Powered by TCPDF ([Link].

org)
 Contents

Preface to the Second Edition iii

Preface to the First Edition v

1 Introduction 1
1.1 Relevance for Climate and Weather . . . . . . . . . . . 1
1.1.1 Solar Radiation . . . . . . . . . . . . . . . . . . 2
1.1.2 Thermal Infrared Radiation . . . . . . . . . . . 3
1.1.3 The Global Heat Engine . . . . . . . . . . . . . 4
1.1.4 Components of the Earth’s Energy Budget . . 6
1.2 Relevance for Remote Sensing . . . . . . . . . . . . . . 7

2 Properties of Radiation 11
2.1 The Nature of Electromagnetic Radiation . . . . . . . 11
2.2 Frequency . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.1 Frequency Decomposition . . . . . . . . . . . . 18
2.2.2 Broadband vs Monochromatic Radiation . . . 19
2.3 Polarization . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4 Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5 A Mathematical Description of EM Waves† . . . . . . 24
2.6 Quantum Properties of Radiation . . . . . . . . . . . . 31
2.7 Flux and Intensity . . . . . . . . . . . . . . . . . . . . . 33
2.7.1 Flux . . . . . . . . . . . . . . . . . . . . . . . . . 33

vii
viii CONTENTS

2.7.2 Intensity . . . . . . . . . . . . . . . . . . . . . . 35
2.7.3 Relationship between Flux and Intensity . . . 46
2.8 Applications . . . . . . . . . . . . . . . . . . . . . . . . 49
2.8.1 Global Insolation . . . . . . . . . . . . . . . . . 49
2.8.2 Regional and Seasonal Distribution of Insolation 50

3 The Electromagnetic Spectrum 55

3.1 Frequency, Wavelength and Wavenumber . . . . . . . 56
3.2 Major Spectral Bands . . . . . . . . . . . . . . . . . . . 57
3.2.1 Gamma Rays and X-Rays . . . . . . . . . . . . 61
3.2.2 Ultraviolet Band . . . . . . . . . . . . . . . . . 61
3.2.3 Visible Band . . . . . . . . . . . . . . . . . . . . 63
3.2.4 Infrared Band . . . . . . . . . . . . . . . . . . . 64
3.2.5 Microwave and Radio Bands . . . . . . . . . . 66
3.3 Solar and Terrestrial Radiation . . . . . . . . . . . . . 68
3.4 Applications . . . . . . . . . . . . . . . . . . . . . . . . 69
3.4.1 UV Radiation and Ozone . . . . . . . . . . . . 69

4 Reflection and Refraction 74

4.1 A Closer Look at N . . . . . . . . . . . . . . . . . . . . 76
4.1.1 The Real Part . . . . . . . . . . . . . . . . . . . 76
4.1.2 The Imaginary Part . . . . . . . . . . . . . . . . 76
4.1.3 The Dielectric Constant† . . . . . . . . . . . . . 79
4.1.4 Optical Properties of Heterogeneous Mixtures† 80
4.2 Refraction and Reflection . . . . . . . . . . . . . . . . . 82
4.2.1 Angle of Reflection . . . . . . . . . . . . . . . . 82
4.2.2 Angle of Refraction . . . . . . . . . . . . . . . . 84
4.2.3 Reflectivity . . . . . . . . . . . . . . . . . . . . 86
4.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . 90
4.3.1 Rainbows and Halos . . . . . . . . . . . . . . . 90

5 Radiative Properties of Natural Surfaces 96

5.1 Natural Surfaces Idealized as Planar Boundaries . . . 97
5.2 Absorptivity and Reflectivity . . . . . . . . . . . . . . 98
5.2.1 Examples of Reflectivity Spectra . . . . . . . . 99
5.2.2 The Graybody Approximation . . . . . . . . . 100
5.3 Angular Distribution of Reflected Radiation . . . . . . 102
5.3.1 Specular and Lambertian Reflection . . . . . . 102
5.3.2 Reflection in the General Case† . . . . . . . . . 105
 CONTENTS ix

5.4 Applications . . . . . . . . . . . . . . . . . . . . . . . . 107

5.4.1 Solar Heating of Surfaces . . . . . . . . . . . . 107
5.4.2 Satellite Imaging at Visible and Near-IR
Wavelengths . . . . . . . . . . . . . . . . . . . . 109

6 Thermal Emission 113

6.1 Blackbody Radiation . . . . . . . . . . . . . . . . . . . 115
6.1.1 Planck’s Function . . . . . . . . . . . . . . . . . 117
6.1.2 Wien’s Displacement Law . . . . . . . . . . . . 120
6.1.3 Stefan-Boltzmann Law . . . . . . . . . . . . . . 122
6.1.4 Rayleigh-Jeans Approximation . . . . . . . . . 123
6.2 Emissivity . . . . . . . . . . . . . . . . . . . . . . . . . 123
6.2.1 Monochromatic Emissivity . . . . . . . . . . . 124
6.2.2 Graybody Emissivity . . . . . . . . . . . . . . . 124
6.2.3 Kirchhoff’s Law . . . . . . . . . . . . . . . . . . 125
6.2.4 Brightness Temperature . . . . . . . . . . . . . 127
6.3 When Does Thermal Emission Matter? . . . . . . . . . 130
6.4 Applications . . . . . . . . . . . . . . . . . . . . . . . . 132
6.4.1 Radiative Equilibrium in a Vacuum . . . . . . 133
6.4.2 Top-of-the-Atmosphere Global Radiation Bal-
ance . . . . . . . . . . . . . . . . . . . . . . . . 136
6.4.3 Simple Radiative Models of the Atmosphere . 139
6.4.4 Nighttime Radiative Cooling . . . . . . . . . . 144
6.4.5 Radiative Cooling at Cloud Top . . . . . . . . 146
6.4.6 IR Imaging from Space . . . . . . . . . . . . . . 148
6.4.7 Microwave Imaging from Space . . . . . . . . 151

7 Atmospheric Transmission 155

7.1 Extinction, Scattering and Absorption Coefficients . . 159
7.2 Extinction Over a Finite Path . . . . . . . . . . . . . . 160
7.2.1 Fundamental Relationships . . . . . . . . . . . 160
7.2.2 Mass Extinction Coefficient . . . . . . . . . . . 163
7.2.3 Extinction Cross-Section . . . . . . . . . . . . . 166
7.2.4 Generalization to Scattering and Absorption . 167
7.2.5 Generalization to Arbitrary Mixtures of Com-
ponents . . . . . . . . . . . . . . . . . . . . . . 168
7.3 Plane Parallel Approximation . . . . . . . . . . . . . . 169
7.3.1 Definition . . . . . . . . . . . . . . . . . . . . . 171
7.3.2 Optical Depth as Vertical Coordinate . . . . . . 173
x CONTENTS

7.4 Applications . . . . . . . . . . . . . . . . . . . . . . . . 174

7.4.1 The Transmission Spectrum of the Atmosphere 174
7.4.2 Measuring Solar Intensity from the Ground . . 185
7.4.3 Transmittance in an Exponential Atmosphere 187
7.4.4 Optical Thickness and Transmittance of a
Cloud Layer . . . . . . . . . . . . . . . . . . . . 194

8 Atmospheric Emission 204

8.1 Schwarzschild’s Equation . . . . . . . . . . . . . . . . 205
8.2 Radiative Transfer in a Plane Parallel Atmosphere . . 210
8.2.1 The Emissivity of the Atmosphere . . . . . . . 211
8.2.2 Monochromatic Flux † . . . . . . . . . . . . . . 212
8.2.3 Surface Contributions to Upward Intensity . . 215
8.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . 217
8.3.1 The Spectrum of Atmospheric Emission . . . . 219
8.3.2 Satellite Retrieval of Temperature Profiles . . . 228
8.3.3 Water Vapor Imagery . . . . . . . . . . . . . . . 233

9 Absorption by Atmospheric Gases† 236

9.1 Basis for Molecular Absorption/Emission . . . . . . . 238
9.2 Absorption/Emission Lines . . . . . . . . . . . . . . . 240
9.2.1 Rotational Transitions . . . . . . . . . . . . . . 243
9.2.2 Vibrational Transitions . . . . . . . . . . . . . . 251
9.2.3 Electronic Transitions . . . . . . . . . . . . . . 256
9.2.4 Combined Energy Transitions and Associated
Spectra . . . . . . . . . . . . . . . . . . . . . . . 258
9.3 Line Shapes . . . . . . . . . . . . . . . . . . . . . . . . 258
9.3.1 Generic Description of Lines . . . . . . . . . . 260
9.3.2 Doppler Broadening . . . . . . . . . . . . . . . 261
9.3.3 Pressure Broadening . . . . . . . . . . . . . . . 263
9.3.4 Comparing Doppler and Pressure Broadening 266
9.4 Continuum Absorption . . . . . . . . . . . . . . . . . . 267
9.4.1 Photoionization . . . . . . . . . . . . . . . . . . 268
9.4.2 Photodissociation . . . . . . . . . . . . . . . . . 268
9.4.3 Continuum Absorption by Water Vapor . . . . 269
9.5 Applications . . . . . . . . . . . . . . . . . . . . . . . . 270
9.5.1 Atmospheric Absorbers in the IR Band . . . . 270
 CONTENTS xi

10 Broadband Fluxes and Heating Rates† 280

10.1 Line-by-line Calculations . . . . . . . . . . . . . . . . . 281
10.2 Band Transmission Models . . . . . . . . . . . . . . . 286
10.2.1 Absorption by an Isolated Line . . . . . . . . . 288
10.2.2 Defining a Band Model . . . . . . . . . . . . . 293
10.2.3 The Elsasser Band Model . . . . . . . . . . . . 294
10.2.4 The Random/Malkmus Band Model . . . . . . 297
10.2.5 The HCG Approximation . . . . . . . . . . . . 298
10.3 The k-Distribution Method . . . . . . . . . . . . . . . . 299
10.3.1 Homogeneous Path . . . . . . . . . . . . . . . . 300
10.3.2 Inhomogeneous Path: Correlated-k . . . . . . . 303
10.4 Applications . . . . . . . . . . . . . . . . . . . . . . . . 306
10.4.1 Fluxes and Radiative Heating/Cooling . . . . 306

11 RTE With Scattering 320

11.1 When Does Scattering Matter? . . . . . . . . . . . . . 321
11.2 Radiative Transfer Equation with Scattering . . . . . . 322
11.2.1 Differential Form . . . . . . . . . . . . . . . . . 322
11.2.2 Polarized Scattering† . . . . . . . . . . . . . . . 324
11.2.3 Plane Parallel Atmosphere . . . . . . . . . . . 324
11.3 The Scattering Phase Function . . . . . . . . . . . . . . 326
11.3.1 Isotropic Scattering . . . . . . . . . . . . . . . . 327
11.3.2 The Asymmetry Parameter . . . . . . . . . . . 329
11.3.3 The Henyey-Greenstein Phase Function . . . . 330
11.4 Single vs. Multiple Scattering . . . . . . . . . . . . . . 332
11.5 Applications . . . . . . . . . . . . . . . . . . . . . . . . 336
11.5.1 Intensity of Skylight . . . . . . . . . . . . . . . 336
11.5.2 Horizontal Visibility . . . . . . . . . . . . . . . 338

12 Scattering and Absorption By Particles 343

12.1 Atmospheric Particles . . . . . . . . . . . . . . . . . . 344
12.1.1 Overview . . . . . . . . . . . . . . . . . . . . . 344
12.1.2 Relevant Properties . . . . . . . . . . . . . . . . 345
12.2 Scattering by Small Particles . . . . . . . . . . . . . . . 347
12.2.1 Dipole Radiation . . . . . . . . . . . . . . . . . 347
12.2.2 The Rayleigh Phase Function . . . . . . . . . . 351
12.2.3 Polarization . . . . . . . . . . . . . . . . . . . . 353
12.2.4 Scattering and Absorption Efficiencies . . . . . 354
12.3 Scattering by Spheres — Mie Theory . . . . . . . . . . 358
xii CONTENTS

12.3.1 Extinction Efficiency for Nonabsorbing Sphere 359

12.3.2 Extinction and Scattering by Absorbing Spheres363
12.3.3 Scattering Phase Function . . . . . . . . . . . . 365
12.4 Distributions of Particles . . . . . . . . . . . . . . . . . 372
12.5 Applications . . . . . . . . . . . . . . . . . . . . . . . . 373
12.5.1 The Scattering Properties of Clouds . . . . . . 373
12.5.2 Radar Observations of Precipitation . . . . . . 376
12.5.3 Microwave Remote Sensing and Clouds . . . . 382

13 Radiative Transfer with Multiple Scattering 387

13.1 Visualizing Multiple Scattering . . . . . . . . . . . . . 389
13.2 The Two-Stream Method . . . . . . . . . . . . . . . . . 392
13.2.1 Azimuthally Averaged RTE . . . . . . . . . . . 392
13.2.2 The Two-Stream Approximation . . . . . . . . 393
13.2.3 Solution . . . . . . . . . . . . . . . . . . . . . . 398
13.3 Semi-Infinite Cloud . . . . . . . . . . . . . . . . . . . . 400
13.3.1 Albedo . . . . . . . . . . . . . . . . . . . . . . . 401
13.3.2 Flux and Heating Rate Profile . . . . . . . . . . 404
13.4 Nonabsorbing Cloud . . . . . . . . . . . . . . . . . . . 406
13.5 General Case . . . . . . . . . . . . . . . . . . . . . . . . 408
13.5.1 Albedo, Transmittance, and Absorptance . . . 409
13.5.2 Direct and Diffuse Transmittance . . . . . . . . 412
13.5.3 Semi-Infinite Cloud as Approximation . . . . . 414
13.6 Similarity Transformations† . . . . . . . . . . . . . . . 416
13.7 Clouds Over Non-Black Surfaces . . . . . . . . . . . . 418
13.8 Multiple Cloud Layers . . . . . . . . . . . . . . . . . . 423
13.9 Accurate solution methods† . . . . . . . . . . . . . . . 424

A Representing the Phase Function 427

A.1 Legendre Polynomial Expansion . . . . . . . . . . . . 427
A.2 δ-Scaling of the Phase Function . . . . . . . . . . . . . 430

B Symbols Used 438

C Further Reading 445

D Useful Physical and Astronomical Constants 447

Ordering Information 459

 CHAPTER 1

Introduction

Electromagnetic radiation is an omnipresent part of our lives. Take

a look around you. The fact that you can see, and interpret in as-
tonishing detail, both your immediate and distant surroundings is
made possible by radiation arriving at your eyes from every pos-
sible direction. Because the properties of this radiation, including
both intensity (brightness) and spectral characteristics (color), are
strongly influenced by its interactions with matter, you are able to
instantly distinguish objects, faces, textures, material compositions,
and many other details, some as small (in relative terms) as the pe-
riod at the end of this sentence or the shape of a bird perched on a
distant telephone wire.

1.1 Relevance for Climate and Weather

By now, if you are a student of atmospheric science, you are proba-
bly already aware of the distinction between adiabatic and diabatic
processes in the atmosphere. It is common — because it is so conve-
nient — to idealize many dynamic and thermodynamic processes as
adiabatic. That is to say, it is often assumed that there is no signif-
icant energy exchange between the air parcel under consideration
and its surrounding environment. In fact, this is not a bad assump-

1
2 1. Introduction

tion for processes taking place in the free atmosphere on time scales
of a day or less. Over longer time periods, and even over short time
periods (hours or less) in close proximity to the earth’s surface, di-
abatic processes cannot be ignored. One such process is thermal
conduction, which is quite slow in air except in the presence of very
large temperature gradients — i.e., a degree per centimeter or more.
Another diabatic process is latent heating and cooling in connection
with the phase changes of water – melting, freezing, evaporation,
condensation, etc. Although latent heating is a very important fac-
tor in cloud and precipitation formation, it operates only intermit-
tently and, conveniently, it can almost always be understood and
computed entirely in terms of local thermodynamic and microphys-
ical processes.
Atmospheric radiation, the subject of this book, is the sole en-
ergy exchange process that operates both continuously throughout
the atmosphere and over long distances. The fact that net heating
or cooling due to radiation depends strongly on nonlocal processes
greatly complicates the problem of computing this diabatic heating
term in weather prediction and climate models. It is also one of sev-
eral reasons why an entire textbook can be devoted to the subject
and still only scratch the surface in terms of both the implications
and applications (e.g., satellite remote sensing) of atmospheric ra-
diative transfer processes.

1.1.1 Solar Radiation

When you go outside and look up at the sky, you are directly ob-
serving one form of atmospheric radiation. In fact, all of the natural
light you can see during the day — and a great deal that you can’t
— originates from the sun, and we refer to this as solar radiation.
It is of course the absorption of solar radiation by the atmosphere
and the earth’s surface that is ultimately responsible for maintain-
ing the atmosphere’s overall temperature structure, including the
horizontal gradients that drive atmospheric circulations. If the sun
were switched off, the world would quickly chill to the point where
it would sustain neither life nor even wind and weather.
Not all of the radiation arriving from the sun is visible to the eye.
For example, a significant fraction of solar radiation arrives in the
 Relevance for Climate and Weather 3

form of infrared (IR) radiation. Although the cloud-free atmosphere

is rather transparent to visible radiation, it is somewhat less so to
solar IR radiation. Furthermore, clouds and snow cover, both of
which are highly reflective to visible radiation and therefore appear
bright white to the eye, would appear rather dark to an eye capable
of perceiving only IR radiation.
Ultraviolet (UV) radiation, another invisible component of solar
radiation, is responsible for the burn you get when you spend too
much time exposed to the sun. It also plays a key role in the forma-
tion of smog in industrialized regions of the world. In its harsher
forms, UV radiation can injure or kill living organisms. Fortunately,
oxygen and ozone in the upper levels of the atmosphere attenuate
the worst components of solar UV radiation, protecting life at the
surface from its most damaging effects. Ironically, ozone itself is a
product of the interaction of UV radiation with ordinary oxygen.
Indeed solar UV radiation plays a decisive role in almost all aspects
of the chemistry of the atmosphere.

1.1.2 Thermal Infrared Radiation

There are other forms of invisible EM radiation that continuously
flood your surroundings, both day and night, but do not originate
directly from the sun. The most important of these is also classified
as infrared, but is often called thermal IR radiation to distinguish it
from the solar variety. You personally experience the effects of long-
wave IR radiation whenever you feel warmth radiating on your face
from a hot oven across the kitchen even though the temperature of
the intervening room air is moderate. Even when we are not directly
conscious of it in this way, every object and surface in our surround-
ings, including the air itself, is constantly exchanging energy in the
form of thermal IR over both short and long distances.
Thermal IR plays a major role in the redistribution of heat en-
ergy within the atmosphere and between the surface and the atmo-
sphere. It is also the mechanism by which the earth-atmosphere sys-
tem invisibly sheds excess heat back to space. Without this mecha-
nism, the earth and atmosphere would continue to heat up indefi-
nitely as they absorbed more and more solar radiation from the sun.
It is also the downward emission of thermal IR by clouds and water
4 1. Introduction

Annual Average Radiative Flux

400
Outgoing Longwave Radiation
350 Absorbed Solar Radiation
Net Radiation
300

250

200
Flux (W m-2)

150

100

50
Surplus
0
Deficit Deficit
-50

-100

-150
-90 -80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90
Latitude

Fig. 1.1: Annual average radiation budget as a function of latitude.

vapor in the lower atmosphere that explains why nighttime tem-

peratures do not fall nearly as sharply on humid or overcast nights
than they do on clear, dry nights.

