Atmospheric Correction

for
Satellite Ocean Color Radiometry
********

A Tutorial and Documentation

of the Algorithms Used by the
NASA Ocean Biology Processing Group

Curtis D. Mobley
Sequoia Scientific, Inc.
2700 Richards Road, Suite 107
Bellevue, WA 98005

Jeremy Werdell, Bryan Franz, Ziauddin Ahmad, and Sean Bailey

NASA Goddard Space Flight Center
Ocean Ecology Laboratory
Greenbelt, MD 20771

June 24, 2016

 Abstract

This tutorial is an introduction to atmospheric correction in general and also documentation of the
atmospheric correction algorithms currently implemented by the NASA Ocean Biology Processing
Group (OBPG) for processing ocean color data from satellite-borne sensors such as MODIS and
VIIRS. The intended audience is graduate students or others who are encountering this topic for the
first time. The tutorial is in two parts. Part I discusses the generic atmospheric correction problem.
The magnitude and nature of the problem are first illustrated with numerical results generated by
a coupled ocean-atmosphere radiative transfer model. That code allow the various contributions
(Rayleigh and aerosol path radiance, surface reflectance, water-leaving radiance, etc.) to the top-
of-the-atmosphere (TOA) radiance to be separated out. Particular attention is then paid to the
definition, calculation, and interpretation of the so-called “exact normalized water-leaving radiance”
and its equivalent reflectance. Part I ends with chapters on the calculation of direct and diffuse
atmospheric transmittances, and on how vicarious calibration is performed. Part II then describes
one by one the particular algorithms currently used by the OBPG to effect the various steps of
the atmospheric correction process, viz. the corrections for absorption and scattering by gases and
aerosols, Sun and sky reflectance by the sea surface and whitecaps, and finally corrections for sensor
out-of-band response and polarization effects. One goal of the tutorial—guided by teaching needs—
is to distill the results of dozens of papers published over several decades of research in atmospheric
correction for ocean color remote sensing. Any subsequent modifications to the originally published
techniques are noted in the documentation. This content of this tutorial is available online as
the Atmospheric Correction chapter of the Ocean Optics Web Book, beginning at http://www.
oceanopticsbook.info/view/atmospheric_correction/chapter_overview. A pdf version of
the report can be downloaded as Mobley et al. (2016) in the publications section of the Ocean
Optics Web Book.

Acknowledgments

This work was supported by NASA Grant NNX14AQ49G to author C.D.M. titled Documentation
of NASA Ocean Color Atmospheric Correction Algorithms in Preparation for the 2015 Summer
Class in Optical Oceanography and Ocean Color Remote Sensing. This report constitutes part
of the final report on that grant. Howard Gordon made useful comments on the computation of
diffuse attenuation, and he and David Antoine helped with the formulation and interpretation of
normalized reflectances.
Contents

List of Figures iv

List of Tables vi

I The Atmospheric Correction Problem 1

1 Problem Formulation 4

2 Example Radiances 8

3 Normalized Reflectances 13
3.1 Normalized Radiances and Reflectances . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.2 The BRDF Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

4 Atmospheric Transmittances 25
4.1 Direct Transmittance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.2 Diffuse Transmittance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

5 Vicarious Calibration 29

II The NASA OBPG Algorithms 34

6 Gases 38
6.1 Nonabsorbing Gases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
6.1.1 Wind Speed and Surface Reflectance Effects . . . . . . . . . . . . . . . . . . . 38
6.1.2 Pressure Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
6.2 Absorbing Gases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
6.2.1 Absorption by Ozone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
6.2.2 Absorption by NO2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

7 Sun Glint 46

8 Whitecaps 48

ii
9 Aerosols 50
9.1 Aerosol Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
9.2 Black-pixel Calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
9.3 Non-black-pixel Calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
9.4 Strongly Absorbing Aerosols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

10 Spectral Out-of-band Correction 60

11 Polarization Correction 66

References 70

iii
List of Figures

1.1 Processes contributing to the TOA radiance . . . . . . . . . . . . . . . . . . . . . . . 6

2.1 Geometry for example calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.2 Example radiances contributing to the TOA radiance . . . . . . . . . . . . . . . . . . 10
2.3 Fractional contributions to the TOA radiance . . . . . . . . . . . . . . . . . . . . . . 11
2.4 TOA radiances for various environmental and viewing conditions . . . . . . . . . . . 12

3.1 Comparisons of Lu and Lw for a zenith Sun, with and without an atmosphere . . . . 15
3.2 Comparison of exact normalized and unnormalized water-leaving reflectances . . . . 22
3.3 Example reflectances contributing to the TOA reflectance . . . . . . . . . . . . . . . 22

4.1 Direct and diffuse transmittance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

5.1 Flowchart of the atmospheric correction process . . . . . . . . . . . . . . . . . . . . . 36

6.1 Rayleigh optical thickness and depolarization ratio . . . . . . . . . . . . . . . . . . . 39

6.2 Transmittance by O2 and H2 O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
6.3 Transmittance by O3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
6.4 Transmittance by NO2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
6.5 Band-averaged optical depth and cross sections . . . . . . . . . . . . . . . . . . . . . 43

8.1 Whitecap reflectance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

9.1 Illustration of aerosol volume, number, and particle size distributions. . . . . . . . . 51

9.2 Relative humidity effects on an aerosol volume and particle size distributions. . . . . 52
9.3 Dependence of (λ, 865) on aerosol model . . . . . . . . . . . . . . . . . . . . . . . . 55
9.4 Remote-sensing reflectances in the NIR . . . . . . . . . . . . . . . . . . . . . . . . . 56
9.5 Qualitative behavior of (λ, 865) for blue-absorbing aerosols . . . . . . . . . . . . . . 56
9.6 Ocean regions for non-black-pixel corrections . . . . . . . . . . . . . . . . . . . . . . 59

10.1 MODIS sensor response; linear ordinate . . . . . . . . . . . . . . . . . . . . . . . . . 61

10.2 MODIS sensor response; logarithmic ordinate . . . . . . . . . . . . . . . . . . . . . . 61
10.3 MODIS 412 nm sensor response vs. an idealized 10 nm FWHM response. . . . . . . 62
10.4 Rrs spectra as functions of the chlorophyll concentration . . . . . . . . . . . . . . . . 63
10.5 Rrs responses for the SeaWiFS 555 nm band vs. an idealized 11 nm FWHM sensor . 64
10.6 Example out-of-band correction factor . . . . . . . . . . . . . . . . . . . . . . . . . . 65

iv
11.1 Geometry for polarization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

v
List of Tables

1.1 Radiance notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

5.1 Processing Levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

5.2 Ancillary data needed for atmospheric correction . . . . . . . . . . . . . . . . . . . . 37

9.1 NIR bands used for aerosol correction . . . . . . . . . . . . . . . . . . . . . . . . . . 54

vi
 Part I

The Atmospheric Correction Problem

1
 In several recent years the NASA Ocean Biology and Biogeochemistry Program supported an
intensive summer course, “Ocean Optics Summer Class: Calibration and Validation in Support of
Ocean Color Remote Sensing,” at the University of Maine (http://misclab.umeoce.maine.edu/
education.php). Those graduate-level classes covered both theory and instrumentation for optical
oceanography and ocean color remote sensing. During those courses, Jeremy Werdell of the NASA
Goddard Space Flight Center, Ocean Biology Processing Group (OBPG) gave lectures on how the
OBPG calibrates, validates, and processes ocean color data from sensors such as SeaWiFS, MODIS,
and VIIRS (Werdell, 2015). His lectures outlined the many complicated steps used for atmospheric
correction of measured at-sensor radiances and inspired this tutorial.
The purpose of this tutorial is to expand upon Werdell’s lectures and review and summarize
in one document the entire process of atmospheric correction as currently implemented by OBPG.
The algorithms and equations presented here rest on several decades of research going all the way
back to the Coastal Zone Color Scanner (CZCS), which was launched in 1978. References are
given to the original literature, which can be consulted for historical perspective and the scientific
underpinnings and details of the current algorithms.
There are many other sources with additional information about atmospheric correction. The
NASA ocean color web site at http://oceancolor.gsfc.nasa.gov/cms/ contains a wealth of
information about how NASA collects, processes, calibrates, validates, archives and distributes
ocean color data from a variety of satellite sensors. That web site has many pages with links
to various technical memos and other information about ocean color, and many of the data files
underlying the atmospheric correction process can be downloaded there.
There are also many non-NASA sources of information on atmospheric correction. The Ocean
Optics Web Book (http://www.oceanopticsbook.info/) presents basic information on optical
oceanography and ocean-color remote sensing needed to understand the present tutorial. The Uni-
versity of Maine website given above links to PowerPoint presentations of lectures given at the Uni-
versity of Maine summer courses, and to videos of the 2015 lectures. The International Ocean Color
Coordinating Group (IOCCG) has hosted summer lecture series during which the lectures were
videoed. The IOCCG lectures delivered by Menghua Wang in 2012 and 2014 (http://www.ioccg.
org/training/SLS-2012/ and http://www.ioccg.org/training/SLS-2014/) cover much of the
material presented here. IOCCG report 10 (http://www.ioccg.org/reports/report10.pdf)
compares the SeaWiFS-MODIS vs. MERIS vs. OCTS-GLI vs. POLDER atmospheric correc-
tion algorithms, but assumes that the reader is already familiar with the general process.

This tutorial is organized as follows. Part I first formulates the atmospheric correction problem
in terms of the various contributions to the top-of-the-atmosphere (TOA) radiance measured by
a satellite-borne sensor. Those contributions come from solar radiance scattered by atmospheric
molecules and aerosols, Sun and sky radiance reflected by the sea surface (either by the water
surface itself or by foam from whitecaps), and finally from water-leaving radiance. The nature and
magnitude of these contributions is then illustrated using numerical simulations from a coupled
ocean-atmosphere radiative transfer model. The computations of various reflectances and atmo-
spheric transmittances are then discussed in detail. Part II then treats the various contributions
in turn, showing how each undesired contribution is estimated so that it can be removed from the
measured TOA value. The end result is an estimate of the water-leaving radiance, or its correspond-
ing exact normalized water-leaving reflectance, which carries information about the water-column

2
itself.
Once obtained, the normalized water-leaving reflectance is the input to algorithms for retrieval
of various quantities of scientific interest. These ocean-color products include—among others—the
Chlorophyll a concentration, the water-column diffuse attenuation for downwelling plane irradi-
ance at 490 nm (Kd490, which is a proxy for water transparency), water-column absorption and
backscatter coefficients, and particulate organic and inorganic Carbon. The algorithms for re-
trieval of specific products, given the normalized reflectance, are given in a series of reports found
at http://oceancolor.gsfc.nasa.gov/cms/atbd. Those retrieval algorithms are not discussed
here.

3
 CHAPTER 1

Problem Formulation

The total radiance Lt measured by a satellite-borne sensor at the top of the atmosphere (TOA)
comes from contributions by atmospheric scattering, Latm ; Sun and sky radiance reflected back
upward by the sea-surface and reaching the TOA, LTOAsurf ; and water-leaving radiance that reaches
the TOA, LTOA
w :
Lt = Latm + LTOA
surf + Lw
TOA
. (1.1)
For brevity, the viewing direction (θv , φv ) and wavelength λ are not shown. Expanding this equation
into further levels of detail requires the definition of many different radiances, and precise notation
is needed to minimise confusion. The atmospheric contribution Latm is always considered to be
at the TOA. However, the surface-reflected radiance and water-leaving radiance can be formulated
either at the sea surface or at the TOA. For these radiances, a superscript TOA will be used to
specify the TOA value. Thus Lw will denote the water-leaving radiance just above the sea surface,
and LTOA
w will denote how much of Lw reaches the TOA. Table 1.1 summarizes the various radiances
introduced in this chapter and used throughout this report.
The atmospheric contribution in Eq. (1.1), usually called atmospheric path radiance, comes
from scattering by atmospheric gases and aerosols, including multiple scattering between gases and
aerosols. The path radiance that comes solely from scattering by atmospheric gas molecules is
usually called the Rayleigh radiance, LR , because scattering by molecules is well described by the
Rayleigh mathematical model of scattering by particles that are much smaller than the wavelength
of light. In the absence of any aerosols, the atmospheric path radiance would equal the Rayleigh
radiance. Let La denote the aerosol contribution, which is the path radiance that would occur if
the atmosphere consisted only of aerosol particles. Let LaR denote the contribution resulting from
multiple scattering between aerosols and gases. The total surface reflectance can be separated into
a contribution due to direct Sun glint from the water surface, LTOA g ; by background sky radiance
reflected by the water surface, LTOA sky ; and by Sun and sky radiance reflected by whitecaps and
TOA
foam, Lwc . Thus Eq. (1.1) can be further partitioned into

Lt = LR + [La + LaR ] + LTOA

g + LTOA TOA
sky + Lwc + LTOA
w . (1.2)

In practice, the aerosol and aerosol-gas contributions are usually grouped together and treated as
one contribution, sometimes denoted LA = La + LaR and often called just the aerosol contribution.

4
Table 1.1: Radiance notation. Spectral radiance L has SI units of W m−2 nm−1 sr−1 ; in practice
mW cm−2 µm−1 sr−1 is often used.
Symbol Definition
Lt total upwelling radiance at the top of the atmosphere
Latm total contribution of atmospheric scattering to the TOA radiance
LTOA
surf total contribution of surface-reflected radiance to the TOA radiance
LR total Rayleigh radiance at the TOA
Lr “standardardized” Rayleigh radiance at the TOA
La TOA radiance due to scattering by aerosols only
LaR TOA radiance due to aerosol-molecule scattering
LA La + LaR ; total aerosol radiance at the TOA
Lw water-leaving radiance just above the sea surface
LTOA
w the part of the water-leaving radiance Lw that reaches the TOA
Lg direct Sun glint radiance just above the sea surface
LTOA
g the part of the direct Sun glint radiance Lg that reaches the TOA
Lsky surface-reflected background sky radiance at the sea surface
LTOA
sky the part of the surface-reflected background sky radiance Lsky that
reaches the TOA
Lwc radiance due to whitecaps and foam just above the sea surface
LTOA
wc the part of the whitecap radiance Lwc that reaches the TOA
Lu upwelling underwater radiance just beneath the sea surface

The sky reflectance term is accounted for as part of the Rayleigh correction, which incorporates
reflectance by the sea surface. For some sensors that were specifically optimized for ocean color
(e.g., CZCS and SeaWiFS), the strongest part of the Sun glint (the Sun’s glitter pattern) is avoided
by pointing the sensor in a direction away from the Sun so that almost no direct glint is present
in the image. However, there is still a correction for residual amounts of Sun glint. Figure 1.1
illustrates these contributions to the TOA radiance.
Most papers (e.g., Wang and Bailey, 2001; Wang, 2002) rewrite Eq. (1.2) as
Lt = LR + [La + LRa ] + T Lg + tLwc + tLw , (1.3)
or something very similar. Now, however, Lg , Lwc , and Lw are all measured at sea level. T is the
direct transmittance between the sea surface and the TOA along the viewing direction, and t is
diffuse transmittance in the viewing direction. These transmittances are discussed in §4.
Yet a third formulation can be found in the literature (e.g., Franz et al., 2007, Eq. 1):
 
Lt = Lr + [La + Lra ] + tdv Lwc + tdv Lw tgv tgs fp . (1.4)

5
Figure 1.1: Qualitative illustrations of the various processes contributing to the total TOA radiance.
The notation corresponds to Table 1.1 and colors correspond to the spectra of Fig. 2.2. The blue
N-N represents a nitrogen (N2 ) molecule, or any other atmospheric gas molecule; the brown blob
represents an aerosol particle. The red glint terms are discussed in §2.

Here tdv is the diffuse transmittance along the viewing path of the sensor. tgv is the transmittance
by atmospheric gases in the viewing direction, and tgs is the transmittance by atmospheric gases in
the Sun’s direction; these transmittances are usually called gaseous transmittances. fp is a known
instrument polarization-correction factor. Comparison of Eqs. (1.3) and (1.4) shows, for example,
that
LR = Lr tgv tgs fp .
Thus the total TOA Rayleigh contribution LR has been factored into a product of terms involving
a Rayleigh term times gaseous transmittances and a polarization-correction factor. The difference
between Eqs. (1.3) and (1.4) is primarily a matter of simplification for presentation purposes. The
fp term came into the nomenclature because MODIS has large polarization sensitivity and this
requires correction. Earlier papers by Gordon and Wang often ignored the gaseous transmission
terms because they were only considering ozone, which could be “taken off the top,” so to speak,
with the remaining problem being effectively formulated below the ozone layer. The Lr term is
computed using a standard atmosphere and only non-absorbing gases N2 and O2 . This allows
“standard” Rayleigh radiances Lr to be computed as a function of Sun and viewing geometry. The
gaseous transmittances are computed by use of absorption coefficients, computed path lengths, and

6
gas concentrations for the various gases. The fp term is computed for each image pixel as a function
of atmosphere and surface polarization states (modeled Rayleigh and glint Stokes vectors) and the
sensor-specific polarization sensitivity with viewing direction.
All of Eqs. (1.2), (1.3), and (1.4) can be found in the literature. They all give the same TOA
total radiance Lt . Which form is used in a particular instance is determined by convenience. Forms
(1.2), (1.3) are often convenient for discussions of theory, whereas form (1.4) is convenient for
operational atmospheric correction algorithms.
The goal of atmospheric correction is to convert a measured top-of-the-atmosphere radiance Lt
into the corresponding sea-level water-leaving radiance Lw . Since only Lt is measured, this requires
estimation of the various atmospheric and surface-reflectance terms seen in Eqs. (1.3) or (1.4) so
that they can be subtracted from Lt in order to arrive at Lw . How this is done is the subject of
Part II of this report.

7
 CHAPTER 2

Example Radiances

Let us illustrate the magnitudes of the various radiances in Eqs. (1.2) and (1.3) with a spe-
cific example. The orbital characteristics of the proposed NASA HyspIRI (Hyperspectral Infrared
Imager; http://hyspiri.jpl.nasa.gov/) satellite were used to obtain the Sun zenith and az-
imuthal angles at the time the sensor would fly over a point at (latitude, longitude) = (28.75
N, 158.00 W) on June 21. This point is north of the island of Oahu in Hawaii and is known
as Station ALOHA (A Long-term Oligotrophic Habitat Assessment). Figure 2.1 shows the rel-
evant angles needed for the simulation. A coupled HydroLight-MODTRAN ocean-atmosphere
radiative transfer code was used to compute the in-water and atmospheric radiances both just
above the sea surface and at the top of the atmosphere. (HydroLight is an underwater radia-
tive transfer code; http://www.hydrolight.info. MODTRAN is an atmospheric code; http:
//www.modtran5.com/about/index.html. Both are widely used for radiative transfer calculations
in their respective geophysical domains.) That code can separate the Rayleigh vs. [aerosol +
aerosol-Rayleigh] contributions, but cannot separate the pure aerosol from the aerosol-Rayleigh
contributions. Like wise, it does not normally separate Sun glint and sky glint contributions, al-
though that separation can be effected with some extra effort (explained below). (The partitioning
of atmospheric radiance contributions in the model simulations is not exactly the same as is done
operationally, but the model simulations can still illustrate the various contributions to the TOA
radiance.)
A simulation was done for the following environmental conditions:

• The water was homogeneous and infinitely deep.

