able to measure the distance between the satellite and the sea surface below it to
within 10 cm, no mean feat when the distance is about 800 km, an accuracy
of one in eight million! When it is realised that the altimeter offers oceano¬
graphers the chance of doing something they have dreamt of for a century,
namely, measuring the absolute slope of the sea surface, the important implica¬
tions of the altimetric achievement become apparent. Add to this the ability of
the altimeter to accurately measure the significant wave height and to estimate
the surface wind speed at the same time, and we have the reasons why the major
space agencies are each planning to fly another altimeter within the 1980 s.
The satellite-borne radar altimeter promises a great deal for dynamical
oceanography, but even the accuracy achieved on Seasat was barely sufficient
to be of real value, particularly because the flight was so short. It will take the
next generation of satellite altimeters to measure distances within a few centi¬
metres before the technique’s exciting potential in oceanography is fully realised.
In this chapter we examine the principles of satellite altimetry and the
various sources of error which must be allowed for. We look at the different
components of the altimetric signal and how the oceanographic part can be
extracted. We explore the rhechanisms of dynamical oceanography which make
possible the interpretation of sea-surface height and slopes in terms of ocean
currents on a variety of length and time scales. Finally we look at other oceano¬
graphic applications of altimetry, including the measurement of ocean tides and
the estimation of ocean bathymetry. We leave until Chapter 11 the consideration
of another separate use of the altimeter to measure significant wave height and
to estimate surface-wind speed. Satellite altimeter data have also been applied
[Sec. 9.1] Introduction 279

enthusiastically to the study of the geoid and in the observation of the snow-
and ice-covered polar regions. Except insofar as these applications overlap with
oceanography they lie outside the scope of this book.
Scientific understanding of how to analyse and apply the observations
of satellite altimetry is rapidly developing. This chapter is intended as an
introduction to the principles of the technique, and not a review of the latest
developments in the recent literature. Many of the key papers have been
published together in collected form, and to explore the field further the reader
is directed towards the final eleven papers in Gower (1981), the first sixteen
papers of Bernstein (1982 c), the first fourteen papers of Kirwan et al. (1983),
and part 4 of Allan (1983).

9.1.2 Principles of satellite altimetry

Satellite altimeters are radars which transmit short pulses towards the earth
beneath them. The return time of the pulse after reflection at the earth’s surface
is measured, and this yields the height of the satellite, provided that the radar
wave speed is known. For a satellite at a height of 500 km the return time
is around 3 milliseconds, but to achieve a 1 cm accuracy of distance measure¬
ment requires the time to be measured accurately to within 30 picoseconds
(3 X 10”“ s). This places demands on both the time resolution and the time
stability of the clock, which must keep time to within 1 part in 10®. The first
task of satellite altimetry is to design a radar transmitter and associated timer
capable of approaching these requirements.
Errors can arise when calculating the distance from the measured time,
because of variability of the wave speed in the ionosphere and in the atmosphere.
Degradation of the pulse shape on reflection from a rough surface also leads to
possible errors, and these deviations must be allowed for so far as possible by
applying corrections.
Measuring the distance between the satellite and the ground is only the first
step in a longer procedure which must be followed before oceanographically-
useful information is obtained. Fig. 9.1 illustrates the problem. The aim for
oceanographic applications is to obtain the sea-surface height relative to the
geoid. The geoid is an equipotential surface, by definition normal to the local
effective gravity force which incorporates earth-rotation forces and the gravita¬
tion of the solid earth and also the ocean and atmosphere. The geoid is the
equipotential surface at mean sea level, that is, if the ocean were everywhere in
stationary equilibrium relative to the earth, its surface would define the geoid.
The height of the ocean surface above or below the geoid is therefore a para¬
meter containing oceanographic information — about ocean currents or about
the tides. In this latter context, the geoid is defined with regard to earth gravita¬
tion only. The gravitational attractions of the sun and moon which give rise to
the tides are thought of as perturbations in addition to the earth’s gravitational
field. Now the form of the earth is fairly regular, and can be thought of in
280 Satellite altimetry of sea-surface topography [Ch.9

general as an ellipsoid with an equatorial radius of 6378 km and a polar radius of

6357 km, the equator being drawn out and the poles squashed together by the
earth’s own rotational forces. However, because of uneven mass distribution
within the earth, and over its surface, the true geoid deviates in height from this
‘reference ellipsoid’ by distances of order 50 m above or below. For example
where there is a seamount protruding above the ocean floor, the geoid will rise
above the ellipsoid, and conversely for an ocean trench. (It is this occurrence
which gives the altimeter its possible application in charting ocean bathymetry.)
To know the sea-surface height relative to the geoid, we therefore need to know
the shape of the geoid relative to the ellipsoid, to the same accuracy as the
satellite altimetry measurement.

SATELLITE ORBIT

Fig.9.1 - Contributions to the satellite-measured altimetry signal

Furthermore, the distance which is actually measured only relates the

sea-surface height to the satellite orbit. The satellite orbit must therefore be
determined relative to the reference ellipsoid, to the same degree of accuracy
as the height measurements being sought. This involves both the modelling of
the satellite orbit and the use of laser ranging from earth observatories, or other
methods of satellite tracking and position fixing.
The whole process is a complex geodetic exercise, since the different
calculations are interrelated. For example, the satellite orbit is itself dependent
on the geoid shape since the satellite responds to spatial variations in gravity.
Fortunately, it is the longer wavelength geoid variations which influence the
satellite. Unfortunately, the small-scale variability of the geoid is not known over
most of the ocean, except where detailed gravity surveys have been performed
at sea. Indeed, in the science of geophysics as a whole the satellite altimeter is
regarded as an instrument for measuring the geoid, and the small perturbation of
sea-surface height above or below the geoid due to tides or ocean currents is
Sec. 9.2] Distance measurement with a radar altimeter 281

regarded as noise to be removed from the geoid signal. Until the geoid is better
defined, marine dynamicists must find ways of extracting oceanographically
useful information from altimeter data which is not distorted by unknown geoid
variability. As in other areas where geodesy impinges on oceanographic science,
such as the measurement of mean sea level at tide gauge stations levelled into a
geodetic network, caution and clarity of thought are required to avoid unsound
assumptions in which the geodesist and oceanographer each rely on the other to
provide an absolute reference level. Otherwise what is believed to be an absolute
measurement will turn out on closer inspection to be entirely arbitrary, and
variable.
Before considering these geodetic problems, we first examine the process of
measuring the distance between the satellite and the sea surface.

9.2 DISTANCE MEASUREMENT WITH A RADAR ALTIMETER

9.2.1 The instrument design
The first hurdle to overcome if high-accuracy altimetry is to be achieved is the
design of a radar capable of the desired time resolution within the constraints of
antenna size, satellite power, and the frequency bands permitted by international
agreements.
For example, in order to measure to within a time resolution of 30 ps (that
is, 1 cm) the leading edge of the emitted pulse should rise to full power within
that time. The sharper a radar pulse is, the wider is the frequency band required
to carry it. In the limit a step function requires an infinite range of frequencies
if it is to be defined exactly by Fourier synthesis. The frequency bandwidth
necessary for a 30 ps pulse would be the inverse (30 X 10“^^)“^ — 30 GHz,
which could not be accommodated in the 3—30 GHz frequency range used for
altimetry. Even if such a sharp pulse could be achieved, reflection at a rough
surface would cause it to be scattered, first reflections of the leading edge
coming from the top of the waves, and the strength gradually building up as
more reflections return from lower parts of the surface (see Chapter 11). This has
the effect of stretching out the return pulse front, and except for a flat calm
sea the 30 ps precision would be swamped by surface-wave noise. Therefore a
longer pulse of about 3 ns, a hundred times longer, is used, and the return signal
which can be sampled approximately twice per nanosecond is fitted to a model
curve (Fig. 9.2). The shape of the curve varies with wave height (enabling the
wave height to be determined), but for altimetry the timing of the midpoint of
the leading slope is required. By comparing it to the known characteristics of the
emitted pulse the travel time can be obtained. Curve fitting leads to errors, but
these can be reduced by averaging over a thousand or more samples to achieve
an accuracy of 150 ps (5 cm). A thousand pulses can be transmitted per second,
so that one height measurement to this accuracy can be obtained every second.
Thus a bandwidth of around 300 MHz can be used, centred at 13.5 GHz.
282 Satellite altimetry of sea-surface topography [Ch.9

c
UJ

TIME (n«)

Fig. 9.2 - Smooth curves fitted through averaged altimeter pulse shapes corres¬
ponding to two different sea states. SWH = significant wave height. (From
Walsh, UUana, & Yaplee (1978))

Another problem confronting the radar designer is how to put sufficient

power into such a short pulse so that the strength of the return signal is
measurable. This is achieved by the process of pulse compression. The sharp
pulse is dispersed by a filter into a very much longer swept-frequency pulse
which conveys exactly the same information, but is able to carry more energy.
On return, the reflected dispersed pulse is fed through the inverse of the
dispersive filter to compress it again. It then has the same form as if it had been
transmitted and reflected as the compressed pulse, because the propagation and
reflection processes are linear. Seasat employed pulse compression by a factor
of 1000.
The next question to consider is the size of the area for which the altimetric
measurement is representative. To achieve a radar beam which is sharply focused
in a small area of sea would require far too large an antenna aperture to be
practical (see Chapter 12 on SAR and radar resolution). Beam-limited geometry
has therefore not been a possibility with present technology, and instead the
effective footprint size is governed by pulse-limited geometry. This is relevant
also to the use of the altimeter to measure wave height, and is illustrated in
Fig. 9.3. Considering the pulse as sharply rectangular, the leading edge will strike
the sea first immediately beneath the satellite and then move out in a circular
front. The trailing edge will do the same a little later, resulting in the illuminated
area being first a circle of growing area and then an annulus of growing radius
but gradually decreasing width. Inspection will show that the circumference of
the annulus is approximately proportional to sin0, and its radial thickness to
l/sin0, resulting in its area being constant. The reflecting area, and hence the
strength of the return signal, increases from zero to a maximum when the rear
Sec. 9.2] Distance measurement with a radar altimeter 283

of the pulse reaches the ground, and then levels out. The timing algorithm fits
a curve to the leading edge of the return signal up to its maximum, and hence
the effective area whose height is sampled is the illuminated circle just before it
becomes an annulus. For a calm sea and a pulse length of fp, this has a radius
of ''a ~ (2 but for a rough sea with significant wave height hij-^ its area
is increased to

''a = (2hcrp)^/^, where

16^]^

With pulse dispersion and compression, the effective in this calculation

is the compressed pulse length.
For Seasat, h = 800 km and tp = 3 X 10~® s, and so was 1200 m for a
calm sea. Thus the altimeter viewed a swath of width 2.4 km (or up to 12 km in
high seas). During the 1 s averaging time for each altimetric value, the satellite
travelled a distance of only about 6.7 km over the ground, giving it quite a high
spatial resolution.

Fig. 9.3 - Interaction of a pulse of duration t with a smooth sea surface. The
lower part shows the annular shape of the illuminated region, and to the right is
shown the strength of the returned signal. (Based on Walsh, Uliana, & Yap lee
(1978))
284 Satellite altimetry of sea-surface topography [Ch.9

9.2.2 Corrections for atmospheric transmission

The height is calculated from the measured pulse travel time tj on the assump¬
tion that the radar signal travels at the speed of light. However, its speed is
reduced through the atmosphere and troposphere, resulting in an overestimate
of the distance.
If c is the speed of light in a vacuum and the pulse speed in a medium,
then the distance is overestimated by A/r = That is.

Ah = VitjCm h{n-\)

where h is the true height and n is the refractive index of the medium
(n = c/cni)- In the estimation of the error Ah, it is sufficiently accurate to use
the approximate height of the satellite (for example its nominal orbit height).
In practice the refractive index of the medium varies with height, in which
case the error is calculated as

Ah = /J'(n-l)dz . (9.1)

In the ionosphere, for radio propagation frequencies around 10 GHz,

{n-\) = ViNar^ (9.2)

where the constant a = 80.5 m^ s"^, N is the number of free electrons per unit
volume, and / is the radar frequency. Thus for the ionosphere, from (9.1) and
(9.2),

Ahj = Ha/"VVdz. (9.3)

Now N varies between day and night and between winter and summer
hemispheres, being less at night and in summer. It increases with the solar
sunspot maximum. Consequently it cannot be directly predicted. It can be
calculated from ground observations looking at its effect on radiation coming
through the ionosphere, and this approach was used to make estimates of Ahj
for the Seasat altimeter (LoreU et al. (1982)). However, ground observations
are widely spaced, and the necessary extrapolations between them can lead to
gross errors in the estimate of N. Fortunately, the influence of the term in
the error reduces the magnitude of Ahi to a maximum of 20 cm at 13.5 GHz,
and for Seasat the correction could be made to within 3 to 5 cm. For higher
accuracy it would be necessary to have more ionospheric observations or else to
exploit the /"^ factor in the correction by using a multifrequency altimeter.
In this approach the difference in travel times at different frequencies could be
used to estimate the transmission error.
In the troposphere the refractive index is influenced by the dry gases
Sec. 9.2] Distance measurement with a radar altimeter 285

present in the atmosphere and by water vapour. This leads to one of the major
altimetric errors, of the order of 2.5 m, which has to be calculated if the
altimeter measurements are to be of any value. The methods used for calculating
the Seasat tropospheric errors are given by Tapley et al. (1982 a).
The dry atmosphere correction, in metres, can be obtained from the
known vertical distribution of gases in the atmosphere, using

^~ 2.277 X 10“^ (1 + 0.0026 cos 20)po (9.4)

where 0 is the latitude of the subsatellite point and po is the surface air pressure
in Pascals. Thus A/Z(j is typically of the order of 2.3 m, but using global inter¬
polations of the surface pressure field from meteorological observations it is
possible to estimate A/z^j to an accuracy of within 1 to 2 cm.
The wet atmospheric correction A/z^ can be calculated from

Ah^ = 2.277 X 10-5 (1255/ro +0.05)/o

where Tq is the surface atmospheric temperature in degrees K, and /q is the

surface partial water-vapour pressure in Pascals. Equation (9.5) makes assump¬
tions about the vertical distribution of water vapour, but can be calculated from
ground measurements of meteorological parameters.
An alternative approach is to use the radiation emitted by water vapour
at frequencies around 22.3 GHz to measure W (kg m"^), the total mass of
precipitable water in a column of atmosphere, using satellite passive microwave
sounders. With Seasat the SMMR was used for this purpose and gave more
satisfactory corrections than by interpolating meteorological data and using
(9.5). Tapley et al. (1982b) present an expression for A/z^ directly in terms of
SMMR brightness temperatures, but also give

Ah^ = 6.36 X 10-3 W

as an approximation in low latitudes. Stewart (1985) suggests

A/z^ = 1.723 HyTa

where is the average temperature (K) of the lower atmosphere.

Typical magnitudes for Ahy, range between 8 cm to 40 cm, but there are
variations of over ±5 cm between estimates based on (9.5) and more complete
integrations of atmospheric effects based on radiosonde data. The SMMR-based
calculations have a better accuracy of less than ±3 cm. Future satellite altimeters
are expected to be flown along with microwave sounders in order to reduce
what could otherwise be a serious source of error in trying to bring altimeter
precision below the 10 cm goal of Seasat.

9.2.3 Surface roughness errors

The fact that the sea surface is corrugated by waves has already been mentioned
in terms of its effect on spreading the return-pulse front, and also on increasing
286 Satellite altimetry of sea-surface topography [Ch.9

the area of the effective footprint. The rougher the sea, the harder it is for
the logic algorithm which tracks the slope of the leading edge of the pulse
to operate accurately, and a tracker bias occurs. In addition, the shape of ocean
waves is not sinusoidal. Troughs tend to be flattened and crests steepened, so
that the radar return for the troughs is stronger than from the crests, leading
to an electromagnetic bias which emphasises the troughs, and overestimates
the distance between the satellite and the mean sea level. The overall effect
of surface roughness is known as sea-state bias and was estimated for Seasat
to be 7% of the significant wave height, which should be subtracted from the
calculated satellite-to-ocean distance (Born et al. (1982)). The electromagnetic
bias which will be common to all altimeters operating at 13.5 GHz appeared
to be between 1.5% and 2% of and the tracker error, specific to Seasat,
was 5% to 5.5%. Even when a 7% bias correction is applied, there remains a
scatter of around 3% of /21/3 in the observations, as checked from repeat orbits
over the same piece of sea under different wave conditions.

9.2.4 Altimeter specifications

The Seasat altimeter has already been mentioned as a typical example of the
present state of the art. In fact there have so far been only three altimeters flown
on civil satellites for scientific use, one on Skylab in 1973, one on the GEOS-3
satellite in 1975 and Seasat in 1978. The characteristic parameters of each are
hsted in Table 9.1.
Table 9.1 Specifications of satellite altimeters
Satellite Skylab GEOS-3 Seasat
Period of operation May 1973 ->■ 14 Apr. 1975 -> 7 July ^
Feb.1974 1 Dec. 1978 9 Oct. 1978
Operating frequency, GHz 13.9 13.9 13.5
Bandwidth, MHz 100 80 320
Pulse compression 13 30 1000
Averaging time, ms 300 200 1000
Range resolution, m 1.0 0.5 0.1
Height of satellite, km 435 845 800
Ground resolution, km 3.6 (at high 2.4 calm sea
resolution) 12 high sea
Orbital period, min 93.1 100.6 100.8
Orbit inclination, deg 50 115 108
Orbit repeat period, days variable variable 3,17, variable
Wave-height prediction
accuracy, % (height range) ±25% (4^ 10 m) ±10% (1^20 m)
Sec. 9.3] Establishing a datum 287

9.3 ESTABLISHING A DATUM

9.3.1 Orbit determination
Having examined how the distance between satellite and sea surface is measured
and corrected for various sources of errors, we now consider how to turn that
measurement into a statement about the ocean surface height relative to the
geoid. The first stage is to determine the ocean height relative to the earth’s
centre or the reference ellipsoid by determining the satellite’s exact position and
height at the time when a given distance measurement was obtained. This is
known as creating an ephemeris of the satellite orbit.
The satellite position is obtained by computer modelling of the orbit, tied in
to ground observations of the satellite’s orbit as it passes close to laser-tracking
stations. The orbit is dependent on the dynamic forces which act on the satellite.
These must be known accurately to achieve accurate orbit determination. The
principal forces which cause deviation of orbit from the pure ellipse discussed in
Chapter 2 are the variations in the gravitational attraction of the earth, moon,
and sun (including tidal forces and the gravitational attraction of earth and
ocean tides), the drag of the atmosphere, and the pressure of direct and earth-
reflected solar radiation. The gravitational variability arises from the spatial
inhomogeneity of mass distribution over the earth, which is reflected in the geoid
shape. Local gravity also varies temporally with the tide-generating potential,
and because the earth and ocean mass is continually being redistributed in
response to tidal forces. The drag and pressure forces depend on the shape, size,
and the orientation of the satellite, and can vary in space and time.
All of these effects contribute sizeable deviations from the nominal orbit,
the major deviation being up to 10 km due to gravitational effects. Errors in the
ultimate sea-surface height calculation arise from inaccuracies in the modelling
of these deviations. Given an incomplete knowledge of the forces on the satellite,
the orbit error must be kept to a minimum by having as many ground-tracking
stations as possible with globally-dispersed coverage. Even then it is not possible
to model the orbit with complete accuracy in between ground checks.
Moreover, the exact altitude and other positional coordinates of the
tracking station itself must be known with great precision if further altimetric
errors are not to be introduced, and this raises the problem that there is not yet
one single geodetic reference standard to which all astronomical and geodetic
observations from different places across the earth can be related. In the future
the use of a global positioning system (GPS) based on position fixing relative to
a series of satellites with precisely determined orbits will help to overcome many
of the problems, and may provide a sufficiently accurate means for fixing the
position of the altimeter satellite itself without the need for laser ranging.
Another source of error relates to the accuracy of the time lag attributed
to each satellite observation. If this is in error, then the measured height is
attributed to a different part of the orbit. If the orbit is sloping relative to the
reference ellipsoid, an altitude error results.
288 Satellite altimetry of sea-surface topography [Ch.9

Given these various sources of error it is remarkable that the Seasat orbit
determinationf resulted in errors of only 1 to 2 m, but this still makes the orbit
tracking a principal source of error for satellite altimetry. One way of objectively
determining the error in orbit calculations is to compare the altimeter measure¬
ments at points where the ground tracks cross on different orbits. The experience
of Seasat has led to an expectation that it will be possible in future to reduce
the radial orbit error to 10 cm provided that an improved geopotential model
is developed (which will require further measurement of the earth’s gravity
distribution by satellites), provided that the satellite is dense and/or in a higher
orbit than Seasat to minimise drag and pressure forces, and provided that the
computation methodology is refined.
It should be noted that the present error level of around 1.5 m applies to
the whole-earth coverage of all orbits. Because the error varies over length scales
of order 10000 km or more along the track, then its effect can be minimised
to effectively less than 10 cm over arc lengths of 1000 km or less. This can
be achieved either by looking only for relative height differences, or by fitting a
Unear error trend between two ground-control points spaced close together,
where this is possible.

Table 9.2 Seasat altimeter altitude error budget

Amplitude of Residual error

phenomena after modelling Wavelength
Type of error Source of error (cm) (cm (1 a)) (km)

1 Noise 5.0
Altimeter
\ Bias 10 2.0
Sea state Waveheight and 7 ± 2% of /2i/3 2% of hy^ 500-1000
related bias tracker biases
Troposphere Mass of air 240 0.7 1000
Troposphere Water vapour 10-40 3.0 50-500
Ionosphere Free electrons 2-20 3.0 50-10000
Liquid water Clouds, rain 10-100 30-50

1 ' Gravity
Drag
10 km
300
140
30.0
40000
10000
Orbital error <
Solar radiation
1 Station location
300
100
30.0
10.0
10000
10 000
Timing Data time tag 5.0 20000

f See ‘SEASAT ephemeris analysis’, special issue of J. Astronaut. Sci. 27 (no. 4), 1980.
Sec. 9.3] Establishing a datum 289

9.3.2 Summary of height errors relative to the reference ellq)soid

So far we have looked at errors arising from the methodology of altimeter
measurements from satellites. Since the geoid errors to be discussed next arise
from attempts to interpret the height measurement in an oceanographically
useful way, it is appropriate to summarise the measurement errors at this point.
Table 9.2 shows the error budget for the Seasat altimeter (Tapley et al. (1982a))
which represents the current state of the art until the next satellite altimeter is
launched.
Two things are worth noting. The first is the variable wavelength of the
different sources of error, that is, the typical length scale along the orbit ground
track over which the error changes from maximum positive to negative and back
to positive again. As we saw for the orbital error, if this is very long, the error
can be reduced in its effect over a much shorter length scale. The second is the
fact that the resulting error is not directly related to the size of the different
corrections which must be made, but depends on how well the physics of the
effect can be modelled. Thus the ionospheric correction of between 2 and
20 cm has a residual error of 3 cm, while the dry-air correction of 240 cm has
a residual error of only 0.7 cm.

9.3.3 The geoid

To obtain ocean dynamics information from sea surface-height measurements, it
is necessary to know the surface height and slope relative to the geoid rather
than the reference ellipsoid. This requires a detailed knowledge of the geoid
shape. The deviations of the geoid from the reference ellipsoid range from
— 104m to -I-64 m. Spatial variability can be large. In areas where there are
ocean trenches or ridges, for example, the geoid height can vary by several
metres over a few kilometres. In contrast, the variation of sea-surface elevation
relative to the geoid is typically ±1.5 m. To resolve ocean-surface topography it
is therefore necessary to know the geoid shape to around 10 cm accuracy over
length scales the size of the ocean length scales being studied.
At present this is not possible. The geoid is measured by three techniques.
The longer wavelength components, varying on length scales over 2000 km or
more, have been determined from the accurate tracking of the orbits of many
different satellites over a long period (see King-Hele (1976)). The geoid can also
be computed from measurements of gravity on land or from ships. This can give
high spatial detail in small areas, but is less able to give larger-scale variability
over large areas. The third technique is to use satellite altimetry, and this at
present provides the best global-wide observations of the geoid. Data from
GEOS-3 and Seasat can probably map the geoid with an accuracy of about
1 m and a resolution of around 200 km, but the error in this estimate is due to
uncertainty in the sea-surface height. Thus we have a situation where the best
geoid estimate is not independent of the variable we are trying to extract from
290 SateUite altimetry of sea-surface topography [Ch.9

the altimeter data, and therefore cannot be used for its correction. Consequently
it is not surprising that most interest has been shown in Seasat altimeter data not
by dynamical oceanographers but by geodesists, determining the marine geoid
(Brown etal. (1983)).
If tidal corrections are applied, the geoid does not change with time,
and it is therefore possible with repeat orbits to detect time variability in sea-
surface topography from the differences between different overpasses of the
same transect of ocean. At present, however, there is no way of confidently
distinguishing the residual long-period average sea-surface topography from the
spatial variability of the geoid. The only prospect of being able to achieve an
independent measure of the geoid to an accuracy of a few centimetres over
length scales of a few hundred kilometres is by using a dedicated gravitational
satellite system in which the distance between two satellites in low orbit would
be accurately measured (Douglas et al. (1980)). Without this, future successors
to the Seasat altimeter will continue to provide information only on time-varying
aspects of ocean dynamics.

9.4 APPLICATION OF ALTIMETRY TO THE STUDY OF OCEAN

CURRENTS
9.4.1 Geostrophic balance
Many dynamical processes in the ocean with a timescale of hours or longer can
be considered to be in geostrophic balance, in which the pressure forces driving
the flow are balanced predominantly by the Coriolis force. This is the inertial
force due to the absolute acceleration which occurs as a body moves relative to
a frame of reference which is itself rotating. In the ocean context, with x and
y coordinates in the east and north directions respectively, fixed relative to
the rotating earth, and the z coordinate vertically upwards (parallel to the local
gravitational force), geostrophic balance is expressed by (see, for example. Pond
& Pickard (1978));

fv = - — (9.6)
p ox

fu =-. (9.7)
P
These are the east and north pressure balances respectively. Here p is the local
fluid pressure, u and v are the horizontal velocities in the x and y directions, p
is the sea-water density, and / is the local Coriolis parameter, defined as

/ = 2^2sin0

where 0 is the latitude, and = 7.272 X 10"^ rad s“^ is the earth rotation rate.
Sec. 9.4] Application of altimetry to the study of ocean currents 291

Equations (9.6) and (9.7) express the physical reality that water flowing
horizontally experiences a horizontal force perpendicular to the direction of
flow, tending to move it to the right of its direction of flow in the northern
hemisphere, and to the left in the southern hemisphere (where / is negative).
For currents at the sea surface, the horizontal pressure gradient is propor¬
tional to the sea-surface slope measured relative to the equipotential surface
(that is, the geoid). If the sea-surface height is f relative to the geoid, (9.6) and
(9.7) become

fv = g — (9.8)
bx

fu = -g — . (9.9)
by
Equations (9.8) and (9.9) show how sea-surface topography measured from a
satellite altimeter can be directly related to ocean currents. Simple inspection
will show that a current of magnitude 1 m s”^ at 43°N (where / = 10"'* s"‘)
will have an associated sea-surface slope of about 10“^, or 1 cm in 1 km. Ocean
circulation velocities are normally of the order of a few cm s"\ with a slope of
less than 1 mm in 1 km, or 10 cm in 100 km. This is the reason why such a
high-accuracy geoid is required for satellite altimetry. Clearly, these typical
geostrophic slopes are much smaller than some geoid slopes.
Nevertheless, their observation is of great importance to oceanographers, for
the following reason. Until now, apart from a few costly deep-sea current-meter
moorings, oceanographers have been unable directly to measure ocean currents
far away from land with any absolute accuracy, since they cannot provide a
stationary platform to which the measured speed can be related. They have,
however, been able to measure the vertical distribution of density in the ocean
from temperature and salinity soundings. This mass distribution is related to the
pressure field by the hydrostatic balance equation
bp
^ = -pg. (9.10)
bz
If the pressure field dictated by the mass distribution results in the horizontal
pressure gradient being different at different depths, then from (9.6) and (9.7)
the horizontal geostrophic velocity must vary with depth. This is expresed by
the ‘thermal wind equations’ (named from the meteorological analogue), one
form of which is:

_^(pv) _
(9.11)
bz f

^(pu) _ £^ (9.12)
bz f by
292 Satellite altimetry of sea-surface topography [Ch.9

Thus oceanographers have been able to determine the vertical shear of

horizontal currents, given the vertical and horizontal distribution of density.
They have not been able to define absolute currents without postulating a‘level
of no motion’ at which u = v = Q. If the satellite altimeter can provide the
surface currents, then integration of (9.11) and (9.12) (or their counterpart in
pressure coordinates — see Pond & Pickard (1978)) yields the complete absolute
velocity field throughout the ocean wherever density measurements have been
made. It must be remembered, of course, that only geostrophically-balanced
currents are capable of computation from surface slopes. Ageostrophic flows
cannot.

9.4.2 The possibilities for ocean-circulation determination

The general characteristics of ocean circulation have been presented in section
3.3.1. We have already noted that until an independent geoid measurement is
made, the mean sea-surface topography (and hence the associated mean surface
circulation) cannot be determined. Only when it is will we be able to discern
the variety of length scales over which the spatial variability of the mean circula¬
tion occurs. However, given our prior general knowledge about certain mean-
circulation features, it may be possible to obtain some information from the
present state of altimetry. Thus, for example, we know that the major ocean
near-surface gyres tend to have their poleward flow concentrated in a narrow
energetic western boundary current. This has a length scale of typically 100 km
across, and it is feasible to make accurate gravity measurements by ship over
such length scales in order to determine the local geoid shape. Hence the surface
slope across the western boundary currents could be obtained in order to deter¬
mine the poleward mass fluxes, and thus the circulation rate around the whole
gyre could be estimated.
The measurement of the time-variable ocean-circulation is much more
readily achieved provided that a repeat cycle is maintained in the orbit to enable
geoid and mean-circulation effects to be eliminated. Fig. 3.3 shows qualitatively
the distribution of circulation energy across different length and time scales in
the ocean. It appears that length scales less than 30 km and time scales less than
a few days are not of great interest, suggesting minimum requirements for orbit
spacing and repeat period. The maximum period of motion that can be resolved
depends on the lifetime of the satellite. The failure of Seasat about 95 days after
altimeter measurements commenced prevented the annual current variability
from being detected, but this should easily be within the grasp of future
satellites. The longer length scales, at about 1000 km, are likely to be hidden by
orbit-error noise unless they have a strong surface-slope signature, or unless a
gravitational satellite programme is able to provide a better geopotential model
for improved global-scale orbit determination. That still leaves much of the
interesting mesoscale eddy fields and boundary-current meandering processes
Sec. 9.4] Application of altimetry to the study of ocean currents 293

capable of observation by the altimeter, provided that it has sufficient height

resolution. Most of the ocean-dynamics applications to which the Seasat data
have been put fall into the spectral window of mesoscale processes.

9.4.3 Examples of ocean circulation phenomena observed with Seasat

altimeter data
For the last 25 days of its useful life, Seasat was flying on a 3-day and 13-minute
repeat orbit. This meant that eight repeat overpasses of the whole ground track
were obtained, and several interesting parts of this dataset have been analysed
to reveal ocean-current variability. The most evident signatures of variability are
for areas where the mean flow is strong, resulting in meandering or the shedding
of eddies which are correspondingly energetic. Cheney (1982) has examined
part of the Gulf Stream, whilst Bernstein et al. (1982) looked at the Kuroshio
current, the western boundary flow past the Pacific coast of Japan.
Figs 9.4 and 9.5 show the results from two segments of orbit over the
Gulf of Mexico (Thompson et al. (1983)). Fig. 9.4 runs from S.W. to N.E. left
to right, whilst Fig. 9.5 runs N.W. to S.E. (The tracks are shown in Fig. 9.6).
In each case (a) is the sea height for all eight overpasses after most of the cor¬
rections listed in Table 9.2 have been applied. It is therefore relative to the
reference ellipsoid and contains the geoid effect. To remove the geoid, the mean
of the eight is subtracted from each, yielding (b), which eliminates a 20 m signal,
which is probably geoidal but may contain a small mean circulation signature.
The 2 m variability in (b) contains some remaining orbit error which is removed
to form the bottom part of (c), in which the eight overpasses have the same
mean height, but show considerable (time) variability from each other. When
these passes are displayed in time sequence (the upper part of (c)) it is evident
that the variability is not random but reflects a definite dynamical process. This
is particularly apparent in Fig. 9.4, where a high region is seen to move gradually
north-eastwards (left to right along the transect). Fig. 9.6 illustrates a general
interpretation of the sea-surface topography in the region, as being due to a
strong current flowing north round the west of Cuba and on to form the Florida
Current, whilst a branch breaks off to form a Loop Current which meanders into
the main part of the Mexican Gulf. The transect of Fig. 9.4 indicates the way in
which the main current pattern shifts north-east during the 24 days. Fig. 9.5
is harder to interpret, but the shaded high probably denotes the centre of the
Loop Current which drifts to east or west to be replaced by a low.
This example illustrates the potential of satellite altimetry to make observa¬
tions which were hitherto not possible. At the same time it shows how sparse
was the spatial sampling of the 3-day repeat orbit. To resolve properly the
length scales of the loops and eddies in this case would require at least three
times more tracks to fill in the gaps. As well as explicitly revealing the move¬
ment of the highs and lows, the data also yield the spatial distribution of the
time variability of surface elevation, as indicated in (d) of Figs 9.4 and 9.5.
 -10.

294 [Ch.
-15.

S 0.0
QJ
or

2.5

£
2.0
C3
LU

I >-5
Li_J
CXI

^ l.o
cr
CD

0.0

longitude

Fig. 9.4 - Analysis of eight collinear altimeter profiles collected along track 2
of Fig. 9.6, by Seasat over the Gulf of Mexico, between September 17 and
October 8, 1978.
(a) Sea height relative to the reference ellipsoid.
(b) Profiles replotted after subtraction of the group mean.
(c) Profiles after removal of orbit error through tilt and bias adjustment, overlaid
(lower), and displaced vertically at 3-day intervals (upper).
(d) Root-mean-square variability about the mean.
(From Thompson, Born, & Maul (1983) published by the American Geophysical
Union)
Sec. 9.4] Application of altimetry to the study of ocean currents 295

SEA HEIGHT (M!

MEAN REMOVED (M)
TILI/BIAS REMOVED (M)
RMS VARIABILITY (CM)

LONGITUDE

Fig. 9.5 - As for Fig. 9.4, but for track 3 on Fig. 9.6. (From Thompson, Born, &
Maul (1983) published by the American Geophysical Union)
296 Satellite altimetry of sea-surface topogrq)hy [Ch.9

30*N

25*N

20*N

96*W 90*W 85*W 80*W

Fig. 9.6 - Tracks of Seasat over the Gulf of Mexico, showing the location (tracks
2 and 3) of the altimeter transects illustrated in Figs 9.4 and 9.5. Also shown is a
typical (but not permanent) current pattern in the Gulf, in which an anticyclonic
eddy is about to separate from the Loop Current. H and L correspond to raising
and lowering of the sea surface and subsurface isothermal surfaces in geostrophic
balance with the current system. (Adapted from Thompson, Born & Maul (1983))

Menard (1983) has developed the techniques for mapping the spatial
distribution of the time variability of sea-surface topography, based on the three-
day repeat orbits. The deviation of the surface height from the mean of the
8 or 9 orbits are calculated at each 0.1° of latitude along the orbit track to
obtain the surface deviation for each (ith) pass at each location. Fig. 9.7
shows a map of surface-height variability in the North Atlantic, drawn by hand
from the values of a^(Af') at each calculation point, where the variance is:

aMAf'l =
n
Menard also calculated the variable velocity field Av„ in the direction n, normal
to the satellite track s, with the geostrophic relation:
 Sec. 9.4]
Application of altimetry to the study of ocean currents

Fig. 9.7 — Sketch of sea-surface topography variability in the N.W. Atlantic, constructed from the Seasat altimeter 3-day repeat data along the
track lines shown. The interpretation of the data is subjective away from the track lines. (From Menard (1983), copyrighted by the American
297

Geophysical Union)
298 Satellite altimetry of sea-surface topography [Ch. 9

the American Geophysical Union)

-70
 m o 3
3 O
X G
a> O
00
r-
H-* -
v-H lya
^
«

(U

O
•M
00
o G d^
O
-i ^
o “ J3
3:
o
s
. ^
<U a>
tG > •O
G
B ^
dj ^
•M (1> 00
a c C
D
C/3 B °
*0
% »o O
d> CO
r*

G
(D

(From Cheney, Marsh, & Beckley (1983))

O d>
^ 5
'S g •
^ >. '
»H ,ti '
(D ;g
'G
d) ^
E
o e ’S'
:G co
co ^
^ *0
CO a> '
tn >*-1 .
CO 3 «
1> o '
c/5 rG
i-i

^1'
«-7
>v >>

73 ^
E S O
o
*00
g ^
CO o cO
o a> o >
Bo^
. O 0)
.■az:5
o
B o
■^ o _ S
l-t IH V3
CO
> .B
cO
<w
G>
.2P O
*S
4=
cO
<u
V)

'B
CJ c
o
a 2^
<u CO Xi
B B ■=«
CO
"eo
Xi - s
.S
--S '-G
O o G
^ 2
>> ^
ON o 00
ON CO *Q
.s
.Op Cu.
d> IH
I-I 00

O O o O O o
(O CM CM (O
I t
300 Satellite altimetry of sea-surface topography [Ch.9

From this the velocity variance

was calculated.
n
On the assumption of an isotropic turbulent eddy field, that is,

o\Av„) = o\Av,)

the eddy kinetic energy E = 14[a^(Av„)-I- a^(Avs)] = o^(Av„) was evaluated

and plotted in Fig. 9.8.
As expected, it shows a maximum value where the Gulf Stream meanders
and sheds eddies. Although the spatial sampling was rather sparse in this
example, and the frequency spectrum of the eddy field was limited to a window
of between 6 and 25 days period, it illustrates the type of observation which
could become commonplace from future satellite altimeters. Cheney, Marsh, &
Beckley (1983) have produced similar maps of surface-height variability over
the global scale. Fig. 9.9 indicates that, as in the Gulf Stream of the North
Atlantic, mesoscale variability is associated with regions of strong currents, such
as the Kuroshio system in the western Pacific Ocean, which supply the necessary
input of kinetic energy. In the southern hemisphere the Agulhas current off
the south-east of South Africa, and the confluence of the Falkland and Brazil
currents off eastern South America are also regions of increased surface-height
variability. The strongest signal is associated with the Antarctic Circumpolar
Current around the southern ocean. In contrast, the equatorial currents give rise
to a much weaker signal, whilst in the centre of the large ocean gyres, away from
eddy shedding by the boundary currents, the variability is very small indeed.