1.1.3 The Global Heat Engine

In broadest terms, the importance of radiation for weather and cli-
mate can perhaps best be appreciated by examining Fig. 1.1. The top
two curves represent the long-term zonally averaged distribution of
absorbed solar radiation and outgoing longwave radiation, as seen
at the top of the atmosphere. The bottom curve depicts the differ-
ence between the input of energy from the sun and the loss of energy
to space. In the tropical belt, more energy is received from the sun
than is lost to space in the form of longwave radiation. If there were
no compensating processes at work, the tropical belt would con-
tinue to heat up to well above its current temperature, and the poles
would plunge to even lower temperatures than exist there now. The
net effect of this radiative imbalance is the creation of a meridional
 Relevance for Climate and Weather 5

temperature gradient.
Horizontal temperature gradients in a fluid such as our atmo-
sphere inevitably create pressure gradients that in turn initiate cir-
culations. These fluid motions serve to reduce the temperature gra-
dient by transporting heat from warm to cold regions. In fact, they
intensify until the net horizontal heat transport exactly offsets (on
average) the imbalance in radiative heating and cooling. All circula-
tions observed in the ocean and atmosphere — from ocean currents to the
Hadley circulation to extratropical cyclones to hurricanes and tornadoes
— can be viewed as mere cogs in a huge and complex machine serving this
higher purpose.
In fact, if you have previously taken a course in thermodynam-
ics, you might recall that a heat engine is defined as a system that
converts a temperature gradient into mechanical work. It does this
by taking in heat energy at a high temperature and discharging the
same amount of heat at a cooler temperature. If you take a second
look at Fig. 1.1 you will agree that that is exactly what is occurring
in the earth-atmosphere system: a net intake of heat energy in the
warm tropics and a net discharge of heat from the cool polar re-
gions.

Problem 1.1: Referring to Fig. 1.1:

(a) Estimate the latitude LC where the net radiation crosses over
from positive to negative in the Northern Hemisphere. Through that
point, use a straightedge to sketch a line that best fits the trend in net
radiation toward the North Pole. Find the value Qnp of the net radi-
ation where your straight line intercepts the right axis. Use the two
pieces of information to find the linear equation that approximately
describes the net radiation (in W m−2 ) as a function of latitude L
between 10◦ N and the North Pole.
(b) The latitude LC is where the northward transport of heat by
the atmosphere and ocean is at a maximum in that hemisphere. Ex-
plain why.
(c) The rate of meridional heat transport at that latitude equals
the total radiation deficit integrated over the surface area of the earth
from LC to the North Pole. Compute this value using the equation
you obtained in part (a), keeping in mind that a unit change in lati-
tude is not proportional to a unit of surface area.
(d) Convert your result to a power per unit distance (east or west
along the line of constant latitude at LC ), using units of kilowatts per
kilometer. The mean radius of the earth R E = 6373 km.
6 1. Introduction

Earth's Energy Budget

Reflected by Reflected Reflected from
atmosphere by clouds earth's surface
6% 20% 4% 64% 6%
Radiated to space
from clouds and
Incoming atmosphere
solar energy
100% Absorbed by
atmosphere 16% Radiated
directly
to space
from earth
Absorbed by
clouds 3% Radiation
absorbed by
Conduction and atmosphere
rising air 7% 15%

Carried to clouds
Absorbed by land and atmosphere by
and oceans 51% latent heat in
water vapor 23%

Fig. 1.2: Components of the globally averaged energy budget expressed as per-
centages of the incoming solar radiation. Redrafted from an illustration by J.T. Kiehl
and Kevin E. Trensberth.

1.1.4 Components of the Earth’s Energy Budget

Just as we looked at how incoming and outgoing radiation is par-

titioned, on average, between high and low latitudes, we can also
break the energy budget down in terms of average exchanges within
the earth-atmosphere system. Fig. 1.2 schematically depicts the fate
of the solar radiation that is intercepted by the earth. Approximately
30% is immediately reflected back to space by the atmosphere,
clouds, and the surface. The remaining 70% is absorbed. Almost
three-quarters of that radiation is absorbed by the land and ocean
surfaces; the remainder by the atmosphere (including clouds).
Of the energy absorbed by the surface, a mere tenth is emitted
right back to space and lost forever. The remainder is transferred
to the atmosphere, either by conduction of sensible heat, evapora-
tion of water (which carries latent heat of vaporization), or via the
emission and reabsorption of radiation.
Ultimately, the atmosphere itself is responsible for radiating
about nine-tenths of the total absorbed solar energy back to space.
 Relevance for Remote Sensing 7

The loss of radiation to space is the major cooling term in the atmo-
spheric energy budget, offsetting the direct or indirect heating from
the ground and (to a lesser extent) the direct absorption of solar
radiation. The magnitude and vertical distribution of this cooling
at any given location and time depends on the profile of tempera-
ture, humidity and cloud cover, among other variables. When it oc-
curs mainly at high altitudes, it helps to destabilize the atmosphere,
making verical convection more likely. When it occurs near ground
level, the effect is to stabilize the atmosphere and suppress convec-
tion.
By now, you should be persuaded that it is impossible to under-
stand, let alone accurately predict, the medium- to long-term evolu-
tion of weather or climate without accounting for radiative transfer
processes in the atmosphere. Even on very short time scales (hours),
radiative absorption and emission at the earth’s surface and at the
bases and tops of cloud layers have noticeable effects on everyday
weather. Key examples are the initiation of convection by solar heat-
ing, and the formation and subsequent evaporation of frost, dew,
and ground fog. In general, the longer the time scale, the more de-
cisive radiative processes become at all levels in the atmosphere.

Problem 1.2: Averaged over the globe, about 342 W m− 2 of solar

radiation is incident on the top of the earth’s atmosphere. Using the
information in Fig. 1.2, compute the average rate at which the atmo-
sphere (including clouds) would be heated by direct absorption of
solar radiation if there were no other processes at work. Hint: You
will need the values of g, p0 , and c p from the back of the book. Com-
bine these with the rate of absorbed solar radiation to find a value
that can be expressed in degrees per day.

1.2 Relevance for Remote Sensing

I have just outlined the role that atmospheric radiation plays in re-
distributing energy over potentially long distances, both within the
atmosphere and between the earth-atmosphere system and outer
space. But that is not the whole story. EM radiation carries not
8 1. Introduction

only energy but also a wealth of information about the environment

within which it originated and through which it subsequently prop-
agated. Since the early 1960s, virtually all areas of the atmospheric
sciences have been revolutionized by the development and appli-
cation of remote sensing techniques — that is, measurements of at-
mospheric properties and processes at a distance, using radiation
sensors placed in space, on aircraft, and/or on the earth’s surface.
As we shall see, the interactions of various forms of EM radia-
tion with the environment are extremely rich and complex. Conse-
quently, there are few important atmospheric variables that cannot
be directly or indirectly estimated from the vantage point of a satel-
lite in orbit if one is clever enough in the design of both the instru-
ment and the analytical techniques. Today, there are large parts of
the globe — especially the oceans, polar regions, and sparsely popu-
lated land areas — where meteorologists depend almost entirely on
satellite observations for up-to-date information about temperature
and humidity structure, wind, cloud cover, precipitation, etc.
The three images in Fig. 1.3 give just a hint of the variety of in-
formation contained in satellite observations of the atmosphere at
different wavelengths. Panel (a) is a snapshot of the Eastern Pacific
at a wavelength in the visible part of the spectrum. This is essen-
tially the view of Earth you would see with the naked eye if you
were on board a spacecraft, except that it’s in black and white. The
source of the illumination is of course the sun, and brightest areas
in the image correspond to highly reflective features, such as clouds
and snow. The ocean, which is not very reflective of sunlight, ap-
pears very dark. Land areas fall in between. Images of this type
are primarily useful for observing the extent and evolution of cloud
features associated with storms and other atmospheric circulations.
Also, by observing the cloud coverage over time, it is possible to es-
timate what fraction of the sun’s energy, on average, is absorbed by
the earth and atmosphere.
Panel (b) shows the same scene but the image was taken at an
infrared wavelength for which the cloud-free atmosphere is very
transparent. The shades of gray in this image give an indication
of the temperatures of the various surface and cloud features visible
from space. Light shades correspond here to cold temperatures, and
dark shades correspond to warm temperatures. Note that many of
 Relevance for Remote Sensing 9

a) Visible, 0.65 µm b) IR window, 10.7 µm

c) IR water vapor band, 6.7 µm

Fig. 1.3: Geostationary satellite images of the Eastern Pacific taken at the same time
but at three different wavelengths.

the bright clouds seen in panel (a) appear dark in panel (b), indi-
cating that their tops occur at a low, warm altitude. Other clouds,
such as those in the upper center of the image, appear bright, indi-
cating that they have high, cold tops. Rain clouds are usually deep
and therefore have cold tops, so infrared images can often help dis-
tinguish deep precipitating clouds from shallow nonprecipitating
clouds. The California lowlands appear as the darkest shade, so we
can conclude that they have the warmest temperatures in this im-
age. The Sierra Nevada mountain range, however, is visible as a
light streak parallel to the coast, because the surface temperatures
at high elevations are much cooler than those in the lowlands.
Panel (c) is very similar to panel (b) except that it was taken at an
10 1. Introduction

infrared wavelength for which water vapor is nearly opaque. The

temperature patterns we see in this case correspond to the tempera-
ture of the atmosphere at some level above the majority of the atmo-
spheric water vapor. Where the image shade is bright, the temper-
ature is cold, and we can infer that the atmosphere is humid over a
fairly deep layer; where it is dark, the atmosphere must be quite dry,
because we are observing radiation emitted from the lower, warmer
levels of the atmosphere.
This book will not delve too deeply into the details of atmo-
spheric radiation as it relates specifically to remote sensing tech-
niques and applications — for that, I recommend the book by G.
Stephens (hereafter S941 ). However, once you have mastered the
basic radiative transfer facts and principles covered in the following
chapters, you will be well prepared for further study in this area.

1 Abbreviations like S94 will be used as a shorthand for the textbooks recom-

mended for further reading. See Appendix C for the full bibliographic informa-
tion.
 CHAPTER 2

Properties of Radiation

The first order of business in a book about atmospheric radiation is

to clarify what radiation is, how it behaves at the most fundamental
physical level, what conventions are used to classify it according to
wavelength and other properties, and how we define the character-
istics (e.g. intensity) that appear in quantitative descriptions of ra-
diation and its interactions with the atmosphere. We will start from
absolute basics, which requires us to at least touch on some topics in
classical physics. Our forays into such matters will be as brief and
descriptive as the subsequent material allows, in keeping with the
title of this book. Here, and throughout the book, students inter-
ested in a more comprehensive treatment should consult the more
specialized textbooks and other sources cited at the end of this book.

2.1 The Nature of Electromagnetic Radiation

Electric and Magnetic fields
Everyone has experienced the effects of electric and magnetic fields.
A nylon shirt pulled from a dryer may be found to have several
socks clinging firmly to it. A magnet may be used to affix a note
or photograph to a refrigerator door. In the first case, an electric

11
12 2. Properties of Radiation

field induced by excess charge on one item of clothing exerts an

attractive force on the excess opposite charge found on the others.
In the second, a magnetic field exerts an attractive force on the iron
atoms in the door.

Both magnetic and electric fields are detectable at some distance

from their source. A plastic balloon that was rubbed on a sweater
may attract your hair from a foot or so away. A refrigerator magnet
may deflect the needle of a compass from a meter or more away.
Taking the magnet away, the needle returns to a position aligned
with the magnetic field of the earth’s iron core, thousands of kilo-
meters beneath your feet.

The basic laws governing static electric and magnetic fields are
well known and probably already at least somewhat familiar to you
from an earlier physics course. According to Coulomb’s Law, the
electric field at any point is determined entirely by the distribution
of electric charge in the space surrounding that point. According to
Faraday’s Law, magnetic fields are determined by the distribution
of electric current (moving electric charge) in the neighborhood.

The latter law is what makes electromagnets possible, without

which electric motors and most audio speakers would not exist in
their current form. A related law tells us that a changing magnetic
field induces an electric field that can drive a current. Thus, an
electric motor that is caused to rotate by an external torque (e.g.,
a steam-driven turbine) can reverse roles and become a generator
instead.

Thus, although static magnetic and static electric fields — as il-

lustrated by the examples given at the beginning — may appear to
have little or nothing to do with each other, they are in fact inti-
mately connected: a changing electric field induces a magnetic field,
and a changing magnetic field induces an electric field. Quantita-
tively, this interplay of electric and magnetic fields is embodied both
completely and remarkably succinctly in Maxwell’s Equations (see
Section 2.5). However, one does not need to look at equations them-
selves to appreciate a few of the implications of this interplay.
 The Nature of Electromagnetic Radiation 13

Electromagnetic Waves
Imagine a refrigerator magnet resting on the kitchen table. It creates
a magnetic field that extends essentially indefinitely (though with
ever-weaker strength at increasing distances from the source) in all
directions. Pick up the magnet and stick it back on the refrigerator
door. Because the magnet has moved (and probably changed its ori-
entation as well), its induced magnetic field has also changed. But
a changing magnetic field creates an electric field, which persists as
long as the magnetic field continues to change. Once the magnet is
stationary again, the magnetic field stops changing and the electric
field must disappear. But wait: this means the electric field is un-
dergoing a change, which leads to the reappearance of a magnetic
field!
You can see where this is leading. A change in either an electric
or magnetic field, however brief, leads to a disturbance that is self-
perpetuating. Less obvious to the casual observer is the fact that this
electromagnetic disturbance propagates away from the source at a
finite speed, just as ripples propagate outward from the point where
a pebble strikes the surface of a pond (in the latter case, the interplay
is not between magnetic and electric fields but rather between the
kinetic and potential energy of the water’s surface).
In the case of both electromagnetic waves and ripples on a pond,
the disturbances carry energy. In the absence of viscosity, the pond
ripples will transport the original energy outward, with no net loss,
until they encounter something that can convert some or all of that
energy to another form, such as heat or/and sound — e.g., through
the breaking of the waves on the muddy bank.
Likewise, in a perfect vacuum, where there is no opportunity to
convert the energy carried by electromagnetic waves into another
form, such as heat, kinetic, or chemical energy (all of which can only
exist in association with matter), the waves propagate indefinitely
without net loss, though distributed over an ever-larger volume of
space. Furthermore, it has been observed that, unlike pond ripples,
electromagnetic waves always travel in a vacuum at an absolutely
constant speed. This speed of light is approximately 3.0 × 108 m s−1 .
By convention, it is represented by the symbol c.
The direction that an EM wave travels in a vacuum is always
perpendicular to the wave crest, again just like pond ripples. And
14 2. Properties of Radiation

because c is a constant in a vacuum, the wave can only propagate

directly away from the source. Any change of direction would im-
ply a slowing down or speeding up of part of a wave crest at some
point during its travels. This can happen in the presence of matter,
but remember we’re still talking about a vacuum here.
Sometimes it is helpful to visualize the propagation of waves in
terms of rays. A ray is an imaginary line that always crosses wave
fronts at right angles. Thus at any point on a ray, the direction of
propagation of the wave is along that ray. In the case of pond rip-
ples, all rays would be straight lines originating at the point where
the pebble strikes the surface. If you like, you can think of each ray
as carrying a unit of energy. Thus, the density of their intersections
with a given wave front is a measure of the energy content of the
wave at that location. If a given wave front propagates outward
from a source without losses, the total number of rays remains con-
stant (implying conservation of energy), but the density of intersec-
tions along the wave front decreases with distance from the source.
This is consistent with the spreading of the wave’s energy over a
larger area and thus of its weakening as it gets further and further
away.
Again, the bending of rays can only occur in the presence of local
changes in wave propagation speed. This phenomenon, known as
refraction, is the subject of Section 4.2.
Imagine throwing two pebbles into a pond at the same time.
Each one produces its own set of ripples that propagate outward
from the source. Sooner or later, the ripples from one pebble en-
counter those from the other. What happens? Do they bounce off
of each other? Do they annihilate each other in a flash of heat and
gamma rays? Of course not; each passes through the other as if it
didn’t exist. At each point on the water surface, the height perturba-
tions associated with each set of waves simply add in a linear fash-
ion. Where two crests intersect, the water surface is (temporarily)
raised to about twice the height of either crest individually. Where
a crest from one wave intersects the trough from another, the two
may partially or completely cancel, leaving the water (temporarily)
at its original level. This effect is called interference — constructive
interference in the first case, destructive in the second. It must be
emphasized however that nothing is created or destroyed; the ef-
 The Nature of Electromagnetic Radiation 15

Magnetic field vector

Direction of propagation

Electric field vector

Fig. 2.1: Schematic depiction of the interplay between the electric and magnetic
fields in an electromagnetic wave.

fect is purely local at the points where the waves overlap and has no
influence on the subsequent propagation of the individual waves.
Exactly the same principle of linear superposition applies to EM
waves. If you shine a flashlight on a wall in a dark room and then
turn on the overhead light, the latter adds its illumination to that of
the flashlight but otherwise has no influence on the propagation of
the flashlight beam.
There is one further analogy to pond ripples that we can use to
illustrate an important property of EM waves. Unlike sound waves
in air, which entail oscillations of air molecules parallel to the direc-
tion of propagation, water waves arise from vertical displacements
of the water surface; i.e., perpendicular, or normal, to the direction of
propagation. Likewise, EM waves are associated with an oscillating
electric field vector that is normal to the direction of propagation
(Fig. 2.1). Both are therefore transverse waves, unlike sound waves,
which are longitudinal.
Now we finally come to a couple of points where the analogy to
pond ripples starts to break down:
• Unlike sound waves, water waves, and all other everyday
kinds of waves, EM waves require no material medium in
which to propagate. You can’t have pond ripples without
a pond; you can’t have sound without air (or some other
medium). But EM waves are quite at home, and indeed easi-
est to understand, in a perfect vacuum. It is when matter, such
as the atmosphere and its various resident particles, gets into
16 2. Properties of Radiation

the picture that the propagation of EM waves become consid-

erably more complex and interesting.

• Whereas pond ripples are confined to the water surface and

therefore propagate in only two dimensions, EM waves prop-
agate in three dimensional space, like sound waves (imagine
if you could only hear your stereo when you put your ear on
the floor!).

• Whereas the transverse oscillations of water particles in pond

ripples are constrained to be vertical, there is no similar con-
straint on the electric field oscillations in EM waves. The ori-
entation of the electric field vector may lie in any direction, as
long as it is in the plane normal to the direction of propaga-
tion. Occasionally it is necessary to pay attention to this orien-
tation, in which case we refer to the polarization of the wave as
vertical, horizontal, or some other direction. A more detailed
discussion of polarization will be taken up in Section 2.3

2.2 Frequency

Up until now, we have imagined an arbitrary EM disturbance and

given no thought to its detailed dependence in time. In principle,
we could assume any kind of EM disturbance we like — a light-
ning discharge, a refrigerator magnet dropping to the kitchen floor,
the radiation emitted by a radio tower, or a supernova explosion in
deep space. In each case, Maxwell’s equations would describe the
propagation of the resulting EM disturbance equally well.
Let’s consider a special case, however. Imagine that we take our
magnet and place it on a steadily rotating turntable. The fluctua-
tions in the magnetic field (and in the associated electric field) are
now periodic. The frequency of the fluctuations measured at a dis-
tance by a stationary detector will be the same as the frequency of
rotation ν of the turntable. But recall that the periodic disturbance
propagates outward not instantaneously but at the fixed speed of
light c. The distance λ that the fluctuation propagates during one
 Frequency 17

cycle of the turntable is called the wavelength and is given by

c
λ= . (2.1)
ν

In the above thought experiment, ν is extremely low (order 1 sec−1 )

and the wavelength is therefore extremely large (order 105 km).
In nature, electromagnetic waves can exist with an enormous
range of frequencies, from a few cycles per second or even far less
to more than 1026 cyles per second in the case of extremely ener-
getic gamma waves produced by nuclear reactions. According to
(2.1), EM wavelengths can thus range from hundreds of thousands
of kilometers or more to less than the diameter of an atomic nucleus.
Because it is a common point of confusion, it bears emphasizing
that the wavelength is not a measure of how far an EM wave can
propagate. In a vacuum, that distance is always infinite, regardless
of wavelength. In a medium such as water or air, wavelength does
matter, but in a rather indirect and highly complex way.