• The water IOPs were simulated using a chlorophyll concentration of Chl = 0.05 mg m−3 in
the “new Case 1” IOP model in HydroLight. This IOP model is based on Bricaud et al. (1998)
for absorption and Morel et al. (2002) for scattering. (This IOP model is described in de-
tail at http://www.oceanopticsbook.info/view/optical_constituents_of_the_ocean/
__level_2/a_new_iop_model_for_case_1_water.)

• The Sun zenith angle was θSun = 17.99 deg and the Sun’s azimuthal angle was east of the
nadir point at 84.34 deg from true north.

8
 Figure 2.1: Sun and view-
ing geometry for a mid-day
HyspIRI pass over Station
ALOHA on 21 June.

• The off-nadir viewing angle was θv = 30 deg, φv = 281.12 deg, which is at right angles to
the satellite’s orbital direction and looking to the west side of the orbit, away from the Sun’s
direction.

• The atmospheric conditions (temperature profile, water vapor, ozone, etc.) were typical of a
tropical marine atmosphere (defined via MODTRAN’s “Tropical Atmosphere” option). The
sky conditions were clear. The aerosols were for an open-ocean marine atmosphere.

• The wind speed was 10 m s−1 .

• The wavelength resolution was 10 nm from 350 to 1500 nm.

Figure 2.2 shows various radiances and irradiances obtained from this simulation. The solid
curves are values at the TOA, and the dotted curves are the corresponding quantities just above
the sea surface. The Ed TOA curve (the solid purple line) is the extraterrestrial solar irradi-
ance (averaged over 10 nm bands) on a surface parallel to the mean sea surface. The dips in the
curve below 700 nm are due to absorption by various elements in the Sun’s photosphere; these are
Fraunhofer lines averaged over the 10 nm bands of this simulation. Above 700 nm, the Sun’s ir-
radiance is close to a blackbody spectrum (see http://www.oceanopticsbook.info/view/light_
and_radiometry/level_2/light_from_the_sun). The purple dotted line shows how much of the
TOA solar irradiance reaches the sea surface. There are large dips in the TOA irradiance that
reached the sea surface in the regions around 940 and 1130; these are due to absorption by water
vapor, as are the smaller dips near 720 and 820 nm. The large opaque region between 1350 and
1450 nm is due to water vapor and carbon dioxide. The dip at 760 nm is due to absorption by
atmospheric oxygen. These absorption features of the Earth’s atmosphere will affect any radiation
passing through the atmosphere.
The solid blue line shows the TOA radiance Lt that would be measured by a satellite looking
in the direction 30 deg West of the nadir point. The orange curve shows how much of the total
is atmospheric path radiance, Latm . The aqua and gray curves respectively show how much of

9
Figure 2.2: Example radiances contributing to the TOA radiance. Solid lines are radiances at the
TOA; dotted lines are at the sea surface (SFC). The geometric, atmospheric, and water conditions
are described in the text.

the path radiance is due to Rayleigh scattering by atmospheric gases and by aerosols (including
aerosol-gas interactions). The green curves in Fig. 2.2 show that the water-leaving radiance at the
TOA (the solid curve) is less than the water-leaving radiance just above the sea surface (the dotted
curve). This makes intuitive sense, because part of the water-leaving radiance would be lost to
atmospheric absorption or scattering into other directions before that radiance reaches the TOA.
The red curves show the total radiances due to surface reflectance, i.e., the sum of the back-
ground sky reflectance and the direct Sun glint. However, the red curves show that the surface-
reflectance contribution is greater at the TOA than at the surface. This seems counterintuitive and
requires explanation. In Fig. 1.1 the arrow labeled Lg represents Sun glint due to the occasional
wave facet that is tilted in just the right direction to create glint that is seen by the sensor. The
arrow labeled Lgs represents the very bright glint in the Sun’s specular direction; the sensor is look-
ing in the direction away from the Sun’s azimuthal direction in order to avoid viewing this specular
glint. However, the specular glint gives a strong reflected radiance, some of which is being scattered
by the atmosphere into the sensor viewing direction; this is illustrated by the Lgs2 arrow in Fig.
1.1. The surface contribution in Fig. 2.2 is the sum of the Lsky , Lg , and Lgs2 contributions. If the
ocean is viewed from just above the sea surface (the red dotted line), the surface-reflected radiance
comes only from reflected sky radiance Lsky and a small amount of direct Sun glint Lg from wave
facets that are tilted in just the right way to reflect the Sun’s direct beam into the direction of
the sensor. (This direct Sun glint Lg is minimal because of the choice of viewing direction.) These
surface-reflected sky and Sun radiances decrease between the surface and the TOA, just as does the

10
Figure 2.3: The fraction of Lt due to various processes, for the particular simulation of Fig. 2.2.
(Need to add a curve for total atmos plus glint.)

water-leaving radiance. However, the total TOA radiance that arises from sea surface reflectance,
as partitioned in the simulation, comes from the sum of the surface reflectance in the viewing di-
rection (decreasing with altitude) and the contribution by atmospheric scattering of specular Sun
glint into the viewing direction (Lgs2 , increasing with altitude). The specular radiance Lgs2 is large
in magnitude, so the atmospheric scattering of this radiance into the viewing direction can be
significant. If the present simulation is run with a level sea surface, for which there is no Lg Sun
glint into the sensor direction, this behavior is still present and, indeed, is even somewhat greater
in magnitude. (To fully isolate the effect of specular Sun glint being atmospherically scattered into
the sensor, a special run was made in which any photon from the Sun’s unscattered beam that was
reflected by the sea surface was killed. That is, all Sun glint was forced to be zero. Only photons
from the background sky that were reflected by the sea surface were allowed to contribute to the
TOA surface-reflected radiance. In that case, the surface-reflected radiance behaves the same as
the water-leaving radiance: less sky glint reaches the TOA than leaves the sea surface.)
Figure 2.3 shows the fractional contributions by these various processes to the total TOA
radiance Lt . For this particular case, the water-leaving radiance is at most 10% of the total TOA
radiance. The atmospheric path radiance contributes 70 to 90% of Lt , with the rest coming from
surface glint. Below 500 nm, Rayleigh scattering by atmospheric gases is the largest contributor to
the TOA radiance. This is also true in the band from 1350 to 1400 nm, where the atmosphere is
essentially opaque, because the aerosols are located mostly near the sea surface. Aerosols are the
greatest contributor between 500 and 1350 nm.
Figure 2.4 shows the Lt of Fig. 2.2, plus the corresponding Lt radiances for wind speeds of 0

11
Figure 2.4: Upper curves: TOA radiances for various environmental and viewing conditions as
labeled. Each TOA radiance has the same water-leaving radiance Lw , which is shown by the green
curve.

and 10 m s−1 , clear and hazy atmospheric conditions, and viewing directions of 0 (nadir) and 30
deg East or West of the flight direction. (These simulations are discussed in detail at http://www.
oceanopticsbook.info/view/remote_sensing/the_atmospheric_correction_problem.) Each
of these much-different TOA radiances has essentially the same water-leaving radiance Lw , which
is shown in green. The atmospheric correction problem can be visually summarized as follows:
Given any of these TOA spectra and the geometry (Sun location and viewing direction), recover
the water-leaving radiance specrum Lw . This is clearly a difficult problem because the needed
atmospheric conditions (aerosol type and concentration in particular) are not obtained as part of
the Lt measurement.

12
 CHAPTER 3

Normalized Reflectances

Ocean-color remote sensing algorithms usually work with remote-sensing reflectances or normalized
water-leaving reflectances. The calculation and interpretation of those quantities are discussed in
detail in this section.
The ratio of water-leaving radiance Lw to incident sky irradiance Ed is an apparent optical prop-
erty (AOP) that has only weak dependence on external parameters such as solar zenith angle and sky
conditions, but which is strongly correlated with water-column inherent optical properties (IOPs).
However, the remote-sensing reflectance Rrs ≡ Lw /Ed still does depend somewhat on the atmo-
spheric and other conditions at the time of measurement and thus, strictly speaking, is tied to the
particular time and location of the observation. (Further discussion and examples are given at http:
//www.oceanopticsbook.info/view/overview_of_optical_oceanography/reflectances.)
It would be desirable to have an AOP that completely removes the effects of solar zenith angle,
viewing direction, atmospheric conditions, and sea state, while retaining a strong dependence on
the water IOPs. It would then be possible to compare this AOP for measurements made at different
times and/or locations, and thereby extract information about the differences in the water columns
for the different measurements. Even for measurements made at the same time and location,
normalization to a common set of conditions is needed, e.g., when comparing in situ measurements
having different viewing directions. Such an AOP is obtained via the concept of the normalized
water-leaving reflectance.

3.1 Normalized Radiances and Reflectances

Let Lu (z, θs , θv , φ) be the in-water, upwelling radiance at depth z, for a Sun zenith angle of θs and a
viewing direction of θv , φ. Polar viewing direction θv = 0 indicates a direction looking at the nadir,
detecting radiance traveling toward the zenith. The azimuthal angle φ is measured relative to the
Sun’s azimuthal direction. Lw (θs , θv , φ) denotes the corresponding water-leaving radiance, which
is measured just above the sea surface. These radiances depend strongly on wavelength, which
is not shown. In practice, Lu is measured by instruments in the water. Lw can be obtained by
atmospheric correction of a radiance measured at the top of the atmosphere, from an above-surface
measurement at sea level after correction for surface reflectance, or from an in-water measurement

13
of Lu extrapolated upward through the sea surface.
One goal of normalization is to transform a satellite-based measurement of top-of-the-atmosphere
radiance Lt into something that can be compared with a standard measurement made in situ, in
the ocean, for whatever Sun zenith angle, viewing direction, atmospheric conditions, and wave state
occurred at the time of the satellite measurement. Let this standard in situ measurement be the
nadir-viewing radiance measured just below the sea surface, Lu (0− , θv = 0). Depth z = 0− refers
to a location in the water just below the sea surface; 0+ refers to a location in the air just above
the sea surface. Dividing Lu (0− , θv = 0) by the downwelling plane irradiance within the water,
Ed (0− ), gives the in-water Remote Sensing Ratio RSR:

Lu (0− , θv = 0)
RSR ≡ . (3.1)
Ed (0− )

The division of Lu (0− , θv = 0) by Ed (0− ) removes the “zeroth order” effect of solar zenith angle θs
and the “first order” atmospheric effects (including aerosol effects) on the magnitude of Lu (0− , θv =
0). We now want to transform Lt into something comparable.
In the early days of satellite remote sensing, it was sometimes assumed that the upwelling
underwater radiance distribution is isotropic. Under that assumption, RSR is approximately what
you would get if the Sun were at the zenith and there were no atmosphere (i.e., the sky were black).
This was the origin of statements that the normalized water-leaving radiance is the water-leaving
radiance “which would exit the sea surface if the Sun were at the zenith and if the atmosphere were
absent” (Gordon et al., 1988, page 10,910).
Fig. 3.1 compares upwelling and water-leaving radiances for “no atmosphere” or “black sky”
vs. realistic sky conditions. The curves of this figure were generated using HydroLight with the
Sun placed at the zenith. The sky was either black (a collimated incident sky radiance) or had a
diffuse radiance angular distribution typical of a clear sky. The water IOPs were determined using
the same “new Case 1” bio-optical model as for the simulations of the previous section. Runs were
made for chlorophyll values of 0.05, 0.5 and 5 mg m−3 . The runs were at a wavelength of 550
nm and the sea surface was level. Each radiance is normalized by its value at the nadir-viewing
direction to isolate the differences in the shapes of the curves. The upper set of curves shows the
shape of the upwelling radiances Lu just below the sea surface as functions of the in-water, off-nadir
viewing angle θv0 at right angles (φ = 90 deg) to the solar plane. The lower set of curves (those
curving downward in the figure) shows the corresponding water-leaving radiances Lw as functions
of the in-air, off-nadir viewing angle θv . For a level sea surface, these in-water and in-air angles are
related by Snel’s law sin θv = nw sin θv0 , where nw ≈ 1.34 is the water index of refraction. Although
the differences in the black-sky and real-sky radiances are only a few percent over the range of
angles relevant to most remote sensing (θv . 60 deg), differences of this magnitude are significant
given the high accuracy requirements for retrieved water-leaving radiances in ocean remote sensing.
The observation that the upwelling radiance distribution depends on the sky radiance distribu-
tion indicates that the idea of removing the atmosphere is too extreme. Morel and Gentili (1996,
page 4852) therefore revised the definition of normalized water-leaving radiance to be “...the radi-
ance that could be measured by a nadir-viewing instrument, if the Sun were at the zenith in the
absence of any atmospheric loss, and when the Earth is at its mean distance from the Sun.” The
distinction between “atmosphere were absent” (i.e., a vacuum) and “absence of any atmospheric

14
Figure 3.1: Comparisons of Lu (z = 0− , θs = 0, θv0 , φ = 90) and Lw (θs = 0, θv , φ = 90) for a zenith
Sun in a black sky (no atmosphere; dashed lines) vs. a zenith Sun in a typical atmosphere (solid
lines). The colors identify the chlorophyll concentrations. The black line at an ordinate value of 1
corresponds to an isotropic radiance distribution.

loss” (i.e., no attenuation by the atmosphere) is fundamental to understanding the developments of

this section.
The normalization proceeds as follows. The first step is to account for solar zenith angle and
atmospheric attenuation effects on Lw (θs , θv , φ) via (e.g., Gordon et al. (1988, page 10,910) or
Gordon and Wang (1994b, Eq. 4)):

Lw (θs , θv , φ)
[Lw (θv , φ)]N ≡ ,
cos θs t(θs )

where t(θs ) is the atmospheric diffuse transmittance for irradiance in the Sun’s direction for the
given atmospheric conditions. Recent papers include an explicit factor to correct Lw for the Earth-
Sun distance at the time of measurement:
 2
R Lw (θs , θv , φ)
[Lw (θv , φ)]N ≡ . (3.2)
Ro cos θs t(θs )

Here R is the Earth-Sun distance at the time of measurement, and Ro is the mean Earth-Sun
distance. The (R/Ro )2 factor corrects the Lw measurement to what it would be at the mean
Earth-Sun distance. (The solar irradiance at the TOA varies by about 8% over the course of a year
due to the Earth’s elliptical orbit.)
[Lw (θv , φ)]N is called the normalized water-leaving radiance. It is the water-leaving radiance
that would occur if the earth were at the mean Earth-Sun distance, the Sun were at the zenith, and
the atmosphere were non-attenuating. Note that although the factors of (R/Ro )2 , cos θs , and t(θs )
largely remove the effects of Earth-Sun distance, solar zenith angle, and atmospheric attenuation,

15
respectively, on the measured Lw , the normalized water-leaving radiance still refers to a particular
viewing direction and depends on the sky angular radiance distribution at the time of observation.
Multiplying the [Lw (θv , φ)]N of Eq. (3.2) by a factor of π/Fo , where π has units of steradian and
Fo is the extraterrestrial solar irradiance at the mean Earth-Sun distance, gives the nondimensional
normalized water-leaving reflectance [ρw ]N (e.g. Gordon and Wang, 1994b, page 7756):
 2
R
π Ro Lw (θs , θv , φ)
[ρw (θv , φ)]N ≡ [Lw (θv , φ)]N = π . (3.3)
Fo Fo cos θs t(θs )

(Radiant energy is a physical quantity that propagates through space and that can leave the water.
Radiance is a derived physical quantity that likewise can leave the water, so it makes sense to speak
of the “water-leaving radiance.” Reflectance, however, is a property of a surface. Reflectance does
not propagate through space and it cannot leave a surface, so it does not make sense to speak of
the “water-leaving reflectance.” However, the term “water-leaving reflectance” is well established
shorthand for “reflectance based on the water-leaving radiance and the incident irradiance,” and
we will use the term even if it is linguistically somewhat improper.)
The remote-sensing reflectance Rrs is usually defined as

Lw (θs , θv , φ)
Rrs (θs , θv , φ) ≡ . (3.4)
Ed (0+ , θs )

In this definition, both Lw and Ed are values for the Earth-Sun distance at the time of measurement.
However, this Rrs is numerically the same as what would be obtained if both Lw and Ed were
corrected to the mean Earth-Sun distance by (R/Ro )2 factors applied to each, because the correction
factors on Lw and Ed cancel out. That is to say, the irradiance at the sea surface for Earth-Sun
distance R is  2
Ro
Ed (0+ , θs ) = Fo cos θs t(θs ) . (3.5)
R
It thus follows that
 2
R
Ro Lw (θs , θv , φ) Lw (θs , θv , φ)
[ρw (θv , φ)]N = π =π = πRrs (θv , φ) . (3.6)
Fo cos θs t(θs ) Ed (θs )

Another way to view [ρw ]N is to think of it as the bidirectional reflectance distribution function
(BRDF) of the ocean normalized by the BRDF of a perfectly reflecting Lambertian surface. The
BRDF of a surface as measured in the laboratory is the radiance reflected by the surface divided by
the incident plane irradiance onto the surface. The BRDF of a Lambertian surface whose irradiance
reflectance is R is R/π, with units of inverse steradian. For a perfect Lambertian reflector, R = 1,
and
BRDFocean Lw /Ed
[ρw ]N = = = πRrs .
BRDFLamb 1/π
This makes it clear that the π carries units of solid angle, so that [ρw ]N is nondimensional.