9.5 TIDAL OBSERVATIONS WITH THE ALTIMETER

9.5.1 The value of satellite altimetry to tidal studies
The principal features of ocean tides have already been discussed in section
3.3.2. Tides are studied by altimetry to determine the tidal elevation for its own
sake rather than in order to calculate tidal currents from the surface slope. In
fact, tides are not in geostrophic balance, although the Coriolis force influences
their propagation, because they are forced at frequencies comparable with the
inertial period (/”^). Thus geostrophic and inertial forces are of comparable
magnitude in tidal dynamics. It is not possible to obtain tidal currents simply
from tidal surface slopes, as is the case with geostrophic ocean currents. The
interest in using altimetry to study tides is that it provides an opportunity
to determine the spatial distribution of tidal elevation over the whole ocean.
If desired, the barotropic (that is, depth-averaged) tidal streams could be
determined from the elevation using mass-continuity considerations.
Sec. 9.5] Tidal observations with the altimeter 301

Satellite altimetry is not the only way to gain an absolute measurement of

offshore tidal elevations, since for over a decade it has been possible to deploy
deep-sea tide gauges on the sea bed to measure tidal rise and fall using pressure
sensors (see, for example, Cartwright et al. (1980)). This is, however, a slow and
costly business. Recent advances in numerical solutions of the equations of tidal
dynamics mean that it is now possible to model the global ocean tides to a
reasonable degree of confidence, provided that the models are constrained to fit
a network of observed tidal elevations. In the context of testing models against
other independent observations a satellite altimeter offers the opportunity to
obtain systematic coverage of the tides of the ocean.
The typical length scale of tidal variability in the deep oceans is over
1000 km. In shallow shelf seas it is less than 100 km, and here the use of
satellites could be extremely valuable in providing sufficient spatial coverage (at
least along orbit tracks) to resolve the tidal-elevation patterns. Shallow seas are
of particular interest to tidal scientists since it is here that most tidal-energy
dissipation occurs. The tidal range grows from a typical oceanic value of less
than one metre to ten or more in some shelf seas, and the tidal streams are
amplified even more. The global tidal-energy dissipation problem will not be
properly solved until we can determine the tidal-energy fluxes in remote areas
like the Patagonian Shelf and the Bering Sea. It is not reasonable to make
estimates based on extrapolating energy-dissipation observations from areas such
as the U.K. shelf seas and the Gulf of Maine/Bay of Fundy which have been
investigated using tide gauges and current meters. The satellite altimeter appears
to be the only realistic technique for providing the necessary global access to
tidal data in remote locations.
Another reason for wanting to know the global distribution of deep-ocean
and shelf tides is to be able to estimate the coastal-loading effect of the weight
of the water. In conjunction with measurements of the solid-earth tidal response,
the modelling of earth tides can yield geophysical information about the crustal
structure. Satellite altimetry will supply regularly-spaced, global-coverage tide
data which are well suited to this application.
The other major reason for determining the global tidal distribution with an
altimeter is to be able to apply predictable tidal corrections to satellite altimetric
observations of the geoid and/or the geostrophic sea surface topography.

9.5.2 Extraction of tidal information from altimeter data

Because of its regular harmonic time variability it is in principle possible to
distinguish tidal data from mean or other time-variable phenomena, provided
that neither the other processes nor the satellite sampling strategy cause aliasing
of the tidal frequencies. Aliasing occurs when a signal is sampled at a regular
interval longer than half its period. Since the important tidal harmonics which
account for most of the tidal amplitude are in frequency bands of one or two
302 Satellite altimetry of sea-surface topography [Ch.9

cycles per day (diurnal and semidiurnal tides), and the satellite in a repeat orbit
has a regular return time of a few days, some aliasing is inevitable. Fig. 9.10
illustrates the effect of aliasing. The observed time series appears to have a
much longer period than the signal being sampled. If the sampling interval is an
exact multiple of the signal period, the signal is aliased to zero frequency, and
no information about the tide at that frequency can be extracted. Thus a sun-
synchronous orbit which repeated its track exactly over an exact multiple of
24 hours would alias the Solar S2 and Si constituents to zero frequency, and the
lunar M2 tide would appear as a 14-day period constituent.

Fig. 9.10 — Aliasing of an oscillating signal by low-frequency sampling

It is important that major tidal constituents are not aliased to zero

frequency or to such low frequencies that they cannot be resolved in less than
half a year. This is not only from the point of view of the tidal scientist who
wishes to use the altimeter to study tides, but also so that the tidal signal will
not corrupt the long-period ocean-circulation topography. The usual way to
correct for the tides when studying the surface topographic changes along a
repeated orbit arc (see section 9.4.3) is at each point to remove any part of the
time series which is coherent with the known tide-generating potential. If several
different tidal frequencies are aliased to the same frequency, provided that it
is shorter than the record length, this presents no problems to making tidal
corrections.
However, if the tides are being studied in their own right, then not only
must aliasing to long periods be avoided, but so must the aliasing of several
constituents into the same frequency, because this would prevent them from
being separated. A sun-synchronous orbit would alias the T2, R2, Pi, and Kj
constituents all to a frequency of one cycle per year. It is the precession rate
(dn/df) of the orbital plane (see section 2.1.2) which controls the repeat
period. The choice of dJ2/dt will be examined further in section 9.7, but it is
obvious that dJ2/dt = 360° per year is to be avoided.
Sec. 9.5] Tidal observations with the altimeter 303

As for ocean-eddy observation, it is important that tides be examined with

a repeat track orbit, in order to remove geoidal and mean-circulation effects, and
fundamentally in order to sample time series at the same points. This implies
that only limited spatial sampling can be achieved. It is desirable in shallow
seas to be able to sample with spatial separations between orbits of not more
than a few hundred kilometres, in order to avoid spatial aliasing of the standing-
wave-like cotidal patterns of amphidromic systems. In the deep ocean, the tidal
wavelengths are of order 1000 km, and spatial aliasing is less of a problem.
Spatial aliasing is also not such an acute problem as time aliasing, because of the
resolution achieved along each orbit track, and because the tracks in a repeat
cycle intersect one another (see Figs 9.13, 9.14). Indeed, if the spatially-varying
altimeter tidal observations are analysed in conjunction with a tidal model, the
differing spatial distribution of different constituents may enable them to be
separated even if they are aliased to the same frequency.

9.5.3 Examples of altimeter applications to ocean-tidal science

Since useful tidal measurement from satellites is dependent on having a repeat
orbit, only the last 25 days of Seasat’s altimeter data have been seriously studied
for tidal purposes. This is not really a long enough record length to perform
accurate tidal analysis, particularly since there were only eight passes over the
repeat orbit, and hence only 16 or 17 data values for each track intersection
point. Cartwright & Alcock (1981, 1983) devised a method of analysing the
tidal elevations at each orbit-intersection point, not in isolation but relative to
the adjacent intersection points. Essentially, the method found the amplitude
and phase of the tidal function at each point necessary to account for the
differences in elevation observed between different intersection points along
each orbit path at different times. These were fitted in a least-squares sense,
together with a time-invariant constant to allow for geoidal variations, under the
added constraint that the tidal and geoidal differences would sum to zero around
closed loops in the network of lines joining intersection points. Consequently
the tidal constituents at each intersection were obtained relative to all the
others. By fixing an amplitude and phase at one intersection point close to a
tide-gauge station, absolute values could be assigned to all the other points in
the network. The result for the Mj constituent is shown in Fig. 9.11, where com¬
parison is made with a cotidal map based on deep-sea tide-gauge observations.
With such a short and sparse dataset the agreement is remarkably good, although
there is scope for much improvement with a larger dataset.
Cartwright & Alcock made full use of the areal coverage of the satellite in
conjunction with the spatial coherency of tides. A simpler approach was taken
by Brown (1983) who analysed the data from a single location, Cobb Seamount
in the North-East Pacific, to extract the M2 tidal signal. Comparisons with tide-
gauge data from the Seamount demonstrate that he was able to obtain very good
304 Satellite altimetry of sea-surface topography [Ch.9

phase agreement, to within one degree, but less satisfactory amplitude agreement
(about 47% low). This is probably typical of the expected performance of
sateUite altimeters for tidal observations. It will be easier to obtain more accurate
phase than amplitude data. The Cartwright & Alcock results also bear this out.
0.5° GRID, 1 M CONTOUR
Ae = 6378137m, 1/f= 298.257
REF. NO. 83402 Sec. 9.5]

played in Marsh & Martin (1982))

306 Satellite altimetry of sea-surface topography [Ch.9

9.6 OCEAN BATHYMETRY FROM SATELLITE ALTIMETERS

So far we have been looking for applications of altimeter data in dynamical
oceanography. One application which arises from examination of the improved
geoidal data is the possibility of determining ocean-bathymetric features from
the geoid shape. This is shown in Fig. 9.12, which contours the mean sea surface
(relative to the reference ellipsoid) from the Seasat altimeter (Marsh & Martin
(1982)), which is essentially geoid information. All the major ocean-bathymetric
features are apparent, including the major trenches and the mid-ocean ridges.
Since gravity increases over a greater crustal mass, the geoid tends to hump up
over the ridges and dip down over troughs directly reflecting the bathymetry.
This is particularly so for bathymetric features with a wavelength between
40 km and 400 km (see, for example, McKenzie & Bowin (1976)). At shorter
and longer wavelengths the effect becomes too small to detect above other noise.
This property of the altimeter geoid has already been used in an attempt
to discover whether previously-undetected bathymetric features may exist in
the ocean. White et al. (1983) have developed an automatic technique to scan
the geoid data to detect the signatures of small seamounts, which might not
be apparent in the noise without the application of matched—filter techniques.
Dixon et al. (1983) describe a technique of searching for bathymetric anomalies
in poorly-surveyed oceanic regions.

9.7 THE SELECTION OF SATELLITE ORBITS FOR ALTIMETERS

Of all the sensors carried on satellites, the altimeter is probably the one which
is most dependent upon its orbit to be capable of successful calibration and
interpretation. Considerable attention has been paid to determining the most
appropriate orbit for a dedicated ocean-altimetry experiment (see TOPEX
(1981)). For ocean applications a repeat orbit is essential, although that may
not be the case if the altimeter is to be used principally for geodetic and gravita¬
tional applications. The most suitable height must be decided, as must the earth
coverage which can be achieved. Given a repeat orbit, the best return period
must be determined, bearing in mind that more frequent temporal sampling
corresponds to sparser spatial coverage. To satisfy these sometimes conflicting
demands, the space engineer has to choose suitable basic orbit parameters
(see section 2.1.2). The orbit parameters influence several of the above design
parameters in interrelated ways, and there are a number of possible parameter
permutations.
Let us start by considering the height of the satellite. There is every
advantage in having as near-circular an orbit as possible, so that the altimeter
is always at the same nominal height with the same footprint size. Low orbits
(Seasat at 800 km was considered to be relatively low) have certain advantages.
The power of the return signal from the circular illuminated footprint just
Sec. 9.7] The selection of satellite orbits for altimeters 307

before it spreads to become an annulus is inversely proportional to the cube of

the height. Thus a lower orbit requires much smaller antennae and less power
for the same signal-to-noise performance as a higher orbit. The footprint of
a lower orbit is also smaller, giving higher spatial precision. A lower orbit is
faster, and gives better spatial/temporal coverage, although this is a much weaker
function of the height. Finally, a lower orbit enables an accompanying micro-
wave radiometer (virtually essential for the atmospheric water correction) to
observe a relatively narrow column of atmosphere despite its wide field of view.
In contrast, a high orbit has a much-reduced atmospheric drag. Above
1300 km this is small enough for the drag error in the orbit calculation to be
less than 3 cm, and the orbit decays more slowly, requiring much fewer orbit
adjustments to maintain a selected ground track. A higher orbit enables a
ground station to track the satellite for much longer during its orbit than a lower
one, and enables the ground station to intercept more overpasses. For a lower
orbit some of the same overpasses would be too close to or over the horizon.
Thus the advantages of height are principally related to the reduction of
satellite orbit error. Since this was the principal limitation with Seasat, any
future dedicated altimetry satellite designed for greater accuracy is likely to be
in a higher orbit than Seasat. For the proposed TOPEX satellite, 1300 km is
recommended.
Once the height is fixed, so is the period of the satellite, but very small
alterations in the height can change the period subtly so that the repeat cycle is
changed. To obtain an exact repeat cycle in approximately D days there must
be an exact number of orbits in exactly D earth revolutions relative to the orbit
plane of the satellite. The actual repeat period depends on the orbit-precession
rate (for example, if the orbit plane were not precessing the repeat period would
be D sidereal days, and if the orbit precession were sun-synchronous it would be
D solar days). A very small change of orbit period can make a big change in
the repeat cycle (Cutting et al. (1978)), enabling repeat intervals to be signi¬
ficantly changed by small course alterations during'the lifetime of the satellite,
as occurred with Seasat.
The choice of repeat period also controls the spatial-sampling interval. Thus
the 10-day repeat cycle proposed for TOPEX gives a 316 km track separation at
the equator, for a 65° inclined orbit, whereas the 3-day repeat cycle of Seasat’s
72° orbit gave a track separation of about 800 km. Figs 9.13 and 9.14 compare
those orbits over the North Atlantic. The final choice rests on the envisaged
apphcations, but the 10-day cycle appears to be a better compromise than
the 3-day if reasonable spatial coverage is to be achieved, although it reduces
the resolvable frequency of variability to a period of 20 days compared with
6 days for Seasat.
The inclination directly affects the high-latitude coverage. Since the
altimeter is of great value to icecap and polar studies (beyond the scope of this
book), there is a demand for a near-polar orbit, but since the inclination affects
308 Satellite altimetry of sea-surface topography [Ch.9

the precession rate, a purely oceanographic mission would be more constrained

by problems of tidal aliasing, and a 65° to 70° orbit which could cover most
of the ocean would be preferable. Another advantage of a low-latitude orbit is
that the angle between northward and southward passes is closer to 90 in
equatorial and mid-latitudes than it is for a more polar orbiting satellite.

Fig. 9.13 — Spatial distribution of the ground track over the N. Atlantic for a
10-day orbit repeat pattern.

Fig. 9.14 - Spatial distribution of the ground track over the N. Atlantic for a
3-day orbit repeat pattern
Sec. 9.7] The selection of satellite orbits for altimeters 309

Fig. 9.15 illustrates how the precession rate is affected by the inclination
and the height of the satellite. The constraints of atmospheric drag (at low
altitude) and lack of ocean coverage (at low inclination) are included, and the
dotted shading indicates the precession rates at which tidal aliasing is worst.
Two possible regions for an ocean-altimeter orbit are indicated. The one around
64 inclination gives less tidal aliasing problems than the one around 72°, but
poorer global coverage.

aliasing

Fig. 9.15 — Orbit inclinations plotted against the resulting precession rates for
different heights of orbit, showing the limits within which a suitable altimeter
orbit is found. The dots indicate that tidal aliasing is a problem. The numbered
regions represent the parameter combinations of promising orbits.

One possible way of satisfying many more applications scientists in oceano¬

graphy and polar research would be to fly two satellites. A dedicated altimetric
satellite in the 1300 km 65° orbit with a 10-day repeat cycle would give high-
accuracy altimetry at low- to mid-latitudes and an improved geoid model in
those latitudes. In addition, a polar-orbiting satellite at lower altitude, with
a long repeat cycle but dense spatial coverage, could use the first satellite to
provide orbit and geoid error correction and improve its performance. This may
well be a possibility within the next decade. Certainly the brief experience with
Seasat has whet the appetite of the oceanographic community for a dedicated
altimeter satellite in the future.
CHAPTER 10

Active microwave sensing

of sea-surface roughness

10.1 INTRODUCTION TO THE REMOTE SENSING OF SEA-SURFACE

ROUGHNESS
In the next four chapters (10 to 13) we focus attention on the remote sensing
of the shape and roughness of the sea surface at short length scales from a few
hundred metres down to millimetres. We have already met the influence of
surface waves and wind-generated roughness in their effect on the surface
reflection of visible light — sun glint and sky glint. Indeed much of the early
work on visible-wavelength satellite imagery sought to use this effect to deter¬
mine wind fields. Remote sensing by visible light is hindered by cloud cover,
and with the development of all-weather microwave instruments, the attention
of oceanographers wishing to study sea-surface roughness has largely switched
to radar techniques. As we have already seen in Chapters, passive microwave
radiometry is strongly influenced by the effect of surface roughness on the
emissivity of the sea surface, enabling wind speed to be recovered from multi¬
frequency sensors. However, it is the active radar instruments which are able to
remotely sense the most information about sea-surface roughness. It is a measure
of their success that the understanding of the physics of sea-surface roughness
phenomena, and of their relationship to other oceanographic parameters, has
become a subject of central importance for satellite oceanography.
Chapters 11 to 13 describe the theoretical principles, the instrument design,
and the oceanographic applications of those active microwave sensors which
depend on surface roughness for their operation — the altimeter in its wave¬
measuring role, the synthetic aperture radar, and the scatterometer. Much of
the recent scientific work published about these types of instrument relates to
Seasat, and there are already several useful collections of papers describing the
techniques and oceanographic application of active microwave satellite sensors,
to be found in Bernstein (1982c), Kirwan et al. (1983), and Allan (1983).
Before looking in detail at particular sensors and their applications, we turn
in this chapter to consider how active radar sensors interact with, and hence
observe, surface roughness. Next, we examine the nature of sea-surface rough¬
ness and the factors which control it. Finally, we look at some of the ways
[Sec. 10.2] Radar reflection from the sea surface 311

in which sea-surface dynamical processes such as internal waves, and artificial

causes such as the passage of ships, can influence the surface roughness, resulting
in a surface signature capable of being imaged by radar.

10.2 RADAR REFLECTION FROM THE SEA SURFACE

10.2.1 Backscatter cross - section
The radar backscatter cross-section, o, of a surface is defined as the ratio
between the radiance of radar energy reflected back towards the source and the
radiance incident on the surface. It is a function of viewing angle and frequency.
After due allowance for the spreading geometry of the radar beam, it is a which
is measured by an active microwave remote-sensing device, and therefore it is
the dependence of o on the environmental properties of the surface which
enables the microwave sensor to be used as a remote-sensing instrument.
The basic physics of radar scattering is presented in texts such as those of
Isimaru (1978) and Uslenghi (1978). Considerable progress has been made in
understanding the various processes of radar backscatter from rough surfaces,
and hence solving the ‘direct’ problem, that is, modelling the backscatter given
the surface properties. The solution of the ‘inverse’ problem, that is, deducing
the surface properties from the observed backscatter, is more difficult, but much
research effort has been devoted to it (see, for example, Boermer et al. (1981)).
The radar backscatter cross-section depends on a variety of factors. The
electrical properties of the surface material affect the penetration depth of
e.m. radiation. At the microwave frequencies used for active ocean remote
sensing, the depth at which the radio energy is reduced to 1/e of its surface value
is between 0.1 mm and 10 cm in water, which is much smaller than for dry land
surfaces. Because of the very shallow penetration it has largely been assumed
that backscatter occurs entirely at the sea surface, and o is related to the shape
of the sea-surface profile. Recently, however, the suggestion has been made
that bubbles of air just below the surface, and foam in the surface, may also
contribute to the backscatter, although there is need for further research to
pursue these ideas.

10.2.2 Specular reflection and scattering

Reflection at the surface is strongly dependent upon the viewing angle. For
near-nadir viewing (looking vertically down on to the surface) specular reflection
is the principal process for returning energy to the sensor. The magnitude of o is
proportional to the area of surface whose inclination is such as to directly reflect
energy back to the sensor. For a flat calm, the surface as a whole will reflect
energy back to a sensor emitting and viewing at nadir, but other incidence angles
will result in no reflection back to the sensor. As it is roughened, the surface will
present many facets reflecting in different directions, so that radar return to a
nadir sensor will be reduced. Conversely, off-nadir sensors will now receive some
312 Active microwave sensing of sea-surface roughness [Ch. 10

reflected energy, but since even the roughest sea is unlikely to have slopes tilted
at more than 20°-25° from the horizontal, specular reflection is important only
for viewing angles between 0 and about 15°.
For viewing angles greater than this, a return signal is only achieved by
scattering at the surface, and this depends on the surface roughness. Fig. 10.1
illustrates different degrees of surface roughness resulting in (a) purely specular
reflection, (b) some scattering but a dominant reflection in the specular direc-

(c)

Fig. 10.1 - Radio-wave scattering from (a) a smooth surface (specular reflection)
(b) an intermediate surface, and (c) a rough surface
Sec.10.2] Radar reflection from the sea surface 313

tion, and (c) complete scattering. The definition of roughness varies with radar
wavelength Xr and viewing angle 6. A smooth surface (a) can be considered to
exist when the vertical height of surface roughness irregularities is

hr Xr cos0/471, or hr < 1/25 Xr cos0. (10.1)

A rough surface (c) is approximately defined as:

hr > Xr cos0/47r, or hr > 1/4 Xr cos0. (10.2)

10.2.3 Resonant Bragg scattering

For a smooth surface, oblique viewing of the surface with active radar yields
virtually no return. If the surface is rough, significant backscatter occurs. In
the far field (many radar wavelengths away from the surface) the interference of
scattering from many different parts of the surface results in a reinforcement of
scattering from periodic structures in the surface roughness which have suitable
wavelengths, and destructive interference of all other reflections. This is known
as resonant Bragg scattering, and is illustrated in Fig. 10.2. The physical principle
is that of a diffraction grating. For radar waves of wavelength Xr, incident
at 6 to the vertical, viewing a train of sea waves of wavelength Xg whose crests
are perpendicular to the radar line of sight, first-order resonant Bragg scattering
occurs if

(10.3)
2 sm0

Fig. 10.2 — Bragg scattering geometry: radar-wave direction and surface-wave

direction in the same plane
314 Active microwave sensing of sea-surface roughness [Ch. 10

For sea waves whose crests are at an angle 0 to the radar line of sight (Fig. 10.3),
the criterion is
XRSin0
(10.4)
2 sin0

Fig 10.3 - Bragg scattering geometry; surface-wave direction at angle </> to plane
of radar waves

In the context of ocean remote sensing, Xr and 6 are fixed within narrow
bands, whilst the sea surface contains a whole spectrum of different wavelengths
as discussed in section 3.4. This results in the radar responding most strongly to
those waves for which
XRsin0

2sin0 ’
and a is therefore an approximate measure of the amplitude of such waves.
The choice of radar wavelength is seen to be crucial in determining the
sea-surface roughness which can be detected, both in controlling the height of
roughness which can be detected (through (10.2)) and also the ocean wave¬
length, if any, which will dominate the backscatter return (10.4). For Seasat,
the L-band SAR had Xr — 230 mm, 6 ~ 20°, and therefore would observe
as ‘rough’ any surface for which 40 mm. The scattering cross-section was
most strongly influenced by waves whose apparent wavelength along the line of
sight was about 300 mm.
Sec. 10.3] Wind-generated surface-wave roughness 315

In an imaging radar, there are several processes which can contribute to the
apparent spatial variation of backscatter cross-section. These include the layover
effect, when the surface is tilted along the line of sight, and the various modula¬
tions which occur because of the interaction of different scales of roughness.
The swell waves being imaged have much longer wavelengths than the short
gravity waves which cause the Bragg resonance, and there are various possible
ways in which the longer waves can modulate the Bragg scattering by the shorter
waves. These will be discussed in more detail in Chapter 12.
For obhque viewing of rough surfaces, shadowing will occur, in which
part of the surface is hidden from the view of the sensor by another part. This is
a further complication which may need to be incorporated into any model of
surface scattering which is constructed.
The whole subject of interaction between ocean and radar waves is reviewed
by Valenzuela (1978 b).

10.2.4 Polarisation
As well as the amplitude of the reflected radiation, other properties are also
changed by reflection, and potentially can be analysed to yield information
about the reflecting surface. Doppler frequency shift and phase information will
be considered further in the chapter on SAR (Chapter 12). Another property is
the polarisation of the radio waves. The transmitted signal is normally horizon¬
tally or vertically polarised. On a single reflection, the polarisation is not changed
very much, but if multiple reflections occur, the polarisation orientation can be
significantly altered. In this case, a radar system employing HV polarisation
(Horizontal polarisation of the emitted energy, and reception of a vertically
polarised return) would encounter a significant return. This occurs when viewing
land surfaces, and useful information can be gained to distinguish between, for
example, grasslands and forest.
Over the ocean, very Httle multiple reflection occurs, and ocean-viewing
sensors are normally HH or W. There is little to be gained by examining the
cross-polarisation signals, but the polarisation ratio P = CTvv/*^hh> the ratio
between scattering cross-section for vertically and horizontally polarised waves,
can contain oceanographic information. P is usually large, because vertically
polarised radar reflects more strongly than horizontally polarised, but P varies
with angle of incidence and radar frequency. It has been found to vary with
wind speed. Valenzuela (1978a) discusses further the possibilities of using
polarisation information for remote sensing purposes.

10.3 WIND-GENERATED SURFACE-WAVE ROUGHNESS

Given radar instruments operating within the frequency range 1 to 30 GHz, and
assuming that Bragg resonance is the principal backscattering mechanism, it
316 Active microwave sensing of sea-surface roughness [Ch. 10

follows that it is the part of the surface wavelength spectrum between about 5
and 500 mm which is important for the creation of radar surface roughness.
Reference to Figs 3.9, 3.11, and 3.12 reminds us that these are high-frequency
waves, with a period much less than 1 s, for which surface tension is usually an
important if not the dominant restoring force. They are also located in the high-
frequency tail of typical ocean-wave spectra, and the energy associated with
them is negligible in comparison with the rest of the energy in the main part of
the spectmm.
For this reason, they have been of very little interest to oceanographers
hitherto, except insofar as they influence the surface-roughness height in the
parameterisation of the turbulent atmosphere—ocean boundary layers. Typical
oceanographic wave-measuring instruments have been designed to measure the
main part of the spectrum, and are unable to resolve such high-frequency, short-
wavelength waves. The shape of this end of the spectrum is therefore not well
documented, and not very well understood. Indeed, most of the information
we have about high-frequency surface waves is based on radar measurements.
If the radar surface-wave interaction problem is to be objectively tackled, new
techniques of high-frequency wave measurement by independent methods, such
as lasers, must be developed.
Until such observations become available, it is being assumed that the high-
frequency tail of the spectrum is a continuation of the —5 power law which is
a very good model of the observed shape of the frequency spectrum up to a
frequency of 12 rads s~* (14-second period). In this region, the spectral energy
density is inversely proportional to the fifth power of frequency. Phillips (1977)
describes this as the saturation range of the spectrum, because it represents
a steady state even when the peak area of the spectrum at lower frequencies
is evolving in response to wind forcing. The assumption is made that once the
wind is stronger than a few cm s"*, sufficient to roughen the surface with high-
frequency ripples, the ripples reach a high enough amplitude for energy to be
lost from this part of the spectrum at the same rate as it is fed in by the wind.
The energy is lost by dissipation processes, and by nonlinear wave—wave inter¬
action it is cascaded to higher and lower frequencies. In this way the high-
freqency tail remains the same whilst the low-frequency peak of the spectrum
develops towards lower frequencies, depending on the strength of the wind.
The hypothesis that this theory holds into the Bragg resonant wavelength,
capillary-wave part of the spectum, still remains to be experimentally tested.
If it does, then it presents a problem for the principles of scatterometry which
assume, with empirical justification, that the radar backscatter increases propor¬
tionately with increased wind speed. If the saturation hypothesis holds, and
Bragg resonance is the only backscatter mechanism, it would be expected that
the radar backscatter would soon reach saturation like the wave energy in the
tail of the spectrum. That this does not happen may mean that mechanisms
other than Bragg scattering contribute to the radar backscatter cross-section. As
Sec. 10.4] Surface films and slicks 317

mentioned earlier, bubbles or surface foam are possible alternative or additional

scattering agents.
On the other hand, the problem may lie in the assumption that Bragg
resonance selects only the appropriate wavelengths at high wavenumber irrespec¬
tive of the longer wavelengths present due to the low-frequency spectral peak.
In other words, even though the amplitude of the high-frequency waves may not
change with increasing wind strength, the surface roughness in a Bragg resonant
sense may increase when those waves coexist with larger waves on whose crests
and troughs they rise and fall. Theoretical models of wave—wave interaction can
help us here to some extent, but we face another fundamental problem when¬
ever we wish to verify theoretical models with observational data. Even when
laser velocimeter techniques have been developed to measure the high-frequency
wave energy, they will still yield only spectral information about frequency.
What is needed is wavenumber data, that is, the spectral decomposition of the
horizontal spatial profile of the sea surface, in contrast to the time variation
of surface height at a point. Only when this can be observed at length scales of
a few to hundreds of millimetres will we be able to understand more objectively
how the radar is actually interacting with the sea-surface roughness, and whether
the Bragg scattering mechanism can account for most of the retrieved radar
signal.
Until then, we must simply accept that surface roughness sufficient to
give a radar return will occur in Hght or stronger winds, and make use of the
modulation of the surface-roughness ripples by longer surface waves or by
surface-current patterns, to enable radars to expHcitly image the larger-scale
features and processes. The way this can be done depends on the particular type
of active radar system being used, and wiU be dealt with in the following chapters
(11—13) discussing individual microwave instruments.

10.4 SURFACE FILMS AND SLICKS

Although the principal cause of sea-surface roughness is the wind, and the
variability of sea-surface roughness is principally a measure of the variability of
the wind and surface-wave field, the physicochemical properties of the sea
surface also affect the response of the waves to the wind. Surface-roughness
patterns may therefore be due also to the presence of surface films and slicks of
oil or organic fluids, even floating plant material or other flotsam.
Surface films are normally not randomly distributed on the sea surface, but
organised in coherent patterns, generally related to the near-surface velocity
field. If there is a surface convergence, as shown in Fig. 10.4, then as the near-
surface water is drawn down away from the surface by the flow pattern, the
floating material is concentrated at the surface along the convergence, and films
get thicker. Conversely, a divergence zone results in a weakening or removal
altogether of any surface film or slick.
318 Active microwave sensing of sea-surface roughness [Ch. 10

surface debris

Fig. 10 4 - Surface convergence thickens surface films and forms a line of surface
debris

Short gravity waves and capillary waves are damped by the dynamic
elasticity of the water surface, that is, by changes in surface tension which occur
when the surface is stretched or compressed. This has the effect of extracting
energy from those waves which depend wholly or partly on surface tension to
provide the restoring force necessary for wave propagation. If a surface film is
present, the surface tension is lower than it would be in the absence of the film,
and stretching and compression of the film due to the presence of waves provides
the dynamic elasticity which enhances the wave damping. Thus capillary waves
and short gravity waves are always damped in the presence of surface films. If
the surface film is spatially patchy, varying in thickness, or lined up in slicks due
to surface convergence, it is to be expected that the capiUary/gravity wave energy
will reflect that patchiness, being greatest where there is no surface film.
The rate of damping is not, however, simply related to the thickness of the
film. Fig. 10.5 shows the variation in damping coefficient with the film pressure
(that is, the reduction in surface tension due to the presence of a surface film).

Fig. 10.5 - Rate of damping of capillary waves (wavelength 0.52 cm) on sea-water
as a function of film pressure (the decrease in surface tension due to a surface
film). (After Garrett (1967))
Sec. 10.5] Dynamical causes of sea-surface roughness patterns 319

measured in the laboratory by Garrett (1967) using a variety of sea-water

samples. As the film was made thicker by surface convergence, the surface
tension decreased, film pressure increased, and the damping coefficient passed
through a maximum and then levelled out at a value ahout three times that
for clean-surfaced sea-water. Scott (1972) has shown also that the presence of
surface films significantly reduces the growth rates of capillary waves when a
wind is blowing over the surface.
Huhnerfuss, Walter, & Kruspe (1977) observed that weak films were always
present during the summer of 1974 in the tropical Atlantic, since the measured
surface tension was always less than 70.5 mN/m, compared with pure sea water
at 73.8 mN/m. In a coastal area of the North Sea they found the surface tension
to be as low as 53 mN/m. They concluded that the requirements for surface
films to develop sufficiently for wave damping to be made visible in surface
slicks are more complex than simple surface convergences. Clearly, if there is a
strong surface convergence, as in a front, then a slick will form even in moderate
surface winds, provided that there is material present to form a film. Once a film
has formed it is harder to disperse since it resists wind stirring. However, once
the sea becomes fairly rough, surface films are broken up and the material
is then either submerged or projected into the marine atmosphere by bursting
air-bubbles. In coastal waters in particular, where biological productivity is high,
sufficient material is generated in the water column to produce natural slicks
without surface convergence to thicken the surface film. Surface-active material
migrates to the surface either through its own buoyancy, or more efficiently
through being scavenged from the upper water column and transported to the
surface by small bubbles of air. In summer, the slicks grow during the daylight
hours as more material is produced, and may tend to be dispersed at night. The
wind speed necessary to destroy the films depends not only on their thickness,
but also on their chemical composition, and much research remains to be
performed in order to understand the composition and dynamics of surface
films.
Artificial slicks are also found in the ocean, and remote-sensing techniques
may possibly be able to contribute to the surveillance and monitoring of oil-
spills. However, given the widespread occurrence of natural surface slicks,
particularly in coastal regions, there is little evidence that satellite radar devices
can make much contribution at present to this operational problem, since they
cannot distinguish between natural and artificial surface films, although airborne
remote sensing is proving to be a promising tool for this work.

10.5 DYNAMICAL CAUSES OF SEA-SURFACE ROUGHNESS

PATTERNS
Surface waves, and the near-surface wind field, are the obvious ocean
phenomena to be studied using active microwave devices. However, it is possible
320 Active microwave sensing of sea-surface roughness [Ch. 10

for other ocean processes to be observed by microwave remote sensing, provided

that they are able to influence the surface roughness in some way.
Principally, it is the near surface velocity field of a deeper process which
enables the surface roughness to be modulated. This may be the velocity
associated with an internal wave field, or the variation in turbulent velocity asso¬
ciated with tidal flow over shallow topographic features. There are several ways
in which the surface roughness can be influenced, and when ocean features
other than surface-wave patterns have been imaged by active satellite radars it is
not always clear which of these modulation processes is occurring.
One is the way convergence and divergence cells influence the surface
films as discussed in the previous section. However, caution is necessary in the
interpretation of slicks which appear on satellite imagery, because there can be
no conclusive one-to-one correlation of slicks with surface convergences. A
strong convergence may be present, without forming a visible slick either because
the sea state is rough enough to destroy the slick and thus overcome its
wave-damping properties, or because there is insufficient surface-active material
available. Similarly, slicks may develop naturally without any surface conver¬
gence, and indeed SAR imagery may be a valuable tool in the further study of
the formation of natural slicks. However when slicks form in coherent patterns,
it is reasonable to look for an underlying hydrodynamic cause of the regularity,
as in the case of a front or internal waves.
Another way in which the surface velocity can influence surface roughness
is through the influence the current exerts on surface waves already present. If a
train of surface gravity waves runs into an opposing surface current, then their
absolute progression is slowed down and the wave energy is concentrated, with
a resulting increase in wave amplitude (Fig. 10.6). The opposite is true if the
current is flowing with the waves. It is the current convergence or divergence
which is the important control of this mechanism. The effect becomes more
marked for shorter waves, since their phase speed is comparable with the speeds
of order 0.1->■ 1.0 m s"‘, which occur in strong surface currents, whereas swell
waves with phase speeds of 10 m s”‘ or more, are less affected. There is clearly
the possibility that an image of surface roughness can show up boundaries
between currents flowing at different speeds, either across shear flows or
strong surface convergences. However, it is not yet clear how important this
mechanism is, nor how well current-distribution patterns can be imaged through
a surface roughness signature. The dynamical theory of wave—current interaction
is examined in detail by Phillips (1977).
A third way in which currents can influence surface roughness is by affecting
the actual wave-generation process. Waves are generated when there is a relative
shear between the water surface and the wind. Horizontal variability of the wind¬
forcing will produce variability of the wind-energy spectrum, with a consequent
variation of backscattered radar intensity. This variability may be due to varia¬
tions in the water velocity. For example, when a tidal current is running against
Sec.10.6] Artificial causes of sea-surface roughness patterns 321

the wind, even if it is only a relatively light breeze, the sea may become quite
rough. If the tidal current varies from place to place, so will the sea-surface
roughness. This potentially provides a mechanism for defining the distribution
of tidal currents, provided that the wind is not so strong that the whole sea
becomes uniformly rough.

wav* diraction taapaning wava braaking

convargent aurfaca currant

Fig. 10.6 - Surface convergence produces wave steepening and possibly breaking,
enhancing surface roughness

10.6 ARTIFICIAL CAUSES OF SEA-SURFACE ROUGHNESS PATTERNS

It should not be forgotten that another way in which the sea-surface rough¬
ness can be given systematic patterns which appear on radar images is by the
direct influence of man-made structures and vessels. Moving ships, or fixed and
moored stationary structures in a moving stream, generate both a surface-wave
pattern and a wake which cover a much larger surface area than the object
causing it. Thus the vessel or structure itself may be barely discernible on a radar
image because of its small size, whilst its surface roughness signature is plainly
resolved.
The surface-wave pattern associated with a ship moving at a constant speed
in a straight line is the same, irrespective of the speed or size of the ship.
It consists of the pattern of crests and troughs shown in Fig. 10.7. Outside the
lines which trail behind the ship at an angle of 19)^° to the ship track, no waves
can penetrate. Fig. 10.7 is drawn from an analytical solution (Lamb (1936))
which models reality extremely well. Close to the ship, of course, the pattern is
complicated by the shape of the ship, and the bow wave does not fit the pattern,
but within about a ship’s length away from the ship the pattern as shown is a
good representation of the real phenomenon. For a small slow ship the scale of
the wave pattern would be too small to be resolved even by a high-resolution
SAR on a satellite, but the overall envelope should show up as a region of greater
surface roughness. For large vessels, however, sufficient energy is fed into the
wave pattern for it to grow and persist for a long time after the ship has passed.
Furthermore, if the ship was travelling fast, then the waves it generates may be
of a wavelength sufficiently long to be individually resolved by a SAR, like swell
322 Active microwave sensing of sea-surface roughness [Ch. 10

waves. A ship steering a curved course, or changing speed, does not produce
such a clear pattern, although Stoker (1957) gives solutions for other types of
ship tracks. It should be remembered that if the sea is rough, the ship-wave
pattern will be lost, and not be capable of being imaged.

Fig. 10.7 - Surface wave pattern generated by a ship moving at a constant velocity

As well as the surface-wave pattern which spreads out in the familiar

V-shape, the ship also leaves a narrow wake of disturbed turbulent water,
produced partly by its own displacement in the water and partly by the rotary
action of its propellors It does not spread out like the surface waves, but
remains as a record of where the ship has been, until the turbulent motion is
eventually damped out. It is not immediately clear what effect the wake has on
the surface roughness. In a calm sea the turbulent motion will tend to increase
the short-wavelength energy. However, if waves are already present through wind
action, the turbulent motion inhibits their propagation and may also tend,
through the generation of bubbles, to increase the concentration of surface-
active agents in the wake once the turbulence has subsided. This consequently
leaves behind a persisting wake in which the gravity/capillary wave energy is
reduced.

The manifestation of dynamic ocean processes in a surface-roughness

signature is a subject not yet well defined scientifically. Mariners of many ages
and civilisations have used a knowledge of surface conditions to guide their
navigation, but this has been more by intuition than by scientific understanding.
Now that remote sensing of the ocean is a reality, there is the impetus for
Sec. 10.6] Artificial causes of sea-surface roughness patterns 323

further research in this area. The Synthetic Aperture Radar in particular has
revealed a wealth of surface-roughness patterns in the images recovered from the
short Seasat flight. We shall return to these examples in Chapter 12, but of
course the subject of radar interaction with sea-surface roughness which has
been introduced in this chapter has relevance to a variety of different active
radar remote-sensing instruments.
CHAPTER 11

The altimeter as a
surface-roughness sensor

11.1 A NADIR-VIEWING RADAR

The principles of satellite altimetry have already been presented in Chapter 9.
It was pointed out there that as well as measuring the sea-surface height relative
to the satellite position, an altimeter is also capable of measuring surface-wind
strength and significant wave-height. The reason for this is straightforward. A
microwave altimeter is fundamentally a nadir-viewing radar, and therefore
measures the radar backscatter cross-section of the surface, which is related
to the surface roughness. Moreover, because it is designed to time the reflected
microwave pulse with great accuracy, the altimeter is able to discern the effect
of the surface-wave height distribution upon the shape of the returned pulse,
providing another source of information about the surface waves.
This chapter presents the general physical principles behind this use of the
satellite altimeter. It describes the accuracies of wave and wind measurement
achieved by the altimeter flown on Seasat, and discusses the wider oceanographic
applications of this type of remote-sensing instrument.

11.2 PHYSICAL PRINCIPLES OF OPERATION

11.2.1 The effect of waves on the return-pulse shape
Fig. 9.3 illustrates the flat-surface reflection of a radar pulse. From the time the
leading edge reaches the satellite sub-point, the return signal grows linearly as
the area of reflecting surface increases with time, until the trailing edge reaches
the sub-point and the illuminated surface becomes an annular region, of constant
area. Because the radar antenna is very direction-specific, the plateau of return
power which is achieved when the annular stage is reached soon starts to decline
as the illuminated annulus moves further away from nadir.
In contrast. Fig. 11.1 shows what happens when the same pulse impinges on a
surface distorted by ocean waves. The first returns begin to arrive earlier than for
the flat case, because the pulse front first reaches the wave crests and is reflected
back from them. Thereafter the illuminated area gradually increases, but it is only
[Sec. 11.2] Physical principles of operation 325

when the trailing edge of the pulse has reached the bottom of the wave troughs at
the nadir point that the illuminated area will equal the area of the illuminated
annulus in the flat case, and thereafter remain constant (on average). In contrast
with the flat case, where the reflection of an emitted step pulse grows to its full
strength within the time of the pulse duration, it takes many times this duration
for the pulse to grow to full strength in the rough-sea case, provided that the pulse
width is small compared to the wave height.