Problem 2.1:
(a) Visible light has a wavelength of approximately 0.5 µm. What
is its frequency in Hz?
(b) Weather radars typically transmit EM radiation with a fre-
quency of approximately 3 GHz (GHz = “Gigahertz” = 109 Hz).
What is its wavelength in centimeters?
(c) In the U.S., standard AC electrical current has a frequency of
60 Hz. Most machinery and appliances that use this current, as well
as the power lines that transport the electric power, emit radiation
with this frequency. What is its wavelength in km?

Problem 2.2:
Whenever we talk about a single frequency ν characterizing an
electromagnetic wave, we are tacitly assuming that the source and
the detector are stationary relative to one another, in which case ν
is indeed the same for both. However, if the distance between the
two is changing with velocity v (positive v implying increasing sep-
aration), then the frequency of radiation ν1 emitted by the source will
18 2. Properties of Radiation

be different than the frequency ν2 observed by the detector. In partic-

ular, the frequency shift ∆ν = ν1 − ν2 is approximately proportional
to v, a phenomenon known as Doppler shift.
(a) Derive the precise relationship between ∆ν and v, by consider-
ing the time ∆t elapsed between two successive wave crests reaching
the detector with speed c.
(b) For the case that v ≪ c, show that your solution to (a) simpli-
fies to a proportionality between ∆ν and v.

2.2.1 Frequency Decomposition

The above discussion of periodic waves is interesting, but what does
it have to do with real EM radiation? After all, the EM disturbance
that arises from a lightning discharge, or from dropping a magnet
on the floor, is clearly not a steadily oscillating signal but more likely
a short, chaotic pulse! What sense does it make to speak of a specific
frequency or wavelength in these cases?
The answer is that any arbitrary EM fluctuation, short or long,
can be thought of as a composite of a number (potentially infinite)
of different “pure” periodic fluctuations. Specifically, any continu-
ous function of time f (t) can be expressed as a sum of pure sine
functions as follows:
Z ∞
f (t) = α(ω ) sin[ωt + φ(ω )] dω (2.2)
0

where α(ω ) is the amplitude of the sine function contribution for

each specific value of the angular frequency ω and φ(ω ) gives the
corresponding phase. If f (t) itself is already a pure sine function
sin(ω0 t + φ0 ), then of course α = 0 for all values of ω except ω0 .
For more general functions f (t), α(ω ) and φ(ω ) may be quite com-
plicated. It is beyond the scope of this book to explain how we find
α(ω ) etc. for any given f (t); it is only important to recognize that it
can, in principle, always be done.1
1 This
so-called Fourier decomposition is extremely useful throughout the physi-
cal and engineering sciences, including other areas of atmospheric dynamics and
climatology, so if you haven’t seen anything like this before, it is certainly worth
taking the time to read up on it.
 Frequency 19

Now recall what I pointed out earlier: individual EM distur-

bances propagate completely independently of one another. They may
intersect, or even travel together, but the presence of one does not
influence the other. This principle applies equally well to the pure
sine wave components of a Fourier decomposition. Thus, not only
can you regard any arbitrary electromagnetic disturbance as a mix-
ture of pure sine waves of differing angular frequencies ω, but you
can then track the propagation of each frequency component completely
separately from all the others.
The implications of this observation are profound for atmo-
spheric radiative transfer. The most basic path to understanding
and/or modeling the interaction of EM radiation with clouds, wa-
ter vapor, oxygen, carbon dioxide, etc., is to consider one frequency
at a time and then, if required, to sum the results over all relevant
frequencies. There are shortcuts that sometimes allow you to forget
you are doing this, but that is nevertheless what is going on under
the surface.

2.2.2 Broadband vs Monochromatic Radiation

Now is a good time to introduce a few more definitions:

• EM radiation composed entirely of a single frequency is

termed monochromatic (“one color”).

• Radiation that consists of a mixture of a wide range of fre-

quencies is called broadband radiation.

As already noted, the transport of broadband radiation in the

atmosphere can always be understood in terms of the transport of
the individual constituent frequencies. Therefore, we will tend to
focus at least initially on the monochromatic radiation.
It is quite common to have to deal with radiation that is
not strictly monochromatic but is nevertheless confined to an ex-
tremely narrow range of frequencies. Such radiation is often called
monochromatic, even though quasi-monochromatic would be a more
precise term. Another way to distinguish between the two cases is
via the terms coherent and incoherent:
20 2. Properties of Radiation

• Coherent radiation is what you get from a single oscillator, like

the magnet on a turntable described earlier, or from a group
of oscillators that are, for whatever reason, in perfect synchro-
nization with one another. Imagine a stadium full of people
doing “the wave” at a football game, or an audience clapping
in unison for an encore at the end of a rock concert. Microwave
ovens, radars, lasers, and radio towers all produce coherent
radiation. Note that these are all artificial sources. As an at-
mospheric scientist, you are most likely to encounter coherent
radiation in the context of the artificial sources used in remote
sensing, such as radar and lidar.
• Incoherent radiation is what you get from a set of indepen-
dent oscillators that may have nearly the same frequency
(quasi-monochromatic) but are not phase-locked to one an-
other. Imagine a large audience applauding at the end of a
speech. There is no synchronization of the claps of different
individuals, even if everyone is clapping with about the same
frequency. Natural radiation in the lower atmosphere is, for
all practical purposes, incoherent.

2.3 Polarization
As mentioned earlier, the orientation of the oscillating electric field vec-
tor in an EM wave can be any direction that is perpendicular to the di-
rection of propagation. In some radiative transfer applications (espe-
cially remote sensing) it is sometimes important to keep track of that
orientation and how it evolves over the course of a complete cycle.
In coherent radiation, there is a unique, repeating pattern to the
oscillating electric field vector when viewed along the direction of
propagation. There are several basic possibilities for this pattern:
1. It may vibrate back and forth in a fixed plane, like a pendu-
lum or plucked guitar string. This is called linear polarization
[Fig. 2.2 (a)–(c)].
2. It may oscillate in spiral fashion about the direction of propa-
gation, either clockwise or counterclockwise, for circular polar-
ization [Fig. 2.2 (f)].
 Polarization 21

(a) (b) (c)

(d) (e) (f )

Fig. 2.2: Examples of different types of polarization, depicted in terms of the vi-
bration of the electric field vector in a fixed plane perpendicular to the direction
of propagation. (a) vertical linear, (b) horizontal linear, (c) linear at 45◦ , (d) ellip-
tical counterclockwise, with vertical major axis, (e) elliptical clockwise, with the
major axis oriented at a 45◦ angle, (f) circular, clockwise. Note that an infinity of
combinations of orientation, ellipticity, and sense of rotation are also possible.

3. Elliptical polarization is essentially a hybrid of the first two.

Note that elliptical polarization can be viewed as including
both linear and circular polarization as limiting cases [Fig. 2.2
(d)–(e)].

Standard weather radar equipment typically transmits coherent

radiation with linear polarization (either vertical or horizontal) and
then measures the backscattered radiation with the same polariza-
tion. More sophisticated radars may transmit in one polarization
but then separately measure the returned radiation in both vertical
and horizontal polarizations in order to gain additional information
about the targets.
In incoherent radiation, a systematic tendency toward one type
of polarization may or may not be discernible. Therefore, in ad-
dition to the above types of polarization, one must specify the de-
22 2. Properties of Radiation

gree of polarization. As a general rule, natural emissions of radia-

tion in the atmosphere are completely unpolarized, but the radia-
tion may become partially or completely polarized in the course of
its interactions with particles and/or the surface. In particular, it
will be shown later that a smooth surface, like calm water, preferen-
tially reflects radiation having horizontal linear polarization. This
is the phenomenon that motivated the invention of polarized sun-
glasses, which block the reflected horizontally polarized radiation
while transmitting the rest. It is also a phenomenon of great practi-
cal importance for satellite remote sensing in the microwave band.
The purpose of this section was just to introduce you to the ex-
istence and qualitative nature of polarization. Having done that, I’ll
point out that polarization is often disregarded in radiative trans-
fer calculations, especially at the level targeted by this book. Where
polarization cannot be ignored, one can often gain at least qualita-
tive insight into its role without getting too deep into the relevant
mathematics. But since some of you may eventually require a more
complete and quantitative understanding of polarization, I will in-
troduce some elements of the mathematical treatment of polariza-
tion at appropriate points along the way.

2.4 Energy
Barely mentioned so far, but central to our interest in atmospheric
radiation, is the fact that EM radiation transports energy. Just as
gravity waves on the surface of the ocean efficiently transport en-
ergy from a North Pacific storm to the sunny beaches of California,
where that energy is violently deposited onto the bodies of inatten-
tive bathers, EM radiation transports vast quantities of energy from
the thermonuclear furnace of the sun to the vinyl seat cover in your
parked car.
In view of this fact, it would seem natural to characterize ra-
diation in terms of its energy content, using the standard SI units
of joules (J). But because natural radiation is not pulsed but rather
continuous, it actually makes more sense to speak of the rate of en-
ergy transfer, or power, in watts (W = J/sec). Furthermore, radia-
tion doesn’t transport its energy through a single point but rather is
distributed over an area. Therefore, the most convenient way to de-
 Energy 23

scribe the transport of energy by radiation at a location is in terms of

its flux density F (commonly, though somewhat inaccurately, short-
ened to just flux in W/m2 . Other names you may see for the exact
same quantity are 1) irradiance, and 2) radiant exitance. The term “ir-
radiance” is often preferred when referring to the flux of radiation
incident on a plane or surface, while the term “exitance” is usually
applied to the flux of radiation emerging from a surface.
The magnitude of the flux depends on the orientation of the ref-
erence surface. If the radiation is due to a single source, such as
the sun, one might measure the flux through an imaginary surface
normal to the direction of propagation. On the other hand, one is
often interested in the flux density of sunlight (or other radiation)
obliquely incident on a horizontal surface. We will refine these ideas
later in this chapter.

Problem 2.3: The atmospheric boundary layer is that region near the
surface that is “well-mixed” by mechanical and/or convective tur-
bulence originating at the surface. Its thickness may range from a
few meters to several kilometers. Heat added by conduction from
the surface is typically distributed quickly throughout the boundary
layer.
At a certain location in the tropics, the sun rises at 06 Local Solar
Time (LST), is directly overhead at noon, and sets at 18 LST (=6 PM).
Assume that, during the twelve hours that the sun is up, the net
flux of solar energy absorbed by dry vegetation and immediately
transferred to the overlying air is F (t) = F0 cos[π (t − 12)/12], where
t is the time of day (in hours), and F0 = 500 W m−2 .
(a) Ignoring other heating and cooling terms, compute the total
solar energy (in J/m2 ) added to the boundary layer over one 24-hour
period.
(b) If the boundary layer depth ∆z = 1 km, its average air density
is ρ a = 1 kg/m3 , and the heat capacity at constant pressure is c p =
1004 J/(kg K), compute the temperature increase ∆T implied by your
answer to (a).
(c) If, instead, the boundary layer depth started out at sunrise
only 10 m deep and remained at that depth throughout the day, what
would be the corresponding change of temperature? Why is it much
more likely that the boundary layer would deepen quickly after sun-
rise?
24 2. Properties of Radiation

2.5 A Mathematical Description of EM Waves†

Maxwell’s Equations
We have been able to come quite far in describing the behavior of
EM radiation without resorting to many equations. Having laid the
groundwork by defining terms and building mental models of elec-
tromagnetic waves, we can now tear through a more rigorous math-
ematical treatment with scarcely a pause for breath.
Just as the Navier-Stokes equation, the hydrostatic approxima-
tion, and the Ideal Gas Law serve as the underpinnings for virtually
all of atmospheric dynamics, the so-called Maxwell equations gov-
ern classical electrodynamics. If one assumes SI units for all vari-
ables, these equations take the following form:
~ = ρ,
∇·D (2.3)
∇ · ~B = 0, (2.4)
∂~B
∇ × ~E = − , (2.5)
∂t
~
∇×H ~ = ~J + ∂D , (2.6)
∂t
where D ~ is the electric displacement, ~E is the electric field, ~B is the
magnetic induction, H~ is the magnetic field, ρ is the density of electric
~
charge, and J is the electric current vector.
If we further assume (quite reasonably, for most materials) that
we are dealing with a macroscopic, homogeneous medium whose
electrical and magnetic properties are (i) directionally isotropic and
(ii) linear with respect to ~E and ~B, then
~ = ε 0 (1 + χ)~E,
D (2.7)
where ε 0 is the permittivity of free space and χ is the electric sus-
ceptibility of the medium, which describes the degree to which the
material becomes electrically polarized under the influence of an
external field. Also,
~B = µH~, (2.8)
where µ is the magnetic permeability of the medium, and
~J = σ~E, (2.9)
 A Mathematical Description of EM Waves† 25

where σ is the conductivity. Finally, we can usually assume that the

charge density ρ is effectively zero.
With the above assumptions, Maxwell’s equations reduce to

∇ · ~E = 0, (2.10)

~ = 0,
∇·H (2.11)
~
∂H
∇ × ~E = −µ , (2.12)
∂t
~
∇×H ~ = σ~E + ε 0 (1 + χ) ∂E . (2.13)
∂t
Note that only two field variables, ~E and H
~ , appear in the above four
equations. The (complex) parameters σ, χ, and µ then determine the
relationship between ~E and H~ in any given medium.

Time-Harmonic Solution
As discussed in Section 2.2.1, we can always decompose an arbitrary
EM disturbance into its various frequency components and consider
each frequency separately from the others. In keeping with this ap-
proach, we will consider a general time-harmonic electric field of
the form
~Ec (~x, t) = ~C(~x) exp(−iωt) (2.14)
where ~C = ~A + i~B is a complex vector field, ~x is the position vector,
and ω = 2πν is the angular frequency (radians per second). An
analogous representation can be written for the complex magnetic
~ c.
field H
Note that the use of complex quantities is strictly a notational
convenience; in physical problems, we are almost always interested
in just the real part of whatever solutions we derive; e.g.,

~E = Re{~Ec } = ~A cos ωt + ~B sin ωt (2.15)

Substituting (2.14) into (2.10)–(2.13), we obtain the following rela-

tionships:
∇ · (ε~Ec ) = 0, (2.16)
∇ × ~Ec = iωµH
~ c, (2.17)
26 2. Properties of Radiation

~ c = 0,
∇·H (2.18)

~ c = −iωε~Ec ,
∇×H (2.19)

where the complex permittivity of the medium is defined as

σ
ε = ε 0 (1 + χ ) + i . (2.20)
ω

Except for ε 0 , which is a fundamental, real-valued physical constant,

all other parameters depend on the medium under consideration as
well as on the frequency ω.

Problem 2.4: Verify that (2.16)–(2.19) follow from (2.3)–(2.9) and

(2.14).

Solution for a Plane Wave

Any harmonic electromagnetic field satisfying the above equations
is physically realizable. However, we will restrict our attention to
solutions describing a plane wave. Such solutions have the form

~Ec = ~E0 exp(i~k ·~x − iωt), ~ 0 exp(i~k ·~x − iωt),

~c=H
H (2.21)

~ 0 are constant (complex) vectors, and ~k = ~k′ + i~k′′

where ~E0 and H
is a complex wave vector. Thus

~Ec = ~E0 exp(−~k′′ ·~x) exp[i (~k′ ·~x − ωt)], (2.22)

H ~ 0 exp(−~k′′ ·~x) exp[i (~k′ ·~x − ωt)].

~c=H (2.23)

These relationships imply that the vector ~k′ is normal to planes of

constant phase (and thus indicates the direction of propagation of
the wave crests), while ~k′′ is normal to planes of constant amplitude.
The two are not necessarily parallel. When they are, or when ~k′′ is
zero, the wave is called homogeneous.
 A Mathematical Description of EM Waves† 27

The term ~E0 exp(−~k′′ ·~x) gives the amplitude of the electric wave
at location ~x. If ~k′′ is zero, then the medium is nonabsorbing, be-
cause the amplitude is constant. The quantity φ ≡ ~k′ ·~x − ωt gives
the phase. The phase speed of the wave is given by
ω
v= (2.24)
|~k′ |

Substituting (2.21) into (2.16)–(2.19) yields

~k · ~E0 = 0, (2.25)

~k · H
~ 0 = 0, (2.26)
~k × ~E0 = ωµH
~ 0, (2.27)
~k × H
~ 0 = −ωε~E0 . (2.28)
The last two equations reveal that, in a plane wave, the oscillating
electric and magnetic field vectors are normal both to each other
and to the direction of propagation of the wave. This property is de-
picted schematically in Fig. 2.1 for a horizontally polarized wave.

Problem 2.5: Verify that (2.25) follows from (2.21) and (2.16) by (a)
expanding (2.21) in terms of the individual components of the vec-
tors ~E0 , ~k, ~x and (b) substituting this expression into (2.16) and apply-
ing the divergence operator (∇·). The remaining equations (2.26)–
(2.28) are derived in an analogous way.

If we now take the vector product of ~k with both sides of (2.27),

~k × (~k × ~E0 ) = ωµ~k × H

~ 0 = −εµω 2~E0 , (2.29)

and use the vector identity

~a × (~b ×~c) = ~b(~a ·~c) −~c(~a · ~b), (2.30)

we see from (2.25) that the first term on the right is zero, thus

~k · ~k = εµω 2 . (2.31)
28 2. Properties of Radiation

In the case of a homogeneous wave, the above simplifies to

(|~k′ | + i |~k′′ |)2 = εµω 2 (2.32)

or
√
|~k′ | + i |~k′′ | = ω εµ. (2.33)

Phase Speed
In a vacuum, ~k′′ = 0, the permittivity of free space ε ≡ ε 0 = 8.854 ×
10−12 F m−1 , and the magnetic permeability µ ≡ µ0 = 1.257 × 10−6
N A−2 . From (2.24), we have the following phase speed in a vac-
uum:
√
c ≡ 1/ ε 0 µ0 . (2.34)
If we substitute the above numerical values of ε 0 and µ0 into this
expression, we obtain the speed of light in a vacuum c = 2.998 ×
108 m s−1 .
In a nonvacuum, we can write

εµ √ ωN
r
~ ′ ~ ′′
|k | + i|k | = ω ε 0 µ0 = , (2.35)
ε 0 µ0 c

where the complex index of refraction N is given by

εµ c
r
N≡ = ′ , (2.36)
ε 0 µ0 c

√
with c′ ≡ 1/ εµ. If N happens to be pure real (i.e., if ~k′′ = 0 and the
medium is therefore nonabsorbing), then c′ may be interpreted as
the phase speed of the wave within the medium. Even in absorbing
media, this is still a reasonable, though approximate, way to view
c′ .
For most physical media, N > 1, which implies a reduced speed
of light relative to that in a vacuum. It is important to keep in mind
that N is not only a property of a particular medium but also gener-
ally a strong function of frequency.
 A Mathematical Description of EM Waves† 29

The Poynting Vector

Any electromagnetic wave (harmonic or not) transports energy

through space at the speed of light. The instantaneous direction
and magnitude of the transported energy is given by the Poynting2
vector, which is defined (for SI units) as

~S = ~E × H
~ . (2.37)

The above is an instantaneous value and gives the power per unit
area transported by the disturbance. In the case of a time-harmonic
wave, we are interested in the average over one complete cycle:

F = h~Si = h~E × H
~i , (2.38)

which is just the flux density of the wave as measure through a sur-
face normal to the direction of propagation. For a harmonic wave,
the above reduces to
1
F = cε 0 E2 , (2.39)
2
where E is the scalar amplitude of the oscillating electric field at
that location. The key point to note here is that the flux density is
proportional to the square of the amplitude of the electromagnetic
wave.