16
3.2 The BRDF Effect
As noted above, the normalizations contained in [ρw (θv , φ)]N or Rrs (θv , φ) remove the effects of
solar zenith angle, atmospheric attenuation, and Earth-Sun distance on a measured radiance Lw .
However, [ρw (θv , φ)]N still refers to a particular viewing direction (θv , φ). This dependence ties
[ρw (θv , φ)]N to the angular distribution of the upwelling underwater radiance and to the transmit-
tance through the sea surface from water to air (which depends on the wave state, i.e., on the wind
speed). The upwelling underwater radiance in turn depends on the angular distribution of the
incident sky radiance, surface transmittance from air to water, and to the absorbing and scattering
properties of the water body (the scattering phase function in particular). The dependence of the
upwelling radiance distribution on the sky radiance distribution, viewing geometry, and water op-
tical properties is commonly called the BRDF effect. The final step is to remove the BRDF effect
to the greatest extent possible.
The BRDF effect was studied by Morel and colleagues in a series of papers (Morel and Gentili,
1991, 1993, 1996) culminating in Morel et al. (2002). They used numerical radiative transfer models
to compute correction factors that would transform a measurement made for a particular Sun zenith
angle, viewing direction, wind speed, atmospheric conditions, and water IOPs into a measurement
that corresponds to a zenith Sun and nadir viewing direction for a typical marine atmosphere and
for Case 1 water with a given chlorophyll value. The correction involves three separate factors, R,
f , and Q, as follows.
Let R(θv0 , W ) be a nondimensional factor that accounts for all transmission and reflection effects
by the wind-blown sea surface when Ed (0+ ) is transmitted downward through the surface to give
Ed (0− ), and Lu (0− , θv0 , φ) is transmitted upward through the surface to give Lw (0+ , θv , φ). Polar
angle θv0 (measured from the nadir) is the underwater angle that is refracted by the surface into the
above-surface viewing direction θv of the water-leaving radiance Lw (θv , φv ). W is the wind speed.
R(θv0 , W ) depends on the wind speed (i.e., the surface wave state) and the water index of refraction
via the Snel’s law mapping of θv0 to θv . However, Gordon (2005) showed that the dependence of
R(θv0 , W ) on wind speed is very weak, and usually R can be computed with adequate accuracy
over a wide range of viewing angles using W = 0. The detailed derivation of R(θv0 , W ) in Morel
and Gentili (1996, Eq. 5 and Appendix D) shows that, strictly speaking, R also depends on the
solar zenith angle and the angular distribution of the incident atmospheric radiance, which affect
how much incident irradiance is transmitted through the sea surface. Likewise, R depends on
the water IOPs via the in-water irradiance reflectance R(0+ ) = Eu (0+ )/Ed (0+ ). However, these
dependencies are weak compared to the directional (θv0 ) dependency and so, for compactness and
consistency with Morel’s notation, are not shown as arguments.
Let Ro (W ) be the reference value of R(θv0 , W ) corresponding to transmission normal to the
mean sea surface (θv = θv0 = 0). Multiplying the [Lw (θv )]N of Eq. (3.2) by Ro (W )/R(θv0 , W )
corrects for surface-transmission effects for the actual viewing direction θv0 and wind speed W . For
θv0 ≈ 0, Ro ≈ 0.53; Ro is greater than 0.52 out to angles of 50 deg. Note that Figure 4 of Morel
et al. (2002), which shows ratios of Ro (W )/R(θv0 , W ), was incorrectly computed; see the discussion
and revised figures in Gordon (2005).
The dimensionless factor f is defined by Eu (0− )/Ed (0− ) ≡ f × (bb /a), where a and bb are the
water absorption and backscatter coefficients respectively. This factor parameterizes how down-

17
welling irradiance in the water is converted to upwelling irradiance by backscatter and reduced by
absorption. That is to say, f relates the irradiance reflectance within the water to the most relevant
IOPs. f values are in the range of 0.3 to 0.5 (Morel and Gentili, 1996, Fig. 2).
The factor Q ≡ Eu (0− )/Lu (0− ) (units of steradian) describes the angular distribution of the
upwelling radiance. Q = π sr for an isotropic upwelling radiance distribution; actual in-water
radiance distributions typically have Q values in the range of 3 to 6 sr (Morel and Gentili, 1996,
Fig. 3).
In practice, f and Q are combined to give a term that has less variability than the individual
factors. The combined factor f /Q ≡ Lu (0− , θv0 , φ)/Ed (0− ) describes how the downwelling irradi-
ance just beneath the sea surface is reflected back upward as upwelling radiance in the direction
(θv0 , φ). The f /Q term thus describes both the efficiency of conversion of downwelling irradiance
into upwelling radiance, and the angular (non-isotropic) distribution of the upwelling underwater
radiance that generates the water-leaving radiance. f /Q values are typically in the range to 0.07 to
0.15 (Morel and Gentili, 1996, Fig. 6). Let fo /Qo refer to the ratio for the nadir viewing direction
and Sun at the zenith.
Multiplying [Lw (θv , φ)]N by (fo /Qo )/(f /Q) corrects for the difference of the actual angular
distribution of the upwelling radiance and what that distribution would be for the Sun at the
zenith, nadir viewing, for the particular atmospheric and oceanic conditions used to compute f and
Q.
Applying these BRDF corrections to the [Lw (θv , φ)]N of Eq. (3.2) gives (Morel et al., 2002, Eq.
13):
 −1
Ro (W ) fo (ATM, W, IOP) f (θs , ATM, W, IOP)
[Lw ]ex
N ≡ [Lw (θv , φ)]N . (3.7)
R(θv0 , W ) Qo (ATM, W, IOP) Q(θs , θv0 , φ, ATM, W, IOP)

Morel et al. call [Lw (λ)]ex

N the “exact normalized water-leaving radiance”. The arguments “ATM”
and “IOP” refer to the specific set of atmospheric conditions and water inherent optical properties
used to compute f and Q. (As previously noted, these arguments are omitted from the R terms
because the ATM and IOP dependencies are small for R.) This [Lw (λ)]ex N is equivalent to the
normalization seen in Franz et al. (2007, Eq. 2), except for a sensor-specific correction factor
for out-of-band wavelength response. It should be noted that the fo /Qo factor has arguments of
(ATM, W, IOP) because those values correspond to the same atmospheric and oceanic conditions
as the f /Q factor; the difference is that fo /Qo corresponds to θs = 0 and θv0 = 0. Values of the
(Ro /R)(fo /Qo )/(f /Q) product are typically in the 0.6 to 1.2 range, depending on the IOPs, solar
zenith angle, atmospheric conditions, wind speed, and wavelength.
It is noted that the wind speed at the time of observation, W , is shown in both the Ro and
(fo /Qo ) terms in Eq. (3.7). It can be argued that these terms should use a reference value of
W = 0, so that all quantities are referred to a level sea surface. However, as presently implemented
by the OBPG, the same wind speed W is used in the reference terms.
The atmospheric conditions used to compute the f /Q factors were typical of marine atmospheres
and are summarized as follows (Morel and Gentili, 1996; Morel et al., 2002):

• The atmosphere was modeled as 50 layers, each 1 km thick

• The aerosol optical thickness was τa = 0.2 at 550 nm.

18
 • The tropospheric aerosols had a relative humidity of 70% and were distributed in the upper
45 atmospheric layers

• The marine aerosols had a relative humidity of 90% and were distributed in the lower 5
atmospheric layers

• The aerosol phase functions were modeled using Mie theory and the data of Shettle and Fenn
(1979)

• The sea surface was modeled using a Gaussian distribution of wave slopes based on the
empirical wind-speed, wave-slope data of Cox and Munk (1954, page 847) for the given wind
speed W . W was taken to be 0 for the Q calculations, although there are still some residual
capillary waves according to the empirical slope data of Cox and Munk.

The water optical properties were obtained from bio-optical models for Case 1 water, for which
the IOPs can all be parameterized by the chlorophyll concentration Chl. In particular, the scat-
tering phase function was computed as a chlorophyll-weighted sum of phase functions for “small”
and “large” particles, which themselves were computed using T-matrix theory for non-spherical
particles with different size distributions. The details of the IOP bio-optical models are given in
Morel et al. (2002).
The radiative transfer calculations were carried out using a Monte Carlo code for calculation of
the R factors, and using HydroLight for the f /Q calculations. The R are tabulated for an exactly
level sea surface and for wind speeds of W = 0, 4, 10 and 16 m s−1 ; the W = 0 values include a
small amount of residual capillary waves because the Cox-Munk mean square sea surface slopes are
not extactly zero for a zero wind speed. The f /Q calculations were done both with and without
Raman scattering by water. None of the calculations included polarization. These codes were run
for

• 7 wavelengths (412.5, 442.5, 490, 510, 560, 620, and 660 nm)

• 6 chlorophyll values (Chl = 0.03, 0.1, 0.3, 1.0, 3.0 and 10 mg m−3 ). The water was homoge-
neous.

• 6 solar zenith angles (θs = 0, 15, 30, 45, 60 and 75 deg)

• 13 azimuthal angles (φ = 0 to 180 deg by steps of 15 deg)

• 17 nadir angles (θv0 = 1.078, 3.411, 6.289, 9.278, 12.300, 15.330, 18.370, 21.410, 24.450, 27.500,
30.540, 33.590, 36.640, 39.690, 42.730, 45.780, and 48.830 deg)

These runs give a total of 7 × 6 × 6 × 13 × 17 = 55, 629 f /Q values, which are organized into
7 × 6 × 6 = 252 tables, each with 13 columns and 17 rows. A separate table gives values of R(θv0 , W )
at θv0 increments of 1 deg and for wind speeds of W = 0, 4, 10 and 16 m s−2 . The tables including
Raman effects are available at ftp://oceane.obs-vlfr.fr/pub/gentili/AppliedOptics2002/.
Use of these tables requires the chlorophyll concentration, which is not a priori known. In the
initial study (Morel and Gentili, 1996), a band-ratio algorithm was used to obtain an initial guess
for Chl. That value was then used in the tables to obtain the BRDF correction factors. Those
factors give a new estimate of [Lw (λ)]ex
N , which can be used to obtain an new value of Chl, and so

19
on. However, the tables are not tied to how the Chl value is obtained, so in practice any algorithm
can be used to obtain Chl values from the reflectances in an iterative process.
Finally, the exact normalized water-leaving radiance of Eq. (3.7) is used in Eq. (3.3) to obtain
the exact normalized water-leaving reflectance:
π
[ρw ]ex
N ≡ [Lw ]ex
N (3.8)
Fo
(  −1 )
π Ro (W ) fo (ATM, W, IOP) f (θs , ATM, W, IOP)
= ×
Fo cos θs t(θs ) R(θv0 , W ) Qo (ATM, W, IOP) Q(θs , θv0 , φ, ATM, W, IOP)
Lw (θs , θv , φ) . (3.9)

The quantities in brackets can all be obtained from pre-computed look-up-tables given the Sun and
viewing geometry, wavelength, atmospheric conditions used to obtain Lw (θs , θv , φ) from Lt (θs , θv , φ)
(used to determine t(θs )), and the chlorophyll concentration. The chlorophyll concentration deter-
mines the IOPs according to the bio-optical models for Case 1 water used in the Morel et al.
calculations.
Although Morel and others call [Lw ]exN the “exact” normalized water-leaving radiance, and [ρw ]N
ex

the “exact” normalized water-leaving reflectance, it must be remembered that these quantities are
exact only if the atmosphere and ocean have exactly the same absorbing and scattering properties
as used in the model simulations upon which the BRDF correction factors are based. That will of
course not in general be the case. In Case 1 waters, the differences between the Morel et al. Case 1
IOP model and the actual ocean IOPs are often small enough that the BRDF-corrected quantities
are sufficiently accurate for remote sensing. However, the differences can become large in Case 2
waters. Research therefore continues on ways to improve the BRDF correction, both to extend its
validity to Case 2 water and to remove the need to estimate the chlorophyll concentration in order to
use the look-up tables (e.g., Lee et al., 2011; Fan et al., 2016). Although further improvements can
be anticipated, the BRDF correction as described above remains the OPBG operational algorithm
at the time of this writing.
The OBPG works with radiance to get to [Lw ]ex N . However, when doing atmospheric correction
on TOA radiances, the various look-up-tables used for Rayleigh correction, etc. are in terms of
reflectances [ρw ]ex
N . Equation (3.8) allows easy conversion from one to the other, depending on
which quantity is most convenient for a given step of the atmospheric correction process.
It is to be noted that the “remote-sensing reflectance” reported by the NASA OBPG as a
standard Level 2 product of the NASA ocean color satellites such as MODIS and VIIRS is the
exact normalized water-leaving reflectance of Eq. (3.9), divided by π:

[ρw ]ex
N [Lw ]ex
N
Rrs (NASA) = = . (3.10)
π Fo
However, the “remote-sensing reflectance” computed by HydroLight is Rrs (θs , θv , φ) as defined by
Eq. (3.4). A given HydroLight run computes Rrs (θs , θv , φ) for all viewing directions θv , φ for the
given θs and other conditions of wind speed, IOPs, and atmospheric radiance distribution. Thus
the HydroLight Rrs outputs for various viewing directions incorporate the BRDF effects. (Indeed,
comparison of HydroLight-computed Rrs (θs , θv , φ) values with HydroLight values for a zenith Sun
and nadir viewing direction is how the f /Q BRDF factors were determined in the Morel et al.

20
studies.) Thus there is no need for an explicit BRDF correction to the HydroLight-computed Rrs .
If a HydroLight run has the Sun at the zenith, then the nadir-viewing HydroLight Rrs times π
corresponds to [ρw ]ex
N:
[ρw ]ex
N = πRrs (HydroLight; θs = 0, θv = 0) . (3.11)
Recalling Eq. (3.10), this gives

Rrs (NASA) = Rrs (HydroLight; θs = 0, θv = 0) . (3.12)

There should be only a very small difference in HydroLight’s πRrs and [ρw ]ex N , attributable to any
differences in the angular distributions of the sky radiances used to compute the Morel BRDF
factors and as used in HydroLight. Morel et al. (2002, page 6295) notes that these differences are
negligible.
It should be noted that HydroLight Rrs values are valid for whatever IOPs were used in the
run; there is no restriction to homogeneous Case 1 water and no need to estimate a chlorophyll
concentration as must be done when applying the Morel BRDF factors. Thus HydroLight can give
a very general [ρw ]ex
N via Eq. (3.11), without the assumptions made by Morel et al. when developing
the BRDF correction factors seen in Eq. (3.9).
Figure 3.2 illustrates the magnitudes of the corrections to πRrs (θs , θv , φ) values. HydroLight
was first run to generate remote-sensing reflectances for a Sun zenith angle of θs = 50 deg, with
the Sun in a clear sky. The water IOPs and chlorophyll values were the same as those used for
Fig. 3.1. The runs included Raman scatter by water and fluorescence by chlorophyll and CDOM.
The dashed lines in Fig. 3.2 show the values of πRrs (θs = 50, θv = 30, φ = 90). These viewing
angles correspond to a sensor viewing the ocean at an off-nadir angle of 30 deg at right angles to
the solar plane. The HydroLight runs were then repeated with the Sun at the zenith. The solid
lines in the figures show the resulting values of [ρw ]ex
N as determined by Eq. (3.11). Depending
on the water IOPs, Sun zenith angle, viewing direction, and wavelength, the normalization can
change a spectrum by tens of percent, or almost not at all. The changes tend to be greatest in
high-chlorophyll waters, at large solar zenith angles, and at large off-nadir viewing angles.
When formulated in terms of reflectances, the partitioning of Eq. (1.3) becomes (e.g., Gordon
and Wang, 1994a)
ρt = ρR + [ρa + ρRa ] + T ρg + tρwc + tρw , (3.13)
where the terms correspond to those of Eq. (1.3). A similar equation applies to the reflectance
form of Eq. (1.4).
Figure 3.3 shows the radiances of Fig. 2.2 recast as normalized reflectances. It should be noted
in Fig. 3.3 that the solar-spectrum features (most noticeable below 600 nm) seen in the TOA Ed
spectrum of Fig. 2.2 are removed by the normalization process. However, the atmospheric absorp-
tion features (most noticeable beyond 600 nm) remain in the TOA reflectances, but are not present
in the surface reflectances. Thus the dotted curves for ρt , ρg , and ρw , are very smooth functions of
wavelength. The surface glint reflectance spectrum ρg is almost independent of wavelength, whereas
the surface glint radiance seen in Fig. 2.2 depends of wavelength in the same manner as the surface
irradiance spectrum. (Note that the glint reflectance ρg considered here is not the same at the
surface radiance reflectance factor ρ of Mobley (1999, Eq. 4) and Mobley (2015), which is a ratio
of incident to reflected sky radiances for the given wind speed, and Sun and viewing directions.)

21
Figure 3.2: Comparison of exact normalized water-leaving reflectances [ρw ]ex
N (solid lines) with
unnormalized reflectances (dashed lines) for θs = 50, θv = 30, φ = 90. The color identifies the
chlorophyll values of 0.05, 0.5, and 5 mg m−3 .

Figure 3.3: Normalized reflectances corresponding to the radiances of Fig. 2.2. These reflectances
are [ρ]N , not [ρ]ex
N.

22
3.3 Summary
The interpretation of [ρw ]ex
N as developed above can be summarized as follows:

• Start with a measured Lw (θs , θv , φ). Then

• The division by cos θs in Eq. (3.2) moves the Sun to the zenith.

• The division by t(θs ) in Eq. (3.2) rescales the radiance magnitude to account for atmo-
spheric attenuation, but the angular distribution of the radiance is unchanged. This diffuse
transmission is for the actual atmosphere at the time of observation.

• The (R/Ro )2 factor in Eq. (3.2) corrects for the Earth-Sun distance.

• The BRDF factors in Eq. (3.7) normalize Lw as measured for the actual atmospheric and
in-water radiance distributions to what Lw would be for a reference atmospheric radiance
distribution and for Case 1 water with the given chlorophyll value.

The proposed atmospheric retrieval accuracy requirements for the PACE (Pre-Aerosol, Clouds,
and ocean Ecosystem) mission are (Franz, 2015)

• for [ρ]ex
N in the 350-400 nm range, a maximum error of 0.002 or 10%

• for [ρ]ex
N in the 400-710 nm range, a maximum error of 0.001 or 5%

In terms of Rrs = [ρ]ex

N /π, this gives

• for Rrs in the 350-400 nm range, a maximum error of 6 × 10−4 sr−1

• for Rrs in the 400-710 nm range, a maximum error of 3 × 10−4 sr−1

Given that the water-leaving radiance Lw is at most 10% of the TOA radiance (e.g., Fig. 2.3), a
requirement for a 5% maximum error in Lw (expressed as the same percentage error in [ρ]ex N or Rrs )
implies a maximum error of roughly 0.5% in the measured TOA radiance Lt . Such a small error
in the TOA radiance cannot be achieved by pre-launch sensor radiometric calibration alone. Thus
ocean color sensors require post-launch “vicarious calibration.” This term refers to the process
of (1) performing atmospheric correction to recover sea-level Rrs spectra from the measured Lt ;
(2) propagating the corresponding sea-level Lw back to the TOA (including the atmospheric path
radiance contributions) to obtain a TOA radiance LAC t for comparison with the sensor-measured
TOA radiance Lt . The difference between the retrieved LAC t and the measured Lt then (3) gives
a final set of sensor calibration correction factors to be applied to the measured TOA radiances so
that they lead, after atmospheric correction, to the proper sea-level water-leaving radiances and
associated reflectances.
The Q part of the Morel et al. correction has been validated against empirical data for Case
1 waters by Voss et al. (2007) and found to give good agreement with measured radiance distri-
butions. As concluded there, “...the bidirectional corrections based on the lookup tables generated
from the model, and presently applied to ocean color imagery, are sound and amply validated for
Case 1 waters...” However, it should be remembered that the above BRDF correction is based
on a particular atmospheric model and on particular bio-optical models for Case 1 water. For

23
different atmospheric conditions, the f /Q correction would be different, although the difference
would probably be small. However, for different water IOPs, in particular for Case 2 waters with
high mineral particle loads or high concentrations of CDOM, the differences in the water-column
scattering and absorption properties could have a significant effect on the f /Q correction. That
variability has not yet been studied, and the above correction remains the current state of the
science and is implemented by the OBPG for operational ocean color image processing.

24
 CHAPTER 4

Atmospheric Transmittances

Equations such as (1.3) and (3.13) involve various direct (T ) and diffuse (t) atmospheric transmis-
sion terms, which require discussion.