Fig. 11.1 — Altimeter pulse incident on a rough sea surface showing extended size
of pulse limited circle. (From Walsh, Uliana, & Yaplee (1978))

The higher the waves the longer it will take for the return signal to grow, so
that the slope of the leading edge of the reflected pulse is inversely proportional
to the significant wave height. To obtain a meaningful measure of wave height
from the slope of the return pulse, it must be possible to measure the pulse-return
times and to sample the growth of the pulse with sufficient precision. Also, the
pulse width (a pulse must be only 1 ns long if its effective length in space is to
be about 0.3 m) must be small compared with the wave height. If it is not, the
leading-edge slope of the reflected pulse will be hard to distinguish from the slope
obtained from reflection at a smooth flat surface.
So far we have assumed that radar power reflected back to the satellite
is proportional to the area of surface illuminated at any instant. Given a rough
surface, there will in general be a random distribution of sea-surface slopes
at each point where the microwaves reflect from a portion of the sea surface.
Since the altimeter depends on direct specular reflection, it will receive only
those reflections for which the surface facet is oriented normally to the incident
radiation. For a single return pulse the random distribution of surface facets
results in a signal which is dominated by noise (Fig. 11.2). It requires the addition
of n return pulses to reduce the noise by a factor of When 1000 pulses
are examined, the resulting standard deviation is only 3%, as illustrated in the
lowest curve of Fig. 11.2.
As mentioned in Chapter 9, the greater the significant wave height, the greater
is the area effectively sampled for altimetric measurements, and the altimetric
measuring system must take account of the fact that over this wider area the
326 The altimeter as a surface-roughness sensor [Ch. 11

average distance from the satellite to the mean sea level is slightly greater than
under calm conditions when the sampled area is a smaller circle at nadir. A more
significant error is introduced into the altimetry by a rough-sea surface because
of the skewed surface-slope distribution. Because the sea waves tend to be slightly
trochoidal in profile, with peaked crests and flatter troughs, the surface-area
elements with zero slope are found preferentially in the trough regions, whilst the
greatest slopes are found nearer the crests. Since the strongest radar reflections
come from the zero-slope areas, there is a tendency for the strongest signal to
come from surfaces below the mean level, and this tends to give an overestimate
of the distance to mean sea level when many samples are averaged from a rough
sea, producing a sea-state-related error in the altimetry.

Fig. 11.2 — Examples of a single and averaged return pulses received by a radar
altimeter after scattering from many wave facets on the sea surface. (From
Townsend, McGoogan, & Walsh (1981))

The measurement of significant wave height is obtained by sampling the

pulse-rise profile in a series of electronic gates. Fig. 11.3 illustrates the sampling
of signals (after averaging) from three different surface-wave conditions. Ideally,
the mid-height position of the pulse should determine the time corresponding to
the initial pulse return from mean sea level. As can be seen, however, by the
time the pulse reaches its maximum the signal is already in the decay part (or so-
called plateau-droop) of the pulse shape, and so the half-strength signal is
Sec.11.2] Physical principles of operation 327

achieved slightly before the true timing of the pulse return from mean sea level.
This is allowed for in the altitude tracker algorithm which positions the time
corresponding to the mean level. In Fig. 11.3, the three curves have been centred
about this point. On Seasat the wave-height algorithms operated on the values
of 60 gates, and so the profiles are centred with the return corresponding to
the mean level fixed between gates 30 and 31. For Seasat the height tracker
algorithm allowed for the plateau-droop by locating the mean sea level assumed
return at a time when the power had a value of 60/53 of the mean of the 60
gates surrounding it.

Fig. 11.3 — Seasat altimeter pulse shapes for three different wave-height condi¬
tions, and fitted solution curves. The data are averaged over 10 000 pulses (10 s),
and have been corrected for gain biases in the sampling gates

The pulse profile (ignoring the plateau-droop effect) is produced by the

convolution of the transmitted pulse with the sea-surface height distribution,
and integrating the result. In principle it should be possible mathematically
to invert this process to obtain the surface-height distribution from the shape of
the observed return pulse. In practice a more empirical approach has been used
with the Seasat and GEOS altimeters.
The average power received by the radar instrument can be expressed as a
function of time for the pulse-limited geometry under consideration, as (Fedor
et a/. (1979)):
Kct r (til 2t'
Pit) = exp (11.1)
p/J
328 The altimeter as a surface-roughness sensor [Ch. 11

where H = height of satellite above mean sea level

r = half power width of transmitted pulse
s = rms total ocean-wave slope
Vi
+ 2h^
.161n2
h = rms ocean wave height
t,= 2H^llc
2
81n2 1 + Hla^
^-2 = +
R
flg = earth radius
= antenna half-power width
K = constant including transmitted power and pulse-compression
gain, antenna gain, propagation losses, and the Fresnel reflection
coefficient.
t = O is defined when the centre of the pulse reaches the radar after
reflection at the mean sea surface.
An expression of the form of (11.1) can be used to calibrate the pulse
shape. The values of the parameters in the expression are adapted to achieve a
best fit to the satellite-observed waveform. As Fig. 11.3 shows, the theoretical
curves fit the observations well, except in the plateau-droop region, and can
therefore be used to interpret the pulse shape in terms of significant wave height.
There are various refinements necessary to allow for such things as the
skewness of the surface-height distibution, and the method can be studied in
detail in Walsh, Uliana, & Yaplee (1978). Fedor et al. (1979) and Townsend,
McGoogan, & Walsh (1981) also provide helpful summaries of the wave-height
measuring procedure. The details of the Seasat altimeter pulse have been
described in Chapter 9, and it should be remembered that in practice pulse
compression is used, but this does not alter the above discussion. The analysis of
the pulse shape for significant wave height was performed automatically by logic
circuits on board Seasat, so that significant wave-height information could be
transmitted directly from the satellite.

11.2.2 The effect of surface roughness on the return-pulse strength

For the nadir-viewing active sensor, the strongest return occurs when the surface
is flat calm. Roughness causes the surface to tilt and so reflect the incident
radiation away from the satellite. There is therefore an inverse relationship
between the received signal strength and the surface roughness. It is expressed
in (11.1) by the s term. Another way of expressing it is as an inverse relation
between the surface roughness and the effective radar cross-section of the
surface.
Sec.11.3] Perfonnance of wave-height algorithms 329

On Seasat, as samples were selected for the 60 gates to enable comparison

with theoretically derived curves in the significant wave-height algorithm, an
automatic gain control operated so that the sum of the powers from all the 60
gates was kept approximately constant. The same procedure was used on GEOS.
The gain setting necessary to achieve this was relayed back to earth along with
the other data, and can be used as an inverse measure of the radar backscatter
cross-section, a“.
Expressions have been developed to relate to the mean square surface
slope (for example, Barrick (1974) Hammond et al. (1977)):

’^li?(0)|2 , -tan^0
o 0 = —=— sec^0 exp — (11.2)
s L s _
where /?(0) is the Fresnel reflectance of the air—sea interface at normal
incidence, and has a value between —2.08 and —2.37 dB at the 13.9 GHz
frequency used by the Seasat and GEOS altimeters, depending on water
temperature and salinity.
Now the relationship between the mean-square slope of the sea surface
and the wind speed has been known for some time, following interest in optical
glitter patterns (Cox & Munk (1954)). The empirical relation is

7^= 0.003 + 0.00512 6/12.5 (11-3)

where t/12.5 l^e averaged wind speed in m s'^ at 12.5 m above sea level. Using
(11.2) and (11.3), it is easy to obtain an estimate of the wind speed (but not
direction) from the radar backscatter cross-section, calculated from the known
automatic gain control setting and other information about the radar power and
geometry. (11.2) shows how the actual viewing angle 6 (known as the satellite
pointing) strongly influences a® and therefore needs to be known accurately
from the satellite orbit and attitude model used in the altimetric calculations
(see Chapter 9), if wind speed is to be accurately estimated.
Conversely, given some knowledge or assumptions about the wind-field, it
may be possible to use the variability of as an indicator of the satellite point¬
ing. The pointing can also be estimated from the slope of the plateau-droop at
the tail of the pulse-return signal. The effect of slightly non-nadir viewing is
more important for wind estimates than for wave-height measurements, so long
as pulse-limited geometry applies.

11.3 PERFORMANCE OF WAVE-HEIGHT ALGORITHMS

Wave-height algorithms have been applied to both the GEOS-3 and the Seasat
altimeters, the latter being a more precise and sensitive instrument with a cor¬
respondingly improved algorithm performance. Whereas the GEOS-3 had a
12.5 ns pulse width (3.7 m width) and 16 gates for sampling the return-pulse
330 The altimeter as a surface-roughness sensor [Ch.ll

front at 6.25 ns spacing, the Seasat altimeter pulse was only 3.2 ns wide and
there were 60 gates at 3.125 ns spacing to sample the return pulse.
Several algorithms were developed to interpret the GEOS-3 data (reviewed
by Fedor et al. (1979)), and intercomparisons between these and synoptic sea
truth led to the following conclusions.
At low sea states, hy-i < 4 m, and the standard deviation
S.D.(;zi/3) < 0.75 m
At medium sea states 4.1 < hy-^ <8.0,
S.D.(/2i/3) < 0.5 m.
Insufficient sea truth was available for calibration at high sea states.

These results required averaging of up to 1600 pulses (over 16 s), and resulted in
a sea-state measurement averaged over 105 km length of ground track.
These achievements approach the limits of resolution of buoy and ship
measurements of waves, and the Seasat results turned out to be even better.
With a short lifetime of three months, there was not a great deal of synoptic
sea truth available for Seasat calibration tests. Fedor & Brown (1982) made com¬
parison between 51 buoy measurements and nearby Seasat overflights off the
North Atlantic and North Pacific coasts of the USA and in the Gulf of Mexico,
‘nearby’ in this context being within 80 km and hours of coincidence. There
was a mean difference of 0.07 m and a standard deviation of 0.29 m within the
available range of observed wave height, 0.5 m < hy^ < 5 m, when an optimum
algorithm was used. The comparisons are shown in Fig. 11.4. The on-board
processor tended to give wave heights biased 0.5 m higher for /z 1/3 > 2 m.
There was a fortunate coincidence between the lifespan of Seasat and the
Joint Air-Sea Interaction experiment in the North-East Atlantic. Webb (1981)
made comparisons between the altimeter measurement and a pitch—roll buoy
and found their ratio to be 0.96 ± 0.04, with a standard deviation of 0.1. The
eight samples were all at low wave heights, less than 2—3 m. A much larger
dataset of visual observations of wave height was compared by Queffeulou et al.
(1981), and although a standard deviation of 87 cm resulted, this was probably
due more to the inadequacies of the ship-observed data. No significant bias was
noted, even though wave heights up to 8 m were sampled. The overall conclusion
to be drawn is that the altimeter was able to measure significant wave height to
a precision and accuracy at least as good as in situ instruments, with the
added advantage for many applications of supplying an areal average. To achieve
the low standard deviations quoted above, the wave height was obtained from
a 10-second average of 10 000 pulses, corresponding to a ground-track length
of 67 km. With such a short lifetime, exhaustive confirmation and fine-tuning
improvements of the wave height algorithm could not be achieved, but there is
no doubt that the radar altimeter is proven as a powerful and accurate means
of measuring significant wave height.
Sec. 11.4] Perfonnance of wind-speed algorithms 331

Fig. 11.4 — Scatter plot comparing significant wave height estimates from the
NOAA buoy network and Ocean Weather Station PAPA with those inferred from
the Seasat altimeter using the Fedor algorithm. (After Fedor & Brown (1982),
copyrighted by the American Geophysical Union)

11.4 PERFORMANCE OF WIND-SPEED ALGORITHMS

As with the sea-state algorithms, the wind-speed algorithms for GEOS-3 were
able to be improved and tested more readily over the longer life of the satellite
than for Seasat. Brown (1979) reports that the GEOS-3 wind-speed algorithms
were able to predict within a r.m.s. precision of 2.6 m s"^ for wind speeds up to
21 m s“^ Because GEOS-3 was still in orbit at the same time as Seasat, inter¬
comparisons between the two altimeters were used to assist in tuning the Seasat
wind-speed algorithm (Fedor & Brown (1982)). Comparisons between Seasat
estimates and buoy measurements are shown in Fig. 11.5. The mean difference
between the two sets of measurements is 0.25 m s“\ and the standard deviation
of the difference is 1.58 m s”\ It can be concluded that the altimeter is able to
give a fair estimate of wind speed, probably within about 1 m s“^ of the true
value, at least over the 0-10 m s"^ range tested in the available dataset.
The wind-speed-measuring capability is not as important for oceanographers
as the wave-height-measurement achievement, because there are other satellite
332 The altimeter as a surface-roughness sensor [Ch. 11

sensors capable of providing more accurate wind-speed estimates along with

direction as well (for example, the scatterometer). Comparison between the
accuracies of the different Seasat sensors (Wentz, Cardone, & Fedor (1982))
will be referred to in Chapter 13, which describes wind measurement using the
scatterometer.

SEASAT-MEA8UREO WIND, m/«

Fig. 11.5 - Scatter plot comparing buoy-measured wind speeds with those
inferred from Seasat Altimeter measurements using the Brown algorithm. (After
Fedor & Brown (1982), copyrighted by the American Geophysical Union)

11.5 OCEANOGRAPHIC APPLICATIONS OF

SATELLITE-ALTIMETER-DERIVED WAVE HEIGHT
It has been mentioned several times already in this book that one of the principal
benefits of satellites to oceanography is their capability of providing repeated
global sampling. This is particularly true of the altimeter-measured significant
wave height. The wave climate of the ocean (that is, how the wave height varies
spatially and seasonally over the ocean) is of interest not only scientifically to
oceanographers, but economically to marine technologists who must design oil¬
drilling platforms and plan efficient ship routes. It would take an extensive and
costly deployment of wave buoys to begin to provide such data, whereas the
satellite can do so with ease. Wave height is not a high-data-rate measurement,
and so can readily be recorded on board the space vehicle for transmission to
Sec. 11.5] Applications of satellite-altimeter-derived wave height 333

earth at a suitable point in the orbit. Wave-climate data does not require a very
high frequency of coverage, and the repeat-orbit frequencies of 3 to 10 days
suggested in Chapter 9 as being appropriate for altimetry can provide adequate
sampling for most purposes. What is important is that the data should continue
to be available over a period of years. With a microwave sensor, the weather does
not interrupt the data series, but unfortunately the failure of Seasat cut short
the database that was beginning to be established. The task will be continued
when the next satellite altimeter is launched.
The results from Seasat provide an enlightening foretaste of the sort of
results which will be achieved with future satellites. Chelton, Hussey, & Parke
(1981) combined all the available data into an average wave-height map for the
world during July, August, and September 1978 (Fig. 11.6a), and also produced a
wind-speed map (Fig. 11.6b), from the altimeter data. Given the sparse coverage
of the southern hemisphere by shipping routes, and because the ships which
are present tend to avoid storms, this map must be the first realistic indication
of storm-wave conditions in the southern winter, albeit for only one winter
season. Waves with an average height of over 5.5 m can be seen to occur in
the south Indian Ocean over a large region. High waves occur as well in the path
of the East African Jet wind system, in the Arabian Sea. These and many
other interesting features of the wave-climate map open up exciting new areas
for oceanographers to study. It is truly remarkable that without ever having
to venture near these inhospitable areas of rough sea we are able to chart them
with all the confidence that comes from the successfully-tested wave-height
algorithms.
Mognard et al. (1983) have taken a closer look at the southern hemisphere
by plotting monthly maps of wave height and wind speed. Figs 11.7 and 11.8.
These clearly show how a region of maximum wind and waves is found in the
South Atlantic in July, but migrates eastward to the Indian Ocean and across
to the South Pacific during August and September. It is interesting to note how
the change to a 3-day repeat orbit with less dense spatial sampling has resulted
in smoother contours for September, since less information is contained in the
map.
Because wind and wave height are measured together, it is possible to infer
the sweU distribution from the satellite data using the following procedure. The
wave energy is calculated from wave-height data. The wave energy due to local
wind generation is also calculated from wave-prediction models, using the
satellite-measured wind field, assuming that a fully-developed sea has arisen in
equilibrium with the wind forcing. When this is subtracted from the observed
wave energy the residual energy must be that which has propagated into the
region from outside, that is, sweU-wave energy. Hence the swell height can be
mapped as in Fig. 11.9. This is a minimum estimate, since the locally wind-
driven wave energy may have been overestimated by assuming a fully-developed
sea.
 334 The altimeter as a surface-roughness sensor [Ch. 11

(b)
Fig. 11.6 - Seasat altimeter data for 7 July-10 October 1978. (a) Wave heights;
(b) Scalar winds. (From Chelton, Hussey & Parke (1981))
 Sec. 11.5]

SIGNIFICANT WAVE HEIGHT (m), SIGNIFICANT WAVE HEIGHT (m), SIGNIFICANT WAVE HEIGHT (m)
Applications of satellite-altimeter-derived wave height

Fig. 11.7 - Seasat altimeter-derived maps of significant wave height in the Southern Ocean for (a) July, (b) August, and (c) September-Octo-
ber 1978. Contours are at 1 m intervals, and the shaded region is over 5 m. (From Mognard et al. (1983), published by the American Geo¬
335

physical Union)
 336
The altimeter as a surface-roughness sensor

Fig. 11.8 - Sesat altimeter-derived maps of wind speed over the Southern Ocean for (a) July, (b) August, and (c) September - October 1978.
Contour intervals are 2 ms"'. The shaded region is greater than 10 m s"‘. (From Mognard et al. (1983), published by the American Geo¬
physical Union)
[Ch. 11
 Se c. 11.5 ]
Applications of satellite - altimeter - derived wave height

Fig. 11.9 - Seasat altimeter-derived maps of swell height in the Southern Ocean for (a) July, (b) August, and (c) September-October 1978.
Contour intervals are 1 m. Heights greater than 4 m are shaded. (From Mognard et al. (1983), published by the American Geophysical Union)
337
338 The altimeter as a surface-roughness sensor [Ch. 11

Mognard (1983) has extended this interesting method for estimating the
swell climate to cover the whole world for the Seasat lifetime (Fig. 11.10).
Furthermore, for the time when the 3-day repeat orbit was in operation,
3-day maps of swell height have been produced for the North Atlantic Ocean
(Fig. 11.11). From these it is possible in conjunction with a knowledge of storm
centres based on the wind- and wave-height data, to study the way in which the
swell propagates out from storms. The promise of similar interesting and use¬
ful oceanographic data from further altimeters suggests that these methods can
be applied in the future to give us an understanding of the continuously-evolving
wave climate.
It is probable that oceanographers will also be able to use repeated mapping
of ^1/3 to test hypotheses concerning the generation, propagation, and decay of
ocean waves, in order to be able to improve swell and wave forecasts based on
wind forecasts. Queffeulou et al. (1981) provide an example of this approach,
by comparing the satellite-observed wave height with that predicted by a
wave-forecast model over the North Atlantic.
Fig. 11.10 - Field of mean significant swell heights in metres deduced from Seasat altimeter measurements for the period 7 July -10 October
1978 (From Mognard (1983))
340 The altimeter as a surface-roughness sensor [Ch.ll

Fig. 11.11
Sec. 11.5 ] Applications of satellite-altimeter-derived wave height 341

Fig. 11.11
342 The altimeter as a surface-roughness sensor [Ch.ll

Fig. 11.11
Sec. 11.5] Applications of satellite-altimeter-derived wave height 343

Fig. 11.11 - North Atlantic swell maps deduced from the Seasat altimeter
measurements for the periods (a) Sep. 19—21, (b) Sep. 22—24, (c) Sep. 25—27,
(d) Sep. 28-30, (e) Oct. 1-3, (f) Oct. 4-6, and (g) Oct. 7-9, 1978. (From
Mognard (1983))
CHAPTER 12

Synthetic aperture radar

12.1 HIGH-RESOLUTION IMAGING RADARS

In the minds of many people, the concept of satellite remote sensing of the
environment is synonymous with pictures of the earth viewed from space. The
daily television screening of satellite weather pictures reinforces this impression.
It is of course a false conception of remote sensing. The environmental scientist,
and particularly the oceanographer, is more concerned to obtain quantitative
measurements of spatial structures than pictures, if satellites are to contribute
significantly to the development of marine science. In all the preceding chapters
this has been the case. Even when an imaging sensor is used as with visible or IR
scanning radiometers, the emphasis must be on the quantitative measurement
rather than the picture which is produced, if serious scientific study is to be
pursued. However, it would be wrong to relegate pictures from satellites to an
inferior status as merely popular products for the layman but of little value to
science. Images from satellites are able to convey to the scientist an immediate
impression of spatial structures and patterns, and whilst it may ultimately be
necessary to decompose an image into, for example, its wavenumber spectrum,
to obtain quantitative spatial data, it would be shortsighted to underestimate the
contribution which suitably-enhanced images can make in stimulating creative
scientific progress. The human brain is skilled and experienced at discerning
patterns and searching for order in visual images, and the oceanographer who has
for so long been limited to a very piecemeal view of the processes occurring in
the ocean may find that an image from the vantage-point of space provides the
new perspective which enables other ground-based measurements to be properly
understood. At the same time, an image may reveal hitherto unsuspected features
which demand further study and investigation at sea level.
The major drawback of visible and infrared imagery of the ocean is the
problem of cloud cover. The land applications scientist may be content to wait
for occasional cloud-free images, whilst for the meteorologist the clouds are what
he wishes to see. For the oceanographer,who would like as many uninterrupted
views as possible of the ever-changing sea surface, the answer has been to move
[Sec. 12.2] Principles of SAR operation 345

from visible and infrared wavelengths to microwaves which can penetrate the
cloud and give all-weather views of the ocean. This has been the rationale behind
the development of imaging radars, and in particular the synthetic aperture radar
(SAR) which has already proved capable of spatial resolution down to 25 m.
An active radar device is necessary to obtain such resolution, and therefore
what is actually being imaged is primarily the roughness of the sea surface.
Assuming a Bragg resonance type of backscattering process, given oblique
surface viewing, it is sea-surface perturbations on a length scale comparable to
the radar wavelength which appear as roughness on radar images. Herein lies the
excitement of working with SAR images of the ocean surface. As mentioned in
Chapter 10, roughness of the sea surface is not a concept with which oceano¬
graphers have much experience, except in the context of ocean waves, and even
there little attention has hitherto been paid to the very short wavelengths which
produce radar roughness. Yet the SAR images which were obtained first from
aircraft, then from Seasat and latterly from the Shuttle Imaging Radar (SIR),
have revealed not only surface waves and swell but other oceanographic processes
which have a (previously unrecognised) surface-roughness signature.
It is beyond the scope of this book to go into great detail concerning
instrument design, but whilst it has been possible to discuss radiometers or even
the altimeter in terms of the broad concepts of their operation, without being
concerned about instrumental operating details, it is necessary to describe the
SAR more thoroughly. This is because the imaging mechanism and its interaction
with oceanographic parameters is closely related to the engineering design of the
instrument. It is helpful to split up the SAR imaging mechanism into three
distinct, though interrelated, processes. The first is the fundamental principle
of the synthetic aperture radar, as applied to stationary targets such as the
land surface. Second is the interaction between the basic SAR operation and
a surface such as the sea which is itself moving. This becomes closely related to
the question of how waves are imaged by the SAR. Third is the problem of
understanding how other oceanographic phenomena are capable of being imaged
by SAR through their influence on the surface roughness and movement. All
three aspects of SAR imaging are discussed in this chapter. It is important
for the oceanographer to understand fundamentally how the radar operates even
if he need not concern himself with the hardware details, and it is equally
necessary for the radar scientist to appreciate how oceanographic processes come
to have a SAR signature, if both are to develop the use of SAR as the powerful
oceanographic tool which it is capable of becoming.

12.2 PRINCIPLES OF SAR OPERATION

12.2.1 SAR geometry

Fig. 12.1 illustrates the geometrical arrangement of a SAR operating in broadside
mode. SAR can also view in an oblique forward- or backward-looking direction
346 Synthetic aperture radar [Ch.l2

known as squint mode. The analysis is rather more complex for squint mode,
but the principle is the same. The distance from the satellite to the ground in the
viewing direction is called the slant range. Its projection onto the sea surface
is called the ground range. The direction parallel to the satellite track is known
as the azimuth direction.

Fig. 12.1 — A synthetic aperture radar in broadside mode

An antenna emits a microwave pulse which spreads out in a beam to

illuminate the sea in the range direction R. Because the viewing angle is oblique,
there is normally no direct specular reflection, and except for very high sea
states it is assumed that Bragg resonance is the mechanism of backscatter. Thus
for a radar emitting radio waves of wavelength Xr, the perturbations of the
sea surface which will contribute most to the backscatter are those with a
wavelength in the ground range direction close to X^where (eqn. (10.3)),

As (12.1)
2sin0 ’
6 being the angle between the slant-range direction and the vertical.
The resolution of a normal antenna is dictated by its aperture width. Dr.
Standard antenna theory gives the beamwidth (Fig. 12.2) as (3 = Xr/Dr in the
Sec. 12.2] Principles of SAR operation 347

far-field away from the antenna. Thus the resolution at range i? is p = /?/XZ)r,
in the plane normal to the range direction. For oblique viewing this becomes
larger in the ground range direction. Typically, radars will operate with a wave¬
length between 50 mm and 1 m. Even with a low satellite orbit having a range R
of about 750 km, a ground resolution of 100 m will require an antenna aperture
of between 375 m and 7.5 km, depending on the wavelength chosen! This is
clearly impracticable with present space technology, and an alternative approach
must be sought which makes use of the phase of the coherent signal to carry
more information and allows a smaller antenna to be used.

Improved resolution in the range direction is obtained by using the time of

the return of pulses as a measure of the distance to the reflector. This can be
achieved with an ordinary side-looking radar, but cannot improve the azimuth
resolution. To improve this, the synthetic aperture principle is invoked, so called
because it uses the travel of the radar along its path to construct the same
resolution image as would be seen by an aperture of very much greater length.
A thorough description of synthetic aperture radar with the ocean-wave
viewing context in mind is given by Tomiyasu (1978), whilst a clear and helpful
introduction to the subject is presented by Lodge (1981).

12.2.2 Range resolution

Ignoring first the question of azimuth resolution, we consider the resolution
possible in the range direction. Fig. 12.3 illustrates the process. A pulse of length
r is emitted and illuminates an area governed by the beam width in the ground-
range direction. This is now preferably as large as possible to enable a large swath
to be covered. After a time the first return comes from the nearest part of
the illuminated area at range R^, followed by a signal which continues until the
tail of the pulse reflected from the farthest illuminated point (Rf) has arrived
back at the satellite at time tf. The signal is in fact a convolution of the pulse
with the radar backscattering cross-section o as it varies across the illuminated
area between R^ and Rf.
348 Synthetic aperture radar [Ch. 12

Fig. 12.3 - Geometry of range resolution of a side-looking radar

The range resolution is governed by the pulse width. Essentially, the

number of pulse lengths r which can be fitted between and tf is the number
of independent measures of o which can be obtained across the swath. The
slant-range resolution is ct/2, which is the difference in range of two points
whose reflected signals differ in arrival time by r. Thus the ground-range resolu¬
tion Pq = cT/(2smd), where 0 is a function of the position across the swath.
As for the altimeter discussed in section 9.2.1, to achieve a range resolution
down to about 25 m would need a pulse which was so short as to require
an unattainably wide bandwidth which would also be incapable of carrying
sufficient energy to raise the return signal above the noise levels, since the return
signal power is proportional to times the output power. Therefore a chirped
pulse is sent out, and pulse compression is used to effectively reduce a long pulse
into a very short one. Fig. 12.4 illustrates a chirped pulse. It consists of a pulse
whose frequency decreases linearly with time. It is not difficult to construct a
linear filter which applies a phase lag to this chirp, the lag being proportional to
the frequency. If this is carefully chosen all parts of the long pulse are lagged into
Sec. 12.2] Principles of SAR operation 349

a very short space of time when a short pulse of very high frequency and high
power occurs. As this is a linear process, this filter can be applied after the chirp
has been reflected from the ground and received at the antenna, to create the
signal which would have been returned if the short pulse itself had been emitted.
This process in the SAR context is termed range compression of the signal.

fraquancy

Fig. 12.4 - A chirped pulse

Another way of viewing the range compression process is illustrated in

Fig. 12.5. The emitted pulse has a single frequency which decreases linearly with
time. The return echo at any one time is composed of the superposition of
reflections from all ranges within a distance limited by the pulse width. Thus at
time after the start of pulse emission, the echo contains the reflection of the
c(?i—r)
tail of the pulse from A whose ground range is “ ... • a—> l^e reflection or
cti
the start of the pulse from B where Rg = . ^ , and reflections of the rest
Z sincfg
of the pulse from points lying between A and B. The reflection from A will
be composed of a signal at the lowest (final) frequency /min of the chirp, and
from B at the highest (initial) frequency /max of the chirp. Thus the contribu¬
tion from each reflecting point between A and B to the signal received at ti,
is coded uniquely by its frequency. All the possible contributions from A (and
hence all the information about A) can be found in the return signal between
times (fi — r) and ti. Similarly, all the information about B is found between
ti and (fi + r). The frequency at which to find information about A varies from
/max (^1“ ’■) to /min at time tx. Thus to extract information about
reflections at A from the total signal between (fi — r) and ty, the return signal
is correlated with a reference signal having a chirped frequency identical to
the emitted pulse. Only contributions from A add constructively in the correla¬
tion process, and all other reflections cancel out destructively. As the correlation
signal is stepped in time through the time series of recorded data it extracts
information about the reflection from different range positions. Thus informa¬
tion about A and B respectively comes from correlations with the reference
signal centered at ti — Vn and ty +
350 Synthetic aperture radar [Ch.l2

From this point of view, range resolution depends on the ability to sample
sufficiently often for the frequency shifts not to be aliased by the sampling rate.
The resolution ceU is much smaller than (which would be the resolu¬
tion if pulse compression were not performed), because points such as C in
Fig. 12.5 can be identified clearly by a separate reference signal, centred at an
appropriate time in the return signal time series.

Fig. 12.5 - Range resolution with a chirped pulse

Sec.12.2] Principles of SAR operation 351

12.2.3 Azimuth resolution using the synthetic aperture principle

We now turn to the technique for obtaining high resolution in the azimuth
direction without using a large aperture. Let us assume first that the range-
resolution processing has been performed, and we have selected a particular
range to examine more closely (Fig. 12.6). The sea-surface reflectors which
contribute to the signal from a particular range for a given pulse lie on a circle
centred at the satellite whose radius is the range. This radius is several hundred
kilometres, and we may therefore consider for simplicity that the reflectors
contributing to the signal for a given range lie on a straight line parallel to the
azimuth direction. Moreover, we can assume that when the satellite has moved a
short distance in the azimuth direction and another pulse has been emitted and
received, if we examine the new signal for the same range interval as before, the
reflections come from the same range line as for the previous pulse (Fig. 12.6).
In practice small ‘range walk’ corrections have to be made for the curvature of
the constant range lines and the curvature of the earth, but these need not
interfere with our understanding of the basic principles.

satellite

Fig. 12.6 — Definition of range cells, neglecting curvature of constant range lines

One way of conceptualising the operation of a very wide real-aperture

radar is illustrated by Fig. 12.7. A coherent signal is emitted by the radar and
illuminates targets ‘a’, ‘b’, ‘c’, ‘d’, etc. These are assumed to reflect the signal
coherently, which radiates out individually from each. The return signal from ‘a’,
opposite the centre of the antenna, reaches the centre of the antenna first, and
the extremities a little later. There is therefore a small phase change along the
aperture of the signal for ‘a’, but the phase distribution is symmetrical about
the centre of the aperture, and overall the summation of the signal across the
352 Synthetic aperture radar [Ch. 12

aperture is constructive. The same is not true of a point such as ‘d’, which is
much closer to one end of the aperture than the other. The change of phase
across the aperture of the reflection from ‘d’ therefore causes destructive inter¬
ference when the signal is integrated, and consequently does not contribute to
the received signal. Points ‘b’ and ‘c’ may or may not contribute to the signal,
depending on the phase geometry. This governs the effective beam-width and
hence the resolution. The larger the aperture, the closer to ‘a’is the point from
which the reflected signal ceases to contribute to the received signal because of a
sufficiently large phase variation between the two ends of the aperture.

Fig. 12.7 — Phase pattern of radar waves reflected from points opposite the centre
(upper diagram) and extremity (lower diagram) of a radar antenna. Note the large
phase change along the antenna in the latter case

The synthetic aperture uses this principle, but instead of integrating along
the length of a long antenna at a single time the reflections from all the targets of
one single pulse, it builds up this result gradually. Thus as it passes‘t’ (Fig. 12.8)
it emits a pulse and receives the reflection from ‘a’, ‘b’, ‘c’,‘d’,‘e’, etc. It does
the same as it passes ‘u’, ‘v’, ‘w’, etc., and stores the results. Finally, at the data-
processing stage, the reflected signal from ‘a’ received at‘t’, ‘u’, ‘v’, etc. is added
up to synthesise the same effect as the automatic integration along a very wide
real-aperture antenna. The same is done for the reflected signals from‘bVc’,‘d’,
etc., so that the result is a measure of the different values of radar backscatter
cross section at ‘a’, ‘b’, ‘c’, etc.
Sec.12.2] Principles of SAR operation 353

satellite ground
scattering
track

Fig. 12.8 — To illustrate aperture synthesis

This begs the question of how it is possible to preserve separately the phase
information of the contributions from individual points ‘a’, ‘b’, ‘c’, etc. to the
signal received when the sensor is at a particular point‘t’. The key to this lies in
the steady motion of the radar itself in the azimuth direction, which produces
a doppler shift in the frequency of the returned signal. Assuming the targets
are all stationary, the target which is broadside on to the satellite will reflect the
incident signal with no doppler shift. The targets forward of the satellite have a
relative velocity towards the satellite, and will tend to increase the frequency of
the reflected signal, imparting a positive doppler shift. Those targets already
passed by the satellite will produce a negative doppler shift. For example, when
the satellite is at ‘w’, the signal from ‘d’ will be at the emitted frequency, the
signal from ‘e’ will be at a slightly higher frequency, and those from ‘f ’ and ‘g’
higher stiU, because the relative velocity between target and satellite has a bigger
component parallel to the radio wave travel directions at ‘g’, than at ‘f ’ or ‘e’.
At ‘d’, the relative velocity is zero, whilst for ‘c’, ‘b’, and ‘a’ it is in the opposite
direction and there is a frequency reduction in the reflected signal. Consequently,
as the radar approaches a target, the contribution of that point to the reflected
signal is first at a higher frequency than the emitted signal and gradually reduces
in a linear way till it is less than the emitted signal when the satellite has passed
by. At any given time, for a given range, the contributions from each point along
the azimuth at that range can be identified uniquely by its doppler-shifted
frequency. This enables the summation of different contributions from different
target points to be performed in the aperture synthesis.
354 Synthetic aperture radar [Ch. 12

The frequency of contributions from a single point as the satellite passes

by is plotted in Fig. 12.9. Although the curve is not continuous, but consists of
discrete points corresponding to the pulse-repetition frequency (PRF) of the
SAR, the resulting contribution is rather like the chirp signal used for individual
pulses. For this reason the integration process is called ‘azimuth compression’
of the signal. In practice, the integration over the synthetic aperture of the
contributions from a particular target point is achieved by correlating the
received signal at each satellite position with the reference frequency appropriate
for the doppler history of the particular target position whose back-scattering
cross-section is being evaluated. As integration proceeds, only the contributions
from the point correlated with the reference frequency add constructively, and
contributions from all other points interfere destructively. The same received
signal, correlated with a different chirped frequency pulse at a slightly higher/
lower frequency, will yield the radar cross-section of a point slightly shifted
forward/back along the azimuth.

Fig. 12.9 - Frequency of signal received after reflection at ‘d’ (as defined in
Fig. 12.8) while sensor travels from position ‘t’ to ‘z’

To obtain a measure of the azimuth resolution which is possible, we return

to the concept of the large antenna which is synthesised by the above process.
Fig. 12.10 illustrates the geometry. The antenna with real aperture Dr and an
azimuthal beamwidth /J^has a coverage on the ground in the azimuthal direction
of Pr ~ which is its real resolution. Now the length of aperture which
can be synthesised depends on the distance the satellite can travel whilst a point
is illuminated. For an airborne SAR the geometry of Fig. 12.10 applies, and the
synthetic aperture is Pr. For a satellite, the earth and orbit curvature complicate
the situation, but the principle is the same. The azimuth resolution p^ achievable
with this synthetic aperture Pr is p^ = XrR/2pr. (The factor of 2 enters here
because the synthetic aperture requires a two-way path for its construction,
and its effective beam-width is therefore half that of a real aperture of the same
size).
Sec. 12.2] Principles of SAR operation 355

Hence p^. = •Or/2, which is at first sight a surprising result. The resolution
is independent of the range and of the radar wavelength, and improves with
a smaller aperture! The reason is that the construction of the synthetic aperture
can proceed, and information about a target can be gained, so long as the target
is illuminated by the real beam. The wider the real beam is (by having a smaller
real aperture) the more information can be communicated about a particular
target. An increase of R increases the time a target is illuminated, and this
offsets the deleterious effects of a more distant target. The reduction in size
of aperture Dr is limited by the need to generate a sufficiently powerful signal,
but clearly a very high resolution is theoretically possible.
For the azimuth-compression process to work, there is a limitation imposed
on the spacing of the successive pulses. The temporal sampling rate must be at
least twice the highest doppler frequency shift in the return signal, if the doppler
history of each point is to be unambiguously recognisable without aliasing in the
processing stage. This means there must be at least one pulse in the time it takes
the real aperture to travel half its length. On the other hand the pulse-repetition
frequency (PRF) cannot be too great, bearing in mind that for a reasonably wide
swath in the ground-range direction the returned pulse cannot be very short,
and pulses must not overlap. The constraint is therefore.
2V V c
— - — <PRF<- (12.2)
Dr Pa 21Psin0
where W is the swath width and V is the platform velocity, again ignoring the
effect of earth and orbit curvature for simplicity. Eqn (12.2) is a severe constraint
on the design of practical SARs, and requires a compromise between improved
azimuth resolution and a wider swath width.
356 Synthetic aperture radar [Ch.l2

12.2.4 SAR processing

The above process can be represented mathematically. The signal received at
time t when the satellite is at ‘x’ (see Fig. 12.11), due to reflection by a target
at position jci along the azimuth of a given range is

___j
Si{t) = ai(xi, ri)exp 1 (12.3)

1
/CO
c
1
where Oy (xi, ri) is the complex amplitude of the echo, containing the phase
and am.plitude of the reflection and carrying information about the backscatter
cross-section at (xi, ry) which is the desired end product, co is the angular
frequency of the radar, and so cj = 2jrc/\R. Now,

r = [rl + ixi-xfX^\ (12.4)

and since (xi — x) < ri, (12.4) can be approximated to

(x-xi)^
r = ri + (12.5)
2ri
(Vt~xif
= r,+
2r,
The total signal received from range ri at time t when the satellite is at x is
the sum of all contributions from targets at (x„, rx) where x„ lies within the real
beam width. Thus
27r(n-x„)ni|
On{Xn> '•l)exp (12.6)
XrTi }\\
The complete signal at S{t) will have contributions from other ranges, as
discussed in section 12.2.2, and is therefore a double summation over azimuth
and range values. The task of SAR processing is to invert this equation to
retrieve a and hence the backscattering cross-section at each (x„, r^).
 East
Anglia

The
English
Channel

Sj
’■^OTV
358 Synthetic aperture radar [Ch. 12

It is important to point out the large amount of information which is

collected by a SAR. Seasat was not able to store any on board but had to
retransmit it to a ground station in real time. In fact it was collecting sufficient
information every second to produce, after processing, the backscattering cross-
section of around 1 million resolution cells. It is only just possible to store the
data at this rate. Processing takes very much longer.
One route towards rapid processing is to use an optical processor. The data
can be stored on photographic film, with the time variations of the signal within
each pulse return being sequenced across the film, and successive pulses being
stored along the film. The use of a coherent light source and optical holographic
technique allows the SAR observing process to be inverted with light instead of
radar waves, and a final image is produced which represents the variation of the
radar-scattering cross-section as light and dark grey tones. This process is rela¬
tively quick and cheap, but suffers several drawbacks. Inevitably, some of the
dynamic range is lost when transferring from digital to photographic analogue
information storage, and it becomes necessary to re-digitise the photographic
product if image enhancement techniques are to be performed on it. Moreover,
the processing is constrained by the optical hardware which must be specifically
designed to achieve a particular operation on the data. Thus corrections for orbit
and earth curvature, for the curvature of the wavefronts, for the doppler shift
due to the rotation of the earth under the satellite, and other minor correc¬
tions can be incorporated in the initial optical processor design, but cannot be
changed later to try to improve the process, or to experiment with alternative
processing strategies. Finally, it is not possible to obtain the available resolution
of 25 m for Seasat, over the whole swath of 100 km, without splitting the pro¬
cessing into four parallel strips, which are very apparent on optically-processed
imagery of wide area. Fig. 12.12 shows an example of optically-processed Seasat
SAR data.
The digital-processing techniques can be much more versatile, and preserve
the radiometric and spatial resolution achieved by the sensor, but are expensive
in computer time. A further introduction to processors and reference to detailed
descriptions will be found in Lodge (1981).