Absorption

The scalar amplitude of our harmonic plane wave at location ~x is

E = |~E0 exp(−~k′′ ·~x)| . (2.40)

From (2.39), it follows that, for a plane wave with initial flux density
F0 at ~x = 0,

F = F0 [exp(−~k′′ ·~x)]2 = F0 exp(−2~k′′ ·~x). (2.41)

2 The name is easy to remember, because the Poynting vector points in the di-
rection of the energy transport by the wave. However the name is simply an
odd coincidence, as it was named after the English physicist John Henry Poynt-
ing (1852–1914).
30 2. Properties of Radiation

If we now consider a plane wave propagating in the x direction, and

note from (2.35) that

ω ωni 2πνni
|~k′′ | = Im{ N } = = (2.42)
c c c

where ν is the frequency in Hz, we have

F = F0 e− βa x , (2.43)

where the absorption coefficient β a is defined as

4πνni 4πni
βa = = , (2.44)
c λ

with λ the wavelength of the radiation in a vacuum. In summary,

the quantity 1/β a gives the distance required for the wave’s energy
to be attenuated to e−1 ≈ 37% of its original value.

Problem 2.6: Within a certain material, an EM wave with λ = 1 µm

is attenuated to 10% of its original intensity after propagating 10 cm.
Determine the imaginary part of the index of refraction ni .

Problem 2.7: For red light (λ = 0.64 µm), ni in pure water is ap-
proximately 1.3 × 10−8 ; for blue light (λ = 0.48 µm), ni ≈ 1.0 × 10−9 .
The deep end of a typical home swimming pool is approximately
2.5 m deep. Compute the fraction of each wavelength that survives
the two-way trip to the bottom of the pool and back, when illumi-
nated (and viewed) from directly above. In light of your findings
(and in view of the appearance of most swimming pools as seen from
the air), comment on the common assumption that water is “color-
less.”
 Quantum Properties of Radiation 31

2.6 Quantum Properties of Radiation

Now that I have succeeded in indoctrinating you with the wave-
based mental model of EM radiation, including even a neat math-
ematical expression for a plane wave, I will demolish your newly
won confidence by asserting that you should view EM radiation not
as waves but as particles! At least some of the time.
It may surprise you to learn that Albert Einstein won his Nobel
Prize not for his famous theory of relativity but rather for his expla-
nation in 1905 of the photoelectric effect. This effect refers to the phe-
nomenon by which electrons are jarred loose from a material surface
exposed to light in a vacuum. A previously mysterious aspect of
the photoelectric effect had been the observation that incident light
with wavelengths longer than a certain threshold, no matter how in-
tense, would not generate free electrons, whereas light with shorter
wavelengths readily dislodges the electrons and continues to do so,
sporadically at least, no matter how weak the illumination.
The essence of Einstein’s explanation was that light falls on a
surface not as a smoothly continuous flux of wave energy, but rather
as a staccato hail of little discrete packets of energy, called photons.
The energy content E of each individual photon, and therefore that
photon’s ability to knock electrons loose from the surface, is deter-
mined solely by the frequency or wavelength of the radiation via
the relationship

E = hν, (2.45)

where ν is the frequency and h = 6.626 × 10−34 J s is Planck’s con-

stant.
Moreover, you can’t have just part of a photon; therefore very
low intensity light deposits discrete packets of energy on a surface
in a manner analogous to the occasional random splashes of fat rain-
drops on your windshield at the early onset of a rain shower. Thus,
if monochromatic radiation of wavelength λ deposits F watts per
unit area on a surface, then this corresponds to
F Fλ
N = = (2.46)
hν hc
photons per unit area per unit time. For fluxes of the magnitude
32 2. Properties of Radiation

typically encountered in the atmosphere, N is rather large, and it

is therefore hard to distinguish the effects of discrete particles, just
as it is hard to distinguish the contributions of each individual rain-
drop to your growing wetness when you get caught out in a heavy
downpour.

Problem 2.8: Only radiation with wavelengths smaller than

0.2424 µm is capable of dissociating molecular oxygen into atomic
oxygen, according to the reaction

O2 + photon → O + O

Based on this information, how much energy is apparently required

to break the molecular bond of a single molecule of O2 ?

Problem 2.9: A small light source emits 1 W of radiation uniformly

in all directions. The wavelength of the light is 0.5 µm.
(a) How many photons per second are emitted by the light
source?
(b) If the light source were on the moon and were viewed by a
telescope on Earth having a 20 cm diameter circular aperture, how
many photons per second would the telescope collect? Ignore at-
mospheric attenuation. Assume a distance Dm = 3.84 × 105 km be-
tween the moon and the earth.

The above quantum description of EM radiation is completely

at odds with the previous wave description. They cannot both be
true. And yet they are! Do not trouble yourself by trying to men-
tally reconcile the two models — countless smart people have tried
throughout the twentieth century, and all have failed to explain this
paradox in terms most people can visualize.
The important things for you to know are: 1) when radiation
must be viewed as waves, 2) when it must be viewed as a shower of
particles, and 3) when it doesn’t matter.
As a general rule of thumb, the wave nature of radiation mat-
ters when computing the scattering and reflection properties of at-
mospheric particles (air molecules, aerosols, cloud droplets, rain-
drops) and surfaces. By contrast, the quantized nature of radiation,
 Flux and Intensity 33

and thus (2.45), comes to the forefront when considering absorption

and emission of radiation by individual atoms and molecules, includ-
ing photochemical reactions. Finally, for calculations of large-scale
transport of radiation in the atmosphere, the effects of both types of
interactions will have already been deeply buried in some generic
extinction and scattering coefficients and can be conveniently put
out of your mind altogether.

2.7 Flux and Intensity

I have already briefly introduced the concept of flux density (or flux
for short3 ) as a measure of the total energy per unit time (or power)
per unit area transported by EM radiation through a plane (or de-
posited on a surface). It is now time to extend our understanding of
this property and to introduce a closely related quantity, called the
radiant intensity , or just intensity for short. In some books, the term
radiance is substituted for intensity.
Flux and intensity are the two measures of the strength of an EM ra-
diation field that are central to most problems in atmospheric science. The
two are intimately related, as we shall see shortly.

2.7.1 Flux
Recall first of all that the flux F refers to the rate at which radiation
is incident on, or passes through, a flat surface. Without further
qualification, flux is expressed in units of watts per square meter.
The surface may be real (e.g., the ground, or the top of a cloud layer)
or it may be imaginary (e.g., an arbitrary level in the atmosphere ).
Often, but not always, it is taken to be horizontal. Other times it
may be assumed to be perpendicular to a single source of radiation,
such as the sun.
Note that a flux of natural (incoherent) radiation expressed sim-
3 Strictlyspeaking, an energy flux has units of W, whereas the flux density is the
flux per unit area and therefore has units of W m−2 . Meteorologists are almost in-
variably concerned with quantities that are expressed per unit area, volume, mass,
etc. Therefore, when a meteorologist says “flux,” it is generally understood that
she means “flux per unit area” or flux density.
34 2. Properties of Radiation

ply in W/m2 must be a broadband quantity.4 That is, it includes en-

ergy contributions from all wavelengths between some specified (or
implied) limits λ1 and λ2 . Those limits might encompass all possi-
ble wavelengths (i.e., λ1 = 0, λ2 = ∞), or they might define a some-
what narrower range. However, the range cannot be zero (λ1 = λ2 ),
because the power contained in that range would then also be zero!
It is, however, possible to define a monochromatic flux (also
known as spectral flux) Fλ as follows:

F (λ, λ + ∆λ)
Fλ = lim , (2.47)
∆λ→0 ∆λ
where F (λ, λ + ∆λ) is the flux in W/m2 contributed by radiation
with wavelengths between λ and λ + ∆λ. The dimensions of the
monochromatic flux are not just power per unit area but rather
power per unit area per unit wavelength. Typical units would thus be
W m−2 µm−1 .
Having defined the monochromatic (or spectral) flux as above,
you get the broadband flux over some extended range of wave-
length [λ1 , λ2 ] by integrating over the appropriate range of wave-
length:
Z λ2
F ( λ1 , λ2 ) = Fλ dλ . (2.48)
λ1

Problem 2.10: The total radiation flux incident on a surface due to

wavelengths between 0.3 µm and 1.0 µm is 200 W m−2 . (a) What
is the average spectral flux within this interval? Give your answer
in units of W m−2 µm−1 . (b) If the spectral flux is constant with
wavelength, then what is the total flux contributed by wavelengths
just between 0.4 µm and 0.5 µm? (c) What is the total flux (in W m−2 )
contributed by radiation of exactly 0.5 µm wavelength?

Let’s illustrate the concept of flux with a concrete example. If

you mark out an area on flat ground out in the open, the amount
4 This is not necessarily true for artificial coherent radiation, for which finite

power may be associated with exactly one wavelength, rather than being dis-
tributed over a range of wavelengths.
 Flux and Intensity 35

of daylight falling on it can be measured in watts per square me-

ter. This is the incident flux of solar radiation, and it determines how
much total radiation from the sun is available to be absorbed. If a
cloud passes in front of the sun, the incident flux temporarily de-
creases because the transmission of radiation from the sun to the
surface is partially blocked. As the afternoon wears on, the inci-
dent flux steadily decreases, this time because the light from the
sun strikes the surface at an increasingly oblique angle, spreading
the same energy over a larger and larger area.
An important point to remember is that the flux makes no dis-
tinction concerning where the radiation is coming from. 100 W m−2
incident on our front lawn is 100 W m−2 , regardless of whether it is
coming from all directions more or less equally on an overcast day
or primarily one particular direction (that of the sun) on a crystal-
clear afternoon. In order to completely characterize the radiation field at
a given location, we must know not only the flux but also the direction(s)
from which the radiation is coming and thus also in which direction(s)
it is going. This directional information is embodied in the radiant
intensity.

2.7.2 Intensity
The radiant intensity I tells you in detail both the strength and direc-
tion of various sources contributing to the incident flux on a surface.
For visible radiation, intensity corresponds roughly to the “bright-
ness” your eyes see looking backward along a ray of incoming ra-
diation. Thus, if you lie flat on your back and look up at the sky,
you can visually map out which regions of the sky are associated
with high radiant intensity and therefore contribute most strongly
to the total incident solar flux at your location. The sun itself, if not
blocked by a cloud, is seen as a very localized high intensity source,
whereas the clear sky is a relatively uniform source of rather low in-
tensity radiation. Isolated clouds, as seen from your vantage point,
may be either brighter or darker than the clear sky, depending on
how thick they are and from what angle they are viewed, relative to
the sun.
Consider again what happens if a small cumulus cloud passes
between you and the sun, casting you and your marked out area
36 2. Properties of Radiation

of ground into shadow. You can now look directly toward the sun
without hurting your eyes. From this you can conclude that the
intensity of radiation from that direction has been greatly reduced.
But the contributions from the rest of the sky are unaffected. The to-
tal incident shortwave flux is therefore reduced but not eliminated.
Although you are now in the shadow of the cloud, it is by no means
pitch dark.
Clearly, there is a very close relationship between flux and inten-
sity. In words, the flux incident on a surface is obtained by integrat-
ing the contributions of intensity from all possible directions visible
from that surface. It is time to delve into the mathematics of this
relationship. We will begin by setting up the necessary machinery.

Spherical Polar Coordinates

In any discussion of radiation, direction plays an all-important role.

We therefore need to adopt a convention for describing directions.
There is no single “right” choice, but one that is very convenient in
atmospheric radiation is based on spherical polar coordinates. Fig.
2.3 depicts the geometry.
In this system, the zenith angle θ measures the angle from some
reference direction, usually the local vertical. Thus, directly over-
head usually corresponds to θ = 0. In this case, the horizon cor-
responds to θ = π/2 radians, or 90◦ . π/2 < θ < π corresponds
to directions below the horizon, with θ = π being straight down
(“nadir”).
The azimuthal angle φ measures the angle counterclockwise from
a reference point on the horizon, so that 0 < φ < 2π. It is not usually
terribly important which point of the compass is used as the refer-
ence in any given application, and it is often chosen to be whatever
is most natural for the problem at hand, such as the direction of the
sun.
Any possible direction above or below the horizon may thus be
described via the two angles θ and φ. Sometimes directions may
be expressed abstractly in terms of a unit vector Ω̂, in which case
no particular coordinate system√ is implied.
√ Thus the same direction
Ω̂ could be represented by [ 2/2, 0, 2/2] in ( x, y, z)-coordinates,
by [π/4, 0] in (θ, φ))-coordinates when θ = 0 defines the vertical
 Flux and Intensity 37

Ωz = cos θ
Ω

Ωy = sin θ sin φ
φ
os
in θc y
=s
Ωx φ

Fig. 2.3: The relationship between Cartesian and spherical coordinates.

direction, or even by [0, 0] if θ = 0 should for some reason be chosen

to coincide with the direction of the sun at a time when the latter is
45◦ above the horizon.

Solid Angle
Another essential concept is that of solid angle. Surprisingly many
new students of atmospheric radiation find this concept confusing,
presumably because they haven’t had occasion to consciously use it
before, unlike the angles we have been measuring with protractors
since third grade. But it’s really very simple: solid angle is to “regular”
angle as area is to length. You can think of solid angle as something
you might measure in “square radians” or “square degrees”, except
the actual unit used is called the steradian. We will give a precise
definition of this unit later.
In absolute terms, the sun has a certain diameter in kilome-
ters and a certain cross-sectional area in km2 . But absolute dimen-
sions are often of secondary importance in radiative transfer, com-
38 2. Properties of Radiation

pared with angular dimensions, which describe how big an object or

source of radiation looks from a particular vantage point. Thus, what
matters for solar radiation reaching the earth is that the sun’s disk
has a particular angular diameter in units of degrees or radians, and
it also subtends (or presents) a certain solid angle in units of steradi-
ans. Solid angle is thus a measure of how much of your visual field
of view is occupied by an object. For example, the sun subtends a
much larger solid angle as viewed from the planet Mercury than it
does from Earth. Also, from our perspective here on Earth, the full
moon subtends nearly the same solid angle as the Sun, even though
the latter body is much larger in absolute terms. A half moon, of
course, subtends half the solid angle of the full moon.

Definition of Steradian

Now that you understand what solid angle is, you can appreciate
a simple definition of the unit steradian (abbreviation sr). Imagine
you are at the center of a sphere of unit radius — it doesn’t matter
whether the unit is a kilometer, a mile, a furlong or what have you.
The total surface area of the sphere is 4π square units. Likewise, the
combined solid angle represented by every direction you can possibly look
is 4π steradians. The surface area of just one half of the sphere is 2π
units squared. Likewise, the entire sky above the horizon (or “celestial
dome”) subtends a solid angle of 2π steradians, as does (separately) the
lower hemisphere of your field of vision, representing everything below the
horizon.
Conceptually, you can determine the solid angle subtended by
any object by tracing its outline on your unit sphere and then mea-
suring the actual surface area of the tracing. (Note the analogy to
radians as a measure of arc length on the unit circle.) This is gen-
erally not a practical approach, however, so we instead invoke our
polar coordinate system so as to be able to define an infinitesimal
increment of solid angle as follows:

dω = sin θ dθ dφ . (2.49)

This relationship is depicted schematically in Fig. 2.4.

 Flux and Intensity 39

dω = sinθ dθ dφ

dθ

dφ

x y

Fig. 2.4: The relationship between solid angle and polar coordinates.

In other words, if you paint an infinitesimal rectangle on the sur-

face of your unit sphere, and it has angular dimensions dθ (zenith
angle) and dφ (azimuth angle) and is positioned at θ, then the above
expression gives you the increment of solid angle subtended. Why
does sin θ appear in there? Simple: for the same reason that a 1◦
latitude by 1◦ longitude box encompasses far less real estate near
the North Pole than near the equator, because of the convergence of
lines of equal φ to a point.
Now let’s demonstrate that we can recover the expected value
of 4π steradians for the entire sphere:
Z Z 2πZ π Z π
dω = sin θ dθ dφ = 2π sin θ dθ = 4π. (2.50)
4π 0 0 0

Note that the notation used in the left-most integral expresses the
abstract idea that we are integrating over the full sphere (4π stera-
dians). This notation makes no assumption about the coordinate
system that we will use to actually perform the integration. The in-
tegral just to the right of it translates the abstract integration over
40 2. Properties of Radiation

R
D

φ
θmax

Fig. 2.5: Geometric framework for computing the solid angle subtended by a
sphere of radius R whose center is a distance D from the observer.

all directions into a specific double integral using our coordinates θ

and φ, based on (2.49). If we were to use a different coordinate sys-
tem to describe directions, then the integrals on the right-hand side
might take a different form, but the final result for this problem, if
everything is done correctly, should still be 4π steradians!

Problem 2.11: Consider a cloud that, when viewed from a point on

the surface, occupies the portion of the sky defined by π/4 < θ <
π/2 and 0 < φ < π/8.
 Flux and Intensity 41

(a) What is the solid angle subtended by the cloud?

(b) What percentage of the sky is covered by this cloud?

Problem 2.12: The moon is at a mean distance Dm = 3.84 × 105 km

from the earth; the Sun is at a mean distance Ds = 1.496 × 108 km.
The radius of the moon is Rm = 1.74 × 103 km; the radius of the sun
is Rs = 6.96 × 105 km.
(a) Compute the angular diameter (in degrees) subtended by the
sun and the moon. Refer to Fig. 2.5 for help visualizing the geometry
of this problem. (b) Compute the solid angle subtended by the sun
and the moon. (c) Which appears larger from Earth, and by what
percentage do the two solid angles differ? (d) If the above values
were constant, would it be possible to explain the occurrence of total
solar eclipses?

Formal Definition of Intensity

Having defined solid angle, we are now in a position to attach a
precise definition to the term radiant intensity. In words, intensity
I (Ω̂) is the flux (measured on a surface normal to the beam) per unit
solid angle traveling in a particular direction Ω̂. Visualize the above
definition as follows:

• Looking in the direction −Ω̂, identify a very small element of

the scene with solid angle δω.

• Measure, normal to the beam, the flux δF of radiation arriving

just from that small region, while excluding all other contribu-
tions.

• The intensity in that direction is then given by

δF
I (Ω̂) = . (2.51)
δω
42 2. Properties of Radiation

This recipe is strictly valid only when δω is vanishingly small, but

it’s a reasonable approximation for finite solid angles as well, as
long as the maximum arc subtended by the region in question is
much less than one radian — say 5◦ or less — and as long as the
intensity is uniform throughout the region.

Problem 2.13: The broadband flux of solar radiation that reaches

the top of the atmosphere is approximately 1370 W m−2 , when mea-
sured on a plane normal to the beam. Combine this with the solid
angle you derived in a previous problem to compute the average ra-
diant intensity of the sun’s surface.

Problem 2.14: A typical laser pointer used in lectures puts out 5

mW of power into a nearly parallel beam with a diameter of 5 mm.
(a) What is the flux density normal to the beam, and how does it com-
pare with the typical clear-sky solar flux (at ground level) on a sur-
face normal to the beam of 1000 W m−2 ? (b) If beam can be assumed
to be confined to a cone of angular diameter 1 milliradian, what is
the intensity of the beam in watts per steradian, and how does this
compare with the intensity of sunlight computed from the above so-
lar flux and an angular diameter of 0.5◦ for the sun’s disk?

Conservation of Intensity

You have probably known since you were small that the ability of a
light source to illuminate an object weakens rapidly with increasing
distance. A flashlight illuminates a book under the bedcovers much
more brightly than it does an animal lurking in those bushes at the
edge of the campsite. Likewise, the earth is much more brightly
illuminated by the sun than is Pluto.
You might hastily infer from these observations that the inten-
sity of radiation associated with a given point at the source is also a
function of the distance of the observer. If you were talking about
the incident flux, you would be right. This is not the case for inten-
 Flux and Intensity 43

sity however. On the contrary, within a vacuum or other transparent

medium, radiant intensity is conserved along any optical path.5
To verify this principle in a simple case, tape a sheet of white pa-
per to the wall at eye level. Look at it from a couple feet away; then
back up and look at it from across the room. Although its appar-
ent size (solid angle) changes with distance, its apparent brightness
does not (assuming you didn’t adjust the room lights!).
The same principle applies to optical systems such as lenses,
mirrors, prisms, etc., as long as we ignore losses due to absorption
and/or partial reflections (e.g., from the surface of lenses) and as
long as we measure the intensity in the same medium (e.g., air) in
all cases. A magnifying glass makes objects appear larger, but it has
little effect on the object’s brightness. The same is true of binoculars
and telescopes.6

Intensity and Polarization†

For many applications, one need only keep track of the total scalar
intensity I of a stream of radiation as defined above. However, we
previously alluded to the polarization properties of EM radiation,
and it is sometimes necessary to keep track of these as well. One
reason might be a need for greater accuracy in radiative transfer
calculations, as disregarding polarization almost always entails an
approximation, especially when scattering by particles or surfaces is
important. Another occasion arises when you are measuring radia-
tion of one particular polarization (e.g, linear vertical or horizontal).
This is often the case for microwave remote sensing instruments.