4.1 Direct Transmittance

Direct atmospheric transmission is used if only one particular path, or a narrow bundle of nearly
colinear paths, connects the source and the observer. This is the case for specular reflection, as
illustrated in the left panel of Fig. 4.1. When the sea surface is viewed by the sensor, only a small
patch of sea surface is seen as the Sun’s specular reflection or direct glint. Other points of the sea
surface would be seen as localized patches of Sun glint for other viewing directions, but not by the
sensor viewing direction as shown. In each case, the reflected radiance is traveling in a very narrow
set of directions determined by the Sun’s location and the law of reflection (θr = θs for a level
surface). The narrow beams of Sun glint for other viewing directions are represented by the dotted
and dash lines in the left panel of the figure. These beams can influence the direction of interest
only via two scatterings: the first out of the reflected beam and the second into the direction of
interest. Such a two-scatter path is shown by the light dashed line. Two scatterings makes the
contributions of the unseen beams of specular reflection to the direction of interest very small.
Let τ be the atmospheric optical depth along a vertical path (the nadir viewing direction for the
sensor). This τ includes all effects of atmospheric absorption and attenuation by all atmospheric
constituents. For an off-nadir viewing direction θv , the direct transmittance is then simply

T (θv ) = exp(−τ / cos θv ) . (4.1)

This geometry is analogous to the Lambert-Beer law for radiance propagation of a beam through a
homogeneous medium: L(r) = L(0) exp(−cr), where c is the beam attenuation coefficient and r is
the distance traveled. In the present case, τ = cra , where ra is the distance though the atmosphere
on a vertical (nadir-viewing) path, and r = ra / cos θv is the atmospheric path length along the
viewing direction.

25
Figure 4.1: Left panel: Illustration of Sun glint as seen from the TOA, which is described by a
direct transmittance. Right panel: Illustration of water-leaving radiance as seen from the TOA,
which is described by a diffuse transmittance.

4.2 Diffuse Transmittance

For water-leaving radiance Lw , every point of the sea surface is emitting an upward distribution
of radiance Lw (θ, φ), as illustrated in the right panel of Fig. 4.1. Radiance from all locations and
various directions can be scattered into the direction of interest via only one scattering, as illus-
trated by the dotted line in the right panel of Fig. 4.1. Radiance scattered out of the beam along
the viewing direction can thus be replaced via a single scattering from the radiance emitted by a
neighboring point propagating in a different direction. The diffuse transmittance therefore depends
not just on the atmospheric properties and viewing direction, but also on the angular distribution
of Lw , which in general is unknown. This situation is analogous to the diffuse attenuation Kd of
downwelling irradiance within the water column of a plane parallel ocean that is illuminated at all
points of the surface. Because radiance absorbed or scattered out of the path between the source
and sensor can be replaced by radiance from other sea surface points and directions, the diffuse
transmittance t(θv ) will be greater than the direct transmittance T (θv ) (just as the the diffuse
transmittance exp(−Kd z) is always greater than than the beam transmittance exp(−cz)). Ac-
counting for the angular distribution of Lw and for the scattering processes makes the computation
of diffuse transmittance more complicated than for direct transmittance.
The diffuse transmittance of the water-leaving radiance along a particular viewing direction
(θv , φv ) is by definition
LTOA (θv , φv , λ)
t(θv , φv , λ) ≡ w , (4.2)
Lw (θv , φv , λ)
where Lw is the water-leaving radiance at the sea surface and LTOA w is the water-leaving radiance
that reaches the TOA. Henceforth, all quantities are assumed to be functions of wavelength, and the
argument λ will be omitted. One way to compute t(θv , φv ) is to run a coupled ocean-atmosphere
radiative transfer model to compute the needed Lw (θv , φv ) and LTOA w (θv , φv ) for a wide range of
atmospheric and oceanic conditions, Sun and viewing geometries, and wavelengths. The values of
t(θv , φv ) would then be obtained via Eq. (4.2) and tabulated for later use. Such tables would need
to be constructed for all possible aerosol types (i.e, aerosol phase function shapes), aerosol optical

26
thicknesses, water IOPs (or at least for chlorophyll concentrations in Case 1 waters), and viewing
geometries. The tables would necessarily be large because of the large number of parameters that
can affect t(θv , φv ).
Yang and Gordon (1997) examined the computation of diffuse transmittance and errors therein
on the retrieved water-leaving radiance. Using a combination of radiative transfer numerical
modeling of the ocean and atmosphere and reciprocity principles, they compared diffuse trans-
mittances computed using realistic modeled and measured Lu (θ0 , φ0 ) distributions versus diffuse
transmittances computed on the assumption that Lu is isotropic. Here Lu (θ0 , φ0 ) denotes the up-
welling underwater radiance distribution just beneath the sea surface. For a level sea surface,
Lw (θv , φv ) = TF (θ0 )Lu (θ0 , φ0 )/n2w , where nw is the index of refraction of the water, the in-water θ0
is related to the in-air θv by Snel’s law sin θv = nw sin θ0 , and TF (θ0 ) is the Fresnel transmittance
of the surface from water to air. When Lu is isotropic, the diffuse transmittance is independent of
the azimuthal angle and is denoted by t∗ (θv ).
Yang and Gordon (1997, Eq. 3) show via a clever use of reciprocity that the diffuse transmittance
of radiance t∗ along an atmospheric path in the direction of the Sun at solar zenith angle θo is
numerically equal to the diffuse transmittance of irradiance from the TOA to a depth just beneath
the sea surface, on the assumption that there is no upwelling radiance within the water. That is,
ER (θo )
t∗ (θo ) = , (4.3)
Fo cos(θo )TF (θo )

where ER (θo ) is the downwelling plane irradiance just beneath the sea surface for an extraterrestrial
solar irradiance Fo incident onto the TOA at angle θo , and TF (θo ) is the Fresnel downward trans-
mission of the sea surface for radiance incident at angle θo from the normal. Since the upwelling
radiance Lu (θ0 , φ0 ) used to obtain this result is assumed to be isotropic, the azimuthal dependence
of t∗ is irrevelant, and the desired diffuse attenuation for radiance at viewing direction θv = θo is
equal to the value of the irradiance transmission at the same polar angle. The great virtue of Eq.
(4.3) is that it allows the efficient numerical computation of t∗ using backward (reverse) Monte
Carlo simulation of downwelling irradiance for a given aerosol type and optical thickness.
The retrieved water-leaving radiance is, by Eq. (4.2),

LTOA
w (θv , φv )
Lw (θv , φv ) = . (4.4)
t(θv , φv )

Let L∗w denote the retrieval when t∗ rather than t is used in Eq. (4.4). The error in the retrieved
water-leaving radiance due to using t∗ rather than the exact t is
∆Lw L∗ − Lw t − t∗
= w = . (4.5)
Lw Lw t∗
Yang and Gordon (1997) found that for viewing angles φv perpendicular to the principle plane (the
plane of the Sun), the errors in the retrieved Lw are no more than 4% for viewing angles θv ≤ 45 deg,
Sun zenith angles θo ≤ 60 deg, and aerosol optical thicknesses typical of clear atmospheres. The
errors in band-ratio algorithms were less; e.g., the error in Lw (443)/Lw (555) (used to retrieve
chlorophyll concentration) is less than 2% except for very clear water and some viewing directions,
for which the error in the ratio is about 3%. These parameter ranges covered most of the needs
for SeaWiFS and errors of this magnitude were deemed acceptable compared to other errors in the

27
retrieval process (i.e., removal of atmospheric path radiance). In that case, the diffuse attenuation
can be obtained by pre-computed functions of the form

t∗ (θv ) = A(θv ) exp[−B(θv )τa ] . (4.6)

where A(θv ) and B(θv ) are tabulated for each aerosol type. Thus, for N angles θv , only 2N numbers
A(θv ) and B(θv ) need to be tabulated for each aerosol type. Moreover, the aerosol type and optical
thickness are determined as part of the process to remove the aerosol contribution to the total path
radiance. This enables selection of the appropriate A, B, and τa , and evaluation of Eq. (4.4) is
operationally feasible. If aerosols are ignored, Eq. (4.6) reduces to t∗ (θv ) = exp[− 12 τR / cos θv )],
where τR is the Rayleigh optical thickness. This is the formula used in the early days for CZCS
atmospheric correction. The value of Eq. (4.6) is that it allows the aerosol optical thickness τa to be
incorporated into the diffuse transmittance calculations via a simple exponential and pre-computed
A and B values.
However, for larger off-nadir viewing angles θv , for azimuthal viewing directions near φv = 0
or 180 deg, and for very clear water, the errors ∆Lw /Lw can be as much as 6%. This could
be significant for the MODIS Aqua sensor, which views a wide range of φv directions. Gordon
and Franz (2008) therefore re-examined the model of Yang and Gordon (1997) and developed a
correction term to t∗ so that Eq. (4.4) becomes

LTOA
w (θv , φv )
Lw (θv , φv ) = ∗
. (4.7)
t (θv )[1 + δ(θv , φv )]

The δ(θv , φv ) factor corrects for the bi-directional effects resulting from the use of an isotropic Lu
in the computation of t∗ , rather than the exact, non-isotropic Lu (θ0 , φ0 ). The Lu (θ0 , φ0 ) needed for
computation of δ was obtained from Lu = (f /Q)(bb /a) and chlorophyll-based models for f /Q and
bb /a. Again, the δ correction term can be pre-computed and tabulated for various aerosol types
and water IOPs. Evaluation of the impact of the δ correction applied to both SeaWiFS and MODIS
Aqua data showed that retrieved water-leaving radiances will be in error by no more than ∼1% if
θv < 60 deg and the δ correction is omitted. For θv > 60 deg, i.e. near the edges of scan lines, the
use of the δ correction is warranted.
In current operational practice, the tabulated A and B functions are applied for both down-
welling (solar irradiance) and upwelling Lw paths. The δ(θv ) correction of Gordon and Franz (2008)
is not applied because it is an added complication with no significant impact in most instances.

28
 CHAPTER 5

Vicarious Calibration

The atmospheric correction techniques described in Part II of this tutorial remove most of the
effects of atmospheric scattering and absorption and sea surface reflectance from a measured TOA
radiance. The end result is an estimate of the water-leaving radiance, which can be converted to
a normalized reflectance as shown in §3.1. However, neither the sensor radiometric calibration nor
the atmospheric correction are perfect, so a satellite-derived water-leaving radiance usually will not
exactly match a water-leaving radiance measured at the sea surface. It is therefore necessary to
make comparisons between satellite-derived radiances and radiances measured at the sea surface
in order to determine the gain or correction factor needed to convert a best estimate of a radiance
into one that agrees with the radiance measured at the sea surface. This process is called vicarious
calibration.
The vicarious calibration methodology employed by the OBPG is described in Franz et al.
(2007). Their procedure simultaneously corrects for residual errors in both sensor radiometric
calibration and in atmospheric correction. The gain factors are therefore specific to a particular
sensor and atmospheric correction procedure, but they are independent of the how the water-leaving
radiance is obtained. It is assumed that the sensor has been calibrated as well as possible, so that
only residual calibration errors need be corrected by the vicarious calibration process.
Operational data processing starts with a measured TOA total radiance Lt and derives the
corresponding water-leaving radiance Lw . The process is based on Eq. (1.4):
 
Lt = Lr + LA + tdv Lwc + tdv Lw tgv tgs fp , (5.1)

where (recall Table 1.1 and the associated discussion)

• Lt is the TOA radiance measured by the sensor,

• Lr is the Rayleigh radiance, which is due to scattering by atmospheric molecules,

• LA is the radiance due to scattering by aerosol particles and aerosol-molecule multiple scat-
tering,

• Lwc is the radiance due to whitecaps and foam,

29
 • Lw is the water-leaving radiance,

• tdv is the diffuse transmittance along the viewing path of the sensor,

• tgv and tgs are the transmittances by atmospheric gases in the viewing and solar directions
respectively, and

• fp is a known sensor-specific polarization-correction factor.

These quantities are all functions of wavelength. No specular-reflection term is included in Eq. (5.1)
because it is assumed that pixels containing a detectable amount of specular reflection are omitted
from consideration. In this equation, the aerosol radiance LA and the water-leaving radiance Lw
are the two primary unknowns. The aerosol-dependent diffuse transmittance tdv can be computed
for a given aerosol type as described in §4.2. The Rayleigh and whitecap radiances and the gaseous
transmittances can be computed for given atmospheric conditions as described in Part II of this
report. The determination of the aerosol contribution is the crux of the atmospheric correction
process and will be described in §9. These computations constitute the atmospheric correction
process. Once these terms have been computed, Eq. (5.1) can be solved for a measured Lt to
obtain the corresponding Lw .
To develop the correction factors, this process is reversed. Let Ltw denote the known or “target”
water-leaving radiance that is to be matched by the satellite-derived value. In most instances, the
target water-leaving radiance is obtained from in-situ measurement, but can in principle be the
value retrieved by another satellite sensor or predicted by a model. In any case, this value can be
propagated to the TOA under various assumptions about the atmospheric conditions to obtain the
corresponding target TOA radiance Ltt which, ideally, would match the satellite-measured TOA
radiance. The ratio
Lt (λ)
g(λ) = t (5.2)
Lt (λ)
is then the correction or gain factor that, when multipled by a measured Lt gives an adjusted TOA
radiance Ltt that, when atmospherically corrected, yields the correct water-leaving radiance Ltw .
Note that the gain factor is different for each wavelength. The gain factors are created via a series
of “match-up” comparisons of satellite and in situ data and then, once determined, are routinely
applied as part of the operational reduction of satellite-measured TOA radiances to water-leaving
radiances.
Now consider the details of the computation of the gain factors. A satellite-derived Lw is
converted to an exact normalized water-leaving radiance as described in §3.1. This process can be
summarized as
Lw
[Lw ]ex
N = , (5.3)
µs fs tds fb fλ
where

• µs is the cosine of the solar zenith angle (cf. Eq. 3.2),

• fs is the Earth-Sun distance correction factor (cf. Eq. 3.2),

• tds is the Rayleigh-aerosol diffuse transmittance in the Sun’s direction (cf. Eq. 3.2),

• fb is the BRDF correction factor (cf. Eq. 3.7), and

30
 • fλ is a band-pass adjustment factor.

A target water-leaving radiance is converted to an exact normalized water-leaving radiance in the

same manner:
Ltw
[Ltw ]ex
N = , (5.4)
µts fst ttds ttgs fbt fλt
where now the superscript t on the terms in Eq. (5.4) indicates that these terms are evaluated for
the Sun and viewing geometry at the time of the measurement of Ltw , which may be different from
the geometry at the time of the satellite observation leading to Lw . A factor of tgs is included in
Eq. (5.4) to account for the diffuse transmittance due to absorption by gases in the Sun’s direction
at the time of measurement of Ltw . This factor does not appear in Eq. (5.3) because that correction
to the total measured radiance Ltt is accounted for in Eq. (5.1). The radiances in Eq. (5.1) are
computed for the full spectral response of each sensor band. The fλ factor converts these full-band
values to nominal band-center wavelengths to remove residual out-of-band response effects. Since
the satellite and in situ instruments usually have different spectral responses, this factor adjusts
the satellite and in situ values to a common wavelength dependence.
Writing Lw in terms of [Lw ]ex ex t ex
N via Eq. (5.3) and then replacing [Lw ]N by the target value [Lw ]N
gives an equation for the target value of the TOA radiance:
 
Ltt = tdv [Ltw ]ex
N (µs fs t ds fb fλ ) + Lr + Lt
A + t dv Lwc tgv tgs fp . (5.5)

The total transmittance along the Sun’s path is the product of the diffuse transmittance for the
Rayleigh and aerosol scattering and the diffuse transmittance for gaseous absorption. The ttds term
depends on the aerosols and is thus an unknown for the calibration target. The total transmittance
for the target could be obtained from axillary measurements (e.g., from a Sun photometer) made
at the time of the target radiance measurement. However, such measurements are not generally
available and, even if available, any error in those measurements would be an additional source of
error in the the target radiance. Therefore, the satellite-retrieved atmospheric and aerosol properties
are used to evaluate the total transmittance for the target measurement via
 
t t µs
tds tgs = exp ln(tds tgs ) t . (5.6)
µs
The total transmittance for the target is therefore the total transmittance for the satellite with
a correction for the difference in the solar zenith angles. Other terms in Eq. (5.5) such as the
Rayleigh radiance and gaseous transmittances along the viewing direction are evaluated for the
atmospheric conditions of the satellite retrieval as described in §6. Thus the tdv that multiplies Lwc
is determined by Rayleigh-scattering calculations based on the sea-level pressure. The whitecap
radiance is modeled as a function of wavelength and wind speed as described in §8. These choices
reference both the target and the satellite radiances to a common atmosphere, which is desirable
for the development of the gain factors.
Finally, the BRDF correction factors fb and fbt must be evaluated. As discussed in §3.2, the
IOPs needed for evaluation of the BRDF correction are parameterized in terms of the chlorophyll
concentration Chl. If a measurement of Chl is made in conjunction with the target measurement,
then that value of Chl can be used to evaluate the BRDF correction. However, chlorophyll mea-
surements are not usually available. Operationally, the chlorophyll concentration is obtained via

31
insertion of the satellite-derived [Lw ]ex ex
N (or the corresponding reflectance [ρw ]N ) into a chlorophyll-
retrieval algorithm. That is an iterative process because [Lw ]ex
N is required to determine Chl, and
ex
Chl is required to determine [Lw ]N . During the determination of gain factors, the target radiance
can be used as input to the operational chlorophyll-retrieval algorithm and no iteration is required.
The final issue is the determination of the aerosol properties. This is a two-step process based
on the “black-pixel” assumption to be described in §9.2. As seen in Table 9.1, satellite sensors
have two NIR wavelengths used for aerosol retrievals. Call the longer of these wavelengths λL and
the shorter λS . (For VIIRS, the NIR bands are at the nominal wavelengths λL = 862 nmm and
λS = 745 nmm.) During determination of the gain factors, it is first assumed that the water-
leaving radiance at these two wavelengths is zero (the black-pixel assumption). This is usually
a good approximation for the mid-ocean, oligotrophic waters used for vicarious calibration. It is
further assumed that the instrument calibration is perfect for the λL band, in which case the gain
factor for the longer NIR band is g(λL ) = 1. The black-pixel assumption means that Eqs. (5.1)
and (5.5) reduce to
 
Lt (N IR) = Lr + LA + tdv Lwc tgv tgs fp (5.7)
 
Ltt (N IR) = Lr + LtA + tdv Lwc tgv tgs fp , (5.8)

respectively, at the two NIR bands. Given the satellite-measured TOA radiances at the two NIR
bands, Eq. (5.7) can be solved for LA at the two NIR bands. The assumption that g(λL ) = 1
means that LtA (λL ) = LA (λL ). Thus Ltt (λL ) is determined via Eq. (5.8) evaluated at λL . The
locations for match-ups are purposely chosen at times and locations where it is reasonable to
assume that the aerosol type in stable and predictable over the image area, e.g. mid-ocean areas
where the aerosols are predominately sea salt and water droplets. The aerosol model derived from
the satellite measurements as described in §9.2 can then be used along with the value of Ltt (λL ) to
determine Ltt (λS ). Both Lt and Ltt are then known at the two NIR wavelengths, and the NIR gain
factor at g(λS ) can be determined by Eq. (5.2).
Values of g(λS ) are computed for various times and locations during the lifetime of the mission.
These values are averaged to determine the mean gain ḡ(λS ). The selection of suitable images for
gain determination is quite strenuous and most candidate pixels are eliminated from consideration
because of glint, inhomogeneous water at the target location, or non-ideal atmospheric conditions.
The details of the selection criteria and statistical determination of the mean gain factors are given in
Franz et al. (2007). Experience shows that 20 to 40 match-ups are required for the determination of
ḡ(λS ) values that are stable to within 0.1% of their long-term values. Once the NIR gains g(λL ) = 1
and ḡ(λS ) have been determined for the given sensor, the extrapolation algorithm described in §9.2
can be used to determine LtA (λ) at all wavelengths. Equation (5.8) then gives Ltt (λ), and the gains
at the remaining visible wavelengths are obtained from Eq. (5.2). Once determined, the gains are
held fixed and applied as part of the operational atmospheric correction process.
For the SeaWiFS sensor the gains ranged from 1.0377 at 412 nm to 0.972 at 765 nm. A correction
of 3 or 4% to the TOA radiance can correspond to a 30 or 40% correction to the water-leaving
radiance because the water-leaving radiance is typically about 10% of the TOA radiance. Thus
the determination of accurate gain factors is critical to the overall retrieval process. It must be
remembered that a set of gains must be determined for each sensor and atmospheric correction
algorithm. As improvements are made to the atmospheric correction algorithms described in Part

32
II, the gains must be recomputed. However, these recomputations can be made using the original
target radiances. Gain recalculation is a part of the standard reprocessing of data sets.