12.2.5 SAR image errors due to moving targets

The theoretical basis of the SAR principle so far presented assumes that the
target field is stationary. Thus range compression is performed on the assump¬
tion that there is no doppler shift introduced on reflection, whilst the doppler
history of a point which is used for azimuth compression assumes that the

Fig. 12.12 - Optically-processed SAR image of the English Channel, 19 Aug.

1978, five hours before high water at Dover. The image was produced at the
Environmental Research Institute of Michigan, and the grid overlaid by Hunting
Surveys Ltd
Sec. 12.2] Principles of SAR operation 359

reflector is stationary. However, the sea surface is in constant motion. It is only

because the sea’s motion is normally relatively slow that the additional effects
introduced by target motion into the aperture-synthesis and image-construction
process can be considered as small perturbations from the basically stationary
state. Thus allowances can be made for these errors, and it will be seen in section
12.3 that the target movement may in some circumstances contribute to the
imaging process for ocean waves. It is still not clear to what extent the random
motions of a very disturbed rough sea render invalid any interpretation of the
SAR image obtained from it.
It is possible to determine explicitly the effect of a steady target velocity
component Wr in the range direction (Fig. 12.13). It produces three possible
effects known as azimuth image shift, range walk, and amplitude reduction.

Fig. 12.13 - Target motion relative to SAR geometry

Azimuth image shift occurs because of the doppler shift introduced into
the reflected signal from a target moving in the range direction. The shift is too
small to be detected by the range-compression process, but becomes apparent in
the azimuth-compression process. It means that every reflection from a target
moving away from the sensor with speed Wr has a frequency reduction of
2mr/Xr in addition to the linearly-decreasing frequency shift due to the
satellite’s motion. In the aperture-synthesis procedure, this results in the target
not appearing in its proper place, but since it has the same doppler history as a
point encountered earlier by the satellite, the location of the moving target is
shifted in the negative azimuth direction. Similarly, a target moving towards the
360 Synthetic aperture radar [Ch. 12

satellite in the range direction appears to be shifted in the positive azimuth

direction. For an airborne SAR for which the ground speed V is the same as the
vehicle speed, the shift Ax is given by
MrR
Ax =-
V
and for a satellite with ground speed V and vehicle speed Vp

Ax J/2

For azimuth image shift to be negligible, |Ax| < therefore

PaV^
(12.7)
RV„
This can give rise to interesting image phenomena when viewing over land.
Trains or road vehicles moving in the range direction appear to be shifted to one
side of the track or road on which they are travelling. It is even apparent in
ocean imaging. Fig. 12.14 shows a ship radar reflection clearly displaced from its
wake and its wave pattern.
Range walk occurs when Mr is sufficiently large to cause the target to pass
through one or more range-resolution cells within the time taken for the beam-
width to pass over it (that is, within the time the target is contributing to
aperture synthesis). The consequence is that full correlation will not occur for
that target within one range cell. Not only will its presence be smeared across the
range cells it has entered, as with an ordinary optical image, but the reduced
correlation within one range cell will cause it to smear in azimuth also. In terms
of the ground-range resolution cell size Pq, and the azimuth resolution p^, we
note that the integration time of a SAR designed to achieve pj^^ is
\R XR

In this time the target must move a distance less than Pq, that is.

2PgPa^
I WrI < (12.8)
XR
for range walk to be avoided.
Amplitude reduction follows from the above interactions reducing the
strength of the signal returned to the satellite, and this further reduces the
ability of the SAR to detect moving targets.
Target movement in the azimuth direction does not have such a striking
effect, but contributes further to image degradation. The effect of is to
change the relative velocity between the satellite and the target, and this there-
 Fig. 12.14 - Seasat SAR image in the English Channel, showing a ship image
(white) displaced from the image of the wake by the effect of its own ipotion on
the SAR processing. Note both the dark ship wake (a smoother surface) and the
characteristic ship-wave pattern. (Reproduced from Raney (1983) by permission
of Marconi Ltd)

fore changes the rate at which the reflected radar frequency decreases as the
target is passed by the satellite. If positive, the frequency decreases more
slowly, and if negative it changes more rapidly. In either case the doppler
history of the reflections from the target are not the same as the reference signal,
and correlation is impaired. The result is termed azimuth defocusing, and the
consequence after processing is for the image to be blurred. An appropriate
criterion for this to be negligible is

2VPa^
«A<
(12.9)
362 Synthetic aperture radar [Ch.l2

Exactly the same effect occurs if there is steady acceleration in the range
direction — as well as a straightforward azimuth shift, defocusing also occurs.
It is clear that the earth’s own rotation must introduce errors due to the above
target motions, but since these are regular, and uniform over an image, it is
possible to make allowances for them, principally by making adjustments to
the reference signal against which the data is correlated. The reference signal can
in fact be thought of as the response of a perfect stationary reflector located at
the point on the earth’s surface whose radar cross-section is being investigated.
A more detailed discussion of the problems of moving targets will be found in
Tomiyasu (1978).

12.2.6 Image degradation due to speckle

When a coherent electromagnetic wave source illuminates a uniform but rough
surface, the resulting image has a speckled, grainy appearance to a greater
extent than if a non-coherent source were used. This phenomenon, known
as speckle, attends the use of SAR and can cause degradation of an image. Its
cause lies in the scattering of the coherent electromagnetic wave over the rough
surface. The backscattered wave consists of contributions from many discrete
scattering elements on the surface. The path lengths to the radar antenna from
the various scattering elements over the area of a pixel can differ by several
wavelengths. The resulting amplitude of the received signal is made up of many
superimposed coherent contributions which have random phase shifts. In some
pixels these will add up constructively, and in others they will cancel, result¬
ing in a random speckle being superimposed over the image, which may hide
variability in the backscatter cross-section which would otherwise be revealed.
Because the scattering facets are randomly oriented, the effect is a truly
random one, and therefore can be reduced by adding together the results of
several independent looks at the same area. This is possible with the SAR. If
the azimuth-compression processing only correlates the received signal with
a portion of the doppler frequency bandwidth, this has the effect of only
synthesising part of the available synthetic aperture. Azimuth resolution is lost,
but the synthetic aperture is divided into several independent segments by
applying separate correlations with different parts of the doppler frequency
range. The resulting SAR images are independent looks at the same scene. They
are independent because the phase fluctuations which cause the speckle change
rapidly with time and look angle, and so give totally different results when
viewing from different parts of the synthetic aperture. This technique is called
‘multilooking’. It is customary in Seasat SAR processing to take four looks at
the same scene, and this gives an acceptable reduction in speckle degradation.
In other branches of physics, optics and radio astronomy for example, the
statistics of the speckle contain information about the scattering surface. It is
possible that SAR images of the ocean could yield more information about
Sec.12.3] SAR imaging of ocean waves 363

surface roughness by this means, but there is no clear evidence of this yet
available. The subject can be examined further in Barber (1983).

12.3 SAR IMAGING OF OCEAN WAVES

12.3.1 A controversial problem
One of the principal applications of synthetic aperture radar in oceanography is
to the study of ocean-surface waves. There is undoubted evidence that SAR is
able to image ocean waves and can give a sensible estimate of wavelengths, as we
shall see in section 12.3.4. However, there is still not a complete consensus
amongst ocean-wave and radar scientists as to the precise mechanisms whereby
waves are imaged, whether waves are always imaged, and what direction and
wavelength of waves are preferentially imaged by SAR. This uncertainty is partly
due to the fact that the information we think we are obtaining from SAR about
wave-fields and the distribution of the troughs and crests of swell is not readily
checked by field measurements. Indeed, the SAR has aroused great interest
amongst oceanographers just because it appears to be able to yield those spatial
data about swell-wave fields which have not hitherto been available from any
other techniques.
Another reason why controversy remains is that there are several possible
processes which may contribute towards the wave-imaging process, and their
relative importance may vary — depending on the radar parameters and the size
and height of waves being imaged. Moreover, to grasp the problem fully it is
necessary both to understand the SAR mechanism in detail, and to be familiar
with recent developments in the study of hydrodynamic wave interactions. It is
partly the shortcomings in the theory of interactions between waves on different
scales which leave room for uncertainty in the theory of SAR ocean-wave
imaging.
The problem of wave interactions arises when the images reveal wavelike
patterns in the sea-surface roughness as seen by the SAR. The roughness is due
to short-wavelength gravity or even capillary/gravity waves, whereas the waves
which are being imaged are of much longer wavelength. Another complication is
the fact that moving targets are involved, with consequent possibilities of
SAR image degradation. The approach that has been taken to get round these
difficulties is to assume that we are dealing with a field of scatterers whose
motion causes only small perturbations to the image that would be received
from a stationary field. The wave-field is represented in a two-scale model,
consisting of short waves of the length scale required for Bragg resonance, and
swell waves of a scale large enough to be resolved by the pixel size of the radar.
The spectrum of wavelengths which exist in between these scales is ignored, and
it is assumed or hoped that they make no coherent contribution to the observed
image.
364 Synthetic aperture radar [Ch. 12

Concerning the hypothesis that it is short gravity waves which produce the
roughness that backscatters the radar waves through Bragg resonance, it is
necessary to assume that the roughness field does not substantially change within
the integration time of aperture synthesis. This is typically between 0.1 and
4 seconds, depending on the radar frequency. It does not matter that several
individual surface ripple troughs and crests may pass a given point in this time,
since what the radar responds to is the presence of a given degree of roughness
within a particular pixel. What is important is that the interaction time of the
whole field of short gravity ripples should be greater than the integration time,
that is, that the ripple field as a whole does not decay, or is not generated, within
the integration time. Tucker (1983) points out that the viscous decay time of
10 cm wavelength waves is around 80 s, of 4 cm waves it is 13 s, of 2 cm waves
it is 3 s, and for 1 cm waves 0.8 s. Thus, provided that the radar requires a Bragg
resonance wavelength of 4 cm or more, the ripples should persist at a nearly
constant ampHtude during aperture synthesis. Under these conditions SAR
images of the sea surface can be regarded as capable of recording meaningful
information about surface roughness. Meanwhile the long swell waves being
imaged have a period of typically 8-20 s, and so although their orbital motions
may influence the SAR, so long as the integration time is around 1 s it is reason¬
able to assume that their troughs and crests stay approximately in the same pbcel
during this time.
There are three distinct types of wave-imaging mechanism proposed:
hydrodynamic interactions, electromagnetic interactions (sometimes called tilt
modulation), and motion effects (velocity bunching and azimuth image smear).
The first of these could be viewed by radar, visible light, or any other method
of observing surface roughness. The second is related specifically to side-looking
radar and Bragg resonance backscattering, but it is not unique to synthetic
aperture radars. The third is intimately related to the aperture synthesis
procedure. We shall consider each process separately in order to gain some
insight into the different mechanisms, but an adequate theory must eventually
consider all the different mechanisms together.
The foundation of any theoretical model is to describe the swell-wave modu¬
lation of the radar backscattering cross-section o in terms of a complex variable,
the Modulation Transfer Function R(ks), where kj is the ocean wavenumber
vector. Thus

a = Oo [1 + /(i?(ks) Z(ks) ~ + complex conjugate) dkj]

(12.10)
Oo is the background value of a in the absence of the waves. The wave-modula¬
tion effect is represented by the product of /?(ks) and Z(ks) which is the
Fourier transform of the surface elevation associated with the longer waves. The
role of the modulation transfer function is simply illustrated in the case of a
Sec. 12.3] SAR imaging of ocean waves 365

monochromatic swell in which there is present a single-frequency train of waves

whose local surface displacement is Then (12.10) reduces to

a = ao(l + |R|f). (12.11)

It should be remembered that the remotely-sensed parameter is o, and an

image should be interpreted as a contour map of a, whilst Z or f is the oceano¬
graphic parameter which we would like to determine using the SAR. Hence the
ultimate aim of theoretical modelling is to determine a form for R so that
(12.10) can be inverted to obtain Z(ks), or in the monochromatic case to obtain
the amplitude and phase of f. R will contain contributions from the different
mechanisms discussed below. Much discussion hinges on whether the proposed
forms of R will give a unique mapping of monochromatic swell-wave phase onto
the image plane, and the extent to which the two-dimensional Fourier transform
of o can be used as a representation of the directional wave spectrum. Alpers,
Ross, & Rufenach (1981) gave a thorough review and critique of current under¬
standing of the problem, updated by Alpers (1983) and Hasselmann et al.
(1984), but it is an area of active research in which new ideas or criticisms of
earlier ones are regularly being proposed.

12.3.2 Hydrodynamic modulation

The ‘hydrodynamic’ mechanism which causes the distribution of radar back-
scatter cross-section to be uneven across the longer swell waves is a term which
incorporates all the fluid-dynamical processes which result in the amplitude
of the surface ripples being modulated by the phase of the swell waves. It is
relatively easy to conceive of possible mechanisms in a descriptive manner.
Fig. 12.15 illustrates the essential features of a two-scale wave—wave interaction
process. It is the fluid ‘orbital’ velocity field associated with the progression
of the swell wave which creates convergent and divergent velocity patterns at
the surface of the sea. These surface movements advect the trains of ripples on
the surface, and the resulting small-scale wave —current interactions increase the
amplitude of ripples near the swell crests following the passage of a convergent
surface velocity field on the rising edge of the wave. There is a corresponding
reduction in ripple energy near the wave troughs. The theoretical solution of this
process has been presented by Longuet-Higgins & Stewart (1964). Since the
patterns of increased and decreased ripple energy are phase locked to the swell,
they give rise to patterns of strong and weak backscatter on a radar image which
correspond approximately to the swell troughs and crests.
This relatively simple theory only applies to low wave heights and represents
weak interactions. As it stands, the theory does not predict a modulation of
surface roughness if the ripples are aligned at 90° to the swell, as would tend
to occur if the local wind were perpendicular to the swell. Since the swell has
propagated long distances its direction need bear no relation to the local wind
366 Synthetic aperture radar [Ch. 12

direction. Valenzuela & Wright’s (1979) second-order theory demonstrates that

modulation still occurs in this situation. The wave—wave interaction model
ignores the effect which the swell has on the wind ruffling of the surface, but
this can be incorporated as a correction to the theory, and tends to increase the
surface ripples on the windward face of the swell (Alpers & Hasselmann (1978)).

SWELL WAVE DIRECTION

Fig. 12.15 — Hydrodynamic modulation of surface roughness by wave—wave

interaction. N.B. the arrows indicate the component of swell-wave orbital
velocity parallel to the surface (ripples not drawn to scale)

The two-scale wave-model approach to studying the surface modulation by

swell makes an assumption about the wave spectrum being in energy equilibrium.
This is more likely to be valid for wavelengths of order 10 cm or more (decimetric
waves) than for shorter centimetric waves, which grow and decay rapidly in
response to gusts of wind or other forcing.
The two-scale wave model is incapable of describing the roughness
modulations in high sea states. Here the wave amplitudes are high, the inter¬
actions are non-linear, and the small ripples may be spontaneously generated by
hydrodynamic instabilities at the peaked crests of almost-breaking waves. There
may still be a correlation between surface roughness and swell phase which
contributes to the radar imaging of the swell, but once wave breaking occurs this
is expected to be destroyed.

12.3.3 Electromagnetic modulation

This mechanism includes any means whereby the presence of the long swell
waves alters the way in which the radar responds to the roughness of the short
ripples of Bragg resonance length. The principal cause of this is the way the
slope of the long waves presents the ripples to the radar at different incidence
angles. Hence this is often called tilt modulation. Fig. 12.16 illustrates the effect,
which occurs irrespective of hydrodynamic modulation and would still occur if
the sea-surface short-wave energy were uniformly distributed over the long-wave
profile. The radar backscatter tends to be strongest from the slope of the wave
facing towards the radar, and weakest from that facing away. Thus an image of
o modulated by tilt alone would represent a plane parallel swell-wave field as
a series of parallel light and dark lines corresponding to slopes facing towards
and away from the radar, that is, displaced by 90° of phase from lines of troughs
and crests. It would not give any modulation at all for waves whose troughs and
crests were parallel to the ground-range direction of the radar, that is, in the
Sec. 12.3] SAR imaging of ocean waves 367

case of a SAR, waves propagating in the azimuth direction should not be imaged
by tilt modulation.

Fig. 12.16 — Tilt modulation

The actual magnitude of tilt modulation depends on the spectral distribu¬

tion of short-wave energy present, since changing the angle of view slightly
changes the length of ripples which will be in Bragg resonance. It will also
depend on the viewing angle of the radar in relation to the mean-surface level.
Fig. 12.17 represents the relative magnitude of the dimensionless tilt-modulation
function calculated from radar interactions with a two-scale wave model
at different radar incidence angles. It shows clearly the variation with incidence
angle, also that horizontally polarised radiation is more sensitive than vertically

Fig. 12.17 — The variation with incidence angle of the dimensionless tilt-modula¬
tion transfer function, for X-band radar, for horizontal (HH) and vertical (W)
polarisation. (After Alpers (1983))
368 Synthetic aperture radar [Ch. 12

polarised to the presence of long waves, and that azimuthally-traveUing wave

components contribute an order of magnitude less to the tilt modulation than
range-travelling waves. Although Fig. 12.17 is for X band only, the L band radar
calculations give a very similar result.

12.3.4 Motion effects

Whilst the a modulations discussed so far would be imaged by any type of
radar, the SAR is sensitive to the motion of the radar-scattering targets. Now
in an ocean-wave field the ripples which scatter the radar are in motion due to
their own phase velocity, due to the surface—parallel velocity field of the swell
waves on which they ride, and to the vertical movement of the surface caused by
the rise and fall of the swell waves (see Fig. 12.18). For typical SAR parameters,
the first two of these are too small to be important, but the third type of motion
is capable of causing image degradation, because of its large component in the
range direction.

Fig. 12.18 — Motion of Bragg scattering ripples on the surface of swell waves due
to (a) the phase speed of the ripples relative to the surface, (b) advection by the
surface component of the swell wave orbital velocity, and (c) the vertical rise
and fall of the swell-wave surface

The effect on the image of target velocity and acceleration in the range
direction has been discussed in section 12.2.5. Target velocity causes azimuth-
image shift. The rising face of a wave is therefore shifted on the image in the
positive azimuth direction, and the falling face in the negative azimuth direction.
For swell waves where troughs and crests lie parallel to the azimuth direction,
this has no imaging effect, but if the direction of troughs and crests has a
component parallel to the ground range direction (that is, propagating in the
azimuth direction), then Fig. 12.19 illustrates how the waves can be imaged. For
the case shown where the waves are propagating in the positive azimuth direc¬
tion, there is a concentration of scattering facets, and hence a brighter image,
above the wave troughs, and a reduction over the crests. For waves propagating
in the negative azimuth direction, the bright and dark on the image would be
Sec.12.3] SAR imaging of ocean waves 369

reversed to correspond to crests and troughs. This imaging mechanism is known

as ‘velocity bunching’, and it can be characterised by a linear transfer function in
the same way as hydrodynamic and tilt modulations.

Fig. 12.19 - Azimuth image shift (velocity bunching) and azimuth-image smear
due to the motion of a single-frequency wave train. (After Alpers (1983))

AZIMUTH DIRECTION
--WAVE DIRECTION

Fig. 12.20 — Nonlinear azimuth image shift. Note that the sequence of points
along the image plane is now different from that along the ocean plane, in
contrast to the linear process shown in Fig. 12.19

Fig. 12.19 offers a powerful imaging mechanism for monochromatic waves

propagating in the appropriate direction. However, it is deceptive in that the
azimuth shifts have been shown to be of optimum size. If the shifts are smaller,
imaging still occurs, but is weaker. If the shifts are larger, and start to overlap
each other, as shown in Fig. 12.20, the imaging mechanism is quickly lost, and
the transfer function becomes highly nonlinear. Given that the azimuth shift has
a magnitude of u^R/V, the appropriate parameter controlling whether velocity
bunching is linear or nonlinear is
370 Synthetic aperture radar [Ch. 12

R du,
^vb ~
Vdxo'

where Uj is the average velocity of surface facets in the range direction during the
SAR integration time, and xq is the azimuthal coordinate in the ocean-surface
plane (N.B. for waves propagating in the range direction, Cyb = 0). Alpers &
Rufenach (1979) suggest that the velocity-bunching mapping is linear only so
long as I Cvbl < 0.3.
The implications of this for the minimum amplitude fg of swell of a certain
wavelength which will result in nonlinear velocity-bunching are illustrated in
Fig. 12.21. fs has been scaled by a geometric factor

[cos0(sin^0 sin^0 -I- cos^0)^/^],

where <f) is the angle between the swell wavenumber vector and the azimuth
direction, and 9 is the radar-viewing zenith angle (as in Fig. 10.3). For
azimuthally-propagating waves, this simplifies to a factor of cos0. For satellite-
borne SARs, R/V is large, and there is clearly a severe limitation on the size of
waves which can be imaged by velocity bunching. In wave-fields with a broad
wavenumber spectrum, nonlinear velocity bunching tends to shift the spectral
peak towards lower azimuthal wavenumbers, because it is able to image better
the longer waves travelling in an azimuthal direction. It is also implied that if the
linearity criterion is not met, imaging by other types of modulation will also be
severely degraded.

128
(SEASAT)

0 100 200 300 400 SOO

SWELL WAVELENGTH.m

Fig. 12.21 — Curves of Cyo— 0-3 as a function of the geometrically-scaled swell

amplitude, Zc = cos(p (sin’(/)sin^0 -i- cos’6)^ and swell wavelength, for different
values of R/V, including Seasat. The region above the curve corresponds to a
nonlinear modulation transfer function for velocity bunching. (After Alpers et al
(1981))

Another motion effect arises from the acceleration of the surface during the
integration time, which produces azimuth image smear. Fig. 12.19 illustrates the
relative amount of smear for different facets, governed by the vertical accelera-
Sec. 12.3] SAR imaging of ocean waves 371

tion field of the swell wave. Whilst it is principally a degrading effect, the
correlation between smear and the wave phase implies that a suitable refocusing
of the aperture-synthesis process could potentially enhance the image to improve
the wave-phase contrasts of light and dark tones. For Seasat, azimuth smear had
an influence on image quality comparable with velocity bunching, but for a SAR
with a shorter integration time its effect is significantly reduced.

12.3.5 High sea states

At high sea states, the above modulation theory cannot be applied. For a start,
degradation by target motion will occur. Furthermore, the relation of the
spatial distribution of Bragg-resonant size wave energy to the long-wave phase is
less predictable in the nonlinear wave interactions which occur, although some
correlation between steeply-peaked crests and increased roughness might be
expected. For waves propagating in the range direction, tilt modulation will stiU
operate, albeit degraded by other effects.
When white-capping and wave-breaking occurs, the sea spray in the air,
and the air bubbles in the sea near the surface, are capable of causing Rayleigh
scattering of the incident radar, and the Bragg scattering mechanism may be
obscured, the more so for shorter-wavelength radars. Foam itself is thought to
be a radar absorber, and so foam streaks may appear dark on a radar image.
Airborne-radar viewing of high sea states reveals strong spikiness in the returns
which is associated with wave breaking, and this is likely to influence satellite
SAR viewing of the ocean at sub-pixel scales. In general it is expected that the
imaging of individual swell waves is less likely to be possible in high sea states,
a conclusion borne out by observations (see Allen & Guymer (1984)).

12.3.6 Directional wave spectra from SAR images

It is relatively easy to produce a two-dimensional Fourier transform of a SAR
image, either digitally or optically, but the question remains whether this bears
much resemblance to the directional wave spectrum. We shall look at actual
Seasat data in the next section, but here we note the various factors which
may influence the image spectra. There is definite directionality in some of
the transfer-modulation functions. The hydrodynamic modulation depends
on the orientation of the ripples, and therefore of the wind, relative to the swell,
although it is independent of the SAR orientation to the swell. It also depends
on the ripple direction relative to the SAR, the Bragg scattering being strongest
when the ripples travel in the range direction. Tilt modulation occurs only
for range-propagating wave components, and so will tend to bias a spectrum
to emphasise such waves. Velocity bunching, on the other hand, only works for
azimuth-propagating waves, and so will enhance azimuth components at the
expense of range components. Moreover, the velocity-bunching mechanism is
372 Synthetic aperture radar [Ch. 12

wavenumber-dependent, and biases the spectrum towards longer wavelengths,

particularly when the amplitudes are high enough for nonlinear velocity bunching
to occur.
Before the launch of Seasat, the expected image degradation due to velocity
bunching led some oceanographers to doubt whether waves could be imaged
at all. With hindsight, we now know that they are imaged in some, but not
all, situations, but it is still not possible to state confidently what is the high
wavenumber cut-off of the imaging process. Because of the complexity of
the modulation-transfer function, it is doubtful if the problem will ever be
completely solved from theoretical principles, and further experience with more
satellite-borne SARs will be needed, along with synoptic wave observations,
before a satisfactory empirical understanding can be gained.

12.4 OBSERVATIONS OF OCEAN WAVES WITH SEASAT SAR

12.4.1 A description of the Seasat SAR
So far we have looked in general terms at the operation of SAR in the oceano¬
graphic context. The actual details of what a SAR can see is strongly dependent
on the design parameters of individual instruments. In particular, the operating
frequency influences many other factors — the size of ripples which generate
Bragg resonance and the aperture-synthesising integration time being two of the
more important ones. The SAR flown on Seasat operated at L Band, whereas the
proposed instrument for the European ERS-1 satellite is to operate at 5.3 GHz
(C Band).

Table 12.1 Characteristics of the Seasat SAR

Satellite altitude 800 km

Satellite orbital speed 6.96 km s"^
Ground speed of satellite sub-point 6.18 km s“^
Angle of incidence of radar beam 19°^ 25°
Imaged swath 100 km wide, centre offset
290 km from nadir
Imaging resolution 25 ->■ 40 m for typical
processing techniques
Radar frequency 1.275 GHz
Radar wavelength 23.5 cm
R.F. bandwidth 19 MHz
Chirp pulse length 33.4 /is
Pulse-repetition rate 1463->>1640 Hz
Sec. 12.4] Observations of ocean waves with Seasat SAR 373

The parameter values for the Seasat-SAR are given in Table 12.1. It is
instructive to discover what these values imply for some of the dependent
quantities discussed already in this chapter. The given view angle of around 20°
and the radar wavelength of 23.5 cm lead to Bragg resonance with waves around
30 cm wavelength, normally described as ripples, but not to be confused with
capillary waves, since 30 cm waves are strongly gravity dependent and are
influenced very little by surface-tension effects.
Inspection wiU show that the pulse length, R.F. bandwidth, pulse-repetition
frequency, resolution ceU size, and swath width satisfy the interrelated
constraints discussed in section 12.2.3, notably (12.2). This is illustrated in
Fig. 12.22. Here it is revealed that the SAR data-recording system must cope
with an added complication. To achieve a pulse-repetition rate so that two pulses
are emitted in the time for the the satellite to travel half its aperture length,
eight pulses must be emitted before the return from the first pulse is received
back at the sensor. The slant range has to be matched to the pulse-repetition
rate so that the emitted and return pulses fit together efficiently.
With a real aperture of about 11m, the SAR is theoretically capable of an
azimuth resolution of 514 6 m. In practice this becomes 25 m when four
independent looks are used to reduce speckle.

Transmitted every Return from 8th

Fig 12 22 - Synchronisation constraints in the Seasat SAR. (Adapted from Beal

era/. (1981))
 374 Synthetic aperture radar [Ch. 12

"O
ctj
(D

O O
00 ’
Week 4: JuM7-23

XO

eg
13 1
<u
Q
H
o
^ ft
Week 3: Jul 10-17 v

-C r
a^
.22 -o
c C
S
a:
z .t:
o >-•

o
C/3 C
cu eg
s
c ^
c
o ^
a S'
.

a> Z
eg
Week 2: Jul 3-9

^ > >
eg r )
*J w fl.
eg
*13
OC
<

bO
ft
C/2 W
a:
Station coverage

o P.
op 5 <wc
eg 00
<D fteg ^
a(
>
o A (U-e
o
(D o
0) cd-c
a ’k‘^
E cdfj
o ft
U O
eg tu.

bo^Ci
rs
„
ft ft1
cgie

• a> H**
Sf o
2 li-
Sec.12.4] Observations of ocean waves with Seasat SAR 375

The SAR operated only when data could be transmitted in real time to one
of five receiving stations at Oakhanger, England; Shoe Cove, Newfoundland;
Merritt Island, Florida; Goldstone, California; and Fairbanks, Alaska. Fig. 12.23
shows the coverage of these stations in the Northern Hemisphere, and also
indicates the ground tracks of all the swaths collected in nearly 500 separate
passes during the 98 days between 4 July and 10 October 1978 on which the
SAR was operated. More detailed descriptions of the SAR data coverage and
examples of the imagery can be found in Beal et al. (1981), Fu & Holt (1982),
and Vesecky & Stewart (1982).

12.4.2 Wave imagery

When Seasat was launched it was well known that waves could be imaged by
SAR from aircraft, but there was some doubt about whether the longer integra¬
tion time of the Seasat SAR was capable of imaging even long low-frequency
swell waves. In the event, a large amount of the processed SAR imagery over the
ocean has revealed linear features at a spacing of several hundred metres which
are assumed to represent the dominant swell-wave patterns. Fig. 12.24 is an
example of such a wave-field observed over the JASIN experimental area 400 km
northwest of Scotland. If the radar image is taken to be a direct representation
of the surface long-wave distribution, it implies a dominant swell wavelength of
about 170 m in a direction 70° E of N. Synoptic observations during the JASIN
experiment observed 152 m waves in a 50° direction. Similarly good agree¬
ment has been obtained for other comparisons with synoptic in situ data. In
Fig. 12.24 the waves are travelling in the range direction. Fig. 12.25 illustrates
azimuth waves, and suggests that it is possible to image some of the azimuth
waves despite, and indeed probably because of, the velocity-bunching effect.
An overall survey by Fu & Holt (1982) of all the available data has led to
the general conclusion that wavelengths greater than 100 m can be detected by
the Seasat SAR provided the significant wave-height is greater than 1 m and the
surface wind speed greater than 2 m s"^ Agreement with surface observations
is within about 15% in wavelength and 25° in direction. Fig. 12.26 shows the
comparisons made between ground and satellite observations from three
independent experiments, and the scatter probably falls within the accuracy
of the ground measurements. However, azimuth waves were generally less well
imaged than range waves, and whilst range waves with amplitudes up to 5 m
could readily be detected, azimuth waves shorter than 200 m could barely be
detected even with m. This is consistent with the increased problems
of velocity bunching for steeper azimuth waves.
On the whole, analysis is limited to those images where swell waves appear.
Where they do not appear on images, it is not always possible to know whether
it is because the swell was too high or too low, in the wrong direction, because
there was no wind to cause ripples, or because the dominant wavelength was
too short. The requirement of wind to ruffle the surface is well illustrated
Fig. 12.24 - An example of range-travelling swell waves in the JASIN area
(Seasat revolution 1049). The satellite track direction is horizontally across the
page. Some internal waves are also evident

Fig. 12.25 - An example of near-azimuth travelling swell waves in the JASIN

area (from Seasat revolution 1087). The satellite track lies up and down the page
378 Synthetic aperture radar [Ch.l2

by Fig. 12.27. The low-wind conditions represented in this image appear to be

transitional. In some areas the wind is sufficient to produce ripples which are
modulated to enable the imaging of the swell-wave field. Elsewhere the surface
is presumably too smooth at the Bragg resonance wavelength, and although the
swell must still be there in the dark patches it is not imaged.
Little direct oceanographic application has yet been made of the wave
images, but there is considerable potential. Fig. 12.28, for example, shows what
is assumed to be the dominant swell pattern around the Island of Foula off the
Scottish west coast. The swell refraction and interference patterns are clearly
visible. Such information can be useful for coastal and offshore engineers since it
shows clearly where wave energy may be focused. It is also potentially capable of
analysis by inverse techniques to obtain the bathymetric profiles which produce
the wave refraction. In remote uncharted areas this could be of significant value.
SAR images of wave fields have also been used to study the wave patterns in
the heart of a hurricane (Gonzalez et al. (1982)) where ground measurement
techniques would probably have failed! The ability to view a wide spatial
field along a swath many hundreds of kilometres long, enables the mesoscale
interactions of waves with currents and eddies to be studied too.

(a) (b)

SURFACE-OBSERVED WAVELENQTH.m SURFACE-OBSERVED WAVE

DIRECTION,deg.true

Fig. 12.26 - Comparison of dominant ocean-wave length (a), and direction (b),
as measured by Seasat SAR and surface pitch-roU buoys. (After Vesecky &
Stewart (1982) copyrighted by the American Geophysical Union)

Fig. 12.27 - SAR image of the Pacific Ocean, 400 km southwest of Vancouver
Island (Seasat revolution 1349), 29 Sep. 1978. Swell waves show up in the brighter
patches where the wind of about 2.5 m s”' is able to roughen the surface suffi¬
ciently to allow the modulation mechanism to occur. Elsewhere the sea is smooth
and appears dark to the SAR. The direction of radar illumination is towards the
top of the image, which is W.N.W.
 Fig. 12.28 - Swell-wave pattern around the Island of Foula, 70 km north west of
Fair Isle in the N.E. Atlantic. The image area is about 30 km square

12.4.3 Wave spectra from the Seasat SAR

A directional wave spectrum is a compact means of presenting the important
physical characteristics of a given wave field. It shows the direction and length
of the dominant train or trains of waves, and when properly calibrated can
indicate the height and energy of the waves too. Directional wave spectra
are difficult to measure at sea since point-measuring wave buoys cannot supply
Sec. 12.4] Observations of ocean waves with Seasat SAR 381

directional information, and the ease with which a SAR image can be Fourier
analysed digitally and optically offers a means of readily obtaining directional
wave spectra.
Fig. 12.29 is an example of the image wavenumber spectrum obtainable
from SAR. It is a polar plot, indicating wave direction, with wavenumber increas¬
ing radially (that is, the longest waves are found closer to the centre). This
particular example shows that there are two distinct wave trains present, since
two distinct peaks of image variability can be detected, one corresponding to
waves of length 177 m in a direction of 290°, and the other, smaller peak cor-

Spacecraft
velocity vector

[!□ IZD □□ ES H
0-2 2-4 4-6 6-8 8-10

Imaye enerijy
density (linear scalel

Fig. 12.29 - An example of a SAR image-derived directional wavenumber

spectrum within the wavenumber range 27r/400m'* to 27r/67 m'*. There is a
200 m swell system coming from the E.S.E. and a less distinct 100 m swell from
E.N.E. (From Beal et al. (1981))
382 Synthetic aperture radar [Ch. 12

responding to waves of length 93 m and direction 270°. We note that the polar
plot is symmetrical, and there is a 180° ambiguity in direction. Generally this
is easily resolved by assuming that waves are travelling in the direction of the
recent surface wind, if known, and are most likely to be travelling from the open
ocean towards the coast, if there is a coastline nearby. In this case, just north¬
east of Cape Hatteras on the U.S. east coast, there was meteorological and
oceangraphic evidence (Beal (1981)) to confirm that the SAR spectrum was
representing the principal features of the actual wave field.
However, there are serious shortcomings in obtaining wave spectra from the
SAR, and much research needs to be performed before SAR-measured spectra
can be confidently used for oceanographic interpretation without supporting
sea measurements. It is tempting to assume that a directional spectrum such as
Fig. 12.29 is the same as a directional wave-energy spectrum, but there remain
questions concerning all aspects of the information content: wavelength, wave
direction, and wave amplitude. This is to be expected, considering the complexi¬
ties of the SAR wave-imaging process which have already been discussed. The
spatial resolution of the SAR severely limits the high-wavenumber resolution,
whilst it is still not clear whether the SAR spectrum should correspond to the
wave-height or the wave-slope spectrum, since the different imaging mechanisms
rely on both to varying extents. Fig. 12.30 shows a comparison between an
omnidirectional SAR image spectrum and wave-slope and wave-height spectra
obtained from a pitch-roll buoy measuring at the same time as a Seasat overflight
during the JASIN experiment (Vesecky& Stewart (1982)). At short wavelengths,
there is better correspondence between the SAR data and the wave slope, but in
the more important peak area of the spectra, we see that the SAR peak is shifted
to a longer wavelength than the wave-height peak, which is itself at longer wave¬
lengths than the slope spectrum. This accords with our general expectations that,
being able to image the longer waves more readily, the SAR spectrum is biased
to lower wavenumbers. However, the shape of the SAR spectrum is broader than
the ground-measured spectrum, which could well be the result of nonlinearity
in the modulation process shifting some spectral energy to higher wavenumbers,
and thus reducing the steepness of the spectral slope at wavenumbers above the
peak values.
The spectra displayed in Figs 12.29 and 12.30 were of wave systems
propagating primarily in the range direction. Are azimuth waves observed as
clearly in spectra? The work of Beal et al. (1983), examining the variation in
SAR image spectra along a single swath where the changing character of the
waves was known or could be readily assumed from the topography and weather,
reached the conclusion that as the waves turned from range to azimuth orienta¬
tion, the SAR was less able to measure them, and consequently the directional
spectra became distorted. Not only was there a bias towards lower wavenumbers
but also the spectral peaks were biased towards the range direction. Thus while
it is still possible to gain some useful oceanographic insights into the variability
Sec.12.4] Observations of ocean waves with Seasat SAR 383

of the wave field and its spectrum over the length scales of mesoscale dynamical
ocean processes, it is best to regard the information as qualitative rather than
quantitative. It requires further theoretical study of the wave-imaging process,
and further experience from synoptic satellite and surface-wave measurements,
before it will be possible to define a nonlinear transfer function which relates the
ocean spectra to the SAR image spectra.
The same is true if wave amplitude is to be determined from the amplitudes
of the image spectra. The problem is greater here because the density of the
image depends on the ripple roughness, and hence the local wind speed, as well
as the swell-wave amplitude. However, an appropriate measure of hij^ may be
obtained by comparing the height of the spectral peak to the low-wavenumber
noise. The peak-to-background ratio (PBR) is indicated as the signal-to-noise
jiatio on Fig. 12.30, and it appears to be possible to relate this empirically to the
significant wave height, to an accuracy of ±0.75 m.