5 The intensity can change when the radiation passes from one medium to an-
other.
6 But wait, you say. Telescopes definitely do make stars appear brighter: many

that are invisible to the naked eye become clearly visible through a telescope. How
can this behavior be reconciled with the previous assertions? The explanation is
that stars subtend a angle far too small for the eye to resolve. As a result, the eye
responds not to the intensity of the star but rather to the total flux integrated over
a finite solid angle. That solid angle is determined by the eye’s resolving power.
Thus, moderately near-sighted individuals will see a few bright stars but will miss
many more that are easily detectable by sharp-eyed people. A telescope increases
the solid angle subtended by a star, and thus the total flux from that direction,
making it more easily visible to everyone.
44 2. Properties of Radiation

When polarization must be considered, we require a representa-

tion of intensity that is capable of providing complete information
about the state of polarization. One such representation gives the
intensity as a four-element vector


I
 Q 
I=
 U  .
 (2.52)
V

The elements of this vector are called the Stokes parameters. The
first element, I, is the same as the scalar intensity we have already
discussed. The remaining elements Q, U, V contain information
concerning the degree of polarization (recall that incoherent radia-
tion can be polarized to any degree, whereas coherent radiation is
always fully polarized), about the preferred orientation of the polar-
ization, and about the nature of the polarization – circular, linear,
or somethingp in between. In particular, the p degree of polarization is
defined as Q + U + V /I. The ratios Q2 + U 2 /I and V/I, re-
2 2 2

spectively, are the degree of linear polarization and the degree of circular
polarization.
Thus, for completely unpolarized radiation, Q = U = V = 0,
and for fully polarized radiation

I 2 = Q2 + U 2 + V 2 . (2.53)

It is beyond the scope of this introductory text to give a detailed

electromagnetic definition of each of the Stokes parameters.7 How-
ever, some illustrative examples are given in Table 2.1.
How is the vector representation of intensity actually used? In
the scalar case (i.e., when we’re ignoring polarization), most inter-
actions of radiation with matter can be described via the multiplica-
tion of the intensity by a scalar coefficient, which we will arbitrarily
denote A. Thus,
Inew = A · Iold . (2.54)
7A good overview is given in Section 2.3 of S94.
 Flux and Intensity 45

Table 2.1: Examples of Stokes parameter values.

Description [I,Q,U,V]

Horizontally polarized [1, 1, 0, 0]

Vertically polarized [1, −1, 0, 0]
Linearly polarized at +45◦ [1, 0, 1, 0]
Linearly polarized at −45◦ [1, 0, −1, 0]
Right circularly polarized [1, 0, 0, 1]
Left circularly polarized [1, 0, 0, −1]
Unpolarized [1, 0, 0, 0]

For example, if the process in question is a reflection from a surface,

the coefficient A might represent a scalar reflectivity value ranging
from 0 to 1.
In the fully polarized case, the scalar coefficient A is replaced by
a 4 × 4 Mueller matrix A, so that the new intensity is described via
the matrix operation
Inew = AIold . (2.55)
Thus, the Mueller matrix describes not only how the overall inten-
sity changes but also how the polarization changes. For example, an
optical device represented by the following Mueller matrix A will
transform a beam of radiation having arbitrary polarization into one
that is 100% right-circularly polarized:
 
1 1 0 0
1 0 0 0 0 
A=  . (2.56)
2 0 0 0 0 
1 1 0 0

Problem 2.15: Demonstrate that (2.56) does what is claimed.

46 2. Properties of Radiation

θ
dω

dA

Fig. 2.6: The flux density of radiation carried by a beam in the direction Ω̂ through
a surface element dA is proportional to cos θ = n̂ · Ω̂.

2.7.3 Relationship between Flux and Intensity

We previously defined the flux F as the total power incident on a

unit area of surface. We then defined intensity in terms of a flux
contribution arriving from a very small element of solid angle dω
centered on a given direction of propagation Ω̂. It follows that the
flux incident on, passing through, or emerging from an arbitrary
surface is given by an integral over the relevant range of solid angle
of the intensity.
Let us start by considering the flux emerging upward from a hor-
izontal surface: it must be an integral of the intensity I (Ω̂) over
all possible directions Ω̂ directed skyward; i.e., into the 2π steradi-
ans of solid angle corresponding to the upper hemisphere. There is
one minor complication, however. Recall that intensity is defined in
terms of flux per unit solid angle normal to the beam. For our horizon-
tal surface, however, only one direction is normal; radiation from all
other directions passes through the surface at an oblique angle (Fig.
 Flux and Intensity 47

2.6). Thus, we must weight the contributions to the flux by the co-
sine of the incidence angle relative to the normal vector n̂. For the
upward-directed flux F ↑ , we therefore have the following relation-
ship: Z
F↑ = I ↑ (Ω̂)n̂ · Ω̂ dω. (2.57)
2π
The above expression is generic: it doesn’t depend on one’s choice
of coordinate system. In practice, it is convenient to again introduce
spherical polar coordinates, with the z-axis normal to the surface:

Z 2πZ π/2
F↑ = I ↑ (θ, φ) cos θ sin θ dθdφ , (2.58)
0 0

where we have used (2.49) to express dω in terms of θ and φ.

For the downward flux, we integrate over the lower hemisphere,
so we have

Z 2πZ π
F↓ = − I ↓ (θ, φ) cos θ sin θ dθdφ . (2.59)
0 π/2

Since I is always positive, the above definitions always yield posi-

tive values for F ↑ and F ↓ .
Key fact: For the special case that the intensity is isotropic — that
is, I is a constant for all directions in the hemisphere, then the above
integrals can be evaluated to yield

F = πI . (2.60)

Key fact: The net flux is defined as the difference between

upward- and downward-directed fluxes:

Fnet ≡ F ↑ − F ↓ , (2.61)

which can be expanded as

Z 2πZ π Z
Fnet = I (θ, φ) cos θ sin θ dθdφ = I (Ω̂)n̂ · Ω̂ dω. (2.62)
0 0 4π
48 2. Properties of Radiation

Note, by the way, that the notation used throughout this subsec-
tion implies that we are relating a broadband intensity to a broadband
flux. Identical relationships hold between the monochromatic inten-
sity Iλ and the monochromatic fluxes Fλ↑ and Fλ↓ .

Problem 2.16: If the intensity of radiation incident on a surface is

uniform from all directions and denoted by the constant I, verify that
the total flux is πI, as stated by (2.60). Note that this approximately
describes the illumination of a horizontal surface under a heavily
overcast sky. It also describes the relationship between the flux and
intensity of radiation leaving a surface, if that surface is emitting ra-
diation of uniform intensity in all directions.

Problem 2.17: As noted above, when the radiant intensity incident

on a flat surface is isotropic with intensity I, the resulting flux is πI.
The power intercepted by a circular flat plate of radius r is therefore
P = π 2 r2 I, assuming that it is illuminated from only one side. How
much power is intercepted by a sphere of radius r exposed to the same
source? Hint: There are various ways to reach the same conclusion,
some of which are more cumbersome than necessary. Try to find a
simple but compelling geometric argument.

Problem 2.18: Compute the flux from an overhead spherical sun,

as seen from a planet in an orbit of radius D, given that the sun has
radius Rs and a uniform intensity Is . Make no assumptions about
the size of D relative to Rs . Consider the radius of the planet itself to
be negligible. Use two different methods for your calculation:
(a) Method 1: Integrate the intensity over the solid angle sub-
tended by the sun, with the usual cosine-weighting relative to the
local vertical. You will need to derive an exact expression for the
solid angle subtended by the Sun’s disk for arbitrary D > Rs (see
Fig. 2.5).
(b) Method 2: Compute the flux density emerging from the sur-
face of the sun, translate that into a total power emitted by the sun,
and then distribute that power over the surface of a sphere of radius
D.
Do your two solutions agree?
 Applications 49

Solar Flux S0

RE

Intercepted Flux Φ=S0πRE2

Fig. 2.7: The total flux of solar radiation intercepted by the earth is equal to the
product of the incident flux density S0 and the area of the earth’s shadow.

2.8 Applications to Meteorology, Climatology,

and Remote Sensing
Of fundamental importance to the global climate is the input of en-
ergy from the sun and its spatial and temporal distribution. This
input is a function of two variables: 1) the flux of solar radiation
incident on the top of the atmosphere, and 2) the fraction of that
flux that is absorbed by either the surface or the atmosphere at each
point in the earth-atmosphere system. The second of these depends
in a complex way on distributions of clouds and absorbing gases in
the atmosphere, as well as on the absorbing properties of the sur-
face. These are all issues that will be taken up in the remainder of
this book. The first variable, however, can already be understood in
terms of the material presented in this chapter.

2.8.1 Global Insolation

The first question that may be asked is, how much total solar radia-
tion Φ is incident on the earth’s atmosphere, on average? This ques-
tion is easily solved by computing how much of the Sun’s output is
intercepted by the earth’s disk. That is, given that the mean solar
flux at Earth’s mean distance from the Sun is S0 = 1370 W m−2 ,
what cross-sectional area is presented to that flux by the earth (i.e.,
50 2. Properties of Radiation

how big of a shadow does the earth cast)? The answer of course8 is
A = πR2E , where the mean radius of the earth is R E = 6373 km (Fig.
2.7). Thus
Φ = S0 πR2E = 1.74 × 1017 W (2.63)
This result is consistent with a mean distance of the earth from
the sun (D s ) of 1.496 × 108 km. The reality, however, is that the
earth’s orbit is slightly elliptical, with the Ds varying from 1.47 ×
108 km near January 3 (perihelion) to 1.52 × 108 km on about July
5 (aphelion). Thus, the top-of-the-atmosphere (TOA) solar flux S
varies seasonally from as little as 1330 W m−2 in July to as much as
1420 W m−2 in January.

Problem 2.19: Derive an expression for S as a function of S0 , D s ,

and Ds . Show that ∆S/S0 ≈ −2∆Ds /D s . That is, a positive 1%
change in Ds leads to a negative 2% change in S.

2.8.2 Regional and Seasonal Distribution of Insolation

At the distance of the earth from the sun, there is a more or less
constant flux of solar radiation of S0 = 1370 W m−2 . As noted
above, there is actually some deviation from this value over the
course of the year, owing to the slightly varying distance of the
earth from the sun. Also, the power output P from the sun itself
varies slightly over time, due to factors such as sunspot activity as
well as other longer term variations that are neither well-measured
nor well-understood.
Even ignoring minor variations in S0 itself, it is clear that solar
radiation is not uniformly incident on the earth. The night side of
the earth receives no solar radiation at all. And even on the daylight
side, the flux of solar radiation measured on a unit horizontal area
at the top of the atmosphere depends on the angle of incidence of
the sun. If the sun is directly overhead (solar zenith angle θs = 0),
then the flux is equal to S0 , but if θs > 0, then a unit area normal to
8 Strictly speaking, this is an approximation that is valid only because the earth’s

radius is far smaller than the radius of the earth’s orbit.

 Applications 51

Local Zenith

Solar Flux
θ
s
a =A Shadow Area = A cos θ
Are s
f ace
Su r

Fig. 2.8: The relationship between local solar zenith angle θs and insolation on a
local horizontal plane.

the sun’s rays projects onto a larger area on the earth’s surface (Fig.
2.8). Thus, the solar flux measured on a unit horizontal area is given
by
F = S0 cos θs . (2.64)
Now consider the total insolation [energy per unit area] at the
top of the atmosphere at a single location over the course of a 24-
hour period. This insolation is given by
Z tsunset
W= S0 cos θs (t) dt. (2.65)
tsunrise

As you can probably guess from this expression (and from every-
day experience), W depends on two readily identifiable factors: (1)
the length of the day tsunset − tsunrise , and (2) the average value of
cos θs (t) during the time the sun is up.
At the equator, the length of a day is 12 hours year-round, but
the maximum elevation the sun reaches in the course of the day
varies with the time of year. Twice a year, at the time of the vernal
and autumnal equinox (approximately March 21 and September 21,
respectively), the sun passes directly overhead at noon. At other
times of year, the minimum zenith angle achieved in the course of
the day is equal to the angle of tilt of the earth’s axis toward or away
from the sun, up to a maximum of 23◦ at the time of the summer and
winter solstices (June 21 and December 21, respectively).
At latitudes poleward of 23◦ , the sun is never directly overhead,
and the minimum zenith angle is always greater than zero. Dur-
ing the summer season, the sun can reach a point fairly high in the
sky, whereas in the winter season, the maximum elevation angle is
52 2. Properties of Radiation

much lower. Moreover, the days are longer in the summer hemi-
sphere than in the winter hemisphere. Indeed, poleward of the arc-
tic or antarctic circles, there is a substantial period of time during
the winter when the sun never comes up at all, while during the
corresponding period of high summer, the sun never sets. At the
poles themselves, the situation is very simple: the sun is up contin-
uously for one half of the year, and the solar zenith angle θs is nearly
constant over a 24-hour period.9
The combined effects of the length of day, of the variation in
cos θs , and of the slight variation of the earth’s distance from the
sun on daily insolation (at the top of the atmosphere) are depicted
in Fig. 2.9. Blacked-out areas depict dates and latitudes for which
the sun never emerges above the horizon. The dashed line (“decli-
nation of the sun”) indicates the dates/latitudes at which the noon-
time sun passes directly overhead. Not surprisingly, this curve co-
incides with the location of maximum daily insolation over most of
the year. However, within a week or two of the summer solstice,
the maximum daily insolation is found instead near the pole, where
there is daylight for a full 24 hours and the sun is a relatively high
23◦ above the horizon for the entire day.
If you integrate the daily insolation at a given latitude over the
entire annual cycle and then divide your result by the number of
days in a year, you get the daily average insolation, as depicted by the
heavy curve labeled “Annual” in Fig. 2.10. Also shown is the daily
insolation for the two solstice dates.
In closing, I would like to remind you that the insolation dis-
cussed above describes only the amount of solar radiation incident
at the top of the atmosphere. It is thus an upper bound on the amount
of solar radiation that is available to be absorbed by the earth and
atmosphere. In reality, a significant fraction of this radiation is im-
mediately reflected back to space by clouds, aerosols, air molecules,
and the underlying surface. A good part of the rest of this book is
concerned with the processes that determine how much radiation is
9 Atmospheric refraction allows the sun to be visible from a location on the

earth’s surface when it is actually about 0.5◦ , or approximately the diameter of the
sun’s disk, below the horizon. Thus, the sun rises somewhat sooner and sets some-
what later than would be predicted from geometric considerations alone. There-
fore, the length of continuous daylight at the North Pole (for example) is actually
somewhat longer than the expected six months.
 Applications 53

Daily Average Insolation [W m-2]

NP
24 hr 24 hr darkness
24 hr light
darkness
500
0
0
60N
100
100

200
200
30N
300
300
n
tio
na
Latitude

cli
400
De 400 400
EQ lar
So
300

30S 500 200 500

100

60S
0

24 hr light 24 hr darkness 24 hr light

SP
MAY
JAN

FEB

MAR

APR

May

JUN

JUL

AUG

SEP

OCT

NOV

DEC

Fig. 2.9: Daily average solar flux at the top of the atmosphere, as a function of
latitude and time of year. Contour values are given in units of W m−2
.

absorbed and how much is reflected.

Problem 2.20: Compute, and compare with Fig. 2.9, the daily aver-
age top-of-the-atmosphere insolation [W m−2 ] for the following two
cases: (a) the North Pole at the time of the Northern Hemisphere
summer solstice; (b) the equator at the time of the equinox. Assume
that the solar flux normal to the beam is a constant 1370 W m−2 , and
note that the North Pole is inclined 23◦ toward the Sun at the time of
the solstice.
54

Insolation
600

21 December
500 21 June

400
Insolation [W m-2]

300 Annual

200

100

0
SP 60S 30S EQ 30N 60N NP
Latitude

Fig. 2.10: Daily average solar flux at the top of the atmosphere as a function of
latitude, for the two solstice dates and averaged over a year.
 CHAPTER 3

The Electromagnetic Spectrum

In the previous chapter, we examined how electromagnetic radia-

tion behaves on a purely physical level, without being concerned
yet with its detailed interactions with matter. One important obser-
vation was that we can treat an arbitrary radiation field as a super-
position of many “pure” sinusoidal oscillations. The clearest every-
day example of this is the rainbow: white sunlight interacting with
raindrops is decomposed into the constituent colors red through vi-
olet, each of which corresponds to a narrow range of frequencies.
Radiation associated with a given frequency and trajectory in space
may be analyzed completely independently of all the others.
We also saw that there is no fundamental constraint on the fre-
quency that EM radiation can exhibit, as long as an oscillator with
the right natural frequency and/or an energy source with the mini-
mum required energy is present (recall from Section 2.6 that a single
photon has a specific energy determined by its frequency and that
an oscillator cannot emit less than that minimum amount).
In a vacuum, the frequency or wavelength of a photon is of lit-
tle practical consequence, as it cannot be absorbed, scattered, re-
flected, or refracted but rather is condemned to continue propagat-
ing in a straight line forever, regardless. In the presence of matter
however, the frequency becomes an all-important property and, to

55
56 3. The Electromagnetic Spectrum

a very great degree, determines the photon’s ultimate fate.

There are several reasons why frequency does matter in the atmo-
sphere. First of all, as already mentioned several times, the energy
of a photon is given by E = hν. The rate of absorption and emission
of photons by the atmosphere is strongly dependent on the precise
value of that energy. Among other things, a physical or chemical
event requiring a minimum input of energy ∆Emin cannot be initi-
ated by a photon with a frequency of less than νmin = ∆Emin /h. Fur-
thermore, the quantum mechanical behavior of matter at the molec-
ular level imposes an even stronger constraint in many cases: to
be absorbed, the energy of a photon must almost exactly match a
certain well-defined set of values associated with allowable energy
levels in that molecule. We will examine these issues in considerable
detail in Chapter 9.
Another reason arises from the wave nature of radiation, which
comes to the forefront when radiation is scattered or reflected by
particles or surfaces. Such interactions arise primarily when the di-
mensions of a particle are comparable to or larger than the wave-
length. Thus, radiation in the visible band is rather weakly scattered
by air molecules but strongly scattered by cloud droplets. Longer
wavelengths in the microwave band (e.g., radar) are negligibly scat-
tered by cloud droplets but rather strongly by raindrops and hail-
stones. Longer wavelengths still (e.g., AM radio, with wavelengths
of order 102 m) may propagate unimpeded through any kind of
weather but may be diffracted around hills and reflected by deep
layers of ionized gases in the extreme upper atmosphere.

3.1 Frequency, Wavelength and Wavenumber

The most fundamental characteristic of a harmonic electromagnetic
field is its frequency ν = ω/2π, which has units of cycles per sec-
ond, or Hertz (Hz). Regardless of where you are and what other
processes affect it, radiation with frequency ν will always have that
frequency until such time as it is absorbed and converted into an-
other form of energy1 .
1 This assumes that you, the observer, are at a fixed distance from the source.

Otherwise the frequency will be shifted by the Doppler effect.