33
 Part II

The NASA OBPG Algorithms

34
 This part of the tutorial discusses, one by one, the various corrections made to the TOA radiance
during the OBPG atmospheric correction process. The entire sequence of data processing beginning
with the measurement of a TOA radiance and ending with the output of a geophysical product
such as a global map of chlorophyll concentration is divided into a number of processing levels.
These steps are defined in Table 5.1. The atmospheric correction process described here takes the
data from Level 1b to Level 2.

Table 5.1: Processing Levels

Level Definition
0 refers to unprocessed instrument data at full resolution. Data are in “engineering
units” such as volts or digital counts.
1a is unprocessed instrument data at full resolution, but with information such as
radiometric and geometric calibration coefficients and georeferencing parameters
appended, but not yet applied, to the Level 0 data.
1b refers to Level 1a data that have been processed to sensor units (e.g., radiance
units). Level 0 data are generally not recoverable from level 1b data.
Atmospheric correction is applied to Level 1b TOA radiance to create
Level 2 data.
2 refers to derived geolocated, geophysical variables (e.g., normalized water-leaving
reflectance [ρw (λ)]ex
N , chlorophyll concentration, Kd (490 nm), and AOT(865 nm))
at the same resolution and location as Level 1 data.
3 are variables mapped onto uniform space-time grids, often with missing points
interpolated or masked, and with regions mosaiced together from multiple orbits
to create large-scale maps, e.g. of the entire earth.
4 refers to results obtained from a combination of satellite data and model output
(e.g., the output from an ocean ecosystem model), or results from analyses of lower
level data (i.e., variables that are not measured by the instruments but instead are
derived from those measurements).

The next chapter shows how to account for absorption and scattering by atmospheric gases.
Sea-surface reflectance by Sun glint and white caps are then discussed, followed by the various
aspects of correction for the effects of atmospheric aerosols. The last two chapters discuss the
sensor-specific corrections for spectral out-of-band response and for polarization. Figure 5.1 shows
the sequence in which the various corrections are applied during the overall process.
It is important to keep in mind that there are severe computational constraints on how atmo-
spheric correction is performed on an operational basis. The MODIS-Aqua sensor, for example,
collects about 1.4 terrabytes of data per day. The requirement to routinely process this amount of
data (along with data from other sensors) requires that various approximations be made in order
to speed up the calculations. Some of the corrections require ancillary information such as sea
level pressure, wind speed, and ozone concentration, which are not collected by ocean color sensors

35
Figure 5.1: Flowchart of the atmospheric correction process. The section numbers refer to the
pages giving the details of each step.

36
themselves. These ancillary data may be inaccurate or missing (in which case climatological values
must be used). The quality of the ancillary information impacts the accuracy of the atmospheric
correction. Table 5.2 shows some of the ancillary data and its sources as used by the various
atmospheric correction algorithms.

Table 5.2: Ancillary data needed for atmospheric correction. Abbreviations: NCEP, National Cen-
ters for Environmental Prediction; OMI, Ozone Monitoring Instrument; TOMS, Total Ozone Map-
ping Spectrometer; SCIAMACHY, SCanning Imaging Absorption spectroMeter for Atmospheric
CHartographY; GOME, Global Ozone Monitoring Experiment; Reynolds/NCDC, Reynolds analy-
sis, National Climate Data Center; NCEI, NOAA National Centers for Environmental Information
(previously NODC); NSIDC, National Snow and Ice Data. Center.
Data Source Use
atmospheric pressure NCEP Rayleigh correction
water vapor NCEP transmittance
wind speed NCEP Rayleigh, Sun glint, white caps
ozone concentration OMI/TOMS transmittance
NO2 concentration SCIAMACHY/OMI/GOME transmittance
sea surface temperature Reynolds/NCDC seawater index of refraction and
backscattering
sea surface salinity NCEI World Ocean Atlas, seawater index of refraction and
Salinity Climatology backscattering
sea ice coverage NSIDC masking

The sea surface temperature and salinity are used to compute the water index of refraction
and water backscattering coefficient as described in Werdell et al. (2013). Operationally, pixels
are masked before atmospheric correction for only a few conditions, namely the presence of land
or clouds, and saturation of the measured radiance. An attempt is made to process all other
pixels. A separate mask is applied during atmospheric correction to pixels with too much Sun
glint. Various flags are incorporated into Level 2 and 3 data after the atmospheric correction
process described below. These identify pixels that may have various problems such as sea ice
contamination, turbid water, bottom effects, or failed atmospheric correction. These flags are
listed at http://oceancolor.gsfc.nasa.gov/cms/atbd/ocl2flags. Other information on flags
is given in Patt et al. (2003, Chapter 6). With the exception of Sun glint, applying masks and flags
is not a part of the atmospheric correction process per se, so this topic is not discussed here.

37
 CHAPTER 6

Gases

This chapter first describes the Rayleigh corrections made for non-absorbing gases, and then the
corrections for absorbing gases.

6.1 Nonabsorbing Gases

A radiative transfer numerical model is used to solve the vector (polarized) radiative transfer equa-
tion for non-absorbing atmospheric gases only. The radiative transfer model includes atmospheric
multiple scattering, polarization, and sea surface roughness modeled analytically by a Cox-Munk
slope distribution with an added analytical wave shadowing function. The Cox-Munk slope dis-
tribution is azimuthally isotropic (no dependence on wind direction); therefore only the relative
angle between Sun and viewing direction matters. This greatly simplifies the Fourier decomposition
described in Wang (2002, Eq. 3). The radiative transfer model is described in Ahmad and Fraser
(1982).

6.1.1 Wind Speed and Surface Reflectance Effects

Background sky reflectance by the rough sea surface is accounted for as part of the Rayleigh
correction. Some sensors (CZCS, SeaWiFS, OCTS) could be tilted to avoid looking at glint near
the Sun’s specular direction. Other sensors (MODIS, VIIRS, MERIS, OLI) do not tilt and therefore
must account for specular reflection.
Wind speed and surface glint corrections are computed as described in Wang (2002): Run the nu-
merical model to compute a look up table (LUT) of TOA Fourier components Lm r (λ, θs , θv , τRo , W )
(his Eq. 3) for the following conditions:

• The sensor wavelength bands (e.g., bands centered at 412, 443, 490, 510, 555, 670, 765, 865
for SeaWiFS). The radiative transfer model is run using band-averaged optical thicknesses
(rather than running at high-wavelength resolution, and then averaging the Lr values over
the band response functions to get the nominal band values of Lr for a particular sensor).

• 45 Sun zenith angles θs from 0 to 88 deg by 2 deg

38
Figure 6.1: The Rayleigh optical thickness and depolarization ratio based on Bodhaine et al. (1999).
[Plotted from NASA data file rayleigh bodhaine.txt]

• 41 viewing nadir angles θv from 0 to 84 deg by roughly 2 deg

• Rayleigh optical thickness τRo for standard sea level atmospheric pressure Po = 1013.25 hPa
(1013.25 millibar).

• 8 wind speeds W = 0, 1.9, 4.2, 7.5, 11.7, 16.9, 22.9 and 30.0 m/s. (These wind speeds
correspond to convenient spacing in the mean square slopes of the sea surface according to
the Cox-Munk equation mss = 0.00512W : mss = 0.0, 0.01, 0.02, 0.04, 0.06,... for W = 0,
1.9, 4.2, 7.5, 11.7, ....) Linear interpolation is used for values between these wind speeds.

The Rayleigh optical thickness τRo (λ) at 1 atmosphere of pressure, Po (1013.25 hPa), temper-
ature of 288.15K, and a CO2 concentration of 360 ppm is given by Bodhaine et al. (1999, Eq.
30):
1.0455996 − 341.29061λ−2 − 0.90230850λ2
 
τRo (λ) = 0.0021520 , (6.1)
1.0 + 0.0027059889λ−2 − 85.968563λ2
where λ is in micrometers. These values and the corresponding Rayleigh depolarization ratio ρ(λ)
are shown in Fig. 6.1. (At the scale of this figure, the Bodhaine values are almost indistinguishable
from the values given by the formula of Hansen and Travis (1974); see Bodhaine et al. (1999, Eq.
15), which were used in earlier calculations.)
The Rayleigh LUTs for Lr contain the I, Q and U Stokes vector components in reflectance units,
as a function of wind speed and geometry. The Stokes vector V component for circular polarization
is assumed to be zero. There is a separate LUT for each wavelength.
During image correction, the wind speed W for a given pixel comes from NCEP 1 deg gridded
data, interpolated to the image pixel.

39
6.1.2 Pressure Effects
The Rayleigh optical thickness at the time of the observation depends on the number of atmospheric
gas molecules between the sea surface and the top of the atmosphere. The number of molecules
is directly proportional to the sea-level pressure P . Thus the Rayleigh optical thickness at any
pressure P is given by
P
τR (P, λ) = τRo (Po , λ) .
Po
The TOA LR is then computed by (Wang (2005, Eq. 5) and subsequent equations)
1 − exp[−C(λ, M )τR (P, λ)M ]
LR [τR (P, λ)] = LR [τR (Po , λ)] , (6.2)
1 − exp[−C(λ, M )τR (Po , λ)M ]
where
1 1
M= + (6.3)
cos θs cos θv
is the geometric air mass factor for the total path through the atmosphere. C(λ, M ) is a coefficient
that is determined so that Eq. (6.2) gives the best fit to LR [τR (P, λ)] as computed by an extremely
accurate atmospheric radiative transfer model when run for values of sea level pressure P 6= Po .
Numerical simulations show that this coefficient can be modeled as
C(λ, M ) = a(λ) + b(λ) ln(M )
a(λ) = − 0.6543 + 1.608τR (Po , λ)
b(λ) = 0.8192 − 1.2541τR (Po , λ) .

6.2 Absorbing Gases

CO, N2 O, CH4 , and CO2 have negligible absorption at the visible and NIR wavelengths relevant
to ocean color remote sensing. However, O2 , O3 , NO2 , and H2 O have absorption bands in the
visible and NIR. The O2 and H2 O bands can be avoided by judicious choice of sensor bands, as
shown in Fig. 6.2 for the MODIS bands. However, as seen in Figs. 6.3 and 6.4, O3 and NO2 have
broad, concentration-dependent absorption bands that cannot be avoided. It is therefore necessary
to account for absorption by these two gases.
The concentrations of absorbing gases are usually measured as column concentrations, i.e.,
the number of molecules per unit area, or as the equivalent in Dobson units. One Dobson unit
refers to a layer of gas that would be 10 µm thick at standard temperature and pressure, or about
2.69 × 1016 molecules cm−2 . 1000 DU = 1 atm-cm; that is, 1000 DU is the number of molecules
that would give a layer of gas 1 cm thick at a pressure of one atmosphere.
For optically thin absorbing gases that are high in the atmosphere (O3 in particular), it is
possible to correct for absorption using just the geometric air mass factor M computed as in Eq.
(6.3) because scattering is not significant. However, for gases near the surface (NO2 in particular),
multiple scattering by dense gases and aerosols is significant and increases the optical path length,
hence the absorption. Thus M is not a good approximation for the total optical path length through
an absorbing gas near sea level.
Note that an absorbing gas reduces the TOA radiance because light is lost to absorption.
Correcting for this loss will increase the TOA radiance or reflectance, with the effect being greatest
at blue wavelengths where multiple scattering is greatest.

40
 Figure 6.2: Transmittance by O2
and H2 O for a moist tropical atmo-
sphere. The resolution is 1 nm. The
MODIS sensor bands are shaded in
gray.

Figure 6.3: Transmittance by O3 for

200, 350, and 500 Dobson units and
a vertical path through the atmo-
sphere.

Figure 6.4: Transmittance by

NO2 for low (2.8 × 1015 ), typ-
ical (1.1 × 1016 ), and high
(6.0 × 1016 molecules cm−2 ) con-
centrations of NO2 and a vertical
path through the atmosphere.

41
6.2.1 Absorption by Ozone
The diffuse transmission by ozone can be written as
  
1 1
tO3 = exp −τO3 + (6.4)
cos θs cos θv
= exp [−τO3 M ] ,

where M is the geometric air mass factor previously defined in Eq. (6.3), and τO3 is the optical
thickness of the ozone for a vertical path through the atmosphere. Scattering by ozone is negligible,
but absorption is significant at some wavelengths. Thus τO3 is the optical thickness for absorption
by ozone, which is given by
τO3 (λ) = [O3 ]kO3 (λ) , (6.5)
where [O3 ] is the ozone concentration (column amount in molecules cm−2 ), and kO3 is the ab-
sorption cross section (in cm2 molecule−1 ). The ozone concentration [O3 ] for a given image pixel
is obtained from the NASA OMI or TOMS sensors (Ozone Monitoring Instrument; Total Ozone
Mapping Spectrometer, now replaced by OMI).
As was seen for transmittance in Figs. 6.3 and 6.4, the absorption cross sections for gases
like O3 and, especially, NO2 can vary with wavelength on a nanometer scale. To fully resolve
the effects of such wavelength dependence on sensor signals, radiative transfer calculations would
require computationally intensive “line-by-line” calculations followed by integration over the sensor
bands. To avoid that computational expense, band-averaged values of the Rayleigh optical depth
and absorption cross sections kO3 and kNO2 are computed for each sensor and tabulated. Radiative
transfer calculations then use the band-averaged values with just one radiative transfer calculation
done for each sensor band. These band-averaged values depend on the sensor even for the same
nominal wavelength band (e.g, the 412 nm blue band) because of different band widths about the
nominal center wavelength and different sensor response functions within a band. Figure 6.5 shows
example band-averaged values for the VIIRS and MODIS Aqua sensors.

42
Figure 6.5: Band-averaged Rayleigh optical depth τR , ozone absorption cross section kO3 , and
NO2 absorption cross section kNO2 for the VIIRS and MODIS Aqua sensors. (Data from http:
//oceancolor.gsfc.nasa.gov/DOCS/RSR_tables.html)

6.2.2 Absorption by NO2

Nitrogen dioxide NO2 occurs both in the stratosphere and near the Earth’s surface. NO2 in the
lower atmosphere is generated primarily by human activities (automobiles, industry, fires), and
the highest concentrations are near the earth’s surface in industrial areas. Numerical simulations
show that failure to correct for absorption by NO2 can give errors of approximately 1% in TOA
radiances at blue wavelengths, which result in ∼ 10% errors in retrieved water-leaving radiances
(Ahmad et al., 2007).
The geometric air mass factor M of Eq. (6.3) and a simple atmospheric transmittance function
like that of Eq. 6.4) are valid for ozone absorption corrections because scattering by ozone in the
upper atmosphere is negligible. However, such functions may be not adequate for NO2 correction
calculations because of multiple scattering in the dense lower atmosphere. Further guidance for the
form of the NO2 correction comes from the observation that the water-leaving radiance Lw sees all
NO2 in the atmosphere, so the total NO2 concentration N must be used to correct Lw for NO2
absorption. However, the upwelling atmospheric path radiance Latm , which is generated throughout
the atmosphere, is not strongly influenced by the absorbing gas very near the surface. Extremely
accurate numerical simulations show that in that case, N 0 , the NO2 concentration between an
altitude of 200 m and the TOA, can be used as a satisfactory measure of NO2 concentration. This
result leads to different corrections for the measured TOA path radiance and for the water-leaving
radiance.
Let ρobs be the uncorrected, observed (measured) TOA reflectance, and let ρcorr be the TOA re-
flectance corrected for NO2 absorption effects. Comparison of numerical simulations and analytical
approximations justifies a correction of the form
  
1 1
ρcorr = ρobs exp αN 0 + ,
cos θs cos θv

43
where N 0 is the NO2 concentration between an altitude of 200 m and the TOA. This simple formula
gives ρcorr values that are within 0.15% of the values obtained by exact numerical simulations that
account for the total column NO2 concentration and multiple scattering.
The correction for water-leaving radiance proceeds as follows. The reflectances can be written
as (Ahmad et al., 2007, Eqs. 1 and 7)

ρt (θs , θv ) = ρpath (θs , θv ) + t3 (θv )td (θs )ρw (θs , θv ) + glint and whitecap terms

(omitting the arguments for wavelength, azimuthal angle, and wind speed). Here t3 is the dif-
fuse transmission along the viewing direction from the sea surface to the sensor, td is the diffuse
transmission of downwelling solar irradiance, and ρw is the water-leaving reflectance at the sea
surface. Consider now only the path and water-leaving terms, and omit the directional arguments
for brevity. Then multiplying this equation by the exponential correction factor for the observed
TOA reflectance gives

ρt exp αN 0 (sec θs + sec θv ) = ρpath exp αN 0 (sec θs + sec θv )

   

+ [t3 exp(αN 0 sec θ)] [td ρw exp(αN 0 sec θs )] .

Numerical simulations show that both the path reflectance and the diffuse transmission term
[t3 exp(αN 0 sec θ)] are accurate to within 0.2% with this correction. However, the error in the
water-leaving factor, [td ρw exp(αN 0 sec θs )] is in error by 0.5 to 1.5%, which is unacceptably large.
The reason for the greater error in this term is that it depends on the downwelling irradiance, which
passes through the entire atmosphere and thus sees the total concentration N , not just the reduced
concentration N 0 that is adequate for correction of the TOA path term. However, the error in this
term also decreases to ∼ 0.2% if N 0 is replaced by N . This error can be reduced still further by
the following empirical procedure.
For bands where NO2 absorption is significant (e.g., at 412 or 443 nm), the atmospheric
correction is determined as always (without NO2 correction) using the NIR bands. However,
rather than subtract these terms from the corrected TOA reflectance, the atmospheric correction
terms (including the Rayleigh reflectance) are reduced for NO2 absorption by applying a factor of
exp[−αN 0 (sec θs + sec θv )]. The computed path reflectance for NO2 is then subtracted from the
observed TOA reflectance to obtain ∆ρobs = t3 td ρw , which is the TOA value for water-leaving
reflectance in the presence of NO2 . The NO2 -corrected value of the water-leaving reflectance is
then obtained by multiplying this ∆ρobs by exp(αN 0 sec θ) exp(αN sec θs ), which gives

exp(αN 0 sec θ) exp(αN sec θs )∆ρobs

= [t3 exp(αN 0 sec θ)] [td ρw exp(αN sec θs )] .