WAVENUMBER,m

Fig. 12.30 - Comparison between a SAR image spectrum and the wave-slope and
wave-height spectra of coincident data from a pitch-roll buoy in the JASlN area,
4 Aug.1978. The SAR image and wave-slope spectra have been normalised so that
their peaks have the same height as the wave-height spectrum. (From Vesecky
& Stewart (1982), copyrighted by the American Geophysical Union)

We may conclude this section on the measurement of waves using SAR by

noting that the Seasat experience has opened up a range of exciting possibilities
and has presented us with new opportunities for studying wave fields, but there
still remain serious questions concerning how to interpret the SAR data in terms
of ocean-wave statistics. Because the images appear visually to look like typical
wave fields, there is the greater danger that environmental scientists will assume
that these image patterns correspond exactly to troughs and crests of real ocean
waves. In fact we can with confidence say only that they nearly do this for range
waves, and occasionally do so for azimuth waves.
384 Synthetic aperture radar [Ch. 12

12.5 INTERNAL WAVES IMAGED BY SAR

12.5.1 Observations of internal waves in Seasat data
Of the oceanographic phenomena which have been discovered to have a SAR
signature, after surface waves the most widespread and commonly occurring
on the Seasat imagery has been internal waves. It might seem surprising that
a surface-viewing satellite sensor should be able to detect dynamical processes
which by definition have their centre at, and owe their existence to, vertical
density gradients within the sea and particularly at the thermocline between
10 m and 200 m deep. However, the evidence is plain on many Seasat SAR
images such as those shown in Fig. 12.31 of the North-East Atlantic, and in
Fig. 12.32 of the sea off Portugal. The swell-wave field is evident, particularly
in Fig. 12.32, but superimposed are groups or packets of dark and light banding
with a wavelike form. Their spacing is far too great to be that of surface waves,
but they have length scales appropriate to internal waves. Since similar features
can be observed in SAR imagery in locations where internal waves are known to
exist (for example, Trask & Briscoe (1983), Hughes & Gower (1983)) then it is
very reasonable to assume that these patterns are indeed the surface signatures
of internal waves.
Surveys of the available imagery indicate these features to be ubiquitous.
They are found in aU depths of water (though often near to bottom topographic
features which might contribute to their generation) and at most wind speeds,
though they are more frequently evident at lower wind speeds. They also appear
to be observed lying parallel to both range and azimuth directions. The mani¬
festation of internal wave trains on SAR images is of considerable interest
to oceanographers. It is very difficult with conventional measuring techniques
to sample internal waves at more than a few isolated points, obtaining temporal
frequency measurements and a measure of the vertical amplitude of internal
motions but not being able to measure wavelengths, orientation, and direction
of propagation or to picture the patterns of constant phase lines and the extent
of the wave train. If, as seems likely, the grey-tone patterns on the SAR
images can be confirmed to correspond to trough and crest patterns of internal
waves, then a great deal can be learnt about internal waves from satellite data.
Of course, internal waves come in many sizes, shapes, and frequencies, and
it is only a small subset of these which appear to be imaged. The principal
features of those internal waves found in SAR imagery can be summarised as:
(a) The waves are found in groups or packets with 4 to 10 crests per group.
(b) The crests/troughs are often parallel to the bottom topography or else
radiate out as if from a source region or point.

Fig. 12.31 - Optically-processed SAR swath of the North-East Atlantic covering

the Wyville-Thompson Ridge and the Hebridean Shelf. Packets of internal waves
can be seen throughout the image
Sec.12.5] Internal waves imaged by SAR 387

(c) The wavelength between individual light and dark bands is typically between
several hundred metres and several kilometres, and usually decreases from the
leading wave in a packet to the trailing edge.
(d) The separate groups of waves are typically tens to a hundred kilometres
apart.
(e) The crests (or surface manifestations of a constant-phase line) are usually
tens to hundreds of kilometres long, and very often the lengths of crests (as
revealed on images) decrease towards the rear of the wave group.
(f) These waves appear either as dark in a light background (presumably under
rough-sea conditions), as light in a dark background (calmer conditions), or as
dark and light bands in the intermediate case, suggesting that internal waves can
be imaged over a broad range of wind conditions.

12.5.2 Internal-wave theory

Many texts describe the dynamics of internal waves (for example, Neuman
& Pierson (1966), Le Blond & Mysak (1978)). Here we assume that for the
internal waves to have a surface signature, they must be essentially long waves
with respect to the depth of the thermocline along which they propagate. If
their frequency is coiw rads s“^ and their horizontal wavenumber is kjw, they
propagate with a velocity field as shown in Fig. 12.33. For a simple two-fluid
case, with layers of density Pi and depth hi above the thermocline, and density
P2 and depth /za below, the dispersion relation is:

2
ghiki^{p2-Pi) (12.12)
COiw
Pi “b Pi ky^hi coth kiy^h2
if the lower layer is deep. If the lower layer is also shallow compared to the
wavelength this simplifies to:
ghih2iP2~ Pi)kiw (12.13)
COiw^ —
P2^1+ Pl^2
ghih2{P2~ Pi)km (12.14)
or <^iw — / ■*■
P2hi+ P\hi
when the frequency is too low for the Coriolis parameter / to be ignored (that
is, for wave periods of order half an hour or more).
These dispersion-relation models provide an approximate framework for
extracting useful dynamical information from the spatial measurements of
internal waves which can be made from SAR images, as discussed later in
section 12.5.4.

Fig. 12.32 - Packets of internal waves revealed by a SAR image off the coast of
Portugal. Some correlation can be seen between the internal wave-crest directions
and the bottom contours shown. Swell waves are also visible at a smaller scale,
and brighter patches occur where they break near the beach. (From Allan (1983))
388 Synthetic aperture radar [Ch. 12

INTERNAL WAVE
DIRECTION
surface convergence slicK
\

Fig. 12.33 - A sketch of an internal wave showing the fluid-particle velocities

and the surface-convergence zone

For the simplest case (12.13), the waves are non dispersive, and the phase
and group velocities are both equal to:

ghih2{p2
Cm — Qw “ (12.15)
P2 ^1+ Pl^2
The surface velocity (that is, the wave-orbital velocity) associated with an
internal wave, at a point where the thermocline is displaced a distance fiw above
its rest position is:

_ fiw<^iw
“iw - ^ (12.16)
hi
Whilst linear internal-wave theory goes some way towards explaining the
observed phenomena, if the waves are strong enough to generate a surface
signature, then it is likely that in many cases they are of such a large amplitude
as to be nonlinear. Because the density gradient along which they propagate may
be very small, very little energy is required to generate a large-amplitude wave.
In this case, the Korteweg de Vries equation must be solved, and an appropriate
solution is a solitary wave or a train of internal solitons (Osborne & Burch
(1980)). Fig. 12.34 illustrates the internal soliton. Its profile is given by the dis¬
placement f js of the thermocline above its mean level, in terms of the maximum
displacement fo below the mean level, as:

= -fosech^ (12.17)

fo(^2 hi)
The speed of the soliton is c - Ciw (12.18)
2 hih2
Sec.12.5] Internal waves imaged by SAR 389

where cjw is the speed of linear waves given in (12.15) and the characteristic
horizontal length scale of the solition given by
V2
4 hi hi
Lc —

INTERNAL SOLITON DIRECTION

Fig. 12.34 — An internal solitary wave, with fluid particle velocities shown by
the arrows

Although they are nonlinear solutions, solitons can pass through each other
without change, and a train of internal solitons can exist as shown in Fig. 12.35,

•urfac* convarganca zonaa^

/ I \

Fig. 12.35 - A train of internal solitons

Their spacing is discussed in detail by Osborne & Burch (1980), but the
significant result of the theory in connection with the SAR images is that the
leading soliton of a group must be of greatest amplitude, and subsequent solitons
390 Synthetic aperture radar [Ch.l2

become both smaller and more closely spaced. This behaviour resembles the
features observed on many of the images, where a packet of internal waves has
a gradual decrease in wavelength towards the back, and the front crests are
longer than the rear ones, suggesting they are probably of greater amplitude.
The wavelengths encountered in the packets imply that the period of the
internal wavelike motion is between a few minutes and an hour. The spacing
of the separate packets of waves can be accounted for by the hypothesis that
the waves are generated by tidal flow over topography. Very plausible theories
have been advanced by Lee & Beardsley (1974) and Maxworthy (1979). They
suggest that tidal flow over a topographic feature sets up a natural standing wave
of large amplitude. When the tidal flow slackens and reverses, the downward
displacement of the thermocline propagates away as a wave, and if it is of
sufficient amplitude, the nonlinear soliton theory predicts that it will split
into a train of solitons. A related suggestion by Osborne & Burch (1980) is that
an internal solitary wave may be produced by tidal interaction a long way away
and propagate into a region of shoaling topography where the new depth
constraints cause the solitary wave to split up into a train of solitons. Either
way, it is reasonable to suppose that the time interval between generation of
separate packets is the tidal period, normally 12]6h, but sometimes 24—25 h
where diurnal tides dominate.

12.5.3 Surface-roughness imaging of internal waves

The discovery of so much internal-wave evidence on SAR imagery was not
completely unexpected. Internal waves had already been observed by their
surface-roughness visible-wavelength signature on occasional Landsat scenes
(Apel et al. (1976)) and from aircraft (Hughes & Grant (1978)). Furthermore,
the existence of surface slicks associated with some internal waves had been
well documented from ship observations (for example. La Fond (1962)). The
probable explanation is that the surface signature is related to the convergence of
surface velocities found at the surface above the falling slope of the thermo¬
cline, behind the internal wave crest (see Fig. 12.33). The problem of modelling
the generation of surface roughness associated with internal waves is not unlike
the two-scale surface wave problem discussed in Section 12.3 where swell
modulates the surface ripples, except that wave—wave interaction is now between
internal and surface waves.
Apart from the straining of the wind-generated surface ripples by the
internal wave-velocity field, which redistributes the ripple energy in phase with
the internal wave, there are other related explanations for a surface signature.
A surface-convergence zone can accumulate flotsam and surface films which
might lead to a smoother rather than a rougher surface, but it could still have
the potential to be imaged by radar. Alternatively, the convergence might be so
Sec. 12.5] Internal waves imaged by SAR 391

vigorous in the case of a large-amplitude internal soliton, and the surface velocity
sufficiently large, as to generate surface ripples by turbulence. In this case any
wind-driven waves or swell already present, might steepen to cause breaking and
white-capping which would have an enhanced roughness signature. A graphic
description of this type of effect, which they describe as a ‘rip’, is given by
Osborne & Burch (1980).
Whilst possible mechanisms clearly exist for imaging internal motions by
surface roughness, there is still a need to find out more about the details of them,
such as the phase relation between the internal-wave crests and the rougher/
smoother patterns, or the wind conditions under which they will be visible, or
the ripple size which is generated.
Given the horizontally variable surface velocity (12.14), the possibility
arises of velocity bunching being an imaging mechanism. The wavelengths
are probably too long for this, and the internal waves have been observed both
in the range and the azimuth direction, but a complete study of the imaging
process will have to consider what effects the surface motions will have on the
SAR. Until the mechanism is fully understood, it remains reasonable to assume
a one-to-one correspondence between light and dark wavelike patterns on the
image, and internal wave crests and troughs, or vice versa. It appears that internal
waves are readily seen by satellite SAR because of the advantages of an elevated
viewing point from which to survey their spatial structure, and because of the
ability of radar to detect subtle changes in surface roughness more readily than
visible wavelength sensors can.

12.5.4 Oceanographic information content of SAR images of internal waves

There is considerable potential to extract useful oceanographic information from
the spatial measurements that can be made of internal waves on SAR images.
It is necessary to assume that a particular linear or nonlinear propagation model
applies, as discussed in section 12.5.2. For example, by measuring the packet
spacing, and assuming a tidal periodicity between them, the wave-group velocity
can be calculated, which enables an estimate to be made of either the depth
of the thermocline, or the density contrast, but not both (see eqn (12.13)).
If the internal soliton theory applies, then the spacing of the crests and
troughs may yield information about the amplitude of the internal waves. Some
attempts have been made to explore such ideas (for example, Apel (1981)), but
what is needed is the observation of internal-wave systems simultaneously with
satellite SAR overflights. When this can be done with future SARs it may open
the way to deducing oceanographic information about thermocline movements
which would at first sight have seemed totally unrelated to the potentialities of
radar. Here is an example of the creative use of satellite data which may make a
contribution to oceanography in an unexpected way.
392 Synthetic aperture radar [Ch. 12

12.6 OTHER OCEANOGRAPHIC PHENOMENA IMAGED BY SAR

Several other oceanographic phenomena were apparent in Seasat SAR imagery,
some anticipated and some not. There is room here only to mention them
briefly, but they each represent the potential for much more research when
another satellite SAR is launched in future.
One of the most spectacular and unexpected of Seasat SAR results was the
occasional image which revealed sea-bottom topography. The clearest examples
are images of the Straits of Dover (reproduced in Fig. 12.12, expanded in
Fig. 12.36), the Bristol Channel, and the Nantucket Shoals. There seems to be
a direct correlation between light and dark patterns and the slopes of the
bedforms, so that the image appears as a relief map of the bathymetry. There is
insufficient evidence from the limited Seasat dataset to conclude over what
range of wind, wave, and tidal conditions the bathymetry is revealed by the radar.
Presumably the SAR is detecting surface-roughness patterns produced by the
straining of surface ripples by horizontal tidal-velocity gradients which occur as
the tide flows over the relatively shallow topography (about 30 m deep or less).
The roughness could also be due to turbulence generated over the topography,
or, less likely, to the motion effects of the tidal streams on the SAR processing.
It is ironic, and a comment on the oceanographic potential of satellite micro-
wave remote sensing, that it should throw up such a fascinating and unexpected
problem, which defies immediate solution, in one of the most-travelled stretches
of sea in the world.
Other phenomena revealed by the Seasat SAR include the detection of
current patterns, eddies, and gyres, principally by their influence on the surface
wave field. Examples are presented in some detail by Beal, De Leonibus, &
Katz (1981). Ships, their wakes, and their associated wave trains are also readily
detected, as in Fig. 12.14. The wake of ships, after the immediate turbulence
following the ship has passed, is usually smoother than the surrounding sea, as
discussed in section 10.6. Hence the wake appears darker on a SAR image.
Superimposed upon this is the classic ship-wave pattern, and the ship image
which is shifted in azimuth because of its motion.
SAR is also capable of recording spatial variability in the surface winds over
the ocean on length scales significantly smaller than revealed by a scatterometer
or scanning microwave radiometer. Since the Bragg resonance ripples are driven
by the wind, the radar backscatter (that is, the image brightness) is related to
wind speed, and under suitable conditions can reveal meteorological disturbances,
as described by Ross (1981). This could turn out to be particularly useful in
coastal regions.

Fig. 12.36 - A portion of the image shown in Fig. 12.12. The letters correspond
to sandbanks; S.F., South Falls; S. Sandettie; O.R., Outer Ruytingen; W.D.,West
Dyck. The tidal current was flowing towards the south-west, with a speed at the
surface of around 0.6 to 1.0 m s"*. The wind speed was about 25 km/h ►
 J,

y 4:>^-$f

Hti.y 3

^ ^ *■

4' **-c3v
€
«S^'*S^'^“I**
K--/ >
^I 4**'*
-A:'|jA3t.<’V

■H.
394 Synthetic aperture radar [Ch. 12]

Although Seasat was an experimental satellite, the prospect of future SARs

being flown raises the question of the potential of the instrument in offshore
marine technological applications. Wave-refraction patterns, wave spectra,
internal waves, tidal-current visualisation, offshore winds, bottom bathymetry,...
all of these are phenomena of direct interest to offshore-exploration and coastal
engineers. It is not surprising that Wadsworth et al. (1983) concluded after a
survey of the possible applications to the oil industry that “the SAR was of
great promise and could quickly become a part of the tools used on a routine
basis in offshore operations”. It will be an even better tool when oceanographic
and radar scientists have discovered exactly how and under what sea-state
conditions it works!
 CHAPTER 13

Microwave scatterometers

13.1 INTRODUCTION
In this, the final chapter to be devoted to a particular type of satellite sensor, we
look at the radar scatterometer. This is the name usually given to an oblique-
viewing active microwave device, which measures the backscattered radar energy
from a fairly broad sea-surface area illuminated by a long pulse of energy
at a particular frequency. No account is taken of the timing, the phase, or
the coherency of the return signal in comparison with the emitted pulse. The
amplitude of the return signal is interpreted empirically as a measure of the sea-
surface roughness. Depending on the frequency of the radar, and hence the wave¬
length of the Bragg resonant surface roughness, as discussed in section 10.2.3,
the magnitude of the return can be related to either the surface-wind speed
and its related stress, or the surface-wave field, leading to the descriptions of
‘wind-scatterometers’ and ‘wave-scatterometers .
Most practical experience of satellite scatterometry to date has been gained
from the Seasat scatterometer, although a scatterometer was flown on Skylab
in 1973 and several aircraft experiments were flown to prove the scatterometer
concept’ prior to Seasat’s launch. The Seasat-A Satellite Scatterometer (SASS)
operated at a frequency of 14.6 GHz (wavelength approximately 2 cm). Thus
Bragg resonant backscattering occurred for waves of about 3 cm length. These
are surface capillary-gravity ripples which are sensitive to the surface-win
stress, and tend to line up with their crests perpendicular to the wind direction.
Experiments have shown that the backscattering cross-section measured at the
SASS wavelength is strongly correlated with the wind-stress vector, and y
comparing the backscatter for the same area of sea viewed from two different
directions at right angles to each other it is possible to deduce the wind direction.
Although surface-wind stress is the most useful parameter for oceanographers
and meteorologists (since it provides a boundary condition for models of ocean
and atmosphere), and although in principle the wind stress is more clo^se y
coupled to surface roughness than is the wind speed, it was m fact surface-
wind speed for which algorithms were developed to interpret the SASS da a.
The reason for this was that to produce a valid empirical algorithm to calibrate
396 Microwave scatterometers [Ch. 13

the SASS required a broad dataset of ground measurements for comparison.

Since surface-wind observations are routinely made for weather reports from
shipping, while stress is extremely difficult to measure even with a fully-
instrumented research vessel, there was no alternative but to base the algorithm
on the surface-wind data.
Because of the experience gained with Seasat and published in the scientific
literature this chapter concentrates largely on wind scatterometers, and the
SASS in particular. However, it should not be overlooked that a scatterometer
operating at lower frequencies supplies information about longer waves in the
ocean, is not so dependent on the surface wind, and will contain a different type
of oceanographic information, principally concerning the spatial distribution of
gravity-wave roughness. A type of wave scatterometer is expected to be flown
on the first European earth-monitoring satellite ERS-1.

13.2 WIND SCATTEROMETRY

13.2.1 Surface-wind stress and the near-surface wind-velocity profile
The difficulty of relating surface stress to the surface-wind speed lies at the heart
of the empirical calibration of radar-wind scatterometers, and we shall therefore
briefly review the way in which ocean—atmosphere boundary-layer scientists
have approached the problem. It is not only in radar viewing of the sea-surface
roughness, but in most other aspects of air-sea interaction science, that this
problem is faced. It is the surface-wind stress on the sea surface which drives
the dynamics of the boundary layer and is therefore expected on physical
grounds to be related most closely to the generation of surface waves, the
production of wind-driven ocean-surface currents, and the stirring processes
which keep the upper ocean well mixed down to the thermocline. However, it is
extremely difficult to measure t^, since it requires accurate observations of the
turbulent motions near the sea surface to enable the measurement of Reynolds
stress terms such as pu'w' (the correlation between the vertical and horizontal
turbulent velocities).
The wind stress can be expressed as

ry, = pU*^ (13.1)

where U* is known as the friction velocity, and is used by oceanographers and

meteorologists to parameterise the effect of surface stress. Since the wind
velocity C/(z) at a height z above the sea surface is routinely measured by
weather ships, oceanographic research vessels, and by commercial shipping
as part of their routine weather reports, it would be useful if U* could be
estimated from U{z).
An alternative is to express the wind stress as

(13.2)
Sec.13.2] Wind scatterometiy 397

where Pa is the air density and Co.h is the dimensionless drag coefficient,
appropriate for calculating stress from a wind velocity measured at height h.
This gives an easily-calculated approximate estimate of since from many
observations it has been established that

Cd,io = 0.0013 ± 0.0003 (13.3)

for wind speed measured at 10 m. However, this does not cater for measurements
made at different heights, and does not allow for the variation of due to the
stability of the air column, which alters the wind-velocity profile. To cater for
this, the Monin-Obukhov equation is used to define the velocity profile;
U*
U{z) - Us (13.4)
K
Here Ug is the sea-surface velocity which is normally small compared to the wind
speed a few metres above. K is vonKarman’s constant (normally assigned a value
of 0.4). zq is a parameter known as the roughness height, which for a solid
surface is dictated by the height of surface-roughness perturbations, but is not
directly related to wave height because the sea surface is in motion. Over the sea,
an accepted empirical expression for zq is
0.0156 U*^
Zq = -m. (13.5)
g

i// is a correction to the logarithmic profile to allow for the stability of the atmos¬
phere. L is the Monin-Obukhov length, and represents the height above which
stability effects are important. In terms of the appropriate bulk coefficients, L
is given by

L =
KgAT
where 7^ is the atmospheric temperature at the same reference height as that
used for U and Cp. AT is the air-sea temperature difference, AT =
degK. Typically, L is between 30 and 100 m. The form of i// is debatable, as
indeed are many of the empirical formulae used in this whole field of study, but
typical form is,
Z—Zo z —Zo
for stable atmospheres, L> 0, i// = 4.7
L L
Z — Zq Z—Zq
for unstable atmospheres, L<0, \p = 3.8
L L
Thus to obtain U* accurately from a measured wind speed at a particular
height, it is necessary to use (13.4) with an estimate of Zq from (13.5), using an
anticipated value for U* which may need subsequent correction. Whilst the
forms of the equations used in air-sea boundary layer theory have a sound
398 Microwave scatterometers [Ch. 13

physical basis, their implementation is based on empirically-derived constants.

This, and the influence of stability on the wind profile, ensure that no simple
relation exists between U* and the measured wind.
Consequently a decision is faced by the algorithm designer who is tasked
with interpreting the scatterometer-measured radar backscatter cross-section in
terms of sea-surface wind or wind stress. Should the algorithm relate radar back¬
scatter to the wind stress? This is physically a more satisfactory choice, since
there is a close causality between wind stress and the ripples which generate the
backscatter, even though the wind stress cannot be readily measured except
through reliance on the rather complex and sometimes imprecise calculations
presented above. Or should the backscatter be related to the surface wind at
a given reference height, which is readily measured with good accuracy? In this
case there is bound to be a scatter of values around any chosen algorithm because
of the influence of atmospheric stability altering the relation between the wind
speed and the friction velocity (and hence the shear stress and the result¬
ing surface ripples). Recent results by Smith (1980) and Large & Pond (1981)
have done much to consolidate the empirical results concerning the wind-speed
profile and its relation to friction velocity. Unfortunately, these results were not
available when the SASS algorithms were developed. Given the wide availability
of wind-speed data, the algorithms were constructed to give the 19.5 m height
wind vector under conditions of near-neutral stability. When in situ measure¬
ments were used to validate the algorithm they were converted to an equivalent
19.5 m near-neutral wind.

13.2.2 The development of wind-speed and wind-stress algorithms

for interpreting radar backscatter measurements
The experience of relating radar backscatter to surface winds and wind stress, on
which the design of the successful scatterometer on Seasat could be based, was
gained over a decade or more of aircraft experiments and the Skylab mission.
The result has been to produce an extensive dataset of the value of o°, the
normalised radar cross-section of the ocean, and its dependence not only on
the wind and the sea state but also on the vertical-incidence angle of view 0, the
radar wavelength, the radar polarisation, and the azimuth angle 0 between the
wind direction and radar look angle. The data have been fitted to empirical
power-law relations of the form a° = aU°^. The most satisfactory fit has
been found between o° and U* rather than U, as shown in Fig. 13.1 (Jones &
Schroeder (1978)). The empirical equation is

logio({^°) = —48.59 + 19.966 logio(f/*) dB, that is, a. = 1.9966.

The fact that a fairly good fit can be achieved demonstrates that a scatterometer
should be capable of measuring wind stress. The steeper the gradient on the
graph (that is, the greater the exponent a), the more sensitive the radar is to
wind-speed variations. Thus the result shown in Fig. 13.2 (also based on data
Sec. 13.2] Wind scatterometry 399

presented by Jones & Schroeder) is useful. It demonstrates that the model is

more sensitive to wind speed when the Bragg resonant wavelength Xg (see
eqn (10.4)) is smaller. In fact a varies inversely with Xg. Xg depends on both the
viewing geometry and the radar frequency. Clearly, the higher-frequency radars
give the best response, those at 13.9 GHz which resonate with 2—3 cm surface
ripples being ideal.

Fig. 13.1 - The variation of the normalised radar backscatter cross-section with
the friction velocity m* of the wind over the sea surface, for a horizontally-
polarised 13.9 GHz scatterometer viewing at a zenith angle of 0 = 40° in the
upwind direction. The fitted line is
logio(a°)= -48.59027 + 19.9658 logioU*dB.
(Based on Jones & Schroeder (1978))

2 Q Frequency.QHz
A 13.9
V 13.3
• 8.9
oc ■ □ 4.6
♦ 1.3
■ 0.4
1.0 - ♦

»_ _1
oLjl-
0.01 0.1 1.0

Fig. 13.2 - Variation of the wind-speed exponent a with the Bragg resonant
wavenumber (1/ X®)* upwind viewing and vertical polari^tion. Different points
for the same radar frequency correspond to different viewing angles. (After Jones
& Schroeder (1978))
400 Microwave scatterometers [Ch. 13

That ripples of 1—5 cm should give a radar return strongly linked to

wind speed over a large range of wind speeds is rather surprising, bearing in mind
(see section 10.3) that the high-frequency end of the ocean-wave spectrum is
thought to saturate at quite low sea states (that is, at low wind speeds). The
fact that the radar backscatter continues to increase is perhaps related to the
growth of longer waves, enhancing the Bragg resonance through tilt effects. A
full theoretical analysis would require investigation using the two-scale models
discussed in the previous chapter in the SAR wave-imaging context. Further
response at higher wind speed might also be due to Rayleigh scattering from
spray and air bubbles, or it may even be due to localised enhancement of the
short ripples near the peaks of nearly-breaking waves, in which case the ideas
about a saturation range in the spectrum are not entirely valid. Whatever
the reasons and whether or not the mechanisms are understood. Fig. 13.1
demonstrates that there is a sound empirical basis for scatterometer algorithms.
The sensitivity of to the viewing direction and to the radar polarisation
is shown in Fig. 13.3. There is a very clear directional response, with a 7 dB
difference in o° between the maximum, when the viewing direction is aligned
with the wind, and the minimum, when the radar views in the crosswind direc¬
tion. There is also a small reduction in a° for downwind viewing compared with
upwind viewing. This anisotropy provides the basis for obtaining wind direction
from a satellite scatterometer. The anisotropy is greatest for the short Bragg
resonant wavelengths in the capillary range of 1—4 cms. Fig. 13.3 does not show
much difference in directional response between horizontally and vertically
polarised microwaves.
-10

-14 . a
e • • •

-18 + a •.K +
+ a e 4.+ *
cr\ dB 4 .• %
••4"
♦ 1-4 4
f 4
-22

a vertical
polarization
4- horizontal
-26

croaawind downwind croaawind

-30 . I ■ -I-1-1-1 I I t I
90 180 270 360
Radar azimuth relative to upwind direction, degreea
Fig. 13.3 — Variation of the normalised radar backscatter cross-section with the
direction of the radar azimuth relative to the upwind direction. Typical vertical
and horizontal polarisation results are shown for a 13.9 GHz radar at 40°
incidence angle. (From Jones & Schroeder (1978))
Sec. 13.3] Experience with the Seasat scatterometer 401

From these and other observations, which can be studied in more detail in
Jones & Schroeder (1978) and Moore & Fung (1979), the conclusion is drawn
that a radar at about 14 GHz, with a viewing-angle range centred at 40°, should
be the most satisfactory device for measuring the speed and direction of the
wind stress, and hence the wind speed.

13.3 EXPERIENCE WITH THE SEASAT SCATTEROMETER

13.3.1 The SASS design and operation
The Seasat Scatterometer was a radar operating at a frequency of 14.6 GHz,
with four antennae emitting fan-like beams as illustrated in Fig. 13.4. The
four main beams pointed along azimuths at ±45 (forward-looking) and ±135
(backward-looking) to the satellite track. Each beam had a spread of approxi¬
mately 0.5° wide X 25° high, so that it illuminated a narrow beam on the ground
at incidence angles between 25° and 55 . The instantaneous illumination from
all four beams was an X shape but, given the satellite’s motion, two swaths of
500 km width were covered, leaving a gap between them of 400 km underneath

Fig. 13.4 - The Seasat scatterometer illumination pattern and swath definition,
showing the constant doppler lines which define the resolution cells along the
beams
402 Microwave scatterometers [Ch. 13

A secondary nadir-viewing beam with a ±8° zenith angle field of view

illuminated a swath 140 km wide centred on the satellite sub-track. This was
able to record surface roughness beneath the satellite, but could give no wind-
directional information. Being nadir viewing it depended on specular reflection,
the backscattered signal being inversely proportional to the roughness. The
main beams extended a further distance to 64° incidence, incorporating another
250 km wide swath on either side, but this was not able to give accurate measure¬
ments of low wind speeds and was not included within the design specification
of the instrument. The antennae were capable of operation in H—H or V—V
polarisation modes.
Within the main beams, twelve doppler filters were used to sub-divide the
X-shaped footprints into resolution cells, each approximately 50—70 km in
length along the footprint. Because of the varying view angle along the footprint,
the component (in the radar propagation direction) of the relative velocity
between the satellite and the ground varied along the beam. Consequently, the
reflected signal had a doppler shift which was characteristic of the reflection
position along the beam footprint. By filtering out all but a particular frequency,
the return from a 50 km length of sea could be resolved. The emitted pulse was
a long one (4.8 ms in duration) so that the leading edge was about two-thirds of
the way back to the radar when the pulse was turned off. It was also of constant
frequency rather than being chirped, so that the SASS doppler range resolution
technique is very different from the range resolution used by a SAR with a
short chirped pulse. The filters identified parts of the ground according to the
hyperbolic pattern of constant doppler lines indicated on Fig. 13.4. The resulting
velocity cells are shown in Fig. 13.5. Each cell was sampled by 64 separate pulses,
taking 1.89 s during which time the footprint pattern had moved about 12 km,
building up a complete resolution cell as shown in the inset to Fig. 13.5. By
averaging over 64 pulses, and subtracting the separately-measured noise power,
the effect of background noise could be significantly reduced.
The nadir view was achieved within the same sequence by sampling at three
further doppler frequencies corresponding to 0°, 4°, and 8° incidence angles. The
four beams were sampled in turn, each taking 1.89 s,with an antenna-switching
cycle therefore being completed every 7.56 s. Different operating modes
switched through different antenna sequences as shown in Table 13.1. The
numbers refer to the numbered antennae beams in Fig. 13.4. Each sequence used
at least one of the forward beams (1 or 4) and the corresponding rear beam (2 or
3), but not necessarily both sides, if one side was being illuminated with both
horizontally and vertically polarised beams. Thus modes 1 and 2 gave data over a
double swath, but with half the along-track sampling rate of modes 3-8 which
only viewed the swath on one side of the satellite.
By using both fore and aft beams, the radar viewed the same piece of sea
twice, from orthogonal directions. The time interval between the two views
was between 1 and 3 minutes, depending on the range along the beam. Exactly
Sec.13.3] Experience with the Seasat scatterometer 403

antenna beam/

Fig. 13.5 — Instantaneous doppler cell defined by the antenna pattern beam
width and the doppler filter response. To the bottom right the resolution cell is
shown to be constructed by integrating the doppler-cell response over time whilst
the satellite (and hence the instantaneously defined cell) moves forward over the
ground

Table 13.1 SASS antenna sequence operating modes

(after Johnson et al. (1980))

Mode Antenna sequence

1 4V IV 3V 2V

2 4H IH 3H 2H

3 4V 4H 3V 3H

4 IV IH 2V 2H

5 4V 4V 3V 3V

6 IV IV 2V 2V

7 4H 4H 3H 3H

8 IH IH 2H 2H
404 Microwave scatterometers [Ch.l3

the same doppler filters were used for each beam. If the doppler patterns were
symmetrical, the 12 range cells in the main beam would lie at the same distance
from the satellite track for both fore and aft beams. However, the rotation of
the earth underneath the satellite orbit caused an extra doppler shift, different
for fore and aft views, which produced an asymmetry in the ground pattern
of constant doppler lines. The effect was most marked at the equator where the
earth’s surface speed is greatest. This resulted in a non-matching of cells from
the fore and aft beams, illustrated in Fig. 13.6, which had to be corrected for in
the data processing.

158®W 156*W 162<’W 149*’W

Fig. 13.6 - An example of the resolution cell locations for beams 1 and 2 on
Seasat revolution 331 over Hawaii. Only a few of the cells have been drawn in,
to make clear the differences between the forward- and rear-looking beams, due
to the asymmetric doppler effect of the earth’s rotation. The swath for beam 2
is about twice as wide as for beam 1 at this near-equatorial latitude

A complete technological description of the SASS will be found in

Grantham et al. (1977) and Johnson et al. (1980).
Sec.13.3] Experience with the Seasat scatterometer 405

13.3.2 SASS data processing

Before it could be used in a geophysical algorithm for the calculation of surface
winds, certain basic processing had to be applied to the measurements of power
received at the antenna. After separation into the doppler frequency ranges,
the measured signal along with measured noise power were input to a set of
algorithms which used the design details of the radar system to evaluate the gain
of the SASS receiving system, and used the antenna gain in order to calculate
the mean power reflected from the surface during the measurement period.
From the geometry of the beam and a calculation of the doppler pattern
appropriate to the particular location and direction of the satellite, the measure¬
ment footprint cell size could be evaluated. This then led to the calculation
of o°, the normalised radar cross-section for the particular footprint cell under
consideration. The details can be found in Bracalente et al. (1980).
One problem discovered with the SASS was a bias in o° between
the different antennae pointing in different directions. It was possible to remove
this by examining o° over the Amazon rain forest, where the backscatter is
isotropic because of the dense foliage and random orientation of the scattering
surfaces.
Where possible an atmospheric correction was also applied to allow for
any atmospheric attenuation of the reflected signal and for increased backscatter
from liquid water in the atmosphere. This would be significant only under
conditions of heavy cloud and precipitation, which could be detected using
a microwave radiometer to sound the atmosphere. For the SASS, the atmos¬
pheric-water-retrieval algorithms of the Seasat SMMR (Chapter 8) were used
to estimate the appropriate correction (Moore et al. (1982)). This correction
was not entirely satisfactory, because the SMMR only viewed to the right-hand
side of the satellite track, and the footprint cells of the SMMR and the SASS,
even where they overlapped, were different sizes and shapes. Thus a SMMR
observation for an area much larger than, but including, the SASS footprint may
have incorporated the effect of localised clouds and precipitation cells which
did not in fact affect the SASS measurement. The ideal atmospheric correction
for a satellite scatterometer in future will come from a dedicated microwave
radiometer designed to view the scatterometer footprint area.
The final data manipulation which was required to enable the geophysical
algorithms to be applied, was the pairing or grouping of o° measurements of the
same area made from two different directions. This cell-pairing routine (Jones
et al. (1982)) operated in one of two ways. The first option matched together a
cell along the forward beam and a cell along the aft beam lying within 37 km of
the forward cell (50 km for the poorer-resolution data obtained when operating
modes 1 or 2 were being used). The second option grouped together all the cells
whose centres feU within a given latitude-longitude grid box. At least one a®
measurement was required from both fore and aft antennae to produce a valid
data value for input to the wind measurement algorithm.
406 Microwave scatterometers [Ch. 13

13.3.3 The SASS wind-vector algorithm

The retrieval of the wind vector from the observed radar cross-section a® is
based on the assumption that there exists a model function

a° = (t/, X, e) (13.6)

in which the only variables affecting are the wind speed U, and direction x
relative to the pointing direction of the radar beam, the radar beam incidence 6
at the sea surface, and e the polarisation state (H —H or V—V) of the measure¬
ment. This assumption is based on the observational programme mentioned in
section 13.2.2, but the empirically-derived form of (13.6) is still being improved
as more data from land-based, airborne, and satellite scatterometers are gathered.
As the database is widened, it may prove necessary to include a dependence on
further variables, such as the wind fetch or the amplitude of the low-frequency
swell wave end of the gravity-wave spectrum.
The model function selected as the basis of the SASS wind-vector algorithm
was one of the form

Foie, X, e, W) = logio(a°) = 0(6. x, e) + H(e. x, e) log^oW (13.7)

where W is the wind speed in m s'^ at a height of 19.5 m in a neutrally stable

atmosphere. (By defining a neutral atmosphere, it is equivalent to relating the
calibration to wind stress, since there is a unique relationship between wind
stress and the wind vector as discussed in section 13.2.1.)
As Jones etal. (1982) explain, the evaluation of the functions G and H, and
the final form of the whole geophysical algorithm package, resulted from a
combination of the most successful aspects of several independently-developed
calibration algorithms after validation using surface-wind measurements made
during the Seasat flight. Fig. 13.7 illustrates the shape of eqn(13.7) for particular
wind-speed values, polarisation states, and wind directions. The radar viewing
upwind corresponds to x = 0°, downwind to x = 180°, and crosswind to x = 90°
or 270°. Full tabulations of (13.7) will be found in Schroeder et al. (1982).
To recover the wind speed and direction from two measurements a°i and
o°2 of radar cross-section, made from fore and aft passes respectively with
known incidence angles di and 62 (di — 62) and polarisation states ej and ej,
eqn(13.7) is inverted to obtain

logio(a°)- G
log 10 (13.8)
H

This cannot be solved by substituting for G and H from the empirical model
tabulations, because Xi and Xa are unknown, but it is known that:
X2 = Xi ~ Htt for viewing on the right of the satellite track and
X2 = Xi + Htt for viewing on the left.
Sec.13.3] Experience with the Seasat scatterometer 407

Fig. 13.7 — Backscatter as a function of incidence angle for the SASS-1 model
function for winds of 23.6 and 4.6 m s“'. The different responses are shown for
upwind (U), downwind (D) and crosswind (C) azimuths, at both horizontal and
vertical polarisations. (From Pierson (1983))

Instead, (13.8) can be plotted as a function of x for the two parameter sets
(a®i, di, ei) and (o°2, ^2. ^2) corresponding to the radar returns from the fore
and aft views of the same patch of sea surface. Fig. 13.8(a) shows typical curves
for the two functions If'i(x) and 1^2 (x) where x is the wind direction measured
clockwise from the radar-pointing azimuth (N.B. the azimuth directions for
and W2 differ by 90°). If, instead, the curves are plotted so that x is now
the wind direction relative to a common direction, in this case the azimuth
of the forward antenna (Fig. 13.8(b)), W\ remains the same but W2 is shifted
90° to the right (for radar beams to the right of the direction of travel). Since
both curves apply to the same small area of sea, the wind magnitude and direc¬
tion are identified by the intersection of the two curves, assuming that the wind
vector has not changed between the two views.
Unfortunately, the curves normally intersect four times, although occasion¬
ally there are only two intersections. The wind speed is approximately the same
for the four possible solutions, leaving four possible directions, the true one and
three aliases. In principle, it should be possible to remove these by adding two
further curves, corresponding to the response at the polarisation states opposite
to the ones already used. This is the purpose of operating modes 3 and 4.
408 Microwave scatterometers [Ch. 13

(b)

Fig. 13.8 — Typical curves of wind speed versus wind direction appropriate to
a particular radar return for a given incidence angle and polarization state.
corresponds to beam 1 and W2 to beam 2. (a) Each curve is plotted against the
wind direction measured relative to the azimuth of the radar beam in question,
(b) Each curve is plotted against the wind direction measured relative to a fixed,
common direction, in this case the azimuth of beam 1. The actual wind direction
must be found at one of the intersections of and W^, that is, x= 40°, 140°,
220°, or 320°. One of these is the true direction, and the other three are aliases

Fig. 13.9 shows three typical families of curves for one polarisation state (solid
lines) and the slightly different curves for the opposite polarisation (dashed
lines). The aliases are removed, leaving a unique solution. However, the curves
in Figs 13.8 and 13.9 assume that the measurements of 0° are noise free.
In practice the curves are not actually plotted, and the solution is not obtained
graphically, but a least-squares fitting method is used (described by Jones etal.
(1982)) which minimises a difference function based on (13.7), summed over all
the available doppler cells falling within the grid square being studied. Given the
noise in the data, the solution is not so clearly defined as the noise-free curves
imply, and it has not yet been possible to achieve a unique solution even when
using both polarisations. The least squares algorithm normally offers four
solutions, but the noise in the data increases the errors in determining the four
Sec.13.3] Experience with the Seasat scatterometer 409

candidate directions. Since the noise has the effect of shifting the curves of
Figs 13.8 and 13.9 up or down, it causes the worst wind errors in candidate
directions close to 0°, 90°, 180°, and 270°. Indeed, in some cases the noise could
result in the curves not crossing at all, leading to no solution being found. When
four solutions are offered, the interpretation and application of the SASS data
must begin with an initial analysis to determine which is the most likely of the
candidate directions to be correct.