 Major Spectral Bands 57

In practice, it is usually more convenient to specify the wave-

length λ rather than the frequency ν. This is because the frequen-
cies of interest to most atmospheric scientists tend to be numerically
large and unwieldy. The two parameters are related by

c = λν . (3.1)

Note that this relationship is valid for the wavelength in a vacuum.

Inside a medium like air or water, the phase speed of radiation is
somewhat slower than c and the actual wavelength is correspond-
ingly shorter. The dependence of the actual wavelength on the
index of refraction of the medium is important for understanding
some effects such as refraction. Normally, if we refer to wavelength
without further qualification, we mean wavelength in a vacuum.
For atmospheric radiation, wavelength is most commonly ex-
pressed using one of the following units, whichever is most con-
venient: nanometers (nm = 10−9 m), micrometers or microns (µm
= 10−6 m), or centimeters (cm = 10−2 m). Other units, such as the
Angstrom (10−10 m) are no longer widely used by meteorologists.
The description preferred by some specialists is neither wave-
length nor frequency but wavenumber ν̃, which is just the reciprocal
of wavelength:

1 ν
ν̃ = = . (3.2)
λ c

Wavenumber is usually stated in units of inverse centimeters

(cm−1 ).

3.2 Major Spectral Bands

The electromagnetic spectrum spans an enormous range of frequen-
cies, from essentially zero to extremely high frequencies associated
with energetic photons released by nuclear reactions. As a matter of
convention, the spectrum has been subdivided by scientists and en-
gineers into a few discrete spectral bands. The frequency and wave-
length boundaries of the major spectral bands are given in Table 3.1
58

Table 3.1: Regions of the electromagnetic spectrum

Region Spectral range Fraction of Remarks

solar output

X rays λ < 0.01 µm Photoionizes all

species; absorbed in
upper atmosphere

Extreme UV 0.01 < λ < 0.1 µm 3 × 10−6 Photoionizes O2

and N2 ; absorbed
above 90 km

Far UV 0.1 < λ < 0.2 µm 0.01% Photodissociates

O2 ; absorbed above
50 km

UV-C 0.2 < λ < 0.28 µm 0.5% Photodissociates O2

and O3 ; absorbed
between 30 and 60
km

UV-B 0.28 < λ < 0.32 µm 1.3% Mostly absorbed by

O3 in stratosphere;
responsible for
sunburn

UV-A 0.32 < λ < 0.4 µm 6.2% Reaches surface

Visible 0.4 < λ < 0.7 µm 39% Atmosphere mostly

transparent

Near IR 0.7 < λ < 4 µm 52% Partially absorbed,

mainly by water
vapor

Thermal IR 4 < λ < 50 µm 0.9% Absorbed and

emitted by water
vapor, carbon
dioxide, ozone, and
other trace gases

Far IR 0.05 < λ < 1 mm Absorbed by water

vapor

Microwave λ >1 mm Clouds

semi-transparent
 59

100 m
HF 10 MHz
10 m Radio
VHF 100 MHz
1m

UHF 1 GHz
Microwave
10 cm

SHF 10 GHz
1 cm

EHF 100 GHz

1 mm
Far 1 THz
IR
0.1 mm

Thermal 10 THz
IR
10 µm Infrared

Near 100 THz

IR
1 µm
Visible
1000 THz
Ultraviolet
0.1 µm

Fig. 3.1: The electromagnetic spectrum.

60 3. The Electromagnetic Spectrum

and Fig. 3.1. It is important to understand that there is nothing spe-

cial about the precise frequencies defining the boundaries between
bands; in most cases, these boundaries were decided more or less ar-
bitrarily and have no real physical significance. There is for example
no abrupt change in the behavior of radiation as one crosses from
the microwave to the infrared band in the vicinity of 1 mm wave-
length. The exception of course is the visible band, whose bound-
aries are defined by the range of wavelengths (approximately 0.4 to
0.7 µm) that the normal human eye can see. Other animal species
might have defined this band differently. Many insects, for exam-
ple, can see well into the ultraviolet band.
Note that there are three rather distinct ways in which a par-
ticular spectral band can make itself “interesting” to atmospheric
scientists:

Diabatic heating/cooling - As pointed out in the introduction, ra-

diative transfer is one of the most important mechanisms of
heat exchange in the atmosphere, and is the sole mechanism
for heat exchange between the earth and the rest of the uni-
verse. For reasons that will become clearer later, not all spec-
tral bands contribute significantly in this category.

Photochemistry - Many of the chemical reactions that take place in

the atmosphere, including those that produce smog, as well
as some that help cleanse the air of pollutants, are driven
by sunlight. In addition, the existence of the ozone layer is
a direct result of photochemical processes. The photon energy
E = hν is a crucial factor in determining which spectral bands
are “players” in atmospheric photochemistry.

Remote sensing - Any frequency of radiation that is absorbed,

scattered or emitted by the atmosphere can potentially be ex-
ploited for satellite- or ground-based measurements of atmo-
spheric properties, such as temperature, humidity, the concen-
tration of trace constituents, and many other variables.

In this book, we shall restrict our attention to radiative processes

relevant primarily to the troposphere and stratosphere. With this
constraint in mind, we may now undertake a brief survey of the
major spectral bands.
 Major Spectral Bands 61

3.2.1 Gamma Rays and X-Rays

Gamma rays and X-rays, which are associated with wavelengths
shorter than ∼ 10−2 µm, are usually produced by nuclear decay,
nuclear fission and fusion, and other reactions involving energetic
subatomic particles. The most energetic of photons, gamma and
X-ray radiation can easily strip electrons from, or ionize, atoms and
decompose chemical compounds. As such, ionizing radiation poses
significant hazards to life. It is therefore fortunate that the strongest
natural sources are extraterrestrial — so-called cosmic rays — and
thus affect primarily the upper levels of the atmosphere. The in-
tensity of gamma and X-ray radiation arriving at the top of the at-
mosphere is typically reduced by well over half for each 100 mb
of atmosphere that it traverses, so that very little of this radiation
makes it to the lowest levels. But airline passengers are exposed to
nonnegligible levels of cosmic radiation.
In the lower troposphere, most natural radiation observed in
this spectral band is traceable to radioactive materials in the earth’s
crust, such as uranium and its daughter isotopes. Although such
sources are widely distributed, most are (thankfully) rather weak.
The gamma and X-ray bands are the only bands that have no
major significance for any of the three processes identified in the
previous section. Fluxes of radiation in these bands are not large
enough to have a measurable effect on the heating or cooling of the
lower and middle atmosphere. For various reasons, including the
absence of strong natural terrestrial sources and the relatively strong
attenuation of ultrashort wavelength radiation by the atmosphere,
remote sensing of the troposphere and stratosphere is not a practical
proposition in these bands. Finally, although these types of radia-
tion can potentially participate in chemical reactions, their role is
minor compared with that of ultraviolet radiation (see below). In
the view of lack of strong relevance of this band to meteorology, we
will not consider it further in this book.

3.2.2 Ultraviolet Band

The ultraviolet (UV) band occupies the range of wavelengths from
approximately 0.01 µm on the X-ray side to approximately 0.4 µm
on the visible-light side. The sun is the sole significant source of
62 3. The Electromagnetic Spectrum

natural UV radiation in the atmosphere. However, the fraction of

sunlight at the top of the atmosphere that falls in this band is small,
only a few percent of the total power output. Nevertheless, this
contribution is very important. The UV band is further divided into
the following sub-bands:

UV-A extends from 0.4 down to 0.32 µm. Radiation in this sub-
band is a significant component of sunlight, comprising close
to 99% of the total solar UV radiation that reaches sea level.
Although UV-A radiation is invisible to the human eye, it
stimulates fluorescence (the emission of visible light) in some
materials — e.g., “Day-Glo” markers, highway safety cones,
and yellow tennis balls. So-called “black lights” used with
fluorescent posters are artificial sources of UV-A radiation. Al-
though the wavelengths are shorter, and therefore more ener-
getic, than those of visible light, UV-A is still relatively innocu-
ous with respect to living organisms. This is fortunate because
the atmosphere is rather transparent to UV-A.

UV-B extends from 0.32 down to 0.280 µm. Because of its even
shorter wavelength, its photons are energetic enough to ini-
tiate photochemical reactions, including injury of tissues (e.g.,
sunburn) and even damage to cellular DNA, leading to in-
creased risk of skin cancer in exposed individuals. Fortu-
nately, most UV-B (approximately 99%) is absorbed by ozone
in the stratosphere. However, thinning of the ozone layer by
human-manufactured chemicals is believed to be responsible
for a significant increase in the amount of UV-B now reaching
the surface.

UV-C extends from 0.280 to ∼0.1 µm. The most energetic UV sub-
band, virtually all UV-C radiation is absorbed in the meso-
sphere and uppermost stratosphere, where much of its energy
is expended on the dissociation of O2 into atomic oxygen. The
remainder is absorbed by ozone.

UV radiation is interesting in all three of the respects outlined

earlier. As we have already mentioned, it is a major player in at-
mospheric photochemistry. Also, satellite remote sensing of ozone
and other stratospheric constituents is possible in this band. Finally,
 Major Spectral Bands 63

Table 3.2: Relationship between color and wavelength

Wavelength interval (µm) Color

0.39–0.46 Violet
0.46–0.49 Dark Blue
0.49–0.51 Light Blue
0.51–0.55 Green
0.55–0.58 Yellow-Green
0.58–0.59 Yellow
0.59–0.62 Orange
0.62–0.76 Red

the absorption of solar UV radiation by ozone is a major diabatic

heating term in the stratosphere and mesosphere.

3.2.3 Visible Band

The visible band extends from approximately 0.4 µm to 0.7 µm. In
addition to its obvious importance for human vision, its significance
for the atmosphere cannot be overstated, despite the fact that it oc-
cupies a surprisingly narrow slice of the EM spectrum.
First, through an interesting coincidence, the visible band includes
the wavelength of maximum emission of radiation by the sun. (Fig. 3.2)
In fact, close to half of the total power output of the sun falls in this
narrow band.
Second, the cloud-free atmosphere is remarkably transparent to all vis-
ible wavelengths. We may take this for granted, but for no other major
spectral band is the atmosphere as uniformly transparent (Fig. 3.3).
This means that the absorption of visible solar radiation occurs pri-
marily at the surface of the earth rather than within the atmosphere
itself. Thus, the atmosphere is largely heated from below and only
secondarily by direct absorption of solar radiation. The thermal
structure of the atmosphere would likely be quite different if the at-
mosphere were less transparent to this component of the solar flux.
Clouds are remarkably reflective in the visible band. Again,
this might seem obvious, but it’s not true for many other spectral
bands. The global distribution of cloud cover has a huge influ-
64 3. The Electromagnetic Spectrum

2.5

6000 K
5500 K
Solar Irradiance [kW m-2 µm-1]

1.5

0.5

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Wavelength [µm]

Fig. 3.2: Spectrum of solar radiation at the top of the atmosphere, at moderate
spectral resolution. Dashed curves represent emission from an ideal blackbody at
the indicated temperatures.

ence on the fraction of total solar radiation that gets absorbed by

the earth-atmosphere system rather than being reflected back out to
space.
Satellites with imaging capabilities in the visible band are able
to easily detect and classify clouds. In the absence of clouds, visible
imagers are able to map surface features, vegetation types, ocean
color (related to biological productivity) and many other variables.

3.2.4 Infrared Band

The infrared (IR) band extends from wavelengths of approximately
0.7 µm up to approximately 1000 µm or 1 mm. This rather broad
range (over three decades of wavelength) encompasses a rich vari-
ety of absorption and emission features in the atmosphere. IR radi-
ation is enormously important as the means by which energy is ex-
changed between lower and upper levels of the atmosphere and be-
tween the earth-atmosphere system and outer space. Not only does
 Major Spectral Bands 65

λBλ(T) (normalized)

(a) 6000 K 288 K

0.1 0.15 0.2 0.3 0.5 1 1.5 2 3 5 10 15 20 30 50 100

100

80
Absorption (%)

60
(b)
40

20

0
0.1 0.15 0.2 0.3 0.5 1 1.5 2 3 5 10 15 20 30 50 100
Wavelength [µm]

Fig. 3.3: Overview of the relationship between solar and terrestrial emission and
the transmission properties the atmosphere. a) Normalized blackbody curves cor-
responding to the approximate temperature of the sun’s photosphere (6000 K) and
a typical terrestrial temperature of 288 K. b) A coarse-resolution depiction of the
absorption spectrum of the cloud-free atmosphere.)

the steady-state climate of the earth depend heavily on the absorp-

tive and emissive properties of the atmosphere in the IR band, but
we now believe that the climate can change in response to human-
induced increases in IR-absorbing trace constituents (“greenhouse
gases”) in the atmosphere, such as water vapor, carbon dioxide,
methane, and chlorofluorocarbons (CFCs).
Because so many major and minor constituents of the atmo-
sphere have distinctive (and often very strong) absorption features
in the IR band, there are countless ways to exploit this band for re-
mote sensing of temperature, water vapor, and trace constituents.
On the other hand, the IR band is unimportant for atmospheric
photochemistry, because photon energies are below the threshold
required to dissociate most chemical compounds.
Atmospheric scientists tend to subdivide the IR band into three
sub-bands: the near IR band, the thermal IR band, and the far IR
band.
The near IR band is in one sense a continuation of the visible
band, in that the primary source of this radiation in the atmosphere
is the sun. It extends from 0.7 to 4 µm. Approximately half of the
sun’s output is found in this band, so that all but 1% of solar radi-
66 3. The Electromagnetic Spectrum

ation incident on the top of the earth’s atmosphere is accounted for

by the UV, visible, and near IR bands together.
Unlike the case for the visible band, however, the atmosphere
is not uniformly transparent to all near-IR wavelengths but rather
exhibits a number of significant atmospheric absorption features.
Thus, a moderate fraction of near-IR radiation from the sun is ab-
sorbed by the atmosphere in this band. For some wavelengths, e.g.,
near 1.3 µm, absorption is nearly total.
The range from 4 µm to 50 µm encompasses what we will re-
fer to as the thermal IR band. Different sources quote various upper
wavelength bounds on the thermal IR band, some as low as 15 µm.
We have chosen the 50 µm bound because significant thermal en-
ergy exchanges via radiative transfer in the atmosphere occur up
to approximately this limit. The thermal IR band is “where the ac-
tion is” in view of both the magnitude of the energy exchanges and
the enormous complexity of the atmospheric absorption spectra in
this band. We will have much more to say about this band in later
chapters.
For our purposes, the far IR band represents wavelengths be-
tween about 50 µm and 1000 µm (1 mm). Energy transfer in the
atmosphere at these wavelengths is insignificant relative to that as-
sociated with the thermal IR, near IR, and visible bands. There are
some potential applications of the far IR band to remote sensing, es-
pecially of cirrus clouds, but otherwise this region of the spectrum
is relatively uninteresting to meteorologists.

3.2.5 Microwave and Radio Bands

Moving through the EM spectrum toward longer wavelengths
(lower frequencies), one leaves the far IR band and enters the mi-
crowave band at a wavelength of about 1 mm, or at a frequency of
about 300 GHz (GHz = gigahertz = 109 Hz). The lower bound (in
frequency) is often taken to be around 3 GHz, or 10 cm wavelength.
Thus, the microwave band encompasses two decades of frequency.
At lower frequencies still, and continuing down to zero, we have
the radio band. Note that both for historical reasons and because the
numbers are low enough to be manageable, it is most common to
use frequency rather than wavelength when describing microwave
 Major Spectral Bands 67

and radio band radiation.

From an engineering point of view, one of the distinguishing
characteristics of the radio band is that the frequencies involved are
low enough to be amenable to generation, amplification, and detec-
tion using traditional electronic components and circuits. By con-
trast, the much higher frequencies and shorter wavelengths of IR
and visible radiation require mirrors, diffraction gratings, and/or
lenses. The microwave band occupies a gray area, as many of the
components in microwave circuits have a quasi-optical character —
e.g., waveguides, resonant cavities, feedhorns, and parabolic reflec-
tors.
The microwave band has risen greatly in prominence in recent
years for its role in remote sensing of the atmosphere and surface.
Radar, which was first developed during World War II, is now the
principal means by which meteorologists monitor severe weather
and study the dynamics of convective cloud systems. Satellites with
sensors operating in the microwave band have proliferated since the
mid-1970s and are now a very important component of our weather
satellite programs, both for research and operationally.
The utility of the microwave band is greatly enhanced by the
relative transparency of clouds, especially at frequencies well below
100 GHz. The properties of the surface and of the total atmospheric
column — can be observed from space under all weather conditions
except rainfall.
The radio band, which by some definitions includes the mi-
crowave band, continues down to zero frequency. Frequencies
lower than around 3 GHz tend to interact very weakly with the
atmosphere and therefore have only limited applicability to atmo-
spheric remote sensing. Also, because of the long wavelengths in-
volved, it is difficult to achieve good directionality with antennas of
managable size (especially on satellites).
Two notable examples of remote sensing in the radio band do
bear mentioning: 1) ground-based Doppler wind profilers oper-
ating near 915 MHz, which observe scattering from turbulence-
induced fluctuations in atmospheric density and humidity, and 2)
lightning detection systems, which are sensitive to low-frequency
“static” emitted by lightning discharges. Apart from these cases,
radio wavelengths are of very limited interest to meteorologists.
68 3. The Electromagnetic Spectrum

3.3 Solar and Terrestrial Radiation

In the previous section, we surveyed the entire electromagnetic
spectrum with an eye toward outlining the relevance of each ma-
jor band to atmospheric science. The two most important facts to
emerge from this survey are the following:

• Over 99% of the energy radiated by the sun and incident on

the top of the earth’s atmosphere is accounted for by just three
bands spanning wavelengths from 0.1 µm to 4 µm: the ultravi-
olet band contributes a few percent, with the remainder more
or less evenly split between the visible band and the near-
infrared band. We collectively refer to these bands as solar or
shortwave radiation. Although solar emission in other bands
may have some significance for remote sensing (e.g., sunglint
from the ocean surface in the microwave band), it is insignifi-
cant for the energy budget of the atmosphere.

• Over 99% of the radiative energy emitted by the earth and

atmosphere is found in the thermal infrared band from 4–
100 µm. We will often refer to radiation in this band as ter-
restrial or longwave radiation. Emission in other bands (prin-
cipally the far-IR and microwave bands) may be important
for remote sensing but is essentially irrelevant for the atmo-
spheric energy budget.

It is an interesting and convenient coincidence that a wave-

length of approximately 4 µm cleanly separates the band containing
most solar radiation from that containing most terrestrial emission
(Fig. 3.3). For a narrow range of wavelengths in the vicinity of 4 µm,
it may sometimes be necessary to consider both terrestrial and solar
sources, but for most wavelengths it is just one or the other. The
physical reasons for this separation will be addressed in Chapter 5;
for now it is sufficient to point out that disparate temperatures of
the sources (approximately 6000 K for the sun versus 250–300 K for
the earth and atmosphere) are responsible.
 Applications 69

Problem 3.1: For the given electromagnetic waves in a vacuum,

compute the frequency ν in Hz, the wavenumber ν̃ in cm−1 , and the
wavelength λ in µm. Also identify the spectral band.
a) λ = 0.0015 cm, b) ν = 37 GHz, c) ν̃ = 600 cm−1 , d) λ = 300 nm,
e) ν = 3 × 1014 Hz, f) ν̃ = 10000 cm−1 .