Note that t3 exp(αN 0 sec θ) is the NO2 -corrected transmission term, and ρ0w = td ρw exp(αN sec θs )
is the desired NO2 -corrected water-leaving reflectance. This equation is then solved to obtain ρ0w :

exp(αN 0 sec θ) exp(αN sec θs )∆ρobs

ρ0w = . (6.6)
t3 exp(αN 0 sec θ)

Note that the exponentials are increasing the magnitude of the water-leaving radiance compared to
the no-NO2 case, which accounts for the loss due to NO2 absorption along the paths of the Sun’s

44
direct beam and the viewing direction. The term for the Sun’s direct beam uses the full column
NO2 concentration N , whereas the viewing-path term uses the reduced concentration N 0 . This is
an artifice that brings the analytical correction of Eq. (6.6) into close agreement with the exact
numerical calculations. The absorption cross section α is a function of wavelength. Computations
are done for α at 18 deg C, and then a temperature correction is made. (Ahmad’s Table 1 gives the
band-averaged NO2 absorption cross sections for SeaWiFS and MODIS bands, which are called σ
in that table. This σ is same quantity as α in his Eq. (4) and in the equations of this chapter.)

45
 CHAPTER 7

Sun Glint

Both direct Sun glint and background sky reflectance are included in the radiative transfer calcula-
tions. However, the Sun glint contribution is removed from the LUTs so as to allow the historical
approach of Gordon and Wang (1994a) to be used. The current Rayleigh LUTs are wind speed de-
pendent, but the aerosol LUTs assume a flat ocean. Thus the LUTs include diffuse sky reflectance
but not specular reflection. There is thus an explicit correction for Sun glint, but not for diffuse
sky reflectance, which is accounted for as part of the Rayleigh correction.
Even for sensors (such as SeaWiFS) that are designed with tilt capabilities allowing them to
be oriented so that they do not look at the Sun’s glitter pattern, there can still be significant
residual glint radiance reaching the sensor, especially near the edges of the obvious glint area. This
is corrected as follows.
Recall Eq. (1.3):
Lt = LR + [La + LRa ] + T Lg + tLwc + tLw ,
where Lg is the direct Sun glint radiance. Lg is computed using the analytical Cox-Munk wind
speed-wave slope distribution and the Sun and viewing geometry. Wang and Bailey (2001, Eq. 2)
write the Sun glint radiance Lg in terms of a normalized Sun glint LGN , which is defined by

Lg (λ) ≡ Fo (λ)T (θs , λ)LGN .

LGN is computed using an azimuthally symmetric analytical form of the Cox-Munk wind speed–
wave slope distribution for the given Sun and viewing directions, and an incident irradiance of
magnitude Fo (λ) = 1 W m−2 nm−1 . LGN thus has the angular distribution of the surface-reflected
radiance, but its units are 1/steradian. Note that LGN is independent of wavelength.
During image processing, pixels with a value of LGN > 0.005 sr−1 are masked out as having too
much glint to be useful. Pixels with LGN ≤ 0.005 sr−1 have a glint correction applied before use.
For the glint correction, atmospheric attenuation occurs first along the path of the Sun’s direct
solar beam as the Sun’s beam travels from the TOA to the sea surface; the associated transmittance
is T (θs , λ). Attenuation then occurs along the viewing direction from the sea surface back to the
TOA; this transmittance is T (θv , λ). These are both direct beam transmittances because only one
particular path connects the Sun with a point on the sea surface that reflects the direct beam into
the sensor (recall the left panel of Fig. 4.1). The total “two-path” transmittance is the product of

46
the transmittances. The glint radiance correction, which is subtracted from Lt , is then (Wang and
Bailey, 2001, Eqns. 4 and 5)

T (θv , λ)Lg (θv , λ) = Fo (λ)T (θs , λ)T (θv , λ)LGN ,

where   
1 1
T (θs , λ)T (θv , λ) = exp −[τR (λ) + τa (λ)] + , (7.1)
cos θs cos θv
and where τR (λ) and τa (λ) are the Rayleigh and aerosol optical thicknesses, respectively.
Wang and Bailey (2001) comment (page 4792, left column) that the effects of ozone absorption
have already been accounted for before this state of processing. This is now also true for NO2
absorption.
Note that the glint correction cannot be computed unless the aerosol optical thickness (AOT)
τa is known. The AOT is obtained in a two-step process. First, the measured Lt (λ) and the wind
(1)
speed W are used to get a first estimate τa (λ) of the AOT. This estimate is obtained using the
aerosol correction algorithms described in Chapter 9. This estimate is then used in Eq. (7.1), and
the glint-corrected TOA radiance is then computed as
0
Lt (λ) = Lt (λ) − Fo (λ)T (θs , λ)T (θv , λ)LGN . (7.2)
(1)0 0
This gives the initial estimate Lt (λ) of Lt (λ). This value is then used again in the AOT algorithm
(2)
to obtain the second estimate τa (λ) for the AOT. The second AOT estimate is then used again in
(2)0
Eqs. (7.1) and (7.2) to obtain an improved estimate Lt (λ). In practice, only two iterations give
(2)
a satisfactory final estimate for the AOT, τa (λ) = τa (λ), and thus for the glint-corrected TOA
radiance. This final τa (λ) is then used to compute the aerosol contribution to the TOA radiance,
as described in Chapter 9.

47
 CHAPTER 8

Whitecaps

The contribution of white caps and foam to the TOA radiance depends on two factors: the re-
flectance of whitecaps per se and the fraction of the sea surface that is covered by whitecaps.
Following Gordon and Wang (1994b), the contribution of whitecaps and foam at the TOA is
(recall Eq. 3.13)
t(θv , λ)ρwc (λ) = [ρwc (λ)]N t(θs , λ) t(θv , λ) ,
where t(θv , λ) is the diffuse atmospheric transmission in the viewing direction, t(θs , λ) is the diffuse
transmission in the Sun’s direction, and [ρwc (λ)]N is the non-dimensional normalized whitecap
reflectance. [ρwc (λ)]N is defined in the same manner as was the normalized water-leaving radiance
[ρw (λ)]N in §3, namely
 2
R
π Ro Lwc (θs )
[ρwc ]N ≡ [Lwc ]N = π , (8.1)
Fo Fo cos θs t(θs )
where Lwc is the whitecap radiance. It is assumed that the whitecaps are Lambertian reflectors, so
that (unlike for Lw ) Lwc does not depend on direction θv , φ. This gives the interpretation (Gordon
and Wang, 1994b, page 7754) that “ρ is the reflectance–the reflected irradiance divided by the
incident irradiance–that a Lambertian target held horizontally at the TOA would have to have to
produce the radiance L.” [ρwc ]N can be interpreted as the average reflectance of the sea surface
that results from whitecaps in the absence of atmospheric attenuation.
The effective whitecap irradiance reflectance is taken from Koepke (1984) to be 0.22 (albeit
with ±50% error bars). This reflectance is independent of wavelength. This gives [ρwc ]N = 0.22Fwc ,
where Fwc is the fraction of the sea surface that is covered by whitecaps. The fractional coverage
is taken from Stramska and Petelski (2003), who give two models for for Fwc :

Fwc = 5.0 × 10−5 (U10 − 4.47)3 for developed seas (8.2)

Fwc = 8.75 × 10−5 (U10 − 6.33)3 for undeveloped seas (8.3)

where W is the wind speed in m s−1 at 10 m. Formula (8.3) for undeveloped seas is used on the
assumption that if the seas are well developed it is probably stormy, hence cloudy, so that remote
sensing is not possible. The blue curve in Fig. (8.4) shows Fwc for undeveloped seas.

48
 Figure 8.1: Whitecap normalized
reflectance as given by Eq. (8.4)
with awc = 1, and the fraction of sea
surface whitecap coverage as given
by Eq. (8.3).

The final model for [ρwc ]N is then taken to be

[ρwc ]N (λ) = awc (λ) × 0.22 × Fwc

= awc (λ) × 1.925 × 10−5 (U10 − 6.33)3 . (8.4)

A whitecap correction is applied for wind speeds in the range 6.33 ≤ U10 ≤ 12 m s−1 . The factor
awc (λ) is a normalized whitecap reflectance that describes the decrease in reflectance at red and
NIR wavelengths. This factor is taken from Frouin et al. (1996, Figs. 3 and 4 ); the values are

λ = 412 443 490 510 555 670 765 865

awc = 1.0 1.0 1.0 1.0 1.0 0.889 0.760 0.645

Linear interpolation is used as needed between these values. Figure 8.1 shows the whitecap re-
flectance as given by Eq. (8.4) when awc = 1.

49
 CHAPTER 9

Aerosols

9.1 Aerosol Properties

Aerosols are solid or liquid particles that are much larger than gas molecules but small enough to
remain suspended in the atmosphere for periods of hours to days or longer. Typical sizes are 0.1 to
10 µm. An aerosol’s optical properties are determined by its composition, usually parameterized
via its complex index of refraction, and its particle size distribution (PSD).
For the purposes of atmospheric correction, aerosol particle size distributions are modeled as a
sum of “fine” (small; radii less than roughly 1 µm) and “coarse” (large; radii greater than roughly
1 µm) particles, with a log-normal distribution for each. (The log-normal distribution is reviewed
in Campbell (1995).) The cumulative volume distribution is then (Ahmad et al., 2010a)
2
"   #
dV (r) X Voi ln r − ln rvoi 2
= √ exp − √ .
d ln r 2πσi 2σi
i=1

Here V (r) is the volume of particles per volume of space with size less than or equal to r; V (r)
is typically specified as µm3 cm−3 . rvoi is the volume geometric mean radius, and σi is geometric
standard deviation for class i. The integral of dV (r)/d ln r over all sizes r = 0 to ∞ (i.e., ln r from
−∞ to ∞) gives V (∞) = Voi . Thus Voi is the total volume of particles of class i per volume of
space.
A similar equation holds for the cumulative number distribution dN (r)/d ln r, where N (r) is
the number of particles per volume of space with size less than or equal to r. The corresponding
parameters rnoi and Noi can be obtained from rvoi and Voi ; see the equations in Ahmad et al.
(2010a). The particle size distribution (PSD) is given by

dN (r) 1 dN (r)
n(r) = = ,
dr r d ln r
where n(r)dr is the number of particles per unit volume in size range r to r + dr. The units for
n(r) are usually expressed as particles m−3 µm−1 .
Figure 9.1 illustrates shapes of the volume V (r), number N (r), and particle size n(r) distri-
butions for an open ocean aerosol, computed using the parameter values of Ahmad et al. (2010a,

50
Figure 9.1: Illustration of aerosol volume, number, and particle size distributions. The parameter
values are for an open-ocean aerosol. The green curve is fine particles; blue is coarse particles; red
is the sum of fine and coarse particles. Each total curve is normalized to 1 at its maximum value.

Table 2). The distributions of the fine aerosols are given by the green lines. The blue lines are the
coarse aerosols, and the red lines are the sums. The two roughly comparable distributions of left
panel of the figure show that the fine and coarse aerosols each contribute a significant amount of
the total particle volume. In the present example the fines are 25.7% of the total volume and the
coarse particles are 74.3% of the total volume. The middle panel shows that the fines dominate the
number of particles; in the present case there are 477 times as many fine particles as coarse. The
right panel shows the PSD. The black dashed line shows the -4 slope of a Junge distribution for
comparison.
The radius parameter rvoi and index of refraction both depend on the aerosol type (dust, sea
salt, soot, etc.) and on the relative humidity Rh. The index of refraction generally depends on
wavelength. Figure 9.2 shows the effect of relative humidity on cumulative volume and particle size
distributions for an open-ocean aerosol (parameter values from Ahmad et al. (2010a, Table 4)).
Note that as Rh increases, the particles absorb more water and increase in size, so the distributions
shift to the right. The shape of the distribution also changes with Rh.
As modeled in Ahmad et al. (2010a), the fine particles are generally of “continental” origin and
include both dust and soot. The fine particles are sometimes absorbing. The coarse particles are
of “oceanic” origin and are assumed to be non-absorbing sea salts. The tables in Ahmad et al.
(2010a) give the PSD parameters and indices of refraction for different aerosol types (dust, sea salt,
soot, etc.) and relative humidities. (Note that Table 1 of Ahmad et al. (2010a) has errors. The
corrected table is given in Ahmad et al. (2010b).)
An aerosol’s physical properties determine its optical properties, namely its mass-, number-,
or volume-specific absorption a∗ (λ) and scattering b∗ (λ) coefficients and scattering phase function
β̃(ψ, λ), where ψ is the scattering angle. If the particles are homogeneous spheres, Mie theory can
be used to compute the optical properties from the physical properties. This is often done, although
the assumption of homogeneous spherical particles may or may not be valid in a given situation. In
any case, once the a∗ (λ) and scattering b∗ (λ) coefficients are known, then given the concentration

51
Figure 9.2: Effect of relative humidity Rh on the cumulative volume and particle size distributions
for a typical open-ocean aerosol. The curves in each panel are normalized to 1 at the maximum
value of the three curves.

profile Conc(z) as a function of altitude z, the extinction coefficient c(z, λ) = Conc(z)[a∗ (λ)+b∗ (λ)]
can be computed. The aerosol optical thickness or aerosol optical depth is then given by
Z T OA
τa (λ) = c(z, λ)dz ,
z0

where z0 is the surface elevation. (Generally z0 = 0 for mean sea level, but may also be the elevation
of a lake.)
For all else held fixed, the aerosol optical thickness at wavelength λ is approximately related to
the value at a reference wavelength λo by
 α
τa (λ) λo
= (9.1)
τa (λo ) λ

The parameter α is known as the Ångström exponent or Ångström coefficient. Smaller (larger)
particles generally have a larger (smaller) Ångström exponent.
The single scattering albedo ωo defined by

b∗ (λ)
ωo (λ) =
c∗ (λ)

is also of use in modeling the optical effects of aerosols on the radiance distribution.
Ahmad et al. (2010a) constructed look-up-tables (LUTs) for 10 aerosol types and 8 relative
humidities, for a total of 80 aerosol tables. The fine fraction was a mixture of 99.5% dustlike
and 0.5% soot particles (not modeled by Shettle and Fenn) for all 10 aerosol types, which gives
good agreement on average with AERONET measurements of aerosol optical properties. The
ten aerosol models have different weights of fine and coarse particles, but the effective radius reff
(reff = r3 n(r)dr/ r2 n(r)dr) and mean radius µo (µo = rn(r)dr/ n(r)dr) are the same. The
R R R R

aerosol types were then defined by letting the fine-to-coarse fraction vary from 0 to 1. For each

52
aerosol type, relative humidities of Rh = 30, 50, 70, 75, 80, 85, 90, and 95% were used. The actual
aerosol LUTs contain the components of the model from which to derive single-scattering aerosol
reflectance ratios in each band (relative to any reference band), plus a set of quadratic coefficients
relating single to multiple scattering (as in Gordon and Wang (1994a)), plus a separate table of
Rayleigh-aerosol diffuse transmittance coefficients of the form t = A exp(−Bτa ) (recall Eq. 4.6).
Mie is theory used to compute the aerosol phase functions for use in the radiative transfer model.

9.2 Black-pixel Calculations

The algorithm developed by Gordon and Wang (1994a) is used, although the original aerosol models
and LUTs have been updated as described in Ahmad et al. (2010a), and model selection is now
partitioned by relative humidity (RH). (Gordon and Wang ignored the glint term since they were
considering SeaWiFS, whose viewing direction was chosen to avoid direct Sun glint.)
Beginning with Eq. (3.13),

ρt = ρR + [ρa + ρRa ] + T ρg + tρwc + tρw , (9.2)

The basic theory in Gordon and Wang (1994a) is developed using single-scattering theory, in which
case the ρRa term is zero because there is no multiple scattering. Multiple-scattering effects are
then added via numerical models using the guidance of the single-scattering theory.
Assume that the corrections for Rayleigh, whitecaps, O3 , NO2 , and Sun glint have all been
made. Then the left hand side of

ρt − ρR − T ρg − tρwc = [ρa + ρRa ] + tρw (9.3)

or
ρAw ≡ ρA + tρw
is known. Here ρAw is just convenient shorthand for the measured TOA reflectance ρt with the
Rayleigh and other effects removed. The next task is to compute ρA , the combined aerosol and
aerosol-Rayleigh reflectance, and move it to the left hand side, after which the desired ρw will be
known.
Low-chlorophyll, Case 1 waters have negligible water-leaving radiance at near-infrared (NIR)
wavelengths, i.e. beyond roughly 700 nm. For such waters, it can be assumed that ρw (λ >
700 nm) = 0, which is known as the “black-pixel” assumption. Let λ1 and λ2 be two NIR wave-
lengths, with λ1 < λ2 . At these two wavelengths, the TOA normalized reflectance (corrected as
shown in Eq. (9.3)) is due entirely to atmospheric path radiance: ρAw (λi ) = ρA (λi ), i = 1, 2. Table
9.1 shows the λ1 and λ2 bands for several sensors.
Now consider the ratio
ρA (λ1 ) ρAw (λ1 )
(λ1 , λ2 ) ≡ = (9.4)
ρA (λ2 ) Black Pixel ρAw (λ2 )
The quantity (λ1 , λ2 ), and more generally the quantity (λ, λ2 ) for any λ < λ2 , depends on the
aerosol type, which is determined by the particle type, PSD, and relative humidity. Figure 9.3
shows the dependence of (λ, λ2 = 865) for the ten aerosol models, for one particular set of solar
zenith angle, viewing direction, and relative humidity.
For the given image pixel being corrected, the process is as follows:

53
 Table 9.1: NIR bands used for aerosol correction.
Band Label Wavelengths [nm] Nominal Wavelength [nm]
SeaWiFS
7 745-785 λ1 = 765
8 845-855 λ2 = 865
MODIS
15 743-753 λ1 = 748
16 862-877 λ2 = 869
VIIRS
M6 739-754 λ1 = 745
M7 846-885 λ2 = 862

• The relative humidity RH is taken from NCEP.

• For each bounding RH value in the family of 8 RH values in the database, the corresponding
family of 10 aerosol types in the database is then searched to find the two aerosols types
whose precomputed values of (λ1 , λ2 ), call them low and high , bracket the measured value
of (λ1 , λ2 ) for the given Sun and viewing geometry.

• This selects two of the curves like those of Fig. 9.3, for which the corresponding (λ, λ2 ) and
τa values have been precomputed and stored in the aerosol LUT.