Fig. 13.9 — Three examples of the same type of plot as Fig. 13.8(b), for x = 0°,
40°, and 80°. In this case, however, two extra curves are included (drawn as
dashed lines) corresponding to the opposite polarisation states on beams 1 and ^
to those represented by the solid lines. The four curves cointersect at only one
point, giving a unique wind speed and direction solution. (From Pierson (1983))
410 Microwave scatterometers [Ch. 13

Future satellite scatterometers will probably try to avoid the aliases by

increasing the number of different ways of looking at the same piece of sea,
possibly with a second set of antennae at 20° to the first set. Even if the aliases
can be reduced from three to one by the addition of just one more radar beam,
this would be a considerable benefit since the two candidates would be roughly
in opposite directions, and the true value should readily be identifiable by
inspection.

13.3.4 Removing the direction alias

Although it must be hoped that the direction alias of the SASS is a temporary
problem in the development of satellite scatterometers, to be overcome with
improved designs, it is worthwhile mentioning the approaches which have been
adopted in attempts to determine the wind direction. The SASS dataset still
represents a unique and valuable archive of wind-vector data which is capable
of much more exploitation by oceanographers in the next few years, and the

Fig. 13.10 - An example of mapped SASS wind vectors (including aliases) over
the JASIN area. (From Wurtele et al. (1982), copyrighted by the American
Geophysical Union)
Sec.13.3] Experience with the Seasat scatterometer 411

question of determining the wind direction will remain until all the data are
analysed. They are presently available in two forms. The sensor data record
(SDR) contains the parameters a° 9, and e for given radar azimuths. The
geophysical data record (GDR) contains the wind speed and direction for all
possible solutions (the true and the aliased). A short period of data (about
two weeks) has had the aliases removed.
The most reliable method of removing the aliases, but the most laborious,
is to plot the wind streamlines by hand with reference to the surface-pressure
charts and wind observations from shipping, if available for the area. The data
can be plotted automatically in terms of the four wind-vector possibilities, as
illustrated in Fig. 13.10, but from there the process is subjective, and its success
depends on the skdl and experience of the analyst in the drawing of meteoro-

Fig. 13.11 - Schematic depicting the solutions produced by the SASS wind
algorithm corresponding to five possible surface-wind vectors, drawn in rela¬
tion to the SASS beam pattern on the ground. (From Wurtele et al. (1982),
copyrighted by the American Geophysical Union)
 412

+
V
•f

+
+
•V

V +
+

-h
+

+
-V

4
\

/
-

r
■4

4r
4-
-f-
-h

-f-
•V
Microwave scatterometers

4
-v \ \
■v ■V
-F

■V

-
+
-V-
Fig. 13.12 - Typical SASS wind vectors (and aliases) corresponding to idealised surface wind-velocity patterns, (a) Cyclonic vortex (northern
hemisphere); (b) Col, or hyperbolic point separating pairs of cyclonic and anticyclonic circulations; (c) Convergence line; (d) Frontal zone.
[Ch. 13

(From Wurtele etal. (1982), copyrighted by the American Geophysical Union)

Sec. 13.3] Experience with the Seasat scatterometer 413

logical charts. Wurtele et al. (1982) have rationalised so far as possible the
subjective choices and decisions which are made in the process. Although
the data are noisy, some help in interpretation can be gained from noting the
SASS wind vectors and aliases which are predicted from given wind vectors in
noise-free conditions. Fig. 13.11 shows these to be related in a systematic way to
the direction of the satellite. Winds at 45° to the track (that is, blowing along one
of the radar azimuths) produce only a single alias at 180° to the true direction,
whilst a wind in the satellite direction produces 3 aliases perpendicular to each
other and the true direction. Wurtele et al. (1982) also present the SASS
vector-field patterns which would be associated with idealised streamline
patterns typical of certain synoptic wind conditions. These are reproduced in
Fig. 13.12. As with any manual interpretation process, an experienced analyst
begins to recognise in the SASS vector plots the patterns associated with given
synoptic meteorological features. However, to achieve a successful interpretation
it is necessary to work with a total field which is at least as large as the typical
synoptic scales of the wind field. A single SASS overpass, as drawn in Fig. 13.10,
would by itself be insufficient to achieve a confident removal of directional
aliases.

13.3.5 Validation of the SASS algorithms

Considerable effort was made to establish the accuracy of wind observations
from the SASS. Fortunately, an extensive programme of ship and buoy observa¬
tions in support of the satellite observations was carried out in the Gulf of
Alaska Experiment (GOASEX) before the brief life of Seasat ended. This
experiment soon revealed systematic errors in the pre-launch calibration
algorithms which were modified to achieve a much closer matching between
the satellite and surface-wind measurements. Details of the GOASEX data
comparisons will be found in the special reports published by the Jet Propulsion
Laboratory (Barrick et al. (1979), Born et al. (1979)).
To test the validity of the updated algorithms a completely independent
dataset was required, and fortunately this was available from the meteorological
measurements made during the JASIN experiment in the North Atlantic.
Guymer (1983 a) presents a thorough analysis of the comparisons, and demons¬
trates that agreement between satellite predictions and ground observations falls
within the design goals of the SASS, and probably within the limits of accuracy
of the ground measurements, bearing in mind the problems of comparing a
time-meaned point observation from a ship or buoy with an instantaneous area
average from a satellite. Fig. 13.13 illustrates the agreement. It is interesting to
note that the one erroneous point labelled 1 on the wind-speed comparison was
probably associated with a thunderstorm which was not resolved by the atmos¬
pheric-correction part of the SASS algorithm. The increased radar backscatter
led to an anomalously high wind estimate from the SASS.
 SASS

414 [Ch. 13

SASS

Fig. 13.13 — Comparison of the Seasat scatterometer winds derived using the
Wentz algorithm-with the surface winds measured at sea in the JASIN experiment.
The latter are based on 60-minute means of automatically-logged ship and buoy
data. Each point represents the mean of aU comparisons on a given overpass. The
anomalous point 1 is referred to in the text. (From Guymer (1983 a))
Sec.13.4] Applications of wind scatterometry 415

Overall, it can be concluded that the SASS data should be able to yield
the surface wind to an accuracy of better than ±1.7 ms“^ in speed and ±17°
in direction, assuming that directional ambiguity has already been removed
successfully. These statistics are based on comparing datasets in which the wind
speed never exceeded 16 m s“\ However, there was opportunity to make a few
isolated comparisons with storms and gale force winds (Jones et al. (1982)),
and in these there was a tendency for the SASS to underestimate wind speeds.
This could be due to problems of poor atmospheric correction in the rain-bands
associated with the storms, inaccuracies in the spatial registration of cells which
show up because the storm-wind field changes over a relatively short length
scale, or inadequacies in the SASS model function for a® at high wind speeds.
Another means of checking the SASS speed data, but not the direction,
was against other Seasat wind-speed measurements from the altimeter and the
SMMR. Wentz, Cardone, & Fedor (1982) found a generally good agreement,
particularly for nadir viewing, and oblique viewing yielded a standard deviation
of only 1.42 m s"^ between SMMR and SASS predictions from 329 comparisons
for winds up to 20 m s"^ SMMR and SASS estimates of hurricane winds were
also in good agreement.

13.4 APPLICATIONS OF WIND SCATTEROMETRY

Although there is a need to improve on the SASS to avoid the direction-aliasing
problem, and both a longer instrument lifetime and dedicated in situ measure¬
ments in known stormy regions are required if validation of the wind vector
algorithms are to be made over a full range of wind speeds, there is no doubt
that the SASS has proved the concept of a satellite wind scatterometer. The
present algorithms produce the wind at a given height in a neutral atmosphere,
which is directly related to wind stress on the ocean surface. Thus quite apart
from the enormous meteorological implications of satellite scatterometry
which are outside the scope of this book there is the promise of being able to
measure the near-synoptic distributions of the shear stress which drives the
ocean surface, on a global scale. Guymer (1983b) has reviewed some of these
applications, and the future possibilities have been set out clearly by O’Brien
(1982).
The greatest oceanographic gains from wind scatterometry are likely to
be found in conjunction with the numerical modelling of dynamical ocean
processes. Numerical modelling of ocean dynamics has reached a fair degree
of sophistication, and is now limited in its ability to model real instead of
hypothetical case studies by the need for an accurate specification of boundary
conditions. One of the most important boundary conditions is the surface-
wind stress field which is the direct or indirect forcing of many of the motions
occurring in the ocean. Coupled ocean-atmosphere models can attempt to by-
416 Microwave scatterometers [Ch.l3

pass the problem by modelling the wind field, but in this case there is just as
much need to know the realistic wind-stress field in order to validate the models.
When surface-wind stress vectors can be prescribed over the ocean from satellite
scatterometers, the way will be open to use the models to explore dynamical
processes which occur on a variety of length and time scales. With global cover¬
age every few days, synoptic coverage over a swath hundreds of kilometres
wide, and the ability to continue supplying this data over a period of years,
the satellite scatterometer should be able to cope with most of the demands
for boundary data made on it by the modellers.
The types of process whose study will benefit from satellite-derived wind-
stress vectors range from localised upwelling to global ocean circulations.
Localised upwelling and downwelling are related to surface horizontal velocity
divergence or convergence. Fig. 13.14 shows schematically how this can be
influenced by the wind stress. If for simplicity we assume that the sea-surface
horizontal currents in the ocean are in geostrophic balance with the wind
stress, then if there is a shear in the wind stress, there will be a corresponding
convergence or divergence in the surface current. Continuity then demands
downwelling or upwelling. Hence the shear, or more precisely the curl, of the
IdTy dTx\
wind stress 1-1 should be directly proportional to the upwelling. The
\dx dyI
ability of a satellite scatterometer to provide the direction as well as the magni¬
tude of the wind stress means that the wind-stress curl can be accurately defined,
making it possible to study the effects of the resulting surface convergence/
divergence on localised mixed-layer and thermocline dynamics. This mechanism
is probably a generation mechanism for quasi-geostrophic wavelike motions
which propagate in the upper ocean and are controlled by both the Coriolis
force due to earth rotation and the buoyancy forces due to density stratification.
With length scales of tens to hundreds of kilometres and periods of tens to
hundreds of days, their modelling would benefit from a knowledge of the
varying wind-stress curl over such length and time scales.

Fig. 13.14 - Schematic of how a cyclonic wind-stress curl is geostrophically

balanced by a divergent surface flow leading to upwelling
Sec.13.5] Wave scatterometry 417

On a larger length scale, the modelling of the major ocean currents requires
the input of wind data. Some currents are strongly wind-dependent. For
example, the Somali current in the Western Indian Ocean increases its flow
during the year as the Indian Ocean Monsoon wind changes. Even for well-
studied flows such as the Gulf Stream, there is still uncertainty as to the
exact relationship between its meandering path and the location of the line
of zero wind-stress curl over the North Atlantic Ocean. Another ocean-wide
phenomenon is the complex ocean—atmosphere interaction which occurs in the
Equatorial Pacific, in which an anomaly in the wind-stress field on the Western
side can trigger off a wave which propagates eastwards and ultimately causes
a significant rise in the sea temperature off Ecuador and Peru, with disastrous
consequences for the local fisheries and a marked change in the coastal weather
patterns. El Nino, as it is called, has attracted a great deal of scientific attention
recently. If the mechanisms which produce it can be adequately modelled it may
be possible to predict the consequences several months ahead. Since wind stress
controls one of the important links in the chain of processes which are thought
to govern this ocean—atmosphere feedback loop, both its modelling and the
ultimate prediction of the phenomenon will benefit greatly from a detailed
knowledge of the time-evolving surface wind stress field.
Finally, as for the SMMR and altimeter estimates of sea state, the wind-
stress data which the SASS has collected for the southern hemisphere (although
mostly not yet processed) represents a very great improvement in data coverage
for the large areas of sea which are rarely traversed by shipping and even less
often sampled for oceanographic and meteorological parameters. Studies of
the circumpolar current and other dynamical ocean processes in the Antarctic,
which may have great bearing on the biological productivity of the area, and on
the formation of bottom water which spreads northwards at great depth, will be
assisted a great deal by the SASS data, and even more when a future satellite
scatterometer is able to supply a longer dataset.

13.5 WAVE SCATTEROMETRY

Although no satellite has yet carried an active radar specially designed as a wave
scatterometer, a considerable amount of research has been performed with land
and airborne radars to study the ocean-wave spectrum, using somewhat longer
wavelength radars than the SASS which measured the wind-driven capillary-
gravity ripples. Since a wave scatterometer is likely to be flown on a satellite
within a few years, the subject deserves at least a brief mention in a satellite
oceanography book. Moreover, it is artificial to distinguish too categorically
between radars of different designs which sense the ocean-surface roughness
principally through the Bragg resonant backscatter mechanism. One type of
wave scatterometer is virtually the same as the wind scatterometer discussed
above, whilst another type is effectively a simplified SAR. Much of the research
418 Microwave scatterometers [Ch. 13]

into wave scatterometry has been linked to the development of the SASS and
the SAR. A lot of the recent development has come from two United States
experiments - the Marineland and the West Coast Experiments (Shemdin
(1980 a, b)).
A single-frequency wave scatterometer can be used as a two-scale wave
probe by illuminating an area of sea which is small compared with the length of
long ocean-gravity waves. If the modulation of the Bragg scattering waves by the
long waves is represented in a known (calculated or measured) modulation
transfer function, it is possible to recover information about the spectrum of the
long waves by analysing the doppler shifts of the returned radar energy. The
doppler shift is induced by the orbital velocity of the long waves. Directional
information can be obtained by illuminating an area which is short in one
direction and much longer than gravity wavelengths in the other. Since the wave
velocities parallel to the long direction will average out over the length of foot¬
print, only those orbital velocities normal to the long direction will be detected.
In a dual-frequency scatterometer, two closely-spaced and coherently-
related radar frequencies /i, /a, are used to illuminate an area of sea much larger
than the long gravity wavelength. The two frequencies may be emitted together,
or one after the other in a short time interval during which the sea-surface
conditions will not change. By adding together the two returns, a signal is
obtained whose overall energy is related to the amplitude of the Bragg scatter¬
ing waves. However, because the two frequencies are coherently related, a beat
frequency (A/ = /i — /2) can be detected, and the amplitude of this spectral
line which stands out above the rest of the background of the return radar
spectrum depends on the modulation pattern of the Bragg ripples which is itself
in resonance with the beat frequency. If /i and f2 are chosen so that the Bragg
resonance wavelength of the beat frequency falls within the wavelength range of
long gravity (swell) waves, then the amplitude of the beat-frequency spectral
line can be related through the modulation transfer function to the amplitude
of the long waves at the interference wavelength. The theory is developed fully
in Alpers & Hasselmann (1978).
 CHAPTER 14

The way forward

14.1 SATELLITE SENSORS FOR OCEANOGRAPHY IN THE FUTURE

14.1.1 Introduction
It would be foolish to try to predict in detail how satellite oceanography
will develop in the future. The way forward is constrained by the limitations
of current technological capabilities in spacecraft, instrumentation, and data
transmission, by the availability of finances, and therefore by the political will
to advance this area of environmental science. Ultimately it is shaped by the
creativity, imagination, and commitment of the scientists engaged in research.
A remote-sensing technique which is showing great promise now may turnout
to be of little real value. The cessation of financial support for a particular
type of sensor may see scientific interest switch to another type of sensor which
continues to be funded. When science is dependent on high-cost technology as
in this field of research, the choice of study area to be funded may often be
controlled by what is perceived to have potential for commercial application.
Unfortunately, if environmental satellites are operated on a commercial basis,
and the selection of sensors to be flown is judged by their profit-making
potential, there will be much less scope for oceanographers to discover those
further possible marine applications of space remote sensing upon which they
have not yet stumbled.
Therefore this final chapter does not attempt to predict the shape of
satellite oceanography in a decade to come, but aims to summarise our achieve¬
ments to date, and to identify those trends which are already apparent in the
planning and designing of new sensors, satellites, and oceanographic programmes.
The first views of the ocean from space were made in the visible wavelength,
with the human eye and cameras. For well over a decade oceanographers have
been able to look at coastal sea areas with a multispectral scanner, but the
greatest value to oceanography in visible remote sensing has come from the
Coastal Zone Colour Scanner. Its value lies in the narrow-width spectral bands,
the high radiometric sensitivity, and the wider-area coverage which are enabling
new perspectives to be gained of the spatial structure of biological productivity
420 The way forward [Ch. 14

in the ocean. Although there is no immediately scheduled foUow-up for the

CZCS, the further exploitation of CZCS data must eventually lead to a demand
for another ocean-colour monitor, with similar resolution but more spectral
bands to assist with the atmospheric-correction problem. The disadvantage of a
colour scanner, which virtually eliminates its use as ah operational sensor in
oceanography and therefore reduces the commercial justification for its funding,
is its inability to penetrate the clouds. The infrared sensors which suffer from
the same drawback for ocean viewing can be more easily justified by their
operational value in meteorology. In the development of the ATSR, with a multi¬
look as well as a multi-wavelength capability, the IR atmospheric correction
problem is likely to be resolved to the satisfaction of most applications.
Given the achievements of the current state of the art it is difficult to
envisage further major technical developments in either colour or IR scanning
radiometers which will open up new horizons for oceanographers. Solid-state
imaging arrays and push-broom imaging will bring improvements, but not new
types of data. In these fields, therefore, the onus is very much on the oceano¬
graphic community to explore all the possible uses and applications of the
remotely-sensed image data now available, rather than waiting for the satellite
technologist to come up with even better sensors. The one area of ocean-colour
research where there remains a need for a technological breakthrough is in
fluorometry. The laser fluorometer promises to be a powerful remote-sensing
tool for marine biologists, chemists, and physicists, but at present it is difficult
to envisage it being flown on anything higher than an aircraft.
In contrast to visible and IR sensors, microwave sensors are still very much
in the development phase. Skylab, GEOS, and principally Seasat have demons¬
trated the exciting potential of microwaves to provide all-weather radiometers,
altimeters, and surface-wave roughness sensors of different kinds. Oceanographers
have begun to explore the applications of the new types of data, but there is still
room for development and refinement of the sensors. We are still at the stage
of making sure we understand how the sensors operate, how the microwaves
interact with the sea surface, and how to interpret the microwave measurements
in terms of oceanographic variables. There are further types of instrument, such
as the wave scatterometers, yet to be tried on satellites. It therefore seems
premature to think of flying truly operational microwave sensors for commercial
oceanographic applications in the near future. However, once experience is
gained with ERS-1 at the end of the 1980s and similar satellites expected from
other space agencies, there appears to be a definite operational role for the
all-weather microwave satellite sensor, stretching into the next century.

14.1.2 The ERS-1 payload

Since it is the one satellite with primarily oceanographic application which
at the time of writing is firmly on course for a proposed launch in 1988 or 1989,
it is worth looking at the sensors which are expected to be carried on the first
Sec. 14.1] Satellite sensors for oceanography in the future 421

ESA (European Space Agency) Remote Sensing Satellite, designated ERS-1.

Although it uses completely new European sensor designs, the oceanographer
can perhaps view ERS-1 in some ways as a follow-on from Seasat. Certainly it
will provide oceanographers with the first opportunity since Seasat to work once
again with satellite radar altimetry, scatterometry, and synthetic aperture radar.
Moreover, the design of the new sensors is, and wUl be, strongly influenced by
the experience gained with Seasat sensors.
ERS-1 will be another polar-orbiting satellite, with the nominal circular
orbit giving a 3-day repeat cycle during the initial calibration phase. However,
as with Seasat the conflicting orbit-repeat demands of different sensors will
require an orbit-manoeuvring capability to alter the repetition pattern. It is
expected to carry three principal sensors, an active microwave instrumentation
(AMI), a radar altimeter (RA), and the Along Track Scanning Radiometer
(ATSR).
The AMI is in effect three oblique-viewing active microwave devices
designed into one single instrument, which will be able to operate in only one
of the three modes at a time. The first mode is a SAR, capable of 30 m X 30 m
resolution over an 80 km wide swath. The second is a wave scatterometer,
measuring the wave-spectral-energy density to within 20% accuracy at 12
discrete wavelengths over a spectral range between 100 m and 1000 m. The third
mode is a wind scatterometer, designed to measure wind speeds of between 4
and 24 m s“^ to an accuracy of 2 m s”^ and direction to within 20°, over a ceU
of side 50 km. The AMI will operate at 5.3 GHz, four times the Seasat SAR
frequency, and about 1/3 the SASS frequency. In this respect it is a compromise,
enabling the same radar instrumentation to be used for both the SAR and
the wind scatterometer. With a Bragg resonant wavelength (at 23° incidence)
of about 8 cm it will be somewhat less sensitive to wind speed than the SASS,
while the SAR is likely to have a modulation-transfer function which is very
different from that of the longer-wavelength Seasat SAR. This means that it
will not be possible to use the SASS model functions for wind-vector algorithms
or to build on the Seasat experience in interpreting SAR imagery, but it wiU be
very interesting to see how much difference the changed Bragg wavelength will
make to the observations. The wave scatterometer will operate essentially as a
low-data-rate version of the SAR, not producing an image but obtaining directly
the spectral information which might otherwise be gained from a Fourier
transform of the SAR image.
The radar altimeter will operate at 13.5 GHz with a 300 MHz bandwidth.
Its design specification is for altitude measurement with a resolution better than
10 cm, although the aim is to achieve 5 cm. This will require careful ground
monitoring, and the satellite will carry a laser reflector for laser ranging. It may
also be possible to track the satellite using the Global Positioning Satellite
system. The altimetric measurements will benefit from the geoidal measurements
mode by Seasat and any other geoid-measuring satellites which may fly before
422 The way forward [Ch. 14

ERS-1. Significant wave-height between 1 and 20 m is hoped to be measured to

within ±10%.
The ATSR is essentially a multichannel infrared radiometer, but its conical
scan mirror will not only generate curved scan lines, but also will result in the
same piece of sea being observed twice, once from immediately above and once
at approximately 60° incidence, in which case it will look through twice the
thickness of atmosphere. As discussed in Chapter 7, this is expected to improve
the atmospheric-correction algorithms, and when averaged over a 50 X 50 km
square an absolute accuracy of better than 0.5 degK is predicted. It will be
capable of producing images with 1 km X 1 km pixel resolution, with a relative
radiometric resolution of 0.1 degK. It will also carry a microwave nadir sounder,
which will assist the IR atmospheric correction, but which will be primarily to
provide an atmospheric correction for the altimeter.
With a design life of three years, ERS-1 promises to provide much interesting
and useful information for oceanographers.

14.1.3 TOPEX
Another ocean-oriented satellite in an advanced programme planning stage
which may receive funding to enable it to fly within the next few years is the
US/French TOPEX mission. The name TOPEX stands for Ocean Topography
Experiment, and at the centre of the mission is a satellite designed particularly
for radar altimetry over the sea surface. The significant difference from the
Seasat and ERS-1 projects will be that the satellite design and orbit will
be tailored to the demands of measuring sea-surface topography, as discussed in
section 9.7, rather than having to compromise between those and the conflict¬
ing demands of radar measurement of sea state. Consequently the proposed con¬
figuration is for a non-sun-synchronous orbit to avoid tidal aliasing, at a height
of 1334 km to reduce atmospheric drag, with an inclination of 63.4° to obtain
optimum orbit-intersection angles at low- to mid-latitudes. It will therefore
exclude the polar regions, and be of less value for providing regular sea-state
statistics, but the design goal is to measure the satellite height above the sea
surface to an accuracy of two centimetres. To allow for ionospheric transmission-
speed effects, a two-wavelength radar instrument will be used. A 10-day repeat
frozen orbit is envisaged as the basic mode of operation, but with the flexibility
to change the ground-track coverage for special purposes.

“The primary goal of the experiment is to measure the surface topography

of the ocean over entire ocean basins for several years, to integrate these
measurements with subsurface measurements and models of the ocean’s
density field in order to determine the general circulation of the ocean
and its variability, then to use this information to understand the nature
Sec.14.2] Developments in data analysis 423

of dynamics, to calculate the heat transported by the oceans, the inter¬

action of currents with waves and sea-ice, and to test the ability to predict
circulation from the forcing by the winds.” (TOPEX 1981)

Five years is therefore the minimum envisaged duration for the experiment,
in order to determine the longer-period changes in circulation such as the inter¬
annual variability. By the end of this time, the chosen orbit configuration should
make it possible to extract the low- to mid-latitude global-variability field of
sea-surface topography on time scales between 20 days and five years, and
length scales from 30 km to the ocean basin widths. To use this knowledge
of the sea-surface topography to develop a scientific understanding of the
wind-driven ocean circulation, it will also be desirable to know the global wind¬
forcing over the oceans during the lifetime of the mission. Ideally, therefore,
a wind scatterometer should be flown on a separate satellite, or possibly on
TOPEX itself. At the same time, the more in situ ship and buoy measurements
of ocean circulation parameters that can be made during the mission, the more it
will be possible to relate the surface topography (that is, the barotropic motions)
to the vertical structure (the baroclinic motions). In this respect TOPEX wiU
form part of the proposed World Ocean Circulation Experiment (WOCE) whose
aim is to determine the heat and freshwater flow between the ocean basins over
a five-year period between 1988 and 1993.
Whilst topographic variability should be determined with TOPEX to an
accuracy of a few centimetres, provided that accurate orbit-determination is
achieved, it will not be possible to obtain a measurement of the mean sea-surface
topography unless the geoid is known accurately to within a few centimetres
over length scales as short as a few tens of kilometres. Whilst TOPEX will
itself contribute significantly to an improved knowledge of the geoid, it cannot
provide an independent geoidal measurement with which to calibrate mean-
ocean topography. This must come from independent gravity measurements,
achievable on a global scale only by satellite systems such as GRAVSAT. An
improved knowledge of the gravity field will also assist in orbit determination,
and reduce the demand for large numbers of ground-tracking stations. Thus,
the usefulness of TOPEX and other altimeters such as that proposed for the
Canadian RADARSAT, is ultimately dependent on gravity-measuring satellites.
This increases the total cost of such exercises, but it should be noted that
the gravity experiment need only occupy a few months of the total life-span
of-the TOPEX mission.

14.2 DEVELOPMENTS IN DATA ANALYSIS

The 1980s have seen the increasing availability of relatively low-cost image-
processing systems. The continuation of this trend is likely to bring modest
image-processing computers within the price range that will make them available
424 The way forward [Ch. 14

to individual scientists, or groups within their own establishment, rather than

being at a few centralised locations to be used only occasionally. This is bound to
make an impact on the use of remote sensing by oceanographers. Of course the
quantitative information contained in remote-sensing data can often be applied
to ocean science without the need for images to be displayed, but images are
able to stimulate fresh ideas and hypotheses, particularly concerning the spatial
structures of many oceanographic processes. This will only be fully appreciated
as remote-sensing imagery is made available to many more ocean scientists
through distributed image-analysis equipment.
One particular approach to processing remotely-sensed data which is ripe
for development is the analysis of image data from more than one sensor for the
same area at approximately the same time. This is a possibility whenever more
than one ocean-viewing satellite is in orbit at the same time, and the possibilities
will increase as several different space agencies launch environmental-monitoring
satellites. In order that two images from different sensors should be properly
compared, so that their combined data may provide more oceanographic
information than the simple sum of the two images separately, it is necessary to
co-register both images in the same geographical coordinates, as discussed in
Chapter 5.
There are various possible combinations which promise useful results. The
combination of CZCS and AVHRR data gives a more complete view of possible
tracers for dynamical processes. In the study of fronts, for example, it is
customary to look at the infrared thermal signature. A colour-scanner image of
the same front may show that the front also has a strong colour signature. This
may be due to the different sediment loads of the different water masses, or the
colour scanner may reveal a highly-productive area, with a strong chlorophyll
signature, on one side of or straddling the front. Sometimes it is possible to see
on the colour image that although a frontal region loses its IR signature during
part of the year, either because of diurnal thermocline effects, or because
the two separate water masses separated by the front have similar temperatues
over part of the year, there is still a sharp boundary between the two masses
maintained all the year, and revealed by their different colour signatures.
It will be particularly interesting to compare microwave image data with
colour and IR signatures. As microwave radiometers improve in sensitivity it will
be useful to compare their SST images with those from IR sensors, and in
the comparison there may be information about the spatial variability of those
parameters which control microwave emissivity. If lower-frequency microwave
radiometers are developed, with a greater surface penetration, comparison with
IR images may yield information about the thermal-skin layer, and possibly even
the surface-heat flux.
Perhaps the most exciting comparisons will be between images from active
microwave devices and passive radiometers. Generally speaking the former are
measuring properties of the surface (its roughness, wave height, etc.) and the
Sec.14.3] Trends in ocean science 425

latter are observing properties of the water itself (temperature, turbidity, colour,
etc.). There will be many instances where there is no correlation at all between
the patterns in images of the same sea area from an imaging radar and from a
radiometer. But where there is obvious correlation, then that in itself is evidence
of some dynamical process which links the water property under surveillance
to the surface roughness. Some possible effects like this have been observed
already — the convergence associated with a front often results in a change
in surface roughness across it, or an accumulation of surface material producing
a smooth slick at the front. Thus the front acquires a radar signature as well
as a thermal and/or colour signature. There may well be other effects like
this waiting to be discovered, which may give oceanographers further insights
into dynamical processes occurring near the ocean surface. ERS-1, carrying
a scanning IR radiometer as well as an active microwave device capable of
producing radar images, offers promise in this direction.
Whilst this approach of comparing images from different sensors is relatively
new for oceanographic applications, it has been used more for land applications,
and is finding increasing value in polar research and the study of sea ice.

14.3 TRENDS IN OCEAN SCIENCE

Oceanography is a relatively new and vigorous field of scientific study, and it
would be wrong to assume that major developments and new advances in marine
science will all be linked to the use of satellites. However, it is possible to notice
some effects which the use of satellites and the availability of satellite data is
having on oceanographic science. Two in particular can be picked out.
Firstly, the availability of a new vantage point in space gives us a new
perspective on the ocean. At last we can obtain a synoptic view over a large
scale, where before we had to piece together the whole from a very insufficient
number of small pieces. This makes possible, and should encourage, the study of
spatial as well as temporal variability. There is therefore increased scope for the
development of theoretical frameworks for understanding the spatial dynamics
of physical, chemical and biological processes in the ocean. If this does not
occur, then oceanographers are failing to grasp the very aspects of satellite
remote sensing which other ocean-monitoring techniques cannot offer.
Secondly, the development of satellite oceanography appears to be fostering
a more multidisciplinary approach to marine science. This is shown in several
ways. For example the dynamical oceanographer, faced with a colour image
which appears to show interesting eddy structures, must turn to the biologist
to explain the chlorophyU signature which is acting as a tracer, and before
long a joint research project has commenced, to study the spatial dynamics of
plankton communities! As another example, satellite remote sensing will enable
numerical ocean modeUers to be drawn increasingly into the mainstream of
oceanography. On the one hand, the broad areal coverage of satellites can
426 The way forward [Ch. 14

provide the coherent input data (for example, surface-temperature fields or

wind-stress fields) required to drive the more realistic models which are now
being produced with greater sophistication as computing power continues to
grow. On the other hand, it will become increasingly necessary to use numerical
models to interpret the spatial data from satellites, and to relate the surface
values to what is going on at depth, and so help to overcome what is the major
shortcoming of satellite remote sensing of the oceans. In particular, satellite
altimetry and ocean-circulation modelling are likely to be closely linked together
in the next decade.
Finally, the growth of a new technology to serve oceanography has brought
many new research workers into marine science from different branches of
the physical sciences — space science, radiometry, radar technology, communica¬
tions engineering, etc. We can expect that if such people take an interest in
the oceans they are viewing, as well as providing the techniques with which
to do so from satellites, there will be interesting new approaches introduced to
oceanography through the cross-fertilisation of ideas from other disciplines.

14.4 CONCLUSION
This book commenced by addressing itself to the question of whether satellite
oceanography could be isolated as a subject for study in its own right. Whilst
a new technique is being developed, there is good reason for studying it on its
own as a separate branch of ocean science, but books like this will not have
done a service to oceanography if they merely encourage the growth of a breed
of ‘satellite oceanographers’. The aim of this book has been to introduce the
techniques of remote sensing to oceanographers in general. It has also attempted
to give remote-sensing specialists a flavour of those branches of oceanography in
which satellites can be useful, and to show both groups of scientists how remote
sensing is being applied to the advancement of oceanographic science. Ultimately,
satellite remote sensing should become just another observational technique
available to the oceanographer, and one which is understood to some degree
by all oceanographers in just the same way that even the most desk-bound
theoretical oceanographer appreciates the importance, and the limitations, of
observations made at sea.
At present, however, many oceanographers are not familiar with the type
of ocean observations and measurements which can be made using satellites. Of
those who have taken an active interest in remote sensing a large proportion have
concentrated on the development and calibration of sensors and the validation
of data. Relatively few have yet managed to make creative scientific advances in
oceanography which could not have been achieved without observations from
space.
The challenge to those pursuing research in satellite oceanography, both
oceanographers and remote-sensing scientists, is to explore theories and concepts
Sec. 14.4] Conclusion 427

and to develop applications which would not have been possible using only con¬
ventional oceanographic observations. These will include looking at windows in
the space—time spectrum of oceanographic processes which require the synoptic
spatial resolution and long-period sampling capabilities of the satellite sensors.
They will also make use of the different types of sea observation (for example,
radar surface roughness) and the spatially-averaged data which are obtained using
satellites. The most fruitful areas of research are likely to be those in which con¬
ventionally-gathered data and satellite observations are used to complement
each other in order to reveal a fuller perspective of oceanographic processes than
either of them is individually capable of providing. Given a coordinated deploy¬
ment of the exciting range of satellite sensor types discussed in this book, and
a continuation of existing ship and buoy sampling programmes, the stage is set
for a steady unfolding of new scientific understanding of the ocean over the next
decade.
References

Alfoldi, T. T. & Munday, J. C. (1978). Water quality analysis by digital

chromaticity mapping of Landsat data. Canadian Jour. Rem. Sens. 4
108-126
Allan, T. D. (ed.) (1983). Satellite microwave remote sensing, Chichester, Ellis
Horwood, 526 pp
Allan, T. D. & Guymer, T. H. (1984). Seasat measurement of wind and waves of
selected passes over JASIN. Int. Journal of Remote Sensing 5 379—408.
Alpers, W. (1983). Imaging ocean surface waves by synthetic aperture radar — a
review. In: Allan, T.D. (ed.). Satellite microwave remote sensing, Chichester,
Ellis Horwood, p. 107—119
Alpers, W. & Hasselmann, K. (1978). The two-frequency microwave technique
for measuring ocean-wave spectra from an aeroplane or satellite. Boundary-
layer Meteorology 13 215—230
Alpers, W. R. & Rufenach, C. L. (1979). The effect of orbital motions on
synthetic aperture radar imagery of ocean waves. I.E.E.E. Trans. Antennas
Propagat. AP-27 685-690
Alpers, W. R., Ross, D. B., & Rufenach, C. L. (1981). On the detectability of
ocean surface waves by real and synthetic aperture radar. J. Geophys. Res.
86C 6481-6498
Amos, C. L. & Alfoldi, T. T. (1979). The determination of suspended sediment
concentration in a macrotidal system using Landsat data. J. Sedimentary
Petrology 49 159—174
Apel, J. R. (1981). Nonlinear features of internal waves as derived from the
Seasat imaging radar. In: Gower, J. F. R. (ed.). Oceanography from space.
New York and London, Plenum Press, p. 525—533
Apel, J. R., Byrne, H.M., Proni, J. R., & Sellers, R. L. (1976). A study of oceanic
internal waves using satellite imagery and ship data. Rem. Sensing Env. 5
125-136
 References 429

Austin, R.W. (1980). Gulf of Mexico ocean-color surface-truth measurements.