3.4 Applications to Meteorology, Climatology,

and Remote Sensing
3.4.1 UV Radiation and Ozone
The Ozone Layer
The absorption of radiation by way of molecular photodissociation
was briefly mentioned in Section 2.6. It turns out that this process
has easily observable consequences for the atmosphere and, for that
matter, for all of life on Earth. In particular, UV-C radiation is re-
sponsible for dissociating molecular oxygen, according to the reac-
tion
O2 + hν (λ < 0.2423 µm) → O + O. (3.3)
The large amount of molecular oxygen in the atmospheric column
absorbs most solar radiation at wavelengths shorter than 0.24 µm
by this mechanism.
The free oxygen atoms from the above reaction then can combine
with O2 to form ozone according to the reaction

O + O2 + M → O3 + M, (3.4)

where M is any third molecule or atom (required in order to carry

away the energy released by the above reaction).
It is fortunate for us that this second reaction occurs, because or-
dinary oxygen by itself would continue to allow dangerous UV-C
and UV-B radiation with wavelengths between 0.24 and 0.32 µm to
reach the surface, posing a deadly hazard to life. But ozone hap-
pens to absorb strongly between about 0.2 and 0.31 µm via elec-
tronic transitions. It therefore “mops up” most of whatever UV-B
70 3. The Electromagnetic Spectrum

and UV-C was not absorbed via (3.3). Of course it does little for UV-
A radiation, with wavelengths longer than 0.32 µm. But radiation
in this band is relatively innocuous, except perhaps when it is used
to illuminate velvet blacklight posters of Elvis.
There remains a small sliver of the UV-B band between about
0.31 and 0.32 µm that manages to reach the surface without com-
plete absorption; it is precisely this narrow sliver that is primarily
responsible for sunburn. Lately, there has been considerable con-
cern over observed declines in ozone layer density. If the decline
continues, then the resulting widening of this narrow UV-B window
could have serious consequences for life on Earth.
In the very process of absorbing harmful shortwave UV ra-
diation, the ozone layer influences our environment in another
very important way. The solar energy that is absorbed by ozone
warms the atmosphere at those levels to a much higher temperature
than would be the case without the presence of ozone. Have you
ever wondered why temperature increases with height in the strato-
sphere, reaching a maximum at the stratopause before decreasing
again in the mesophere? The ozone layer is responsible!
In an atmosphere without free oxygen, and therefore without
ozone, the temperature structure would be much simpler: we’d
have a very deep troposphere (temperature generally decreasing
with height) transitioning directly to the thermosphere. The strato-
sphere and mesosphere would be missing. This is in fact what you
find on Mars, whose atmosphere consists mainly of CO2 . On Earth,
the temperature structure of the lower stratosphere serves as a very
important “lid” on tropospheric convection and other circulations.
If you already have some background in atmospheric dynamics, try
to imagine how different our weather might be if the tropopause
were near 50 km altitude rather than its present 5-15 km!

Photochemical Smog

We have just surveyed the role of oxygen and ozone in the middle
atmosphere (stratosphere and mesosphere) in absorbing UV-C and
UV-B radiation. The UV-C, we saw, is mostly absorbed by photodis-
sociation of O2 ; the UV-B was then mostly absorbed by the resulting
O3 . This leaves mainly UV-A radiation to reach the troposphere.
 Applications 71

Although less energetic than UV-B and UV-C radiation, UV-A

radiation is a key player in tropospheric chemistry. Among other
things, photochemical reactions involving organic molecules (e.g.
unburned fuel vapors) and nitrogen oxides (produced by high tem-
peratures in automobile engines) can lead to the formation of ozone
in surface air. Although ozone in the stratosphere is highly desirable
because of its UV-blocking characteristics, it is considered a serious
pollutant in near-surface air where we live, because it is a strong
chemical oxidant which attacks most organic substances, including
the lining of your lungs. Ozone is thus one of the main ingredients
of photochemical smog.
In its most basic form, the chemical sequence leading to ozone
pollution goes like this:

1. Automobile engines and other industrial processes emit pri-

mary pollutants, which include unburned hydrocarbons (a.k.a.
volatile organic compounds, or VOC) and nitrogen oxides —
nitrogen monoxide (NO) and nitrogen dioxide (NO2 ), collec-
tively known as NOx . The NO oxidizes, further increasing the
NO2 concentration.

2. The NO2 is photolyzed (photodissociated) according to the re-

action
NO2 + hν (λ < 0.4 µm) → NO + O⋆ , (3.5)
where O⋆ is a highly reactive free oxygen atom that immedi-
ately combines with an ordinary oxygen molecule to form one
of several secondary pollutants2 , in this case ozone:

O2 + O ⋆ + M → O3 (3.6)

3. A third reaction completes the cycle, bringing us back to our

starting point:
O3 + NO → NO2 + O2 (3.7)

Graphically, the cycle is shown in Fig. 3.4. Although each pass

through the cycle doesn’t yield a net increase of ozone, the fact that
2 Another major secondary pollutant, also produced by photochemical action

on the primary emissions, is peroxyacetyl nitrate, or PAN.

72 3. The Electromagnetic Spectrum

NO2

photolysis hν

NO O*
O2

O3

Fig. 3.4: Photolysis of NO2 and generation of O3 .

there is (during the daytime) a continuous input of UV radiation im-

plies that there will be constant generation of new ozone as earlier
ozone molecules are destroyed, leading to a finite steady state con-
centration in the atmosphere. Although we won’t go into the pro-
cess here, one role of VOC molecules is to create additional sources
of NO and NO2 , which increases the equilibrium concentration of
O3 .
The location in the U.S. most stereotypically associated with
photochemical smog is the Los Angeles basin, with its high quota
of sunshine, high concentration of automobiles, and shallow pool
of stagnant air hemmed in by mountains.

The Hydroxyl Radical

So far we have seen an example of a photochemical process that
is “good” (ozone production in the stratosphere) and one that is
“bad” (ozone production in the troposphere). UV-A and visible ra-
diation are responsible for other photochemical processes in the tro-
posphere besides smog, not all of which yield harmful byproducts.
One in particular is quite beneficial: the production of short-lived
hydroxyl (OH) radicals by the following pair of reactions:

O3 + hν (λ < 0.34 µm) → O2 + O⋆ (3.8)

H2 O + O⋆ → 2OH (3.9)
 Applications 73

where the ozone in the first of these equations occurs at low levels
even in unpolluted air. What is interesting about the hydroxyl radi-
cal is that it is highly reactive and acts to break down a wide range
of undesirable pollutants in the atmosphere, such as carbon monox-
ide and methane. In fact, OH radicals are sometimes referred to as
the atmosphere’s “detergent.” Without the daily action of hydroxyl
radicals, the air would be much dirtier, on average, than it is.
Among the very few pollutants that are impervious to break-
down by OH radicals are the synthetic compounds known as chlo-
rofluorocarbons (CFCs). These molecules are so exceptionally sta-
ble that they can persist in the atmosphere for as long as it takes
— often a year or more — for them to get circulated from the tropo-
sphere into the stratosphere. Only there are CFC molecules exposed
to enough UV-B and UV-C radiation to get broken down. Unfortu-
nately, one byproduct of the photolysis is a free chlorine atom. A
single Cl atom has the capacity to catalytically destroy many ozone
molecules. Therefore, the release of CFCs into the atmosphere over
the past few decades has led to a marked decrease in the steady state
concentration of ozone in the stratosphere. This is especially the
case in the polar regions during springtime, where complex chemi-
cal processes involving polar stratospheric cloud particles and sun-
light greatly accelerates the destruction of ozone. As noted above,
this thinning of the ozone layer is a point of great concern, due to
the hazards of UV-B radiation to life. Fortunately, progress has been
made in slowing the release of CFCs into the atmosphere, and the
ozone layer is now expected to gradually recover over the next few
decades.
 CHAPTER 4

Reflection and Refraction in a Homogeneous Medium

In this chapter, we undertake our first quantitative examination of

the interaction of radiation with matter. We will focus initially on
the behavior of EM waves when they encounter, and propagate
through, a homogeneous medium.
By “homogeneous,” we mean a medium that is smooth and uni-
form on scales comparable to the wavelength of the radiation. Wa-
ter, glass, air, red wine, maple syrup, liquid mercury, and solid gold
are all effectively homogeneous with respect to radiation in the visi-
ble, infrared and microwave bands, because the size and spacing of
the individual molecules and other irregularities are much smaller
than the wavelength of the radiation. If this were not the case, then
glass and water, for example, would have a cloudy or milky ap-
pearance rather than being perfectly transparent to the eye. In fact,
milk looks “milky” precisely because of the presence of suspended
particles that are larger than the wavelengths of visible light.
To radiation in the X-ray and gamma bands, these and all
other substances appear quite lumpy, or inhomogeneous, because the
wavelength of the radiation is comparable to or smaller than the
spacing between the molecules. Likewise, centimeter- and meter-
scale turbulent eddies and fluctuations in humidity may make even
the atmosphere appear inhomogeneous with respect to radio and

71
72 4. Reflection and Refraction

microwave radiation. On the other hand, consider a cloud that is

made up of 10 µm-diameter water droplets: it behaves like a ho-
mogeneous medium with respect to the long wavelengths of mi-
crowave radiation even though it is quite inhomogeneous with re-
spect to visible and infrared radiation.

At the microscopic level, interactions of radiation with even ho-

mogeneous media are quite complex, involving the constructive
and destructive interference of waves scattered by the countless in-
dividual atoms or molecules in the medium. From a macroscopic
point of view, however, all of this complexity is hidden in a sin-
gle parameter, the complex index of refraction N = nr + ni i of the
medium. As discussed in Chapter 2, the real part nr controls the ef-
fective phase speed of EM waves propagating through the medium
while the imaginary part ni describes the rate of absorption of the
wave.

The phase speed of radiation in a medium may seem uninter-

esting to the uninitiated; after all, even in the densest medium, it’s
too fast to directly observe without highly specialized equipment.
However, it is the sudden change in this phase speed at boundaries
between media such as air and water that gives rise to the phenom-
ena of reflection and refraction.

It is important to keep in mind that N is not a constant for

any substance but rather depends strongly on wavelength and, to
a lesser degree, temperature, pressure and other variables. It is the
variation of nr with wavelength in raindrops that gives rise to rain-
bows; it is a sharp variation of ni with wavelength that makes wine
red.

It is also worth pointing out in passing that, in any real mate-

rial, nr and ni are not free to vary independently of one another but
rather are tightly coupled to one another via the so-called Kramer-
Kronig relations. Although the equations describing these relation-
ships, and their physical interpretation, are beyond the scope of this
book, suffice it to say that knowledge of the value of ni over the full
range of frequency is sufficient to compute nr at any frequency, and
vice versa. See BH83, Chapter 2 for a detailed discussion.
 A Closer Look at N 73

4.1 A Closer Look at N

4.1.1 The Real Part
Recall from Chapter 2 that, in a nonabsorbing medium, the imagi-
nary part of the index of refraction ni = 0 and N = nr , where nr is
real and the phase speed c in the medium is given by

c
c = , (4.1)
nr

where c is the speed of light in a vacuum. For virtually all real

substances nr > 1, so that the phase speed of light in a physical
medium is somewhat slower than that in a vacuum.
The values of nr for both water and ice are given as functions
of wavelength in the top panel of Fig. 4.1. In the visible band (0.4 <
λ < 0.7), the value of nr for water is approximately 1.33, though it
is slightly larger for the shorter wavelengths (e.g. blue and violet)
than for longer wavelengths (red and orange).
For air at standard temperature and pressure, nr ≈ 1.0003 in the
visible band and is thus so close to unity that the difference between
air and a vacuum may be ignored for some purposes. Nevertheless,
the difference does sometimes matter: it is variations in nr associ-
ated with changes in atmosphere density that give rise to mirages
and the twinkling of stars, for example.

4.1.2 The Imaginary Part

When the imaginary part ni of the index of refraction is nonzero, ab-
sorption of an EM wave occurs as it passes through the medium. In
fact, ni is sometimes referred to as the absorption index. Values of ni
for water and ice are shown as functions of wavelength in the bot-
tom panel of Fig. 4.1. Note that ni for both substances is extremely
small in the visible band but increases sharply as you move into
either the ultraviolet or infrared bands.
In Section 2.5, we found that the rate of power attenuation per
unit distance is given by the absorption coefficient β a (dimensions of
74 4. Reflection and Refraction

inverse length), which is related to ni by

4πni
βa = , (4.2)
λ

where λ is the wavelength in a vacuum. It was shown that, for an

initial intensity I0 at position x = 0 within the medium and propa-
gating in the x direction,

I ( x ) = I0 e− βa x . (4.3)

(a) Index of Refraction of Water and Ice (Real Part)

1.8
1.7 Water (10o C)
Ice (-5o C)
1.6
1.5
nr

1.4
1.3
1.2
1.1
1
0.1 0.15 0.2 0.3 0.5 0.7 1 1.5 2 3 4 5 7 10 15 20 30 50
Wavelength [µm]
(b) Index of Refraction of Water and Ice (Imag. Part)
1
0.1
0.01
0.001
0.0001
ni

1e-05
1e-06
1e-07
Water (10o C)
1e-08 Ice (-5o C)
1e-09
0.1 0.15 0.2 0.3 0.5 0.7 1 1.5 2 3 4 5 7 10 15 20 30 50
Wavelength [µm]

Fig. 4.1: The complex index of refraction of water (solid curves) and ice (dotted
curves). (a) The real part, which largely influences the reflection and refraction
properties of a medium. (b) The imaginary part, which determines absorption
within the medium.
 A Closer Look at N 75

We can rewrite this as

I (x)
t( x ) ≡ = e− βa x , (4.4)
I0

where the transmittance t is the fraction of radiation that survives

the trip over the distance x. That fraction of the initial radiation not
transmitted must have been absorbed and therefore converted to
another form of energy, such as heat.
The above equation is the first instance we have encountered of
a very important relationship called Beer’s law.1 It states that the
intensity of a beam of monochromatic radiation falls off exponen-
tially as it traverses a uniform medium. The rate at which it falls off
is proportional to β a .
It is important to note that even very small values of ni can imply
strong absorption within a medium. For example, if a 1 mm thick
sheet of a material transmits only 1% of radiation having a wave-
length of 0.5 µm, equations (4.2) and (4.3) imply ni = 1.8 × 10−4 .
We can also define a quantity called the penetration depth, which
is the value of x for which t( x ) = e−1 ≈ 37%. It is just the reciprocal
of the absorption coefficient:
1 λ
D≡ = . (4.5)
βa 4πni
The penetration depth D for water and ice is shown as a function
of wavelength in Fig. 4.2, based on (4.5) applied to the values of ni
shown in Fig. 4.1. Whereas D is on the order of tens of meters in the
visible band, it rapidly falls to only millimeters or even micrometers
in the thermal infrared band!

Problem 4.1: Show that the total transmission ttotal of radiation

through two sheets of material stacked together is equal to the prod-
uct of the individual transmittances t1 and t2 . Ignore reflections.

1 In fairness to other scientists who derived physically equivalent relationships,

Beer’s law is cited by some authors as the Beer-Bouguer-Lambert law. In this book,
we will sacrifice fairness for convenience and opt for the traditional short form.
Other authors avoid the thorny issue of attribution altogether and refer to it as the
extinction law.
76 4. Reflection and Refraction

Radiation Penetration Depth in Water and Ice

100 m
10 m Water (10o C)
Ice (-5o C)
1m
Depth [m]

10 cm
1 cm
1 mm
0.1 mm
10 µm
1 µm
0.2 0.3 0.5 0.7 1 1.5 2 3 4 5 7 10 15 20 30 50
Wavelength [µm]

Fig. 4.2: The penetration depth of radiation in water and ice.

4.1.3 The Dielectric Constant†

In some situations, it is convenient to utilize the complex dielectric
constant  of a medium (also known as the relative permittivity) rather
than the complex index of refraction N to describe the absorption
and refraction properties of a substance at a particular wavelength.
There is no need to be dismayed by the introduction of yet another
parameter, as N and  are very closely related.2 In fact, for nonmag-
netic materials (µ = µ0 ),
ε
= = N2, (4.6)
ε0

where the permittivity of the medium ε and the permittivity of

free space ε 0 were previously introduced in connection with (2.36).
Please keep in mind that, despite the name, the dielectric “constant”
is really a function of frequency and, to a lesser extent, of variables
like temperature and pressure.
Expanding in terms of the real and imaginary components, we
find
 = (nr + ni i )2 = n2r + 2ni nr i − n2i , (4.7)
so that the real and imaginary parts of  are given by

 = Re{} = n2r − n2i ,  = Im{} = 2nr ni . (4.8)

2 Specifically,  ≡ ε/ε 0 , which appears in the definition of N given by (2.36).
 A Closer Look at N 77

Though requiring somewhat more involved algebra, one may find

analogous expressions for nr and ni in terms of  and  :


 2 +  2 + 
nr = (4.9)
2


 2 +  2 − 
ni = (4.10)
2
Of course, in any programming environment (e.g., Fortran) that
provides for manipulation of complex numbers, there is no need√to
bother with the above equations; rather, one just computes N = 
or  = N 2 .
I introduce the dielectric constant here because, for certain types
of electrodynamic computations, it is more convenient than the in-
dex of refraction. The subject of the next section is a good example.

4.1.4 Optical Properties of Heterogeneous Mixtures†

At the beginning of this chapter, we stated that we would focus
on radiative interactions with media that are homogeneous on the
scales of the wavelength of the radiation. It is possible for even a decid-
edly heterogeneous mixture of different substances to satisfy this
criterion if the particles of each material are much smaller than the
wavelength.
For example, a layer of fallen snow consists of loosely aggre-
gated ice crystals having dimensions of approximately 1 mm. Sun-
light falling on snowpack has wavelengths small enough to be af-
fected by the small-scale structure of the snow. If we look closely,
we are able to resolve individual snow crystals with our eyes. But
if microwave radiation having wavelengths of 10 cm or more en-
counters the same snowpack, it is no more influenced by the indi-
vidual particles of snow than an ocean wave crashing on the beach
“cares” that the beach is not a solid surface but rather made up of
grains of sand. Thus, in the microwave band, the snowpack be-
haves more nearly like one that is homogeneous and therefore sub-
ject to the same principles discussed in this chapter. Likewise, even
complex structures like falling snowflakes are sometimes be treated
78 4. Reflection and Refraction

as equivalent homogeneous particles for the purpose of computing

radar backscatter.
Several formulas have been derived for computing the effective
index of refraction or dielectric constant of such mixtures. The pre-
cise form of these relationships depends on what assumptions are
made concerning the structure of, and relationship between, the
constituents. In a two-component mixture, one component may be
viewed as the matrix and the other as disconnected inclusions em-
bedded within that matrix. For example, a cloud is best viewed as
water inclusions (droplets) embedded in a matrix of air, whereas
foam patches on a stormy ocean surface consist of air inclusions
(bubbles) in a matrix of sea water. In other cases, such as snowpack,
it may be difficult to decide whether the ice should be considered
the matrix or the inclusion, because neither the air nor the ice tends
to exist in isolated, disconnected pockets.
One common formula used for computing effective dielectric
constants of heterogeneous mixtures is the Maxwell Garnett for-
mula   
3 f +−2m

av = m 1 +  ,
m
(4.11)
 − m
1 − f +2m
where m and  are the dielectric constants of the matrix and the
inclusions, respectively, and f is the volume fraction occupied by
the inclusions.
Another widely used relationship is that of Bruggeman:
1 − av 2 − av
f1 + (1 − f 1 ) = 0, (4.12)
1 + 2av 2 + 2av
where 1 and 2 are the dielectric constants of the two components,
and f 1 is the volume fraction occupied by component 1. Note that
this formula treats both components symmetrically; it does not dis-
tinguish between matrix and inclusion.
It is straightforward to extend either of the above formulas to
mixtures of three or more components. This is desirable in the
treatment of radar backscatter by melting snowflakes, for example,
which may be regarded as mixtures of air, ice, and liquid water. In
the case of the Maxwell Garnett formula, one first computes av for
two components, choosing one to be matrix and the other inclu-
sion. The result may then be combined with a third component by
 Refraction and Reflection 79

applying (4.11) a second time. Unfortunately, there are 12 distinct

possibilities for the order in which three components are combined,
and each may yield decidedly different values of av for the mixture!
Even the Bruggeman formula, which does not distinguish between
matrix and inclusion, yields slightly different results depending on
the order in which the constituents are included.
In general, the Bruggeman formula tends to produce results that
fall between the extremes of the various Maxwell Garnett combina-
tions. Unfortunately, it is not obvious in most cases which, if any, of
the many possibilities is “best.”
Ultimately, any formula for the effective dielectric constant of a
heterogeneous mixture is derived based on approximations or as-
sumptions that may or may not be valid for the case under consid-
eration. Wherever possible, the chosen method should be validated
for the application in question by comparing predictions with actual
measurements.