• It is assumed that the difference in the precomputed (λ, λ2 ) is at all wavelengths in the
same proportion as the measured (λ1 , λ2 ) is to the bracketing values at the NIR reference
wavelengths. Thus let
(λ1 , λ2 ) − low (λ1 , λ2 )
∆= (9.5)
high (λ1 , λ2 ) − low (λ1 , λ2 )

• The aerosol reflectance at all wavelengths is then computed from the measured ρAw (λ2 ) and
the tabulated low and high values using

ρA (λ) = [(1 − ∆)low (λ, λ2 ) + ∆high (λ, λ2 )] ρAw (λ2 ) (9.6)

Now that ρA (λ) is known, Eq. (9.3) gives

tρw = ρAw − ρA = ρt − ρR − T ρg − tρwc − [ρa + ρRa ] (9.7)

Recall that t is the diffuse transmission in the viewing direction, and that tρw is the contribution
of water-leaving radiance (in reflectance form) at the TOA. The desired ρw at the sea surface is
thus obtained from
ρAw − ρA
ρw = . (9.8)
t
The aerosol optical depth is computed at all wavelengths using the values of τ (λ2 ) and the
Ångström exponent in Eq. (9.1).
This technique rests on two main assumptions:

54
Figure 9.3: Dependence of (λ, λ2 = 865) on aerosol model. The aerosol model is determined by the
value of the Ångström exponent α and the effective particle size reff . All curves are for a particular
sun zenith angle θs , viewing direction θv , φv , and relative humidity RH, as shown.

• The water-leaving radiance is negligible at the NIR reference wavelengths. This is valid
only for optically deep, Case 1 waters, with a chlorophyll concentration of 0.3 mg m−3 or
less. Waters containing higher chlorophyll concentrations or mineral particles will violate this
assumption. Figure 9.4 shows an example of very turbid water for which the remote-sensing
reflectances at NIR wavelengths is not negligible.

• The aerosols are not strongly absorbing. Some mineral aerosols absorb strongly at blue
wavelengths but not in the NIR. Their (λ, λ2 ) functions look like the black curve in Fig. 9.5.
Thus their presence cannot be detected from the NIR TOA radiances.

Note that these are unrelated assumptions: the water can have non-zero NIR reflectance and
the atmosphere can have non-absorbing aerosols, or there can be zero NIR reflecance but absorbing
aerosols. If the water-leaving radiance is not zero at the λ2 reference wavelength, then the ρw
contribution to ρAw will be interpreted as a larger aerosol concentration. This leads to over-
correction for the aerosol, i.e, subtracting too much ρA from ρTOA . The resulting ρw is then too
small, and can even be negative at blue wavelengths. Likewise, if the aerosol is blue-absorbing,
over-correction occurs at blue wavelengths and, again, ρw is too small or even negative at blue
wavelengths.

55
Figure 9.4: Examples of Rrs for Case 1 and Case 2 waters. Rrs is not zero in the NIR for the
Mississippi River Delta water.

Figure 9.5: Qualitative behavior of (λ, λ2 = 865) for a blue-absorbing aerosol (black curve).

56
9.3 Non-black-pixel Calculations
Many ways to treat non-black pixels have been developed; the method currently implemented by
the OBPG is described in Bailey et al. (2010). This algorithm works as follows.
It is necessary to estimate Rrs (λ) (or equivalently ρw (λ)) at the NIR reference wavelengths so
that the non-zero water-leaving radiance can be removed from the TOA signal, leaving only the
aerosol reflectance as the contribution to ρAw , from which the aerosol type (i.e., (λ, λ2 )) can be
determined. However, Rrs (N IR) can’t be estimated until the aerosol contribution is removed. Thus
an iterative solution must be used to obtain both Rrs (N IR) and the aerosol type.
The remote-sensing reflectance can be written
f (λ) bb (λ)
Rrs (λ) = . (9.9)
Q(λ) a(λ) + bb (λ)
As discussed in detail in §3, the f /Q factor describes the angular distribution of the water-leaving
radiance, i.e., the BRDF of the ocean. This factor depends on the Sun and sky radiance distribution
(parameterized by the solar zenith angle θs and AOT τa ), water-column IOPs (parameterized by
the chlorophyll concentration Chl in Case 1 waters), sea state (wind speed W ), viewing direction
(θv , φv ), and wavelength. It has been extensively studied and numerically modeled by Morel et al.
(2002), who present tabulated values as functions of θs , τa , θv , φv , Chl, W , and λ. Although the
Morel et al. (2002) f /Q table was generated for Case 1 waters, analysis shows (Bailey et al., 2010)
that it is often adequate for Case 2 waters as well. The f /Q factor is thus considered known for
the present calculations.
The iterative correction for waters where the black-pixel assumption cannot be made has the
following steps:

1. Assume that Rrs (765) and Rrs (865) are both 0, i.e. make the black-pixel assumption for both
NIR reference bands.

2. Complete the atmospheric correction process as described in §9.2. This gives the initial
estimate of ρw (λ), or equivalently Rrs (λ).

3. Use Rrs (443) and Rrs (555) from the initial estimate of Rrs (λ) to get η by the empirical
relationship (Lee et al. (2010, Eq. 8); Bailey et al. (2010, Eq. 3))
  
Rrs (443)
η = 2 1 − 1.2 exp 0.9 . (9.10)
Rrs (555)

4. Use the initial Rrs (λ) to get an initial estimate of the chlorophyll concentration Chl. The
particular algorithm used to obtain Chl from Rrs (λ) depends on the sensor.

5. Use this Chl to obtain a(670) via the empirical relationship (Bailey et al., 2010, Eq. 4)

a(670) = exp[0.9389 ln(Chl) − 3.7589] + aw (670) . (9.11)

where the aw (670) = 0.439 m−1 is the absorption by pure water.

6. Use a(670) and Rrs (670) in Eq. (9.9) to solve for bb (670) = bbw (670) + bbp (670), where
bbw (670) = 4.26 × 10−4 m−1 is the backscatter coefficient for pure sea water.

57
 7. Use η from Eq. (9.10) and (Bailey et al., 2010, Eqs. 2b,3) to compute bb (765):

670 η
 
bb (765) = bbw (765) + bpb (670) (9.12)
765

where bbw (765) = 2.38×10−4 m−1 . bb (865) is computed in the same manner using bbw (865) =
1.41 × 10−4 m−1 .

8. Use this bb (765) and a(765) = aw (765) = 2.85 m−1 to get Rrs (765) from Eq. (9.9). Similarly,
compute Rrs (865) using bb (865) and aw (865) = 4.61 m−1 .

9. Use the new, non-zero value of Rrs (765) (i.e. ρw (765)) to remove the non-zero ρw (765)
contribution to ρt (765). Do the same calculation for 865 nm.

10. Return to Step 2 and repeat the atmospheric correction using the black-pixel algorithm. This
will give a new (hopefully better) estimate of Rrs (λ), thus an new estimate of the other
parameters, and finally new estimates of Rrs (765) and Rrs (865) at Step 8. After using the
new values of ρw (765) and ρw (865) to correct for the non-zero water contribution to ρt (765)
and ρt (865), return to Step 2 for a new iteration. Continue iterating until the change in
Rrs (765) from one iteration to the next is less than 2%, which typically takes 2-4 iterations,
or when 10 iterations have been made.

If this iteration process fails to converge within 10 cycles, then re-initialze with ρa (N IR) = 0,
i.e., set the NIR aerosol contribution to zero. This implies that all of the NIR reflectance is due
to the water (after correction for Rayleigh and Sun glint). Repeat the iteration until convergence
is reached. If convergence is still not reached, do one more calculation with ρa (N IR) = 0 and flag
the pixel as “atmospheric correction warning.” However, even in this case, the retrieval may still
be useful.
The above iteration is not done if the initial Chl estimate is less that 0.3 mg m−3 , and it is
always done if the initial Chl estimate is greater that 0.7 mg m−3 . To prevent discontinuities in
the final results, the Rrs (765) estimate is linearly weighted from 0 to 1 for 0.3 ≤ Chl ≤ 0.7. Figure
9.6 shows the regions of the ocean where the non-black-pixel algorithm is likely to be applied.
Finally, rather than the exact NIR reference wavelengths of λ1 = 765 and λ2 = 865 shown above,
in practice band-averaged IOP values are used for the particular sensor. Thus for VIIRS-NPP, the
reference bands are centered at 745 and 862 nm, with the band-averaged aw (745) = 2.806 m−1 ,
and so on. Band averaged IOPs for different sensors are given at http://oceancolor.gsfc.nasa.
gov/DOCS/RSR_tables.html.

58
Figure 9.6: Regions of the ocean where the non-black-pixel correction is likely to be applied, based
on SeaWiFS climatology. Land is black, gray is Chl < 0.3 mg m−3 (where the correction is not
applied), and white is Chl > 0.3 mg m−3 (where the correction is applied). [Fig. 3 of (Bailey et al.,
2010)]

9.4 Strongly Absorbing Aerosols

The aerosol models of Ahmad et al. (2010a) discussed in Section 9.2 include a small fraction of
absorbing soot for the fine part of the aerosol size distribution. These aerosol models are used for
routine correction for aerosols. These models, however, cannot account for the presence of strongly
absorbing aerosols, which frequently occur in coastal regions downwind from continents, and even
in mid-ocean regions because winds can transport fine dust particles long distances.
Strongly absorbing aerosols (especially at blue wavelengths) have been a topic of much research
over the years (Gordon et al., 1997b, e.g.). However, as was illustrated in Fig. 9.5, there is no
reliable way to detect the presence of absorbing aerosols from the NIR bands of Table 9.1 during
the atmospheric correction process. Therefore, all pixels are processed using the algorithms for
non- or weakly absorbing aerosols as described in the previous sections.

59
 CHAPTER 10

Spectral Out-of-band Correction

Figure 10.1 shows the sensor response functions (SRF) of the MODIS-Aqua bands used for ocean
color remote sensing; each response function is normalized to 1 at its maximum. Visually, in this
plot with a linear ordinate axis, the bands appear to be well defined and narrow, with full-width,
half-maximum (FWHM; the wavelengths at which the function is one-half of its maximum) widths
of 10 to 15 nm. However, when plotted with a logarithmic ordinate as in Fig. 10.2, it is seen that
there is significant “out-of-band” (OOB) sensitivity; i.e., a non-zero response outside the nominal
band width. In each plot, the black curve represents a TOA radiance with a wavelength dependence
of λ−4 .
Rayleigh scattering with a λ−4 dependence dominates the total TOA radiance. If a radiance
with such a wavelength dependence is measured by sensors having the response functions shown in
Figs. 10.1 and 10.2, the total measured radiance in the ith band over the 380-1100 nm range shown
in the figures will be (Gordon, 1995, Eq. 8)
R 1100
LTOA (λ) SRFi (λ) dλ
Li (total) = 380 R 1100 .
380 SRFi (λ) dλ

Define the “in-band” part of the total signal to be the part detected between chosen lower (λlow )
and upper (λup ) wavelengths. The OBPG uses the wavelengths at which the SRF drops to 0.1%
of its maximum value to define the lower and upper boundaries of the in-band region. For the
nominal 488 nm band, for example, λlow = 460 nm and λup = 503 nm. (The FWHM boundaries
for the 488 nm band are FWHMlow = 482 nm and FWHMup = 493 nm.) The part of the measured
radiance that comes from in-band wavelengths is then
R λup
λ Lt (λ) SRFi (λ) dλ
Li (in band) = low R 1100 , (10.1)
380 SRFi (λ) dλ

with similar equations for the out-of-band contributions at wavelengths less than λlow and greater
than λup . Numerical integration shows that for a λ−4 TOA radiance and the nominal 488 nm
band, 99.24% comes from the in-band wavelengths, 0.55% comes from out-of-band response at
wavelengths less than λlow = 460 nm, and 0.20% comes from out-of-band response at wavelengths
greater than λup = 503 nm. For the 866 nm band, the corresponding numbers are 99.22% in-band,

60
Figure 10.1: Relative sensor response functions (SRF) for the MODIS-Aqua bands used for ocean
color remote sensing. The black curve represents a TOA radiance proportional to λ−4 . (Data from
from http://oceancolor.gsfc.nasa.gov/DOCS/RSR_tables.html)

Figure 10.2: Same as Fig. 10.1 except that the ordinate axis is logarithmic to show the out-of-band
response.

61
Figure 10.3: The MODIS-Aqua relative sensor response (purple) and a perfect sensor with a 10-nm
FWHM (gray shading). The black line represents a radiance proportional to λ−4 .

0.63% from wavelengths less than λlow = 843 nm, and 0.16% comes from wavelengths greater than
λup = 891 nm. Thus, for the MODIS-Aqua bands, almost 1% of the TOA radiance attributed to
a nominal bandwidth actually comes from outside that band. This magnitude of misattribution of
radiances between bands is significant and requires correction for proper interpretation of measured
data.
Gordon (1995) points out that the OOB corrections must be applied separately to the individual
components of the TOA radiance because the OOB response depends on the spectral shape of the
radiance. That is, separate corrections must be applied to the Rayleigh, aerosol, and water-leaving
radiances. Those corrections are built into the sensor-specific Rayleigh and aerosol LUTs described
above. This section describes how the OOB correction is applied to the remote-sensing reflectance
Rrs after the preceding steps of the atmospheric correction process have been carried out.
It should be noted that OOB corrections are also required when comparing measurements made
by sensors having different spectral responses. This happens, for example, when comparing a nom-
inal MODIS 412 nm band value with an in-situ measurement made by a multispectral radiometer
having a nominal 412 band. The filters in the MODIS and in situ radiometer instruments will
not have exactly the same spectral responses or nominal bandwidths. In general, it is desirable to
reference any measurement to what would be obtained by an ideal sensor with a perfect response
function defined by the nominal FWHM. This is illustrated in Fig. 10.3 for the MODIS-Aqua 412
band and a sensor with a perfect “top hat” response for 407 ≤ λ ≤ 412 nm. For a radiance with a
λ−4 wavelength dependence, 18.1% of the MODIS nominal 412 band response comes from λ < 407,
60.5% comes from within the nominal 10 nm bandwidth of the perfect sensor, and 21.4% comes
from λ > 412 nm.
Figure 10.4 shows Rrs (λ) as computed for Case 1 water using a model of the type developed in
Morel and Maritorena (2001) for Kd (λ) and R((λ).

62
 Figure 10.4: Rrs (λ) spectra as functions of the chlorophyll concentration.

For each nominal sensor band labeled by λi , i = 1, ..., Nbands , and for each chlorophll value
Chlj , j = 1, ..., NChl , the Rrs (λ) spectra of Fig. 10.4 are used in equations of the form of (10.1)
with appropriate integration limits to compute:
11 (λ , Chl ) = The mean R
• Rrs i j rs over idealized 11-nm bandwidths (center wavelength ±5 nm)
corresponding to the nominal satellite bands (as illustrated by the gray 412 nm band in Fig.
10.3)
full (λ , Chl ) = R
• Rrs th sensor
i j rs computed using the full spectral response function for the i
band.

• The ratio
11 (λ , Chl )
Rrs i j
r(λi , Chlj ) = full
. (10.2)
Rrs (λi , Chlj )

Figure 10.5 illustrates the results of these calculations for the SeaWiFS nominal 555 nm band. This
figure shows chlorophyll values only for Chl <= 3 mg m−3 , which was felt to be the upper limit
of reliability of the chlorophyll-based Rrs (λ) model of Fig. 10.4. A similar figure can be drawn for
each sensor band. If the chlorophyll concentration were known, a rearrangement of Eq. (10.2) and
ratio curves like that of Fig. 10.4 could be used to compute the correction to the measured Rrs full

for the ith band.

However, the chlorophyll concentration is not yet known. To proceed, the r(λi , Chlj ) curves like
the one shown by the blue line of Fig. 10.5 are used to compute correction factors r(λi ) as functions
of the ratio of the uncorrected Rrs (490 nm) to the uncorrected Rrs (555 nm). Each chlorophyll value
shown in figures like 10.5 for the various bands gives a point like those shown in Fig. 10.6 for r(555)
vs. the uncorrected Rrs (490)/Rrs (555). A corresponding set of r(λi ) points is computed for each
band, but in each case as a function of Rrs (490)/Rrs (555). (This choice of the ratio of 490 to 555 nm

63
Figure 10.5: Rrs11 (λ full
555 , Chlj ) (red dots), Rrs (λ555 , Chlj ) (black squares), and the ratio r(λ555 , Chlj )
(blue line, right ordinate axis) for the SeaWiFS nominal 555 nm band.

for the independent variable traces back to SeaWiFS, for which these were the most trustworthy
bands.) A best-fit function to the set of points so generated is then found for each of the r(λi ) vs.
Rrs (490)/Rrs (555) functions.
These functions are then used to apply the OOB correction to the measured Rrs (555) values
as follows. Given the measured full-band (uncorrected) Rrs (i), the value of Rrs (490)/Rrs (555) is
used to evaluate the functional fit to the points of Fig. 10.6 in order to obtain the correction to be
applied to the Rrs (555) value. The corresponding functions for the other bands are used to correct
those bands. For example, the fit to r(412) vs. Rrs (490)/Rrs (555) is used to correct the 412 nm
band, and so on. If the Rrs (490)/Rrs (555) value is outside the range of the points used for the data
fit as illustrated in Fig. 10.6, the value of the nearest point is used, rather than extrapolating with
the fitting function beyond the range of the underlying data.
The Rrs (i) functions were developed using a Case 1 model for Rrs (λ) as shown in Fig. 10.4.
Case 2 waters can have much different Rrs (λ) spectra and therefore should have different OOB
corrections. However, in practice, Case 2 waters have the same correction applied as for Case 1
waters.
A final comment is warranted regarding the use of this out-of-band correction when comparing
satellite-derived and in-situ measurements of Rrs (e.g, when doing vicarious calibration):

• If comparing a multispectral satellite band Rrs with in an in situ multispectral measurement,

perform the OOB adjustment to the satellite data. (Of course, an adjustment should also
be made to the in situ values based on the relative spectral responses SRF (λ) of the in situ
radiometer.)

• If comparing a multispectral satellite band Rrs with in an situ hyperspectral measurement

that has been filtered with a 10 nm bandpass filter, perform the adjustment.

64
Figure 10.6: Out-of-band correction factor for the SeaWiFS 555 nm band as a function of the ratio
of the uncorrected Rrs (490) to Rrs (555).

• If comparing a multispectral satellite band Rrs with an in situ hyperspectral measurement that
has not been filtered, do not perform the OOB adjustment to the satellite data. However, the
hyperspectral in situ spectrum should processed using the satellite sensor SRF (λ) spectra.
That is, replace the spectrum used in integrals of the form of Eq. (10.1) with the hyperspectral
Rrs (λ).

Performing the OOB adjustment is the default for processing imagery at OBPG. Therefore,
if a user wants to compare satellite data with unfiltered hyperspectral data as in the third bullet
above, the standard Level 2 files cannot be used. The user would need to begin with the Level 1b
TOA radiances, disable the OOB correction in the atmospheric correction software, and reprocess
the TOA radiances to Level 2.