Boundary-layer Meteorology 18 269—285
Austin, R.W. & Petzold, T. J. (1981). The determination of the diffuse attenua¬
tion coefficient of seawater using the Coastal Zone Colour Scanner. In:
Gower, J. F. R. (ed.). Oceanography from space, New York, Plenum Press,
p. 239-256
Baker, K. S. & Smith, R. C. (1982). Bio-optical classification and model of
natural waters. 2. Limnol. Oceanogr. 27 500—509
Barber B. C. (1983). Some properties of SAR speckle. In: Allan, T. D. (ed.).
Satellite microwave remote sensing, Chichester, Ellis Horwood, p. 129—145
Barrett, E. C. & Curtiss, L. F. (1976). Introduction to environmental remote
sensing. Chichester, J. Wiley, 316 pp
Barrick, D. E. (1974). Wind dependence of quasi-specular microwave sea scatter.
I. E.E.E. Trans. Ant. Prop. AP-22 135—136
Barrick, D. E., Wilkerson, J. C., Woiceshyn, P. M., Born, G. H., & Lame, D. B.
(eds) (1979). Seasat Gulf of Alaska Workshop II report. Rep. 622—107,
Jet Propulsion Lab. Pasadena, California
Baylis, P.E. (1981a). Dundee University meteorological satellite ground receiving
and data archiving facility. In: Gower, J. F. R. (ed.). Oceanography from
space. New York and London, Plenum Press, p. 49—58
Baylis, P. E. (1981b). Guide to the design and specification of a primary user
receiving station for meteorological and oceanographic satellite data. In:
Cracknell, A. P. (ed.). Remote sensing in meteorology, oceanography and
hydrology, Chichester, Ellis Horwood, p. 81—96
Beal, R. C; (1981). Spatial evolution of ocean wave spectra. In: Beal, R. C.,
De Leonibus, P. S., Katz, I. (eds), Spaceborne synthetic aperture radar for
oceanography. Baltimore and London, The Johns Hopkins University Press,
p. 110-127
Beal, R. C.. De Leonibus, P. S., & Katz, 1. (1981). Spaceborne synthetic aperture
radar for oceanography, Baltimore and London, The Johns Hopkins
University Press
Beal, R. C., Tilley, D. G., & Monaldo, F.M. (1983). Large- and small-scale spatial
evolution of digitally processed ocean wave spectra from Seasat synthetic
aperture radar. J. Geophys. Res. 88C 1761—1778
Bernstein, R. L. (1982 a). Sea surface temperature estimation using the NOAA 6
sateUite AVHRR. J. Geophys. Res. 87C 9455-9465
Bernstein, R. L. (1982 b). SST mapping with the Seasat Microwave Radiometer.
J. Geophys. Res. 87C 7865—7872
Bernstein, R. L. (ed.) (1982 c). Seasat Special Issue 1: Geophysical Evaluation.
Reprinted from J. Geophys. Res. 87 (C5) 3173-3438, American Geo¬
physical Union, Washington
Bernstein, R. L., Born, G. H., & Whritner, R. H. (1982). Seasat altimeter deter¬
mination of ocean current variability. J. Geophys. Res. 87C 3261—3268
430 References

Bernstein, R. L., & Morris, J. H. (1983). Tropical and Mid-latitude North Pacific
sea surface temperature variability from the Seasat SMMR. J. Geophys.
Res. 88C 1877-1891
Blume, H-J. C., Kendall, B. M., & Fedors, J. C. (1978). Measurement of ocean
temperature and salinity via microwave radiometry. Boundary-layer
Meteorol. 13 295-308
Boermer, W. M., Jordan, A. K., & Kay, I.W. (eds) (1981). Special issue on inverse
methods in electromagnetics. Trans. I.E.E.E. Antennae and Propagation,
AP-29 (2) 185-417
Born, G. H., Wilkerson, J. C., & Lame, D. B. (eds) (1979). Seasat Gulf of Alaska
Workshop Report. Vols 1 and 2. Rep. 622—101. Jet Propulsion Lab.
Pasadena, California
Born, G. H., Richards, M. A., & Rosborough, G. W. (1982). An empirical deter¬
mination of the effects of sea state bias on Seasat altimetry. J. Geophys.
Res. %1C 3221-3226
Bracalente, E. M., Boggs, D. H., Grantham, W. L., & Sweet, J. L. (1980). The
SASS scattering coefficient Oq algorithm. I.E.E.E. J. Oceanic Eng. OE-5
145-154
Brown, G. S. (1979). Estimation of surface wind speeds using satellite-borne
radar measurements at normal incidence. J. Geophys. Res. 84B 3974—3978
Brown, R. D. (1983). M2 Ocean tide at Cobb Seamount from Seasat altimeter
data. J. Geophys. Res. 88C 1637—1646
Brown, R. D., Kahn, W. D., McAdoo, D. C., & Himwich,W; E. (1983). Roughness
of the Marine Geoid from Seasat altimetry. J. Geophys. Res. 88C
1531-1540
Bukata, R. R, Bruton, J. E., Jerome, J. H., Jain, S. C., & Zwick, H. H. (1981).
Optical water quality model of Lake Ontario. 2: Determination of
chlorophyll a and suspended mineral concentrations of natural waters from
submersible and low altitude optical sensors. Applied Optics. 20 1704—1714
Bullard, R. K. (1983a). Land into sea does not go. In: Cracknell, A. P. (ed.).
Remote sensing applications in marine science and technology, Dordrecht,
D.Reidel,p. 359-372
Bullard, R. K. (1983b). Detection of marine contours from Landsat film and
tape. In: Cracknell, A. P. (ed.). Remote sensing applications in marine
science and technology, Dordrecht, D. Reidel, p. 373—381
Bunker, A. F., Charnock, H., & Goldsmith, R. A. (1982). A note on the heat
balance of the Mediterranean and Red Seas. Jour. Marine Research 40
(supplement) 73—84
Campbell, W.J., Gloersen, R, Nordberg,W., & Wilheit,T. T. (1974). Dynamics and
morphology of Beaufort Sea ice determined from satellites, aircraft and
drifting stations. Proc. COSPAR symp. on approaches to earth survey
problems through use of space techniques. Constanz, F. R. G., May 1973.
p. 311-327
 References 431

Campbell, W. J., Ramseier, R. 0., Zwally, H. J., & Gloersen, R (1980). Arctic
sea-ice variations from time-lapse passive microwave imagery. Boundary-
layer Meteorology. 18 99—106
Cartwright, D. E. (1983). Detection of large-scale ocean circulation and tides.
Phil. Trans. Roy. Soc. Lond. A309 361—370
Cartwright, D. E. & Alcock, G. A. (1981). On the precision of sea surface
elevations and slopes from Seasat altimetry of the Northeast Atlantic
Ocean. In: Gower, J. F. R. (ed.). Oceanography from space, New York and
London, Plenum Press, p. 885—895
Cartwright, D. E. & Alcock, G. A. (1983). Altimeter measurements of ocean
topography. Chapter 19. In: Allan, T. D. (ed.). Satellite microwave remote
sensing, Chichester, Ellis Norwood, p. 309—319
Cartwright, D. E., Edden, A. C., Spencer, R., & Vassie, J.M. (1980). The tides of
the northeast Atlantic Ocean. Phil. Trans. Roy. Soc. Lond. A298 87—139
Castleman, K. R. (1979). Digital image processing. Englewood Cliffs, N. J.,
Prentice-HaU
Charnock, H. «fe Pollard R. T. (1983). Results of the JASIN project. London,
The Royal Society, 229 pp
Chelton, D. B., Hussey, K. J., & Parke, M. E. (1981). Global satellite measure¬
ments of water vapour, wind speed and wave height. Nature 294 529—532
Cheney, R. E. (1982). Comparison data for Seasat altimetry in the western North
Atlantic. J. Geophys. Res. 87C 3247—3253
Cheney, R. E., Marsh, J. G., & Beckley, B. D. (1983). Global mesoscale variability
from collinear tracks of Seasat altimeter data. J. Geophys. Res. 88C
4343-4354
Citeau, J. & Domain, F. (1981). A short review of an oceanographic use of
Meteosat data at ORSTOM remote sensing service. In: Application of
remote sensing data on the continental shelf. Proceedings of an EARSeL-
ESA Symposium. ESA SP-167, Paris, European Space Agency, p. 145-156
Clark, D. K. (1981). Phytoplankton pigment algorithms for the Nimbus-7 CZCS.
In: Gower, J. F. R. (ed.). Oceanography from space. New York, Plenum
Press, p.221—231
Colwell, R. N. (ed.) (1984). Manual of remote sensing, (2nd edn). Falls Church,
Virginia, American Society of Photogrammetry, 2440 pp
Cox, C. S. (1974). Refraction and reflection of light at the sea surface. In: Jerlov,
N. G. & Steemann Nielsen, E. (eds). Optical aspects of oceanography,
London and New York, Academic Press, p. 51—75
Cox, C. & Munk, W. (1954). Measurements of the roughness of the sea surface
from photographs of the sun’s glitter. J. Opt. Soc. Am. 44 838—850
Cracknell, A. R, MacFarlane, N., McMillan, K., Charlton, J. A., McManus, J., &
Ulbricht, K. A. (1982). Remote sensing in Scotland using data received from
satellites. A study of the Tay Estuary region using Landsat multispectral
scanning imagery. Int. J. Remote Sensing 3 113—137
432 References

Cutting, E., Born, G. H., & Frautnich, J. C. (1978). Orbit analysis for Seasat-A.
J. Astronautical Sci. 26 315—342
Deschamps, P.Y. & Frouin, R. (1984). Large diurnal heating of the sea surface
observed by the HCMR experiment. J. Physical Oceanogr. 14 177—184
Deschamps, P. Y. & Phulpin, T. (1980). Atmospheric correction of infrared
measurements of sea surface temperature using channels at 3.7, 11 and
12 Atm. Boundary-layer Meteorology 131—143
Dickey, T. D. & Simpson, J. J. (1983). The influence of optical water type on the
diurnal response of the upper ocean. Tellus 35 142—154
Dixon, T. H., Naraghi, M., McNutt, M. K., & Smith, S. N. (1983). Bathymetric
prediction from Seasat altimeter data. J. Geophys. Res. 88C 1563—1571
Domain, F., Citeau, J., & Noel, J. (1980). Sea surface temperatures studied by
Meteosat data along the coast of Senegal and Mauritania. In: Coastal and
Marine Applications of Remote Sensing, Proc. 6 th Annual Conf of Remote
Sensing Society, 1979, Dundee (Cracknell, A.P. (ed.)),The Remote Sensing
Society, p. 59—67
Douglas, D. C., Goad, C., Morrison, C., & Foster, F. (1980). A determination of
the geopotential from satellite-to-satellite tracking data. J. Geophys. Res.
85B 5471-5480
Duntley, S. Q. (1965). Oceanography from manned satellites by means of visible
light. In: Ewing, G.C. (ed.). Oceanography from space. Woods Hole Oceano¬
graphic Institution, p. 39—46
Dyer, K. R. (1973). Estuaries, a physical introduction. Chichester, J. Wiley
Ewing, G. C. (ed.) (1965). Oceanography from space. Ref. No. 65—10. Woods
Hole Oceanographic Institution, Mass., 469 pp
Ewing, G. & McAlister, E. D. (1960). On the thermal boundary-layer of the
ocean. Science, New York. 131 1374—1376
Fedor, L. S. & Brown, G. S. (1982). Waveheight and wind speed measurements
from the Seasat radar altimeter. J. Geophys. Res. 87C 3254—3260
Fedor, L. S., Godbey, T. W., Gower, J.F. R., Guptill, R., Hayne, G. S., Rufenach,
C. L., & Walsh, E. J. (1979). Satellite altimeter measurements of sea state —
an algorithm comparison. J. Geophys. Res. 84B 3991—4001
Flatte, S. M., Dashen, R. D., Munk, W. H., Watson, K. M., & Zachariasen, R.
(1979). Sound transmission through a fluctuating ocean. London,
Cambridge University Press
Fleming, E. A. & Le Lievre, D.D. (1977). The use of Landsat imagery to locate
uncharted coastal features on the Labrador Coast. Proc. 11th Int. Symp. on
Remote Sensing of Environment. Michigan, April 1977, p. 775—782
Francis, C. R., Thomas, D. P., & Windsor, E. P. L. (1983). The evaluation of
SMMR retrieval algorithms in satellite microwave remote sensing. In: Allan,
T. D. (ed.). Satellite microwave remote sensing, Chichester, Ellis Horwood,
p. 481-498
Fu, L.-L. & Holt, B. (1982). Seasat views oceans and sea ice with synthetic
 References 433

aperture radar. NASA—JPL Publication No. 81—120. Jet Propulsion Lab.,

Pasadena, Claifornia
Gallagher, J. J., Philippe, M., & Wannamaker, B. (1981). Satellite monitoring of
ocean surface temperature variability in the Mediterranean Sea. In: Gower,
J. F. R. (ed.), Oceanography from space, New York and London, Plenum
Press, p. 175—182
Garrett, W. D. (1967). Damping of capillary waves at the air—sea interface by
oceanic surface-active material. J. Marine Research 25 279—291
Gautier, C. (1981). Daily shortwave energy budget over the ocean from geo¬
stationary satellite measurments. In: Gower, J. F. R. (ed.). Oceanography
from space. New York and London, Plenum Press, p. 201—206
Gloersen, P. (1981). Summary of the status of the Nimbus-7 SMMR. In: Gower,
J. F. R. (ed.). Oceanography from space, London and New York, Plenum
Press, p. 665—672
Gonzalez, F. I., Thompson, T.W., Brown,W.E., &Weissman, D.E. (1982). Seasat
wind and wave observations of Northeast Pacific Hurricane Iva, August 13,
1978. J. Geophys. Res. 87C 3431-3438
Gonzalez, R. C. & Wintz, P. (1977). Digital image processing. Reading, Mass.,
Addison-Wesley
Gordon, H. R. (1978). Removal of atmospheric effects from satellite imagery of
the oceans. Applied Optics 17 1631—1636
Gordon, H. R. (1981). A preliminary assessment of the Nimbus-7 CZCS atmos¬
pheric correction algorithm in a horizontally inhomogeneous atmosphere.
In: Gower, J. F. R. (ed.). Oceanography from space, New York and London,
Plenum Press, p. 257—265
Gordon, H. R., Brown, O. B., & Jacobs, M. M. (1975). Computed relationships
between the inherent and apparent optical properties of a flat homogeneous
ocean. Applied Optics 14 417—427
Gordon, H. R., Clark, D. K., Brown, J.W., Brown, O. B., & Evans, R.H. (1982).
Satellite measurement of the phytoplankton pigment concentration in the
surface waters of a warm core Gulf Stream ring. J. Marine Research 40
491-502
Gordon, H. R., Clark, D. K., Mueller, J. L., & Hovis, W. A. (1980). Phytoplankton
pigments from the Nimbus-7 Coastal Zone Colour Scanner: comparison
with surface measurements. Science 210 63—66
Gordon, H. R. & Morel, A. Y. (1981). Water color measurements - an introduc¬
tion. In: Gower, J. F. R. (ed.). Oceanography from space. New York and
London, Plenum Press, p, 207—212
Gordon, H. R. & Morel, A. Y. (1983). Remote assessment of ocean color for
interpretation of satellite visible imagery. A review. Lecture notes on
coastal and estuarine studies, 4, New York, Springer-Verlag
Gower, J. F. R. (ed.) (1980). Passive radiometry of the ocean. (Reprinted from
Boundary-layer Meteorology 18 (Nos 1, 2, and 3). Dordrect, Boston and
434 References

London, D. Reidel
Gower, J. F. R (ed.) (1981). Oceanography from space. New York and London,
Plenum Press, p. 978
Gower, J. F. R., Denman, K. L., & Holyer, R. J. (1980). Phytoplankton patchiness
indicates the fluctuation spectrum of mesoscale oceanic structure. Nature
288 157-159
Grantham, W. L., Bracalente, E.M., Jones, W. L., & Johnson, J.W. (1977). The
Seasat-A satellite scatterometer. I.E.E.E. J. Oceanic. Eng. OE-2 200—206
Grassl, H. (1976). The dependence of the measured cool skin of the ocean on
wind stress and total heat flux. Boundary-layer Meteorology 10 465—474
Guymer, T. H. (1983a). Validation and applications of SASS over JASIN. In:
AUan, T. D. (ed.). Satellite microwave remote sensing, Chichester, Ellis
Horwood, p. 87—104
Guymer. T. H. (1983b). A review of Seasat scatterometer data. Phil. Trans.
Roy. Soc. London A309 399—414
Hammond, D. L., MenneUa, R. A., & Walsh, E. J. (1977). Short pulse radar used
to measure sea surface wind speed and S.W.H. I.E.E.E. Trans. Ant. Prop.
AP-25 61-67
Hansen, B. & Meincke, J. (1979). Eddies and meanders in the Iceland-Faroe
ridge area. Deep Sea Res. 26 1067—1082
Hasse, L. (1971). The sea surface temperature deviation and the heat flow at
the sea-air interface. Boundary-layerMeteorology 1 368—379
HasseImann, K. et al. (1973). Measurements of wind-wave growth and swell
decay during the Joint North Sea Wave Project (JONSWAP). Erganzungsheft
zur Deutschen Hydrographischen Zeitscrift. Reiche A8 Nr 12, 95 pp
Hasse Imann, K., Raney, R.K., Plant, W.J., Alpers,W., Shuchman, R. A., Lyzenga,
D. R., Rufenach, C. L., & Tucker, M. J. (1984). Theory of SAR ocean wave
imagery: A MARSEN view. J. Geophys. Res. (In press)
Hill, R. H. (1972). Laboratory measurement of heat transfer and thermal
structure near an air-water interface. J. Phys. Oceanography 2 190 — 198
Hojerslev, N. K. (1974). Inherent and apparent optical propteries of the Baltic.
Rep. Inst. Physical Oceanography, University of Copenhagen, No. 23,70 pp
Hojerslev, N. K. (1980). On the origin of yellow substance in the marine environ¬
ment. Rep. Inst. Physical Oceanography, University of Copenhagen, 42
39-56
Holligan, P. M., Viollier, M., Dupouy, C., & Aiken, J. (1983). Satellite studies on
the distributions of chlorophyll and dynoflagellate blooms in the western
English Channel. Continental Shelf Research 2 81—96
Horstmann, U. & Hardtke, P. G. (1981). Transport processes of suspended
matter, including phytoplankton, studied from Landsat images of the
Southwestern Baltic Sea. In: Gower, J. F. R. (ed.). Oceanography from
space. New York and London, Plenum Press, p. 429—438
Hovis, W. A., Clark, D.K., Anderson, F., Austin, R.W., Wilson, W.H., Baker, E.T.,
 References 435

Ball, D., Gordon, H. R., Mueller, J. L., El-Sayed, S. Z., Sturm, B., Wrigley,
R. C., & Yentsch, C. S. (1980). Nimbus-7 Coastal Zone Colour Scanner:
system description and initial imagery. Science 210 60—63
Hughes, B. A. & Gower, J. F. R. (1983). SAR imagery and surface truth com¬
parisons of internal waves in Georgia Strait, British Colombia, Canada. J.
Geophys. Res. 88C 1809—1824
Hughes, B. A. & Grant, H. L. (1978). The effect of internal waves on surface
wind waves, 1. Experimental measurements. /. Geophys. Res. 83C 443—454
Huhnerfuss, H., Walter, W., & Kruspe, G. (1977). On the variability of surface
tension with mean wind speed. J. Phys. Ocean.l 567—571
Isimaru, A. (1978). Wave propagation and scattering in random media. Academic
Press, New York
Jerlov, N.G. (1976). Marine optics, Amsterdam, Elsevier
Johnson, R. W. (1975). Quantitative sediment mapping from remotely-sensed
multi-spectral data. In: Shahrokhi, S. (ed.). Remote sensing of earth
resources. Vol. IV, University of Tenessee,Tullahoma,Tenn., p. 565—576
Johnson, J. W., Williams, L. A., Bracalente, E. M., Beck, F. B., & Grantham, W. L.
(1980). Seasat-A satellite scatterometer instrument evaluation. I.E.E.E. J.
Oceanic Eng. OE-5 138—144
Jones, W. L. & Schroeder, L. C. (1978). Radar backscatter from the ocean:
dependence on surface friction velocity. Boundary-layer Meteorology 13
133-149
Jones, W. L., Schroeder, L. C., Boggs, D. H., Bracalente, E. M., Brown, R. A.,
Dome, G. J., Pierson, W. J., & Wentz, F. J. (1982). The Seasat-A satellite
scatterometer: the geophysical evaluation of remotely sensed wind vectors
over the ocean. J. Geophys. Res. 87C 3297—3317
Katsaros, K. B. (1977). The sea surface temperature deviation at very low wind
speeds: is there a limit? Tellus 29 229—239
Katsaros, K. B. (1980). The aqueous thermal boundary layer. Boundary-layer
Meteorology 18 107—127
Katsaros, K. B., Taylor, P. K., Alishouse, J. C., & Lipes, R. G. (1981). Quality
of Seasat SMMR atmospheric water determinations. In: Gower, J. F. R.
(ed.). Oceanography from space. New York and London, Plenum Press,
p. 691-706
King-Hele, D. (1959). The effect of the earth’s oblateness on the orbit of a near
satellite. Proc. Roy. Soc. London 247A 49-72
King-Hele, D. (1964). Theory of satellites in an atmosphere. London,
Butterworth, 165 pp
King-Hele, D. (1976). The shape of the earth. Science 192 1293-1300
Kinsman, B. (1965). Wind waves. Prentice-Hall, 676 pp
Kirwan, A. D., Ahrens, T. J., & Born, G. H. (eds) (1983). Seasat special issue II:
Scientific results. Reprinted from J. Geophys. Res. 88 (C3), Washington,
American Geophysical Union, p. 1529—1952
436 References

Klein, L. A. & Swift, C.T. (1977). An improved model for the dielectric constant
of sea water at microwave frequencies. I.E.E.E. Trans. Antennae and
Propagation AP-25 104—111
Klemas, V. (1980). Remote sensing of coastal fronts and their effects on oil
dispersion. Int. J. Remote Sensing 1 11—28
Klemas, V., Bartlett, D., Philpot, W., & Rogers, R. (1974). Coastal and estuarine
studies with ERTS-1 and Skylab. Remote SensingEnvir. 3 153—174
Krauss, E. B. (ed.) (1977). Modelling and prediction of the upper layers of the
ocean. Pergamon, Oxford
Kropotkin, M. A., Verbitskiy, V. A., Sheveleva, T. Y., & Tarashkevich, V. N.
(1978). Radiation temperature of a water surface with an oil slick.
Oceanology 18 730—731
La Fond, E. C. (1962). Internal waves. In: Hill, M. (ed.). The sea. Vol. 1,
Interscience, New York, p. 731—751
Lamb, H. (1936). Hydrodynamics. Cambridge University Press
Lane, J. A. & Saxton, J. A. (1952). Dielectric dispersion in pure polar liquids at
very high radio frequencies. Ill The effect of electrolytes in solution. Proc.
Roy. Soc. London 214 531—545
Large, W. G. & Pond, S. (1981). Open ocean momentum flux measurements in
moderate to strong winds. J. Phys. Oceanogr. 11 324 — 336
LeBlond,P. A. & Mysak, L. A. (1978). Waves in the ocean. Elsevier, 602 pp
Lee, C.Y. & Beardsley, R. C. (1974). The generation of long nonlinear internal
waves in a weakly stratified shear flow. J. Geophys. Res. 79 453—462
Le Fevre, J., Viollier, M., Le Corre, P., Dupouy, C., & Grail, J. R. (1982). Remote
sensing observations of biological material by Landsat along a tidal thermal
front and their relevancy to the available field data. Estuarine, coastal and
shelf science 16 37—50
Legeckis, R. (1979). A survey of world wide SST fronts detected by environ¬
mental satellites. J. Geophys. Res. 83C 4501—4522
Lillesand,T. M. & Kiefer, R.W. (1979). Remote sensing and image interpretation.
Chichester, J. Wiley, 612 pp
Lintz, J. & Simonett, D. S. (1976). Remote sensing of environment. Addison-
Wesley Publishing Co., Advanced Book Program, Reading, Mass.
Lipes, R. G. (1982). Description of Seasat radiometer status and results. J.
Geophys. Res. 87C 3385-3395
Lodge, D.W. S. (1981). The Seasat-1 synthetic aperture radar: introduction, data
reception and processing. In: Cracknell, A. P. (ed.). Remote sensing in
meteorology, oceanography and hydrology, Chichester, Ellis Horwood,
p. 335-356
Longuet-Higgins, G. S. & Stewart, R. W. (1964). Radiation stresses in water
waves, a physical discussion with applications. Deep Sea Res. 11 529-562
Lorell, J., Colquitt, E., & Anderle, R. J. (1982). Ionospheric correction for
Seasat altimeter height measurement. J. Geophys. Res. 87C 3207-3212
 References 437

Lowman, P. D. (1965). Space photography: a review. In: Ewing, G. C. (ed.),

Oceanography from space. Woods Hole Oceanographic Institution,
p. 73-90
McAlister, E. D. & McLeish,W. (1970). A radiometer system for measurement of
the total heat flux from the sea. Appl. Optics 9 2697—2705
McAlister, E. D., McLeish, W., & Corduan, E. A. (1971). Airborne measurements
of the total heat flux from the sea during BOMEX. /. Geophys. Res. 76
4172-4180
McClain, E. P. (1982). Operational implementation of AVHRR-only multi¬
channel sea surface temperature products. Unpublished document presented
to 32 nd Meeting of NOAA/NESS SST research panel (30 Sept. 1982)
McClain, E. P., Pichel, W. G., Walton, C. C., Ahmad, Z., & Sutton, J. (1982).
Multi-channel improvements to satellite-derived global sea tempera¬
tures. Reprint for XXIV Cospar Meeting, Ottawa, 22-29 May, 1982.
p. A3.l-A3.il
McClatchey, R. A., Fenn, R.W., Selby, J. E. A., Volz, F. E., & Gaving, J. S. (1972).
Optical properties of the atmosphere (3rd edn) AFCRL-72-0497. Environ¬
mental Research Papers No. 411
MacFarlane, N. & Robinson, I. S. (1984). Atmospheric correction of Landsat
MSS data for a multidate suspended sediment algorithm. Int J. Remote
Sensing 5 561—576
MacIntyre, F. (1977). The top millimetre of the ocean. In: Menard, H.W. (ed.).
Ocean Science, San Francisco, W.H. Freeman, p.145—155 (first published in
Scientific American, May, 1974)
McKenzie, D. P. & Bowin, C. (1976). The relationship between bathymetry and
gravity in the Atlantic Ocean. J. Geophys. Res. 81 1903—1915
Marsh, J. G. & Martin, T.V. (1982). The Seasat altimeter mean sea surface model.
J. Geophys. Res. 87C 3269—3280
Massey, H. (1964). Space physics. Cambridge University Press, London, 237 pp
Maxworthy, T. (1979). A note on the internal solitary waves produced by tidal
flow over a three-dimensional ridge. /. Geophys. Res. 84C 338—346
Menard, Y. (1983). Observations of eddy fields in the Northwest Atlantic
and Northwest Pacific by Seasat altimeter data. J. Geophys. Res. 88C
1853-1866
Miyakoda, K. & Rosati, A. (1982). The variation of sea surface temperature in
1976 and 1977. l.The data analysis. J. Geophys. Res. 87C 5667-5680
Mognard, N. M. (1983). Swell propagation in the North Atlantic ocean using
Seasat altimeter. In: Allan, T. D. (ed.). Satellite microwave remote sensing,
Chichester, Ellis Horwood, p. 425-437
Mognard, N.M., Campbell, W. J., Cheney, R.E.,& Marsh, J.G. (1983). Southern
ocean mean monthly waves and surface winds for winter 1978 by Seasat
radar altimeter. J. Geophys. Res. 88C 1736—1744
Montgomery, D. R. (1981). Commercial applications of satellite oceanography.
438 References

OCEANUS 24 56—64, Woods Hole Oceanographic Institution, Woods Hole,

Mass.
Moore, R. K., Birrer, I. J., Bracalente, E. M., Dome, G. J., & Wentz, F. J. (1982).
Evaluation of atmospheric attenuation from SMMR brightness temperature
for the Seasat satellite scatterometer. J. Geophys. Res. 87C 3337—3354
Moore, R. K. & Fung, A. K. (1979). Radar determination of winds at sea. Proc.
I.E.E.E. 67 1504-1521
Morel, A.Y. & Prieur, L. (1977). Analysis of variations in ocean color. Limnology
and Oceanography 22 709—722
Munday, J. C. & Alfoldi, T.T. (1979). Landsat test of diffuse reflectance models
for aquatic suspended solids measurement. Rem. Sens, of Envir. 8 169—183
Neumann, G. & Pierson, W. I. (1966). Principles of physical oceanography.
Prentice-Hall, 545 pp
Nicholls, S. (1979). Intercomparisons between the MRF Hercules and surface
observations from the JASIN meteorological ships. JASIN News. No. 19,
9-11
Njoku, E. G. (1980). Antenna pattern correction procedures for the SMMR.
Boundary-layer Meteorology 18 79—98
NOAA/NESS (1978). The TIROS-NjNOAA A-G satellite series. NOAA
Technical Memorandum NESS 95. Washington, U.S. Dept, of Commerce
NOAA/NESS (1979). Data extraction and calibration of TIROS-NjNOAA
radiometers. NOAA Technical Memorandum NESS 107. Washington, U.S.
Dept, of Commerce
O’Brien, J. J. (ed.) (1982). Scientific opportunities using satellite wind stress
measurements over the ocean. Fort Lauderdale, Florida: Nova University/
N.Y. I.T. Press
Officer, C. B. (1976). Physical oceanography of estuaries and associated coastal
waters. New York, J. Wiley, 465 pp
Osborne, A. R. & Burch, T. L. (1980). Internal solitons in the Andaman Sea.
Science 208 451 — 460
Parker, C. E. (1971). Gulf Stream rings in the Sargasso Sea. Deep-Sea Res. 18
981-993
Paulson, C. A. & Parker, T.W. (1972). Cooling of a water surface by evaporation,
radiation and heat transfer. J. Geophys. Res. 11 491—495
Paulson, C. A. & Simpson, J. J. (1981). The temperature difference across the
cool skin of the ocean. J. Geophys. Res. 86C 11044—11054
Pedlosky, J. (1979). Geophysical fluid dynamics. New York, Springer-Verlag,
624 pp
Phillips, 0. M. (1977). The dynamics of the upper ocean. (2nd edn), Cambridge
University Press
Pierson, W. J. (1983). Highlights of the Seasat—SASS program: a review. In:
Allan, T. D. (ed.), Satellite microwave remote sensing, Chichester, Ellis
Horwood, p. 69—86
 References 439

Pingree, R. D. (1978). Physical oceanography of the Celtic Sea and English

Channel. In: Banner, F.T., Collins, M. B., & Massie, K. S. (eds). The North-
West European shelf seas. The sea bed and the sea in motion: II Physical
and chemical oceanography and physical resources. Amsterdam, Elsevier,
p. 415-465
Pingree, R. D. (1984). Some applications of remote sensing to studies in the Bay
of Biscay, Celtic Sea and English Channel. In: Nihoul, J. C. J. (ed.). Remote
sensing of shelf seas hydrodynamics. Amsterdam, Elsevier
Pingree, R. D., Forster, G. R., & Morrison, G. K. (1974). Turbulent convergent
tidal fronts. J. Marine Biological Assoc, of U.K. 54 469—479
Platt, T. & Herman, A.W. (1983). Remote sensing of phytoplankton in the sea:
surface-layer chlorophyll as an estimate of water column chlorophyll and
primary production. Int. J. remote sensing 4 343—351
Pond, S. & Pickard, G. L. (1978). Introductory dynamical oceanography.
Oxford, Pergamon Press, (2nd edn. 1983)
Preisendorfer, R. W. (1961). Application of radiative transfer theory to light
measurements in the sea. Union Geod. Geophys. Int. Mon. 10 11—29.
Reprinted in: Tyler, J. E. (ed.). Light in the sea. Stroudsburg, Pennsylvania,
Dowden, Hutchinson and Ross Inc. (1977) p. 46—64
Queffeulou, P., Braun, A., & Brossier, C. (1981). A comparison of Seasat-derived
wave heights with surface data. In: Gower, J. F. R. (ed.). Oceanography
from space, New York and London, Plenum Press, p. 637—644
Raney. R. K. (1983). Synthetic aperture radar observations of ocean and land.
Phil. Trans. Roy. Soc. London A 309 315—321
Reeves, R. G. (1975) (Editor in Chief). Manual of remote sensing. American
Society of Photogrammetry, Falls Church, Virginia, U.S.A. Vols I & II,
2144 pp
Rhines, P. B. (1977). The dynamics of unsteady currents. In: Goldberg, E. D.,
McCave, I. M., O’Brien, J. J., & Steel, J. H. (eds) The sea,Vo\.5, New York,
J.Wiley,p. 189-318
Robinson, I. S. (1983). Satellite observations of ocean colour. Phil. Trans. Roy.
Soc. London A309 415—432
Robinson, I. S. & Srisaengthong, D. (1981). The use of Landsat MSS to observe
sediment distribution and movement in the Solent coastal area. In: Applica¬
tion of remote sensing data on the continental shelf (Longdon, N. & Levy,
G., eds), European Space Agency SP-167, Paris, p. 221—232
Robinson, 1. S., WeUs, N. C., & Charnock, H. (1984). The sea surface thermal
boundary layer and its relevance to the measurement of sea surface tempera¬
ture by airborne and spaceborne radiometers. Int. Jour. Rem. Sens. 5
19-45
Rosenfeld, A. & Kak, A. C. (1982). Digital picture processing (2nd edn). New
York, Academic Press
Ross, D. B. (1981). The wind speed dependency of ocean microwave backscatter.
440 References

In: Beal, R. C., De Leonibus, P. S., & Katz, I. (eds), Spaceborne synthetic
Aperture radar for oceanography. Baltimore and London, Johns Hopkins
University Press, p.75—86
Rouse, L. J. & Coleman, J. M. (1976). Circulation observations in the Louisiana
Bight using Landsat imagery. Remote Sensing of Environment 5 55—66
Sabins, F.F. (1978). Remote sensing: principles and interpretation. San Francisco,
Freeman
Sathyendranath, S. & Morel, A. (1983). Light emerging from the sea - interpre¬
tation and uses in remote sensing. In: Cracknell, A.P. (ed.), Remote sensing
applications in marine science and technology, Dordrecht, D. Reidel,
p. 323-358
Saunders, P. M. (1967 a). Aerial measurement of sea surface temperature in the
infrared. J. Geophys. Res. 12 4109—4117
Saunders P. M. (1967b). The temperature at the ocean—air interface. J. Atmos.
Sci. 24 269-273
Saunders, P. M. (1973). The skin temperature of the ocean — a review. Mem.
Soc. Roy. des Sci. de Liege. Ser. 4 93—98
Schooley, A. H. (1977). Temperature of ocean skin related to cloud shadows.
J. Phys. Oceanogr. 1 486—487
Schroeder, L. C., Boggs, D. H., Dome, G., Halberstam, 1. M., Jones, W. L., Pierson,
W. J., & Wentz, F. J. (1982). The relationship between wind vector and
normalized radar cross section used to derive Seasat-A satellite scattero-
meter winds. J. Geophys. Res. 87C 3318—3336
Scott, J. C. (1972). The influence of surface-active contamination on the initia¬
tion of wind-waves. J. Fluid Mech. 56 591—606
Shemdin, O. H. (1980a). The West Coast experiment: an overview. EOS. 61
649-651
Shemdin, 0. H. (1980b). The marineland experiment: an overview. EOS 61
625-626
Simpson, J. H. (1981). Sea surface fronts and temperatures. In: Cracknell, A.P.
(ed.). Remote sensing in meteorology, oceanography and hydrology,
Chichester, Ellis Horwood, p. 295—311
Simpson, J. H., Hughes, D.G.,& Morris, N. C.G. (1977). The relation of seasonal
stratification to tidal mixing on the continental shelf. In: Angel, M. (ed.),
A voyage of discovery, Oxford, Pergamon Press, p. 327—340
Simpson, J. H. & Pingree, R. D. (1978). Shallow sea fronts produced by tidal
stirring. In: Bowman, M. J. & Esaias, W. E. (eds). Oceanic fronts in coastal
processes. New York, Springer-Verlag, p. 29
Simpson, J. J. & Dickey, T. D. (1981). The relationship between downward
irradiance and upper ocean structure. J. Phys. Oceanogr. 11 309—323
Simpson, J. J. & Paulson, C. A. (1980). Small scale sea-surface temperature
structure. /. Phys. Oceanogr. 10 399—410
Singh, S. M. & Warren, D. E. (1983). Sea surface temperatures from infrared
 References 441

measurements. In: Cracknell, A. P. (ed.), Remote sensing applications in

marine science and technology, Dordrecht, D. Reidel, p. 231—262
Slater, P. N. (1980). Remote sensing: optics and optical systems. Addison-
Wesley Publishing Co., Advance Book Program, Reading, Mass. 575 pp
Smith, S.D. (1980). Wind stress and heat flux over the ocean in gale force winds.
J. Phys. Oceangr. 10 709—726
Smith, R. C. & Baker, K. S. (1978 a). The bio-optical state of ocean waters and
remote sensing. Limnol. Oceanogr. 23 247—259
Smith, R. C. & Baker, K. S. (1978b). Optical classification of natural waters.
Limnol. Oceanogr. 23 260—267
Smith, R. C. & Baker, K. S. (1981). Optical properties of the clearest natural
waters (200—800 nm). Applied Optics. 20 177—184
Smith, R. C. & Baker, K. S. (1982). Oceanic chlorophyll concentrations as deter¬
mined by satellite (Nimbus-7 Coastal Zone Colour Scanner). Marine Biology
66 269-279
Smith, R. C., Eppley, R. W., & Baker, K. S. (1982). Correlation of primary
production as measured aboard ship in Southern California coastal waters
and as estimated from satellite chlorophyll images. Marine Biol. 66
281-288
Smith, R.C. & Wilson, W.H. (1981). Ship and satellite bio-optical research in the
California Bight. In: Gower, J. F. R. (ed.). Oceanography from space, New
York and London, Plenum Press, p. 281—294
Sommerville, D. M. Y. (1934). Analytical geometry of three dimensions.
Cambridge University Press
Spence, T.W. & Legeckis, R. (1981). Satellite and hydrographic observations of
low-frequency wave motions associated with a cold core Gulf-Stream ring.
J. Geophys. Res. 86C 1945—1953
Staelin, D. H., Kunzi, K. F., Pettyjohn, R. L., Poon, R. K., Wilcox, R. W., &
Waters, J.W. (1976). Remote sensing of atmospheric water vapor and liquid
water with the Nimbus-5 microwave spectrometers. J. Applied Meteor¬
ology. 15 1204-1214
Stewart, R. H. (1985). Satellite oceanography. University of California Press
Stogryn, A. (1967). The apparent temperature of the sea at microwave
frequencies. I.E.E.E. Trans. Antennas and Propag. AP-15 278-286
Stoker, J. J. (1957). Water waves. New York, Interscience, 567 pp
Stommel, H. (1958). The Gulf Stream. Berkeley, University of California Press,
248 pp
Sturm, B. (1981a). The atmospheric correction of remotely sensed data and the
quantitative determination of suspended matter in marine water surface
layers. In: Cracknell, A. P. (ed.). Remote sensing in meteorology, ocean¬
ography and hydrology, Chichester, Ellis Horwood, p. 163—197
Sturm, B. (1981b). Ocean colour remote sensing using Nimbus CZCS Data. In:
Gower, J. F. R. (ed.), Oceanography from space, New York, Plenum Press,
442 References

p. 267-280
Sturm, B. (1983). Selected topics of Coastal Zone Colour Scanner evaluation.
In: CrackneU, A.P. (ed.), Remote sensing applications in marine science and
technology, Dordrecht, D. Reidel, p. 137—168
Swift, C. T. (1980). Passive microwave remote sensing. Boundary-layer
Meteorology 18 25—54
Tapley, B. D., Born, G.H., & Parke, M. E. (1982a). The Seasat altimeter data and
its accuracy assessment. J. Geophys. Res. 87C 3179—3188
Tapley, B. D., Lundberg, J. B., & Born, G. H. (1982b). The Seasat wet tropo¬
spheric range correction. J. Geophys. Res. 87C 3213—3220
Taylor. P. K. (1983). The scanning multichannel microwave radiometer — an
assessment. In; Allan, T. D. (ed.). Satellite microwave remote sensing,
Chichester, Ellis Horwood, p. 463—480
Thomas, D. P. (1981). Microwave radiometry and applications. In: Cracknel!,
A. P. (ed.). Remote sensing in meteorology, oceanography and hydrology,
Chichester, Ellis Horwood, p. 357—369
Thompson, J. D., Born, G. H., &Maul, G.A. (1983). Collinear track altimetry in
the Gulf of Mexico from Seasat measurements, models and surface truth.
J. Geophys. Res. 88C 1625—1636
Tomiyasu, K. (1978). Tutorial review of synthetic-aperture radar (SAR) with
applications to imaging of the ocean surface. Proc. I.E.E.E. 66 563—583
TOPEX (1981). Satellite altimetric measurements of the ocean. Report of the
TOPEX Science Working Group. NASA. Jet Propulsion Laboratory,
California Institute of Technology, California
Townsend, W. F., McGoogan, J. T., & Walsh, E. J. (1981). Satellite radar alti¬
meters — present and future oceanographic capabilities. In; Gower, J. F. R.
(ed.). Oceanography from space. New York and London, Plenum Press,
p. 625-636
Trask, R. P. & Briscoe, M. G. (1983). Detection of Massachusetts Bay internal
waves by the synthetic aperture radar on Seasat. J. Geophys. Res. 88C
1789-1799
Tucker, M. J. (1983). Observations of ocean waves. Phil. Trans. Roy. Soc.
London A309 371—380
Uslenghi, P. L. E; (ed.) (1978). Electromagnetic scattering. Academic Press,
New York
Valenzuela, G. R. (1978a). Scattering of electromagnetic waves from the ocean.
In; Lund,T. (ed.), Surveillance of environmental pollution and resources by
electromagnetic waves. Hingham, D. Reidel, p. 199—226
Valenzuela, G. R. (1978b). Theories for the interaction of electromagnetic and
oceanic waves - a review. Boundary-layer Meteorology 13 61-85
Valenzuela, G. R. & Wright, J. W. (1979). The modulation of short gravity¬
capillary waves by longer scale periodic flows. A higher order theory. Radio
Science 14 1099—1110
 References 443

Vesecky, J. F. & Stewart, R. H. (1982). The observation of ocean surface

phenomena using imagery from the Seasat synthetic aperture radar: an
assessment. J. Geophys. Res. 87C 3397—3430
Wadsworth, A., Robertson, C., & De Staerke, D. (1983). The use of Seasat-SAR
data in oceanography at the IFP. In: Allan, T. D. (ed.). Satellite microwave
remote sensing, Chichester, Ellis Horwood, p. 235—245
Walsh, E. J., Uliana, E. A., & Yaplee, B. S. (1978). Ocean wave height measured
by a high-resolution pulse-limited radar. Boundary-layer Meteorology. 13
263-276
Webb, D. J. (1981). A comparison of Seasat altimeter measurements of wave
height with measurements made by a pitch-roll buoy. J. Geophys. Res.
86C 6394-6398
Wells, N.C. (1979). The effect of a tropical sea-surface temperature anomaly in a
coupled ocean—atmospheric model. J. Geophys. Res. 84C 4971—4984
Wentz, F. J. (1983). A model function for ocean microwave brightness tempera¬
tures. J. Geophys. Res. 88C 1892—1908
Wentz, F. J., Cardone, V. J., & Fedor, L. S. (1982). Intercomparison of wind
speeds inferred by the SASS, altimeter, and SMMR. J. Geophys. Res. 87C
3378-3384
Wesely, M. L. (1979). Heat transfer through the thermal skin of a cooling pond
with waves. J. Geophys. Res. 84C 3693—3700
White, J. V., Sailor, R. V., Lazarewicz, A. R., & Le Schack, A. R. (1983).
Detection of seamount signatures in Seasat altimeter data using matched
filters. J. Geophys. Res. 88C 1541—1551
Wilheit, T. T. (1978). A review of applications of microwave radiometry to
oceanography. Boundary-layer Meteorology 13 277—293
Wilheit, T.T., Chang, A.T. C., & Milman, A. S. (1980). Atmospheric corrections
to passive microwave observations of the ocean. Boundary-layer Meteor¬
ology 18 65—77
Woodcock, A.H. & Stommel, H. (1947). Temperatures observed near the surface
of a fresh water pond at night. J. Met. 4 102 — 103
Woods, J. D. (1977). Information theory related to experiments in the upper
ocean. In: Krauss, E. B. (ed.). Modelling and prediction of the upper layers
of the ocean. Oxford, Pergamon Press, p 263—283
Wurtele, M. G., Woiceshyn, P.M., Peteherych, S., Borowski, M., & Appleby, W. S.
(1982). Wind direction alias removal studies of Seasat scatterometer-derived
wind fields. J. Geophys. Res. 87C 3365—3377
Yentsch, C. S. & Garfield, N. (1981). Principal areas of vertical mixing in the
waters of the Gulf of Maine, with reference to the total productivity of the
area. In: Gower, J. F. R. (ed.). Oceanography from space, New York and
London, Plenum Press, p. 303—312
Zwally, H. J. & Gloersen, P. (1977). Passive microwave images of the polar
regions and research applications. Polar Record 18 431—450
Index

absorption distance measurement, 281-286

of light in sea water, 155—157 geoid information, 289—290
of radiation, 43 instrument design, 281-283
of thermal infrared radiation, 201—204 orbit determination, 287—288
acoustic remote sounding, 20 principles of, 279-281
Active Microwave Instrument (AMI) on surface roughness corrections, 285—286
ERS-1, 421 Amazon rain forest, 405
active sensors, 43,45,52,61,62, amphtude reduction effect (SAR), 359, 360
310-418 Antarctic sea ice, 265—266
Advanced Very High Resolution Radiometer, analogue signals, 54
195, 204-205 {see also AVHRR) analogue-to-digital conversion, 54
aerosol optical thickness, 151,153 Angstrom exponent, 151,171
aerosols, 43,95 angular response of sensor, 47
aerosol scattering of hght, 151,152 anomaUes of SST, 225, 232
African coast (North-West), 74 antenna,
Agena, 30 gain, 250
airborne IR radiometry, 207 passive radiometer, 250—252, 260—261
airborne remote sensing, 23, 104 radar, 47
aircraft sampling, 102 temperature, 252, 254
air-sea gas exchange, 226 aperture synthesis, 352-353
air-sea interaction, 20,225 apogee, 31,32
alias, directional in Scatterometer wind Apollo Programme, 18
retrieval, 407-413 apparent temperature, 201
aliasing, 302 applications of satelhte remote sensing to
ALT (see under Seasat Altimeter) the ocean, 60—64
altimeters, 278—309, 324-333 APT, 56,194
ERS-1, 421 area averaging, 22,49
GEOS-3, 286 area coverage, 21,47-49
general oceanographic applications, 62, archiving of data, 57, 59
63 Arctic sea ice, 265—266
orbit selection for, 306-309 Aries, 33
pulse timing, 279-281 ascending node, 33
Seasat, 281-283,286 aspect control, 26, 30
Skylab, 286 Atlantic Ocean,
spatial sampUng characteristics, 40 altimetry, 297, 298
surface roughness sensors, 324-343 circulation, 68
altimetry, 63,72,278-309 CZCS imagery, 177
appUcation to ocean currents, 290-300 JASIN area, 75
apphcation to tidal studies, 300-304 Landsat data, 191
applied to bathymetry, 306 internal waves on SAR image, 384, 385
atmospheric corrections, 284—285 SAR images of sweU waves, 376,377
 Index 445

thermal IR imagery, 234-235,238-239 direction, 345-346

tides from Seasat Altimeter, 303, 304 image shift, 359
atmospheric, due to waves, 369
absorption coefficient, 253 image smear, 369
corrections, 92-96 resolution, 351-355
in altimetry, 284-285 travelling waves, 377
in image pro cessing, 131
need for ground measurements, 103 backscatter of microwaves {see radar
of AVHRRdata, 215-224 backscatter)
ofCZCSdata, 168-171 backscatter of visible light, 60, 62
of Landsat data, 185 baroclinic eddies, 235
of scatterometer data, 405 baroclinic instability, 70
od SMMR data, 266-270 bathymetry, 61,62,63
of thermal IR measurements, from Landsat MSS, 191—192
215- 219 from SAR imagery, 392
of visible wavelength measurements, beam-limited geometry, 282
152-154 beam transmittance, 149
drag, 30, 32 binary data, 54, 55
effect, on infrared radiation, 201 —204 binary image, 119,120
on microwave radiation, biological processes, remote sensing of, 20,
252-254 175-179
on visible wavelength radiation, Biscay, Bay of, 179
148-151 bit error rate (BER), 55
emission spectrum, 43 bit rates, 55
microwave emission, 253 black body, 199
response to SST anomalies, 225 blackbody emission radiation, 43
sounding, 263 bloom (phytoplankton), 178
transmission, 41,61,202 boundary currents, 68,86
ATSR, 49,95,222,420,422 Bragg scattering, 313, 371
attitude of satellites, 30 Brewster angle, 256
automatic gain control (AGO, 329 brightness temperature, calibration of
automatic picture transmission (APT), 56, AVHRRdata, 205-207
194 infrared, 201
AVHRR (advanced very high resolution microwave, 252, 256-257
radiometer), 195,204—205 broadside mode (SAR), 346.
black-body cavity, 205 bucket temperature, 211
data, bulk SST, 209,210,211
atmospheric correction algorithms, buoy sampling, 20,22,36,40,101
216- 219 comparison with SAR data, 383
atmospheric corrections, 95, comparison with Seasat Altimeter
215-216 waveheight, 331
brightness temperature calibration, comparison with Seasat Altimeter winds,
205-207 332
cloud removal techniques, 219—221
examples of applications, 232—247 calibrated data, 57,131
geometrical distortion of image, 50, calibration algorithms, 90, 91, 99,100,103,
116 131
geometrical rectification, 116-119 calibration data, 54, 91,107
potential uses of, 224-231 calibration models, 99
rate, 55,195 calibration,
field of view, 48,195 of AVHRR data, 205-207
noise, 205 of CZCS data, 166-175
spectral characteristics, 195,203 of CZCS sensor, 166,168
swath width, 40,48,195 Californian coast, 175,176,180,246
visible channels, 192 capillary waves, 81,318
azimuth, compression, 354—355 case 1 water, 161
defocussing, 361 case 2 water, 161
446 Index

CCT, 57,107 currents, 20,62,63,67

celestial shell, 33 curvature of the earth, 51, 98,116
Celtic Sea, 179 dark-water pixels, 169
chemistry of sea water, 20 data, analysis, 89-131
Chesapeake Bay, 265 analysis, developments in, 423 — 425
chirped pulse, 348—349 disk storage, 112
chlorophyll, calibration of CZCS data, 173, dissemination, 51,56-58
174 dropout, 125
calibration of Landsat data, 185 file, 107,108
spectral signature, 158,161 format, 107
chromaticity transform, 130,186,187 organisation, 109-111
circular orbit, 32 rate, 51—55
circulation (ocean), 66-71, 293-300 SAR, 358
climate, 21,64-66,224 recording, 56
climatology, 224 retrieval, 51—58
cloud cover, 19, 20,40 stream from satellites, 108
cloud-free conditions, 101 transmission, 54-58,90
cloud liquid water, 254 datum (in altimetry), 287-290
cloud-removal techniques, AVHRR data, daytime passes, 37
219-221 deep convection, 227
coastal fringe fronts, 244—246 density (of sea water), 20,66,72
coastal geomorphology, 192 depth variability, 20, 75,178
coastal waters, 18,76,77 destriping, 57,184
Coastal Zone Colour Scanner (CZCS), 139 dielectric constant of sea water, 255
application in biological oceanography, diffuse attenuation coefficient, 141,144
175-180 calibration algorithm, 174
application in dynamical oceanography, digital data, filters, 122,123
180-184 format, 107
area coverage, 50 look-up tables (LUT), 126
atmospheric correction, 95,168-173 masks, 119,120,127
calibration for water parameters, organisation, 109-111
173-175 overlays, 119,120
calibration of sensor, 166—168 smoothing, 123
design details, 140 stream, 108
field of view, 48 digital noise, 121
orbital characteristics, 35, 40 digital processing of SAR data, 356-358
radiometric resolution, 54 digital signals, 54
coastline delineation, 120 digitisation interval, 54,121
Cogg Seamount, 303 digitisation levels, 92
CoccoUthophores, 178,191 dinoflagellate bloom, 178
colour, 19,45,46,62,76 direction alias of scatterometer wind
colour density slice, 128 retrieval, 407 — 413
colour sensors 19,21,46 {see also visible directional response of sensors, 47
wavelength sensors) directional wave spectra, 62
colour signature 62,72,74,103 {see also from SAR images, 371
spectral signature) observed by Seasat SAR, 380-383
commercial use of satellites, 57,419 diurnal thermocline, 209, 210
communication system, 26, 56 diurnal tide, 71
computer compatible tape, 57,107 DMSP, 30
contrast stretching, 126 doppler history, 359
convergence, 74 doppler shift, SAR operation, 46, 353
Coriolis force, 68,290 Scatterometer operation, 401-404
Coriolis parameter, 68,70 Dover Strait, 393
cosmic radiation, 253 downwelling, 416
cost of satellite data, 57,419 drift of instrument responses, 91
cumulative histogram, 125 dry atmospheric correction (altimetry),
current meters, 69 284,288
 Index 447

dual frequency scatterometer, 418 Florida Current, 68

dual window AVHRR atmospheric fluxes in the ocean, 70
correction algorithm, 218 foam, effect on microwave emissivity, 257
dumping in the ocean, 65,66 effect on SAR imagery, 371
Dundee University Satellite Receiving footprint, microwave radiometer, 260
Station, 57,109,194 SAR, 346
duration (of wave generation), 82 Foula Isle, radar image, 380
dynamical processes, 61-63,65—77 Fresnel reflectivity, 155
influence on SST, 227 — 228 friction velocity, 396
fronts, 72,86,237-246
Earthnet, 57,58,109 Fimdy, Bay of, 186,187
earth rotation, 33,34 future sensors on satellites, 419—423
eccentricity of orbit ellipse, 33
eddies revealed by SST, 233-235 galactic radiation, 253
eddy kinetic energy from altimetry, gas exchange across sea surface, 226
296-298 Gemini, 18
electromagnetic bias (altimeter), 286 geodesy, 281
electromagnetic modulation (SAR), geoid, 62-63,279-280,289-290
366-368 geometrical transformations, 117—119
electromagnetic radiation, 41-47 geometric correction, 97,114
electromagnetic spectrum, 41, 42, 60 geophysical calibration, 99—100
elliptical orbit, 31,34 GEOS-3 altimeter, 286,327—332
El Quichon volcano (Mexico), 95, 223 Geostationary Meterological Satellites
emission spectra, 41,43,200 (GMS), 35,37,196
emissivity, definition, 199 geostationary orbit, 30,33—35
microwave, 249,254—258 geostationary satellites, 20, 30, 35, 36, 49
sea surface, 61,62 geostrophic balance, 290—292
emittance, 199 global changes in SST, 231
encoding of data, 52, 54 global coverage, 36
English Channel, colour imagery, 180,183, global maps of SST, 230—233
191 Global Positioning System (GPS), 287,421
infrared imagery, 245 global sea height variability, 299
SAR imagery, 357 GOASEX, 413
ephemeris, 287 GOES, general description, 27, 30
equator-crossing latitude, 118 ground coverage, 37
equator-crossing time, 34, 35 orbital characteristics, 35, 37
equatorial bulge, 34 sensors, 192,196,197
equatorial cold tongue, 273—276 GOSSTCOMP, 222
equatorial ocean dynamics, 273,277 gravity surveys, 280
errors, 89 gravity waves, 79
ERS-1, 20,30,420-422 GRAVSAT, 423
grey-levels, 111,128
ESA, 57
ESMR, 263-264,266 ground control points, 97,119
estuaries, 18,76—77,85,188—191 ground receiving stations (SAR), 374
ground segment, 51,57
exitance, 141,142,199
ground track, 37, 39, 308
ground tracking stations, 287
false colour composite, 129
group velocity, 79, 388
fetch, 82
Guinea, Gulf of, 75
films, 61,74,81
Gulf Stream, 68,70,73,234—235
effect on microwave backscatter,
Gulf Stream rings, 70,176,177
317-319
film pressure, 318 gyres, 66, 68,72, 86
filters (digital), 123
financial considerations, 57,419 hardware in space, 22,26—51
haze-removal, 187
fish, 20,65,74
fisheries, uses of ocean colour, 179,180 HCMM, 198
HCMR, 52,62,197,198
uses of SST, 246, 247
448 Index

headland eddies, 72,180 internal waves, 384—391

heat flux, 211,213 generation of, 390
height of satellites, 32, 35 imaged by SAR, 384—391
high resolution imaging radar, 62, 344, 345 observations by Seasat SAR, 384—387
high sea states, effect on SAR, 371 oceanographic information content of,
HIRS, 197 391
histogram, digital image data, 125 packets, 384, 390
histogram equalisation contrast stretch, 126 scales, 86
housekeeping data, 54,108 solitors, 389
hurricane, SAR wave images in, 378 surface signature, 389-391
hydrodynamic modulation (SAR), theory, 387—389
365-366 ionic conductivity, 256
ionospheric error (altimetry), 284
ice, effect on microwave emissivity, 257 irradiance, 141,142
measurement by microwave radiometers, downweUing, 142,143
265 upwelling, 142,143
Iceland-Faroes front, 238—240
IFOV, 47-51,102 JASIN, 75,76,375,383,413
image, atmospheric correction, 131 JPL, 180,246
binary, 119,120
calibration, 131 kinetic energy, 69,70, 298
chromaticity transform, 130
classification techniques, 106,129 Lambertian source, 144
colour density slice, 128 Landsat, development, 18
contrast stretch, 126 infrared sensors, 198
data, 107-112 Multi-Spectral Scanner (MSS), 135-138
degradation (SAR), 358-363 area coverage, 38,109
distortion, 97, 98 atmospheric correction, 185
enhancement, 107,124-131 bathymetry applications, 191
false colour composite, 129 calibration for chlorophyll, 185
fflters, 115,121-123 calibration for suspended sediment,
geometric processing, 114-119 186-188
histograms, 125 coastal applications, 191,192
noise reduction, 121-123 data dissemination, 57,109
‘pretty pictures’, 21,107,131,344 data rates, 56
processing, 106-131 destriping, 184
computers, 106,112,113 field of view, 48,135,136
software, 107,109,123 instrument description, 135
recalibration, 127 oceanographic uses, 184-192
smoothing, 121-123 ocean turbulence, 191
spectral characteristics, 129 pixel geometry, 136,137
warping, 115 pollution, 191
imaging arrays, 420 sediment dynamics, 188-190
imaging radar, 344-345 spectral characteristics, 135,136
inclination of orbital plane, 33-35, 309 Return Beam Vidicon (RBV), 52
in-flight calibration, 91 orbital characteristics, 35, 36-39,135
infrared (IR), absorption, 95,201 satellite description, 27, 28
emission, 95,199-201 spin rate, 30
radiometers, (see also thermal infrared Thematic Mapper (TM), 138-139
sensors) data rate, 56,138
comparison with microwave field of view, 48,138,139
radiometers, 270-272 instrument description, 138
physics, 199-207 spectral characteristics, 135,136
sensors, 18,68,74,75,194-247,422 laser fluorometer, 420
wavebands, 41-43 laser instruments for remote sensing, 61
instantaneous field of view (IFOV), 47-51, laser-ranging, 280,421
102 length scales of oceanographic processes, 86,
 Index 449

87,102 of waves, 368—371

lifetime of satellites, 27, 30 MSS, 135—138 {see also Landsat
light, {see optical) Multi-Spectral Scanner)
light scattering at the sea surface, 164-166 multi-look atmospheric correction of IR
line-of-sight transmission, 55-56 data, 216-219,420,422
location of images, 96 multiple reflections, 257 — 258
longwaves, 79 multiplexing of data, 55
look angles, 50,51
look-up tables (LUT), 126 nadir view, 36, 95
madir-viewing radar, 62,324
Maine, Gulf of, 177 navigation (of image), 96
mapgrids, 56 navigation satellites, 23
Marineland Experiment, 418 NASA, 180,246
masks, 119,120 near-infrared sensors, 61
matched-filter techniques in altimeter near-surface suspended particulates, 62
bathymetry, 306 near-surface thermal structure of the ocean,
MCSST, 222-223 207-215
Mediterranean Sea, 240—241 near-surface wind profile, 396-398
Mercury Programme, 18 NEMS (Nimbus-E Microwave Spectrometer),
mesoscale eddies, 70,73,86 263-264
methodology, 89,182 night-time overpasses, 40
Meteosat, ground coverage, 37 Nimbus-5, 263-264
orbital characteristics, 35, 37 Nimbus-6, 263-264
radiometer characteristics, 196 Nimbus-7, CZCS, 166-184 {see also
satellite description, 27, 30 Coastal Zone Colour Scanner)
sensors, 197 development, 19
visible sensors, 192 orbital characteristics, 35,40
Mexico, Gulf of, 68,293-296 satellite description, 27, 29
microwave antennae, 250—252 SMMR, 261-277 {see also SMMR)
microwave atmospheric sounding, 95 spin rate, 30
microwave bands, 41,42 NOAA, 30,217,232
microwave emission, 62,249-258 NOAA satellite series.
microwave radiometers, 43,45,61, 62, Advanced Very High Resolution
248-277 {see also SMMR) Radiometer, 195,204—205 {see
design constraints, 260—261 also AVHRR)
instrument specifications, 261—263 APT, 194
microwave radiometry, 248-277 data dissemination, 56
atmospheric effects, 252-254 historical development, 18
comparison with IR radiometers, orbital characteristics, 35
270-272 satellite description, 27, 30
instrument design, 260-264 sensors, 197
oceanographic applications, 264-270, Very High Resolution Radiometer
272-273 (VHRR), 195
physical principles, 248-260 non-scanning sensors, 40,47
radiation received by antennae, numerical models, 22,64,425-426
250-252 nutrients, 64,66,74
skin depth in the sea, 258—260
solar radiation effects, 252-254 objections to the use of satellites, 19,20
thermal emission, 249-250 oblique viewing, 51,95,98,116
microwave scatterometers, 395-418 observations at sea, 87
{see also scatterometers) ocean, basins, 85
microwave sensors, 19,20,40,62,63 bathymetry from altimeters, 306
mineral extraction, 65 circulation, 62-64, 66-71,293-300
mixed-layer dynamics, 75 colour, 18,62,76
mixing, 73-15,17 interpretation of, 157-164
modulation transfer function (MTF), 364 pure sea water, 157-158
motion effects (SAR images), 358-362 sensors, 134-193
450 Index

coverage by altimeter, 307-309 theory, 141-166

currents, 290-292 thickness, 150,151
current variability, 293-295 transmittance, 149,150
dynamics, 65—76 scattering in sea water, 155 — 157
influence on surface roughness, water cases, 161
319-320 orbital dynamics, 31-36
eddies, 70,233-235 orbital elements, 31
eddy kinetic energy, 296—297 orbit, characteristics, 20
fluxes, 70 corrections, 30, 34
fronts, 237-240 determination, 287 — 288
heat budget, 226 repeat cycle, 35, 39, 307, 308
heat flux, 209,211,226 selection for altimeters, 306-309
mixed-layer dynamics, 75 overlap, 119,120
resources, 64 overlap zone, 38, 39
sampling techniques, 100—105 ozone optical thickness, 151
science, future trends, 425-426 ozone transmittance, 154
in the 1980 s, 64
multidisciplinary approach, 425-426 Pacific Ocean, 175,176,272—277
theoretical framework, 21,182,425 passive sensors, 43,46,52,53,60
shortwave energy budget, 192—193 passive microwave sensors (see microwave
skin temperature, 207 radiometers)
skin temperature deviation, 210-214 penetration depth, light, 175,193
thermal structure, 207-215 microwave, 258-259
thermocline, 208-210 perfect emitter, 199
tides, 71-72 perigee, 31,33,34
topography, 84, 85 period of orbit, 31,33,35
Topography Experiment (TOPEX), permittivity, 256
422-423 Peruvian coast, 74
upwelling, 74, 237 phaeophytin, 158
waves, 19,62,77-88 phase speed, 78,80,81,388
wave spectra and Bragg scattering, photography from space, 18
315-317 photon backscatter probability, 154
oceanographic applications, altimeter data, phytoplankton, 62,66
209-306 measured with CZCS, 174
CZCSdata, 175-184 optical properties, 158,161
remote sensing data, 60 — 64 picture transmission, 56
SARdata, 391-394 pitch (of satellite vehicle), 96
SMMRdata, 272-277 pixel (picture element), 49,50,54
thermal IR measurements of SST, location, 119
232-247 resampling, 115,116
oceanographic length scales, 64, 69, 71 Planck’s Law, 199,249
oceanographic parameters measured from plankton, 60, 62, 66
space, 60-64,99 plateau-droop effect, 326-327
oceanographic processes, 65 platforms in space, 26-31
oceanographic time scales, 64,71 point samples, 22,49
offshore industry, 65 point-spread function, 49
oil films, 61 polarization, 46,47
effect on microvare emissivity, 257 microwave radiometry, 255
oil slicks, 61 reflection of microwaves, 315
operational algorithms, 99 polar orbit, 34—36
operational satellites, 18, 56, 57,196 pollution, 77, 228
operational SST products, 221-224 population dynamics, 64
optical, absorption in sea water, 155 -157 Portland Bill, 181
pathways in the atmosphere, 148-151 Portuguese coast, internal waves imaged by
processing of SARdata, 358 SAR, 386
properties of the sea surface, 164-166 positional registration, 96-98
quantities, 141-144 position-fixing at sea, 104
 Index 451

power pattern of microwave antennae, 251 reflected radiation, 60,62

power reflection coefficient, 253-255 refraction, of light, 155 — 156
power requirements of satellites, 27, 28 of swell waves, 380
precession, 34, 36 remote-sensing principles, 89-105
rates, 309 repeat cycle of orbit, 35, 39, 307, 308
pre-flight calibration, 91 residual flows, 76
productivity, 20,64,74 retrograde orbit, 34
measured with CZCS, 175—178 Return Beam Vidicon (RBV), 52
patches, 178 Reynolds stress, 396
pulse, chirped, 348-349 ripples, 81
compression, 282 interaction with longer waves, 317
limited geometry, 282,325 roll of satellite vehicle, 96
repetition frequency, 354, 373 Rossby radius, 70
return shape, 324-328
return strength, 62 salinity, 20,61,62,66,72
return time, 46,62-63,279,281 effect on microwave emissivity,
shape, 62,281 — 283 256-258
push-broom imaging, 420 measured by microwave radiometers,
265
radar, antennae, 47 samples (scan line), 48,49
backscatter, 46,311 sampling interval or period, 38-40,87
backscatter cross-section, 311,328—329 sampling strategy of sensors, 52, 53,103
beamwidth, 351—352 SAR, 344-394 (see also synthetic aperture
frequencies, 41,42 radar and Seasat SAR)
reflection from the sea surface, Sargasso, 70,73
311-315 SASS, 401-415 {see also Seasat
scattering, 311—315 Scatterometer)
specular reflection, 311-313 satellite, ephemeris, 287
wavebands, 41,42 oceanographer, 89,426
Radarsat, 423 oceanography, 17,20,22,425-427
radiance, 45,141,143—148 sensors, 27
radiant energy, 141,142 descriptions, 52-53
flux, 141,142 applications, 62
intensity, 141,143 tracking, 96
power, 141,142 vehicles, 26—30
radiative transfer equation, microwave, 252 SCAMS, 263-264
radiometers, 45 Scan geometry, 48,49
radiometric resolution, 54,92 Scanning Multichannel Microwave
rainfall measurement, 263—264 Radiometer, 260—277 (see also
rain rate algorithm for SMMR, 268 SMMR)
range, compression, 349 scanning sensors, 38 — 40,47,48
direction, 345—346 SCAT, 401-415 {see also Seasat
resolution, 347 — 350 Scatterometer)
travelling swell, 374 scattering of light in sea water, 155-157
walk, 359, 360 scatterometers, 395-396
walk corrections, 351 applications, 40,43,52,62,63
Rayleigh-Jeans approximation, 250 dual frequency, 418
Rayleigh optical thickness, 151 ERS-1, 421
Rayleigh path radiance, 152,171 Seasat, 401-415 {see also Seasat
Rayleigh scattering, visible light, 151 Scatterometer)
radar, 371 wave, 417-418
RBV, 52 wind, 396-401
record length, 40 applications, 415 — 417
reflectance ratio, 155,157 sea, bed, 61,62
reflectance spectra, case 1 water, 161-162 data, 100-105
case 2 water, 162-164 ice, measured by microwave radiometry,
reflectance spectral ratio, 156,157 265
452 Index

Seasat, 27, 29, 35 (VIRR), 198

Altimeter (ALT), 286 seasonal thermocline, 209
achievements, 278 sea state, 18,19,62
applications, 62 sea state bias (altimeter), 286
application to ocean circulation sea-surface, height-, 19, 20, 287-290
phenomena, 293-300 height variabiUty, 296,297,299
applied to surface roughness optical reflection, 164—166
measurement, 326-343 reflection of radar, 311—315
atmospheric corrections, 284-285 roughness {see surface roughness)
comparison of wind retrieval with slope, 291
SASS and SMMR, 415 temperature (see SST)
dry atmospheric correction, 285,288 topography, 292,297
electromagnetic bias, 286,288 sea truth, 100—105
error summary, 288,289 sedimentation, 77
instrument design, 281—283 self-calibration data, 54
ionospheric correction, 284,288 semidiurnal tides, 71
sea-state bias, 286,288 sensor (s), applications, 62
specification, 286 calibration, 90-92
surface roughness error correction, degradation, 91
285-286 general descriptions, 52-53
tracker bias, 286,288 geometric properties, 146-148
wave-height algorithms, 329—331 on which satellites, 27
wet atmospheric correction, 285, 288 spectral response, 145-146
wind-speed algorithms, 331-332 shallow sea fronts, 240-244
coincidence with JASIN, 76 shallow seas, 72, 73, 77
description, 27,29,30 shallow sea topography imaged by SAR,
Ephemeris, 287 392-393
historical context, 19 shelf seas, 73,85
orbit characteristics, 35 ship of opportunity, 101
research programme, 57 ship-sampling, 40,49,87
SAR, 372 ship wake, 361,392
data coverage, 374 ship-wave patterns, 321-322, 361, 392
data rate, 56 short-wave radiant energy, 193
imagery of other phenomena, Shuttle Imaging Radar (SIR), 345
392-394 side-looking radar, 62
instrument description, 372-375 signal-to-noise ratio, 55
internal wave imagery, 384-391 significant wave height (hi/,), 62, 83
parameter values, 372 algorithms for altimeter, 329-331
synchronisation constraints, 373 effect on altimeter pulse shape,
wave imagery, 375-379 325-327
wave spectra, 380-383 effect on altimeter pulse strength,
Scatterometer (SASS or SCAT), 328-329
401-415 measured by altimeter, 332-335
antenna sequence, 403 size of satellites, 26, 27
algorithm validation, 413-415 skin, depth of microwave emission,
comparison with SMMR and 258-260
altimeter, 415 effect (thermal), 210-214
data processing, 405 layer, 21,210,259
direction alias, 408-413 temperature, 101,207-208
doppler cells, 403 temperature deviation, 210-214
footprint, 402,403 Sky lab Altimeter, 286
instrument design, 401-402 slicks, 61, 74
introduction to, 395—396 associated with internal waves, 388,
operation, 402—404 390
wind-vector algorithm, 406-415 effect on microwave backscatter,
SMMR, 261—262 (see also SMMR) 317-319
Visible and Infra-Red Radiometer effect on microwave emissivity, 257
 Index 453

thermal effects, 214,215 measurement, 270-272

SMMR, 260-277 from infrared sensors, 194-247
algorithms for environmental parameters, from microwave sensors, 248-277
266-270 potential uses, 224-231
comparison of winds with SASS and retrieval algorithms for SMMR,
Altimeter, 415 266-270
environmental model, 269-270 stratification, 73,75
general description, 261-263 parameter, 73
geometric configuration, 261-262, 267 subsurface reflectance ratio, 155
oceanographic applications, 272—277 sun glitter, 18
use in altimeter atmospheric corrections, sun-synchronous orbit, 34-36,38-39
285 surface convergence, 74,318,320
use in scatterometer atmospheric internal waves, 390
corrections, 405 wind-stress curl, 416
solar cells, 30 surface film (see films)
solar heating, 73, 76 surface roughness, 18-20,46, 62,72,74
solar illumination, 46,60 active microwave sensing, 310—323
solar irradiance, 154 artificially caused, 321-323
solar radiation, 30,34,41,62,76 dynamically produced, 319-321
Solent, the, 187,189 microwave emission, 258
solitons, internal, 389—391 SAR imaging of waves, 363-364
Southern Ocean, 65 wind-generated, 315-317
Southern Hemisphere wind and waves from surface skin, 20, 21
altimeter, 335-337 surface slick (see slicks)
space platforms, 26-40 surface tension, 81,82,318
spatial averaging, 103 surface wave pattern, 74
spatial resolution, 40, 46, 47 — 51,100 suspended sediment, 20,60,62
imaging radars, 344 calibration of CZCS data, 174
microwave antennae, 260 calibration of Landsat MSS data,
Seasat Scatterometer, 401—404 186-188
spatial sampling, 21, 36 — 40,47 — 51, CZCS imagery, 181,183
84-88,101,131 Landsat imagery, 189
altimeter orbits, 308 movement, 188,190
spatial variability, length scales, 86,102 spectral signature, 160,162
measurement, 124 swath width, 38-40,48
speckle (SAR images), 362—363 swell waves, 19,80
spectral quantities, 141,199 height from altimeter data, 333, 337,
spectral signature, 46,129 339-343
chlorophyll, 158,159 imaged by Seasat SAR, 375-380
pure sea water, 157,158 synoptic measurements, 76,100
suspended particulates, 160,161 synoptic view, 19,21
yellow substance, 159,160 synthetic aperture radar (SAR), 43,46,52,
spectral windows, 87 63,72,74,344-394
specular reflection, radar, 311—313 azimuth resolution, 351-355
visible light, 164-165 directional wave spectra, 371
split-window algorithm for AVHRR geometry, 345-347
atmospheric correction, 218 image degradation, 358—363
squint mode (SAR), 346 imaging mechanism, 345
stability, atmospheric, 397 imaging of ocean waves, 363-372
Stefan’s constant, 199 electromagnetic modulation,
Storm surges, 86 366-368
SST, 18, 19,57,61,62 high sea states, 369-370
anomalies, 225,232 hydrodynamic modulation, 365—366
applications of satellite measurement, mechanisms, 363-365
232-247 motion effects, 368—371
appropriate definition, 223 tilt modulation, 366
comparison between IR and microwave internal waves image, 384-391
454 Index

modulation transfer function, 364 242-244

moving targets, 358-362 upwelling, 74,237,416
principles of operation, 345—363 upwelling radiation, 60,143
processing, 356-358 users of satelhte data, 57
range resolution, 347 — 350 uses of the ocean, 64
Seasat, 372 (see also Seasat SAR) Ushant Front, 179
Shuttle Imaging Radar, 345
speckle, 362—363 varibUity, length scale, 86,102,182
synthetic aperture principle, 351-355 sea surface topography, 296—297
time scale, 86,100
Tape recording, 56 VAS, 197
TDRSS, 56 velocity bunching, 369,371
temperature, 20,66,72,74—75 vertical sampling, 20
temporal sampling, 20,21,36-40,84-88 vertical structure, 20
test ramp (digital), 54 productivity, 178
Thematic Mapper (TM), 138-139 (see also thermal, 75,76,193
Landsat Thematic Mapper) vessel, research, 20,22,101
theoretical base, 21,182,425 VHRR, 195
thermal control of satellites, 30 application of data, 235
thermal emission, 62 viewing angle, 50,51
infrared, 199-201 visible wavebands, 41,42,45
microwave, 249-250 visible wavelength sensors, 134-193
thermal infrared sensors, 45,61, 62, 68 analogy with human vision, 45 — 56
74,194-247,422 development of, 18
thermal wind equations, 291 directional response, 47
thermocline, 73,75,208-210 general applications, 60,62,68,74
depth determined from SAR images, 391 geometric properties, 146—148
thermohaline circulation, 66,74 spectral response, 145,146
threshold response of sensors, 47 VISSR, 196,197
tidal aliasing, 302,309 VISSR Atmospheric Sounder, 197
tidal constituents, 302
tidal fronts, 73,240—244 wakes, 63,321-322,361,392
tides, 62,63,71,72,86 warping of images, 97
from altimetry, 300-304 water - leaving radiance, 148,152,155,169,
tilt modulation, 366-367 170
time sampling (see temporal sampling) water masses, 66,68,72
time scales of oceanographic processes, 86 water mass formation, 227
TIROS satellite series, AVHRR, 195, water vapour, 43
204-205 (see also AVHRR) effect on microwaves, 254
historical development, 18 measured by SMMR, 266, 268
orbital characteristics, 35 wavebands (electromagnetic), 41,42
satellite description, 27, 28, 30 wave (sea surface), 20,65,74,77-88
TOPEX, 69,306-309,422-423 breaking, 371
tracker bias (altimeter), 286 climate data from altimeters, 333-339
transmittance, 44,149 current interaction, 320
triple-window algorithm for AVHRR dispersion, 79
atmospheric correction, 218 height, 62,63,86
tropospheric error (altimetry), 284, 285 applications of altimeter
turbidity, 18,188 measurements, 332-343
turbulence, 69,71,73 measured by altimeter, 326-331
in CZCS imagery, 183 Southern Hemisphere, 335
in Landsat MSS imagery, 191 imaging from Seasat SAR, 375-379
turbulent mixing, 64 length, 78,80,81,86
two-scale wave interaction theory,' 365 number, 78
two-scale wave probe, 418 period, 78, 80, 86
prediction, 84
U.K. shelf seas, summer thermal structure. refraction, 380
 Index 455

spectra, 62, 83, 316 measured by passive microwave

directional, from SAR, 371, radiometers, 265—270
380-383 Southern Hemisphere, 336
wave interaction, 316—317 stress, 396-398
weather, 65 algorithms for Seasat scatterometer,
prediction, 225-226 398-401
weight of satellites, 26,27 vector algorithm for Seasat scatterometer,
West Coast Experiment, 418 406-410
wet atmospheric correction (altimetry), velocities profile, near-surface, 396-398
284,288 World Ocean Circulation Experiment
Wien’s displacement law, 199,200 (WOCE), 423
wind-driven ocean circulation, 66 Wyville-Thompson Ridge, SAR image of
wind, generation of surface waves, 82 internal waves, 385
measurement, 62,63,65
spatial variability revealed by SAR, 392 Yaw of satellite vehicle, 96
speed, algorithms for altimeter, yellow substance, 60, 62
331-332 spectral signature, 160,163
algorithms for scatterometers,
398-401 Zooplankton, 66
 ...

.■ -Jr •'■• *■■ ■’ '-

fT> -
; ^- ■ ,..ii‘, ii£l * yMiff .»* • '- ^-■

'W? ^ I*-'
- MtK ‘i*'** , ' - ^ <|0 '.jr,; ^

^P,*' .V *'- '.nl ."<>1 ,v4.-,Tyi">^

ii^-; ■ ;^'',rtvifcr'.<. ■;•(* '■'• , if iM^Tvai-n

” - ; .i •!_ ‘4
' .i^jyy, ,.,.«i;i( n-iWiii**^ . .£.'■£. hXAWi*:^ VI
'.:’ c'.v'%(*»*.*'■■'I.'
; ‘ tii.it.>^-.-r ■■• iNR. .•,. '■' ■'-r, ■ /»

.!f'ry 1 ^ ITw. 'sJa* i^’r((rt ”* •

-,-;:• ^4.¥.Vvi ' •■

.HI ■-. 'i^wv- I.3-*. ■*J.*>|

»«».' ^-if^i^.; »x ■‘tlM.J

ii.
II ji... t.■'.»• <■' ... -i.i
 Robinson.
R661s Satellite oceanography.

-^^^yyicClDENTAL COLLEGE LIBHAHV

551.46«ij||^jan„„raohV: an.i/Rpbinsonj
REMOTE SENSING IN METEOROLOGY, OCEANOGRAPHY AND HYDROLOGY
Editor; A. P. CRACKNELL, Carnegie Laboratory of Physics, University of Dundee
"timely and valuable ... an excellent book" — J. O. Thomas, Imperial College of Science and Technology,
University of London, in International Journal of Remote Sensing.
"a book with a unique history . . , undeniably handsome" — E, C. Barrett, University of Bristol, in Journal oi'<
Climatology.

SATELLITE MICROWAVE REMOTE SENSING

Editor; T. D. ALLAN, Research Scientist, Institute of Oceanographic Sciences, Wormley
"Tom Allan, who has for many years been a leadirtg figure in remote sensing research in this country,
especially in regard to SEASAT, has done a splendid iob in editing these papers into a coherent volume"
— David Bowers in British Book News.

PHYSICAL OCEANOGRAPHY OF COASTAL WATERS |

K. F. BOWDEN, Emeritus Professor of Oceanography, University of Liverpool i
"a sound and thorough account of the complicated physics of an important and increasingly studied oceanic :|
domain and an encouraging start to a new series in marine science" — Henry Charnock, Professor of
Oceanography, University of Southampton, in The Times Higher Education Supplement

NEW PERSPECTIVES IN MARINE GEOLOGY

R. C. SEARLE and R. B. KIDD, Principal Scientific Officers, Institute of Oceanographic
Sciences, Wormley
Presents highly detailed sonar images of a wide range of sea-floor features showing structure andq
detail on a scale not achievable with other instruments. A unique book which will appeal to
students and researchers of marine tectonics, sedimentology, geomorphology, and volcanology. 1
Also for professional specialist geologists.

PRINCIPLES OF COMPUTER COMMUNICATION NETWORK DESIGN

J. SEIDLER, Institute of Fundamentals of Informatics, Polish Academy of Sciences
Translation Editor; R. J, DEASINGTON, Computer Centre, University of Strathclyde
"of valu« to statisticians working in the design and utilisation of computer communication networks, in
particular transportation networks. Those working more generally in information theory could find it a
welcome addition to their library" — The Professional Statistician.

ENVIRONMENTAL AERODYNAMICS
R. S. SCORER, Professor of Theoretical Mechanics. Imperial College of Science and
Technology, University of London
"authoritatively written by an eminent expert... excellently illustrated" - Choice (USA).
"authoritative and vary readable ... will become a wall read text for those interested in a quantitative
explanation to hydrodynamical problems of the great outdoors" — Garry Hunt in New Scientist.

THE PHYSICAL ENVIRONMENT

B. K. RIDLEY, Department of Physics, University of Essex
"a handy source book of environmental data" — H. Egan in Chemistry and Industry.
"should find a place on the library shelves ... I cannot recommend this book too highly" — R, Leigh in
The School Science Review.

published by distributed by : j
ELLIS NORWOOD LIMITED HALSTED PRESS a division of \
Publishers Chichester JOHN WILEY & SONS
New York Chichester Brisbane Toronto

Halsted Press Edition ISBN 0-470-20148-7