65
 CHAPTER 11

Polarization Correction

Radiance leaving the top of the atmosphere can be strongly polarized even though the sunlight
incident onto the TOA is unpolarized. This is because scattering by atmospheric constituents,
reflection by the sea surface, and scattering within the water all can generate various states of
polarization from unpolarized radiance. Although remote sensing as considered here is based on
the total TOA radiance without regard to its state of polarization, many instruments are sensitive
to polarization. Therefore, the total TOA radiance they measure may depend on the state of
polarization of the TOA radiance and the orientation of the instrument relative to the plane of
linear polarization. Correction for these effects is required so that instruments give consistent
measurements of the total TOA radiance.
The MODIS sensors are polarization sensitive. MODIS radiance measurements vary by up to
±5.4% for totally linearly polarized radiance, depending of the orientation of the sensor relative to
the plane of polarization. This amounts to about ±3% differences in measured TOA radiances for
typical values of atmospheric polarization (Meister et al., 2005). These effects must be accounted
for in order to achieve the desired 0.5% accuracy in measured TOA radiance. Similarly, the VIIRS
instrument polarization sensitivity is 1-2% and requires a polarization correction. SeaWiFS was by
design not very sensitive to polarization (< 0.25%), and no polarization correction was applied.
The state of polarization is described by the four-component Stokes vector [I, Q, U, V ]T , where
superscript T denotes transpose, I is the total radiance without regard for its state of polarization,
Q specifies the linear polarization resolved in planes parallel and perpendicular to a conveniently
chosen reference plane, U specifies the polarization resolved in planes oriented ±45 deg to the
reference plane, and V specifies the right or left circular polarization. The choice of the reference
plane for specification of the Q and U components is arbitrary and can be made for convenience.
The direction of propagation of the radiance is given by a unit vector î, so that the Stokes vector can
be written as I = [I, Q, U, V ]T î when it is desired to indicate both its components and direction.
Direction î can be specified by the polar (θ) and azimuthal (φ) directions in a spherical coordinate
system as shown in Fig. 11.1. In that figure, θ̂ and φ̂ are unit vectors specified by the directions
of increasing θ (θ = 0 at the pole or ẑ direction in a cartesian coordinate system) and increasing φ
(φ = 0 in a conveniently chosen azimuthal direction such as x̂ pointing east or toward the Sun).
In the geophysical setting, it is customary to define the Stoke vector components Q and U with

66
reference to a plane defined by the normal to the sea surface and the direction of propagation of
the radiance. This plane is known as the meridional plane and is partly shaded in light blue in
Fig. 11.1. Following the notation and choices of Gordon et al. (1997a), let r̂t = φ̂ be the reference
direction perpendicular to the meridional plane, and let ˆlt = −θ̂ be the reference direction parallel
to the meridional plane. The direction of propagation of the radiance to be measured is then
î = r̂t × ˆlt . The total TOA radiance resolved in these directions is denoted I t = [It , Qt , Ut , Vt ]T î.
This radiance is being measured by a sensor illustrated by the red rectangle in Fig. 11.1. The
Stokes vector measured by that sensor has its Q and U components resolved along perpendicular
(r̂) and parallel (ˆl) directions chosen for convenience relative to the orientation of the sensor. The
sensor will measure the TOA radiance as a Stokes vector I m = [Im , Qm , Um , Vm ]T î. For an incident
radiance I , the optical system comprising the sensor itself and any associated optical components
(mirrors, lenses, etc.) will convert the incient radiance into a measured value given by I m = M I ,
where M is the 4 × 4 Mueller matrix that describes the optical properties of the sensor optical
system. M is defined relative to the sensor reference directions r̂ and ˆl. In order for M to operate
on the TOA radiance I t , which is defined with reference directions r̂t and ˆlt , I t must first be
transformed (rotated) from the meridional-based r̂t , ˆlt system to the sensor-based r̂, ˆl system.
Let α = cos−1 (ˆlt · ˆl) be the angle between the parallel reference directions for I t and for the
sensor. With the choice of α being positive for clockwise rotations from ˆlt to ˆl as seen looking “into

Figure 11.1: Angles and directions used in specification of Stokes vectors. The blue-shaded wedge
indicates the meridional plane used to define the TOA Stokes vector I t defined via the blue unit
vectors; the red-shaded rectangle represents the sensor that measures I m defined via the red unit
vectors.

67
the beam” (looking in the −î direction), the transformation is given by the 4 × 4 rotation matrix
 
1 0 0 0
0 cos(2α) sin(2α) 0
R(α) =  . (11.1)
 
0 − sin(2α) cos(2α) 0
0 0 0 1

Thus, the radiance as measured by the sensor is

I m = M R(α)I t . (11.2)

It is important to note that the radiance measured by the sensor, I m depends both on the “true”
TOA radiance I t , the sensor optical properties via M , and the orientation α of the sensor relative
to the local meridional plane. M is fixed for a given sensor, but I t and α change from moment to
moment as the sensor orbits and views the TOA radiance in different locations and directions. (To
be exact, there are long-term changes to M caused by degradation of the sensor optical surfaces.
These changes are monitored on-orbit and corrected by a cross-calibration technique.)
The quantity of interest here is the measured TOA radiance magnitude, which is given by the
first element of the Stokes vector. Using (11.1) in Eq. (11.2) gives this to be

Im = M11 It + M12 [cos(2α)Qt + sin(2α)Ut ]

+ M13 [− sin(2α)Qt + cos(2α)Ut ] + M14 Vt . (11.3)

Clearly, if M11 = 1 and all elements of M other than the M11 element are zero, then the sensor is not
sensitive to the state of polarization and Im = It . Physical arguments and numerical simulation
show the circular polarization of the TOA radiance is very small: |Vt | ≤ 10−3 It . This term is
therefore neglected in the present correction algorithm. It is customary to define the elements of
the reduced Mueller matrix by mij ≡ Mij /M11 . Similarly defining reduced Stokes vector element
by qt ≡ Qt /It and ut ≡ Ut /It , Eq. (11.3) becomes

Im = It {1+ m12 [cos(2α)qt + sin(2α)ut ]

+ m13 [− sin(2α)qt + cos(2α)ut ]} . (11.4)

Gordon et al. (1997a) gives the general procedure for measuring m12 and m13 in the laboratory.
Meister et al. (2005) gives the details of these measurements for the MODIS sensors. These quanti-
ties, which specify the polarization sensitivity of the instrument, are determined before the instru-
ment is launched; they are thus known. Angle α is determined by the orbit and pointing geometry
of the sensor. It remains to determine the elements of I t .
Following Eq. (1.3), the total TOA polarized radiance can be decomposed as

I t = I R + I a + I Ra + T I g + tI wc + tI w . (11.5)

Here, as before, the Rayleigh (R), aerosol (a), and Rayleigh-aerosol (Ra) radiances are at the TOA;
the glint (g), whitecap (wc), and water-leaving (w) radiances are at the sea surface. The surface
values are transmitted to the TOA via the appropriate direct (T ) and diffuse (t) transmittances.
According to Eq. (11.2), each of these radiances must be known in order to predict what the sensor

68
will measure for a given TOA radiance and, thereby, to determine the correction needed to account
for sensor polarization effects.
The surface-glint and atmospheric polarization contributions to the TOA signal are computed
separately. Sea-surface glint can be highly polarized. This glint contribution to the TOA signal is
computed using a vector radiative transfer code assuming a Rayleigh-scattering atmosphere above
a rough Fresnel-reflecting ocean surface (Gordon and Wang, 1992; Wang, 2002; Meister et al.,
2005). The water-leaving radiance is at most 10% of the TOA total, and the whitecap contribution
is generally even less. These two terms are therefore ignored in the present development. The
effect of aerosols and Rayleigh-aerosol interactions depends on the particle size distribution and
concentration of the aerosols, which are unknown during atmospheric correction. Fortunately,
numerical simulations show that the Rayleigh contribution to the TOA polarization is usually
much greater than the aerosol-related contributions. Therefore, the aerosol contributions are also
ignored and the polarization correction is based on the TOA Rayleigh radiance. The total TOA
Stokes vector is then modeled as the sum of the glint and Rayleigh contributions.
Meister et al. (2005, Eq. 15) define the polarization correction via pc ≡ Im /It . Equations (11.3)
and (11.4) allow this to be written as
1
pc = .
1 − m12 [cos(2α)Qt + sin(2α)Ut ]/Im − m13 [− sin(2α)Qt + cos(2α)Ut ]}/Im

As applied during atmospheric correction, the unknown total TOA radiance components Qt and
Ut are replaced by the corresponding TOA Rayleigh components QR and UR . The Rayleigh com-
ponents are precomputed and tablulated for use during the first step of the atmospheric correction
process, namely the removal of the Rayleigh contribution as described in §6. The end result is that
the actual measured value Im is used along with the Rayleigh radiance for the given atmospheric
conditions and viewing geometry to obtain an estimate of the TOA radiance via

It = Im − m12 [cos(2α)QR + sin(2α)UR ]

− m13 [− sin(2α)QR + cos(2α)UR ] . (11.6)

Gordon et al. (1997a) show that this approximate polarization correction is acceptably accurate
(errors ∆It /It < 0.01 so long as the error has the same sign throughout the spectrum) when m12 is
independent of wavelength and less than about 0.1 in magnitude. If m12 depends on wavelength,
the approximation does not perform well for m12 as small as 0.02. Application of this polarization
correction to MODIS Aqua imagery shows (Meister et al., 2005) that the polarization correction
pc is largest at blue wavelengths (the MODIS 412 nm band), where pc lies in the range of 0.978 to
1.032.

69
References

Z. Ahmad and R. S. Fraser. An iterative radiative transfer code for ocean-atmosphere systems. J.
Atmos. Sci., 39:656–665, 1982.

Z. Ahmad, C. R. McClain, J. R. Herman, B. A. Franz, E. J. Kwaitkowska, W. D. Robinson, E. J.

Bucsela, and M. Tzortziou. Atmospheric correction for NO2 absorption in retrieving water-leafing
reflectances from the SeaWiFS and MODIS measurements. Appl. Optics, 39:6504–6512, 2007.

Z. Ahmad, B. A. Franz, C. R. McClain, E. J. Kwaitkowska, J. Werdell, E. P. Shettle, and B. N.

Holben. New aerosol models for the retrieval of aerosol optical thickness and normalized water-
leaving radiances from SeaWiFS and MODIS sensors over coastal regions and open oceans. Appl.
Optics, 49:5545–5560, 2010a.

Z. Ahmad, B. A. Franz, C. R. McClain, E. J. Kwaitkowska, J. Werdell, E. P. Shettle, and B. N.

Holben. New aerosol models for the retrieval of aerosol optical thickness and normalized water-
leaving radiances from SeaWiFS and MODIS sensors over coastal regions and open oceans:
publisher’s note. Appl. Optics, 50:626, 2010b.

S. W. Bailey, B. A. Franz, and P. J. Werdell. Estimation of near-infrared water-leaving reflectance

for satellite ocean color data processing. Optics Express, 18:7521–7527, 2010.

B. A. Bodhaine, N. B. Wood, E. G. Dutton, and J. R. Slusser. On Rayleigh optical depth calcula-

tions. J. Atmos. Oceanic Technol., 16:1854–1861, 1999.

A. Bricaud, A. Morel, M. Babin, K. Allali, and H. Claustre. Variations of light absorption by

suspended particles with chlorophyll a concentration in oceanic (case 1) waters: Analysis and
implications for bio-optical models. J. Geophys. Res., 103:31033–31044, 1998.

J. W. Campbell. The lognormal distribution as a model for bio-optical variability in the sea. J.
Geophys. Res., 100:13237–13254, 1995.

C. Cox and W. Munk. Measurement of the roughness of the sea surface from photographs of the
sun’s glitter. J. Opt. Soc. Amer., 44:838–850, 1954.

Y. Fan, W. Li, K. J. Voss, C. K. Gatebe, and K. Stamnes. Neural network method to correct
bidirectional effects in water-leaving radiance. Appl. Optics, 55:10–21, 2016.

B Franz. Atmospheric correction discussion: Where we are today and where we might
be in 2.5 years, 2015. Presentation at the PACE Science Team Meeting; Jan 2014.

70
 Available at http://decadal.gsfc.nasa.gov/PACE/PACE_STM_01_2015/Presentations_day_
2/BryanFranz_AC.pdf.

B. Franz, S. W. Bailey, P. J. Werdell, and C. R. McClain. Sensor-independent approach to the

vicarious calibration of satellite ocean color radiometry. Appl. Optics, 46:5068–5082, 2007.

R. Frouin, M. Schwindling, and P. Y. Dechamps. Spectral reflectance of sea foam in the visible
and near infrared: In situ measurements and remote sensing implications. J. Geophys. Res., 101:
14361–14371, 1996.

H. R. Gordon. Remote sensing of ocean color: a methodology for dealing with broad spectral bands
and significant out-of-band response. Appl. Optics, 34:8363–8374, 1995.

H. R. Gordon. Normalized water-leaving radiance: revisiting the influence of surface roughness.

Appl. Optics, 44:241–248, 2005.

H. R. Gordon and B. Franz. Remote sensing of ocean color: Assessment of the water-leaving
radiance bidirrectional effects on the atmospheric diffuse transmittance for SeaWiFS and MODIS
intercomparisons. Rem. Sens. Environ., 112:2677–2685, 2008.

H. R. Gordon and M. Wang. Surface-roughness considerations for atmospheric correction of ocean

color sensors. I: the Rayleigh-scattering component. Appl. Optics, 31:4247–4260, 1992.

H. R. Gordon and M. Wang. Retrieval of water-leaving radiance and aerosol optical thickness over
the oceans with SeaWiFS: a preliminary algorithm. Appl. Optics, 33:443–452, 1994a.

H. R. Gordon and M. Wang. Influence of oceanic whitecaps on atmospheric correction of ocean-color

sensors. Appl. Optics, 33:7754–7763, 1994b.

H. R. Gordon, O. B. Brown, R. H. Evans, J. W. Brown, R. C. Smith, K. S. Baker, and D. K. Clark.

A semianalytical radiance model of ocean color. J. Geophys. Res., 93:10909–10924, 1988.

H. R. Gordon, T. Du, and T. Zhang. Atmospheric correction of ocean color sensors: analysis of
the effects of residual instrument polarization sensitivity. Appl. Optics, 36:69386948, 1997a.

H. R. Gordon, T. Du, and T. Zhang. Remote sensing of ocean color and aerosol properties: resolving
the issue of aerosol absorption. Appl. Optics, 36:8670–8684, 1997b.

J. E. Hansen and L. D. Travis. Light scattering in planetary atmospheres. Space Sci. Rev., 16:
527–610, 1974.

P. Koepke. Effective reflectance of oceanic whitecaps. Appl. Optics, 23:1816–1824, 1984.

Z.-P. Lee, R. A. Arnone, C.-M. Hu, P. J. Werdell, and B. Lubac. Uncertainties of optical parameters
and their propagations in an analytical ocean color inversion algorithm. Appl. Optics, 49:369–381,
2010.

Z. P. Lee, K. Du, K. J. Voss, G. Zibordi, B. Lubac, R. Arnone, and A. Weidemann. An inherent-

optical-property-centered approach to correct the angular effects in water-leaving radiance. Appl.
Optics, 50:3155–3167, 2011.

71
G. Meister, E. J. Kwiatkowska, B. A. Franz, F. S. Patt, G. C. Feldman, and C. R. McClain.
Moderate-resolution imaging spectroradiometer ocean color polarization correction. Appl. Optics,
44:55245535, 2005.

C. D. Mobley. Estimation of the remote-sensing reflectance from above-surface measurements.

Appl. Optics, 38:7442–7455, 1999.

C. D. Mobley. Polarized reflectance and transmittance properties of wind-blown sea surfaces. Appl.
Optics, 54:4828–4849, 2015.

A. Morel and B. Gentili. Diffuse reflectance of oceanic waters: its dependence on Sun angle as
influenced by the molecular scattering contribution. Appl. Optics, 30:4227–4438, 1991.

A. Morel and B. Gentili. Diffuse reflectance of oceanic waters: II: Bidirectional aspects. Appl.
Optics, 32:6864–6879, 1993.

A. Morel and B. Gentili. Diffuse reflectance of oceanic waters: III: Implication of bidirectionality
for the remote-sensing problem. Appl. Optics, 35:4850–4862, 1996.

A. Morel and S. Maritorena. Bio-optical properties of oceanic waters: A reappraisal. J. Geophys.

Res., 106:7163–7180, 2001.

A. Morel, D. Antoine, and B. Gentili. Bidirectional reflectance of oceanic waters: accounting for
Raman emission and varying particle scattering phase function. Appl. Optics, 41:6289–6306,
2002.

F. S. Patt, R. A. Barnes, Jr. R. E. Eplee, B. A. Franz, G. C. Feldman, S. W. Bailey, J. Gales,

J. Werdell, M. Wang, R. Frouin, R. P. Stumpf, R. A. Arnone, Jr. R. W. Gould, P. M. Martinolich,
V. Ransibrahmanakul, J. E. O’Reilly, and J. A. Yoder. Volume 22, Algorithm updates for the
fourth SeaWiFS data reprocessing. In S. B. Hooker and E. R. Firestone, editors, SeaWiFS
Postlaunch Technical Report Series, pages 1–74. NASA, 2003.

E. P. Shettle and R. W. Fenn. Models for the aerosols of the lower atmosphere and the effects of
humidity variation on their optical properties, 1979. Environ. Res. Paper 676, AFGL-TR-79-0214.

J. Stramska and T. Petelski. Observations of oceanic whitecaps in the north polar waters of the
atlantic. J. Geophys. Res, 108, 2003. doi: 10.1029/2002JC001321.

K. J. Voss, A. Morel, and D. Antoine. Detailed validation of the bidirectional effect in various Case
1 waters for application to ocean color imagery. Biogeosciences, 4:781–789, 2007.

M. Wang. The Rayleigh lookup tables for the SeaWiFS data processing: accounting for the effects
of ocean surface roughness. Internat. J. Remote Sensing, 23:26932702, 2002.

M. Wang. A refinement for the Rayleigh radiance computation with variation of the atmospheric
pressure. Internat. J. Remote Sensing, 26:56515653, 2005.

M. Wang and S. W. Bailey. Correction of sun glint contamination of the SeaWiFS ocean and
atmosphere products. Appl. Optics, 40:4790–4798, 2001.

72
P. J. Werdell. Ocean color satellite atmospheric correction, 2015. Lecture at the 2015 Univ. of Maine
Summer Class in Optical Oceanography and Remote Sensing. Video of the lecture is available at
https://www.youtube.com/playlist?list=PLQesc2lgL_DVxIS4AlkFigT_jq_5HIChX.

P. J. Werdell, B. A. Franz, J. T. Lefler, W. D. Robinson, and E. Boss. Retrieving marine inherent

optical properties from satellites using tempreature and salinity-dependent backscattering by
seawater. Optics Express, 21:32611–32622, 2013.

H. Yang and H. R. Gordon. Remote sensing of ocean color: Assessment of the water-leaving radiance
bidirectional effects on the atmospheric diffuse transmittance. Appl. Optics, 36:7887–7897, 1997.

73