4.2 Refraction and Reflection

When an EM wave encounters a planar boundary between two ho-
mogeneous media having different indices of refraction, some of the
energy of the wave is reflected, while the remainder passes through
the boundary into the second medium (Fig. 4.3b, Fig. 4.4). In ad-
dition, the direction of the transmitted wave in medium 2 may be
altered from the original direction in medium 1, a phenomenon
known as refraction (Fig. 4.3a). The nature of both reflection and
refraction at an interface between two homogeneous media follows
directly from Maxwell’s equations, combined with appropriate con-
tinuity constraints imposed at the boundary, and is covered exten-
sively in other texts. We will not repeat the derivations here but
simply summarize the key results.

4.2.1 Angle of Reflection

Consider an EM wave incident on a plane interface between two
media. If the local normal unit vector is n̂ and the direction of the
incident ray is Ω̂i , then the reflected ray Ω̂r lies in the same plane as
n̂ and Ω̂i but on the opposite side from Ω̂i . Furthermore, the angle
80 4. Reflection and Refraction

a) Refraction

Θi
Air λi

λt

Water

Θt

b) Reflection

Θi Θr
λr λi
Air

Water

Fig. 4.3: Geometry of (a) refraction and (b) reflection of plane waves at an air-water
interface.

Θe between Ω̂r and n̂ equals the angle of incidence Θi defined by n̂

and Ω̂i .
In simple terms, a ray of light (or other EM radiation) reflects
from a smooth surface much like an ideal elastic ball thrown at
the floor: the component of its motion perpendicular to the surface
abruptly reverses, while the component parallel to the surface re-
mains unchanged. Reflection obeying this rule is termed specular
reflection. The basic requirement in order for reflection to be spec-
ular is that any irregularities on the surface must be much smaller
than the incident wavelength. In the visible band, this is generally
 Refraction and Reflection 81

Incident Reflected

N1=n1 (real)

N2=n2+ik2

Transmitted with absorption

Fig. 4.4: Reflection and transmission of normally incident radiation.

true of glass, polished metals, most liquids (including water), and

many plastic materials.

4.2.2 Angle of Refraction

When an incident ray of radiation falls on a smooth surface, reflec-
tion is usually not total. The part of the beam that is not reflected
passes into the second medium. In general, the transmitted ray
changes direction according to Snell’s Law

sin Θt sin Θi
= , (4.13)
N1 N2

where N1 and N2 are the indices of refraction of the first and second
medium, respectively, and Θt is the angle of the transmitted ray Ω̂t
relative to n̂. Although N1 and/or N2 may be complex, the above
law is most easily interpreted when the both are real, or nearly so
(i.e., weak absorption).
Case 1: Θi = Θt = 0. If a ray is normally incident on a surface,
there is no change of direction as it enters the second medium.
82 4. Reflection and Refraction

Case 2: Θi > 0; N2 > N1 . A ray incident obliquely on a medium

with larger index of refraction will bend toward the local normal;
i.e., Θt < Θi . This situation describes sunlight falling on a smooth
water surface, such as the surface of a pond or the exterior surface
of a raindrop.
Case 3: Θi > 0; N2 < N1 . A ray incident obliquely on a medium
with smaller index of refraction will bend away from normal; i.e.,
Θt > Θi . This situation arises, for example, when a ray of sunlight
that has already entered a raindrop attempts to exit on the far side.
Case 3 above includes an interesting and important special case.
Consider the possibility that

Θ i > Θ0 , (4.14)

where the critical angle is defined as

 
N2
Θ0 ≡ arcsin , (4.15)
N1

provided that N2 < N1 . But (4.13) would then imply sin Θt greater
than one, a mathematical impossibility! The way out of this appar-
ent paradox is to recognize that waves incident on the interface at
an angle greater than the critical angle simply cannot pass through
the interface at all but rather experience total reflection.
This is a good opportunity, by the way, to point out that the path
taken by a beam of light is invariant with respect to reversal of di-
rection. Viewed one way, Θ0 defines the threshold for total internal
reflection within the denser medium; viewed another, it describes
the maximum possible value of Θt when light is externally incident
on the medium at the largest possible angle (Θi = 90◦ ).
In the visible band, N1 ≈ 1.33 for water and N2 ≈ 1 for air;
hence, in water, Θ0 ≈ 49◦ .

Problem 4.2: Show that, for real N, Snell’s Law (4.13) can be de-
rived geometrically by requiring that the intersection of wave fronts
with a planar boundary between two media match on both sides of
the boundary (see Fig. 4.3).
 Refraction and Reflection 83

(a) Reflectivity of Water (Visible) (b) Reflectivity of Water (Microwave)

100 100
Vertical Pol. Vertical Pol.
Horizontal Pol. Horizontal Pol.
80 80
Reflectivity [%]

Reflectivity [%]
60 λ=0.5 µm 60

40 40

ν=37 GHz
θ
20 B 20 λ=1.5 cm

0 0
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90
θ [deg.] θ [deg.]

Fig. 4.5: Examples of the specular reflectivity of water as a function of incidence

angle. (a) Visible band. (b) Microwave band.

4.2.3 Reflectivity
We have just addressed three aspects of how EM waves are affected
by a planar interface between two homogeneous media: 1) the angle
of reflection (Θi = Θe ), 2) the angle of refraction Snell’s Law), and 3)
the critical angle for total reflection. We now turn to the following
slightly more complicated question: given that a beam of radiation
is incident on a surface at an angle Θi < Θ0 , what fraction of the beam
is reflected?
As before, the answer follows from the equations for a plane EM
wave, with suitable constraints on the continuity of the magnetic
and electric fields at the boundary. Also as before, we will not re-
produce the derivation here but rather summarize the key results,
in the form of the Fresnel relations:

 
 cos Θt − m cos Θi 2

Rp =   , (4.16)
cos Θt + m cos Θi 

 
 cos Θi − m cos Θt 2
Rs =   , (4.17)
cos Θi + m cos Θt 
84 4. Reflection and Refraction

Reflectivity (Normal Incidence) of Water and Ice

10
Water
8 Ice
Reflectivity [%]

0
0.2 0.3 0.5 0.7 1 1.5 2 3 4 5 7 10 15 20 30 50
Wavelength [µm]

Fig. 4.6: The reflectivity (in air) of water and ice at normal incidence, based on
(4.21) applied to the complex indices of refraction plotted in Fig. 4.1.

where Snell’s Law can be manipulated to yield

 2
sin Θi
cos Θt = 1− , (4.18)
m

and the complex relative index of refraction m is defined as

N2
m= . (4.19)
N1

R p and Rs are reflectivities. They give the fraction of an incident

beam of radiation that is reflected from a smooth interface, given
the local angle of incidence Θi and the relative index of refraction
m.
But why do we have two expressions for reflectivity? It turns
out that this is one of those cases when the polarization of the inci-
dent radiation matters. R p defines the reflectivity when the electric
field vector is parallel to the plane of incidence; Rs is valid when
the electric field vector is perpendicular to the plane of incidence.
Since any EM wave can be decomposed into parallel and perpen-
dicular polarized components, the total reflectivity can always be
found by taking an appropriate average of R p and Rs . For example,
if the incident radiation happens to be unpolarized, then parallel
 Refraction and Reflection 85

and perpendicular polarizations are present in equal amounts, in

which case the total reflectivity is given by

1
R= ( Rs + R p ) . (4.20)
2

Note, however, that the reflected radiation is, in general, no longer

unpolarized, so it can be dangerous to disregard polarization when
considering the net outcome of multiple reflections from surfaces.
For the special case that Θi = 0, both (4.16) and (4.17) collapse to
the following single expression for the reflectivity at normal incidence:

 
 m − 1 2
Rnormal =   . (4.21)
m + 1

At normal incidence, there is no physically important distinction

between the two polarizations, hence the need for only one formula.

In atmospheric applications, smooth reflecting surfaces are of-

ten horizontal, the most common example being a water surface,
such as the ocean or a lake. In this case, parallel polarization is
often known as vertical polarization, and perpendicular polarization
equates to horizontal polarization. The corresponding reflectivities
are then written as Rv and Rh . This terminology is common in the
field of microwave remote sensing, where the rather large differ-
ence between Rv and Rh is of great practical importance. Figure
4.5a gives examples of Rv and Rh as functions of incidence angle for
water in the visible band. The following features are of particular
interest:

• In general, the reflectivity is quite low (2%) for light at near-

normal incidence (Θi ≈ 0) but increases sharply to 100% for
near-grazing angles (Θi → 90◦ ). In other words, a smooth
water surface is a rather poor reflector of sunlight at high noon
but an excellent reflector of the setting sun. This is of course
consistent with everyday experience.
86 4. Reflection and Refraction

• Except at near-grazing and near-normal incidence, the reflec-

tivity for vertical polarization is much lower than that for hor-
izontal polarization. It is this fact that led to the development
of polarizing sunglasses, which block the largely horizontally
polarized glare from water and other reflecting surfaces while
transmitting vertically polarized light from other sources.
• There is a single angle Θ B , known as the Brewster angle, at
which the reflectivity for vertically polarized radiation van-
ishes completely, implying that only the horizontally polarized
component of incident light survives reflection at that angle.
By setting the numerator in (4.16) equal to zero and solving
for sin Θi , we find that

m2
Θ B = arcsin . (4.22)
m2 + 1

For water in the visible band, the Θ B = 53◦ .

All of the above features can be found in the reflectivities of
most nonconducting materials; i.e., those for which ni is zero or at
least very small. Larger values of the real part of m lead to greater
overall reflectivities, larger values for the Brewster angle Θ B , and
smaller values for the critical angle for total internal reflection Θ0 .
Diamonds, with their unusually large N = 2.42, owe their alluring
sparkle to all three properties.

Problem 4.3: For (a) glass with N = 1.5 and (b) a diamond with
N = 2.42, find the values of the reflectivity at normal incidence, and
the critical angle for total internal reflection Θ0 . Compare these val-
ues with their counterparts for water.

In the case of conducting materials, e.g., metals, as well as liquid

water at microwave frequencies (Fig. 4.5b), the imaginary part of m
is significantly greater than zero and also contributes to increased
reflectivity. However, although the vertically polarized reflectivity
still has a minimum at some angle Θ B , that minimum is no longer
zero. Therefore, (4.22) cannot be used to find Θ B in such cases.
 Applications 87

4.3 Applications to Meteorology, Climatology,

and Remote Sensing
4.3.1 Rainbows and Halos
Geometric Optics

In the previous section, we tacitly assumed that we were dealing

with EM waves incident on a planar (flat) boundary between two
homogeneous media. However, the above rules for reflection and
refraction can be applied not only to planar boundaries, but to any
surface whose radius of curvature is much greater than the wave-
length of the radiation. In this case, the angles Θi , Θt , Θr , Θ0 , etc.,
are measured relative to the local normal where the ray intercepts
the surface. With this generalization, we have the ability to analyze
the scattering and absorption properties of a variety of atmospheric
hydrometeors via the straightforward technique of ray tracing, also
known as geometric optics.
Unfortunately, most particles in the atmosphere are not much
larger, and may even be smaller, than the wavelength of interest.
This is true for air molecules, aerosols and cloud droplets in the vis-
ible and infrared bands and even raindrops in the microwave band.
Geometric optics cannot be used for these cases; rather, more so-
phisticated solutions to the wave equation must be derived. These
solutions and their interpretation will be outlined in Chapter 12.
Nevertheless, there are a number of interesting cases for which
the particle size is much larger than the wavelength. This condition
applies for example to the scattering of visible sunlight (λ < 0.7 µm)
by large cloud ice particles (>50 µm) and raindrops (100 µm < r <
3 mm). In fact, a number of common optical phenomena, such as
rainbows, halos, and parhelia (sundogs) can be explained by geo-
metric optics, simply by considering how rays of light refract and
reflect as they encounter the surface of the particle.

The Rainbow

To a reasonable approximation, a falling raindrop is spherical. If a

spherical droplet is uniformly illuminated, then the geometry of the
path of each incident ray depends only on x ≡ r/a, where r is the
88 4. Reflection and Refraction

External reflection

θi
Incident θi

θt
θt θi Θ
r a
θi θt
Double Internal
Reflection Direct
Transmission
θt θt

θi

Single Internal Reflection

Fig. 4.7: Ray tracing geometry for a spherical water droplet of radius a, for a ray
incident at distance r from a parallel line passing through the drop center. θi and
θt denote the incident and transmitted angles relative to the local normal and are
related by Snell’s law. Θ is the angle of scattering relative to the original direction
of the ray, in this case for a ray that has undergone a single internal reflection.

distance of the incident ray from the center axis of the drop, and a
is the radius of the drop (Fig. 4.7). So x = 0 corresponds to a ray
that is incident “dead center,” while x = 1 corresponds to a ray that
barely grazes the edge of the sphere.
Now let’s follow the path of a single incident ray after it inter-
cepts the drop:

1. A fraction of the energy in the ray will be reflected upon its

first encounter with the surface of the drop. If the incident
radiation is unpolarized, then that fraction will be given by the
average of the Fresnel relations (4.16) and (4.17), evaluated for
the local angle of incidence θi . Fig. 4.5 reveals that this fraction
is typically only a few percent, except when the ray strikes the
 Applications 89

sphere at a near-grazing angle.

2. Whatever is not reflected is transmitted into the drop and re-

fracted to an angle θt relative to the local normal, as required
by Snell’s Law.

3. The above ray now encounters the back side of the drop,
where a small fraction (a few percent) is reflected internally,
as determined again by the Fresnel relations. The remainder
exits the drop again at an angle θi relative to the local normal.

4. That portion of the original ray that was internally reflected

now encounters the surface from the inside again. As before,
a fraction is reflected internally (now for a second time), while
the remainder is transmitted to the exterior. It is the part that
escapes at this point that is responsible for the primary rain-
bow.

5. The above process is repeated for each additional internal re-

flection. However, after only two internal reflections, very lit-
tle of the original energy in the incident ray remains inside the
drop. The part that exits the drop after exactly two internal
reflections is responsible for the secondary rainbow.

As noted above, the primary rainbow is associated with radi-

ation that undergoes a single internal reflection before exiting the
droplet again, so let’s take a closer look at that case. Figure 4.8a de-
picts the full range of possible paths for rays undergoing a single
internal reflection. If the incident ray encounters the droplet dead
center (x = 0), it of course gets reflected exactly backward from
the rear surface of the drop, so the scattered direction for that ray
is 180◦ . A ray that encounters the droplet just slightly off-axis will
undergo a slight degree of refraction, reflect off the rear surface at a
slightly non-normal angle, and ultimately exit the drop at an angle
near to, but not quite equal to 180◦ .
Starting at Θ = 180◦ on the right edge of the plot in panel (b), we
see how Θ for each subsequent ray (increasing x) initially decreases
at a fairly steady rate. Also shown is the relative intensity of the ex-
iting ray, based on the fraction that “survives” this particular path
90

x=1

Incident

x=0

(a)

Scattered

Ray Tracing Results - Single Internal Reflection

0.08

0.06
Relative Intensity

(b) 0.04

x=0
0.02 increasing x

0
130 140 150 160 170 180
Scattering Angle (deg.)
Rainbow

Fig. 4.8: Illustration of the ray-tracing method applied to a spherical water drop.
(a) Diagram tracing the rays that undergo a single internal reflection, assuming
uniform incident illumination. (b) Angles and relative intensities of the scattered
rays.
 Applications 91

according to the Fresnel relations. The overall intensity in a partic-

ular direction is proportional to the intensity of the individual rays
times the density of rays per increment of Θ.
Although the trend initially is toward ever-smaller Θ, there
comes a point where Θ reaches a minimum, which we’ll call Θ0 ,
and starts increasing again. For water, which has an index of refrac-
tion of approximately 1.33 in the visible band, Θ0 ≈ 137◦ .
Because the reversal is gradual, there is a fairly significant range
of x for which the scattered rays all bunch up rather close to Θ0 . It
is this “focusing” of energy on a narrow range of Θ that gives rise to
the bright ring that we call a rainbow. Of course, a rainbow is only
visible when a rainshower is illuminated by a directional source of
bright light — e.g., sunlight.
The precise value of Θ0 depends of course on the index of re-
fraction: increasing nr has the effect of increasing Θ0 . A rainbow
exhibits the characteristic separation of colors for which it is best
known because nr for water increases slightly from the red end to
the violet end of the visible spectrum (Fig. 4.1).
As already mentioned, a similar process is behind the much
weaker secondary rainbow, which arises from two internal reflec-
tions. The scattering angle for the second rainbow is approximately
130◦ , which puts it about 7◦ outside the primary rainbow, when you
are viewing it with the sun at your back.

Halos and Related Optical Phenomena

Ray tracing can also be used to explain optical features like halos
, which are bright rings that appear around the sun in conjunc-
tion with a thin cirrostratus cloud layer, and parhelia (or sundogs),
which are bright iridescent spots positioned on either side of the
sun, usually in connection with cirrus clouds, when the sun is fairly
low in the sky. In both cases, the most common angle separating
the halo or parhelion from the sun is 22◦ . This scattering angle is
associated with refraction (without internal reflection) through two
faces of a hexagonal ice crystal whose extensions form a 60◦ angle.
In contrast to the case for the rainbow, ray tracing analysis of var-
ious optical phenomena associated with ice crystals is complicated
by the fact that they are not spherical. Therefore, results for all pos-
92 4. Reflection and Refraction

sible orientations of the crystal must be obtained and then averaged

together. Halos are associated with randomly oriented crystals, but
most other optical phenomena in cirrus clouds, including sundogs,
require ice crystals falling with a preferred orientation.
A fuller discussion of the optics of rainbows and haloes may be
found in BH83, section 7.2 and 7.3, and L02, section 5.3.
 CHAPTER 5

Radiative Properties of Natural Surfaces

In the previous chapter, we looked at what happens when radiation

encounters a smooth surface separating two homogeneous media,
such as air and water. We saw that a handful of relatively simple for-
mulas are able to describe all important aspects of this interaction,
including the angles of reflection (Θr = Θi ) and refraction (Snell’s
law) and the fraction and polarization of the radiation that is re-
flected (the Fresnel relationships).
The above relationships are sufficient to understand a handful of
significant radiative phenomena in the atmosphere – the existence
of rainbows, halos, and parhelia, or the glint of the setting sun off
a calm water surface. Unfortunately, most surfaces encountered in
nature are not so simple. Most land surfaces are covered by soil,
sand, vegetation, rough rocks, or snowpack, none of which are ei-
ther smooth or homogeneous. Water surfaces are usually rough-
ened substantially by wind waves, so that even though the laws of
specular reflection apply, at least approximately, at each individual
point on the surface, the pattern of reflected sunlight is considerably
more diffuse and complex when viewed over a larger area.
Most natural surfaces do not lend themselves to a precise the-
oretical treatment. Therefore, we must be satisfied with character-
izing their radiative properties in a more empirical way; e.g. via

96
 Natural Surfaces Idealized as Planar Boundaries 97

Incident radiation

Reflected radiation

Imaginary plane

Fig. 5.1: Example of how one treats an irregular surface as an equivalent plane
surface.

direct measurements of how much radiation they reflect and absorb

at various wavelengths and in what direction(s) the radiation is re-
flected as a function of the incidence angle.

5.1 Natural Surfaces Idealized as Planar Bound-

aries
As already pointed out, most natural surfaces cannot be viewed as
planar boundaries but rather are highly irregular and inhomoge-
neous. As seen from very high altitude, a forest may appear as a
fairly smooth green surface, but up close, it is anything but smooth.
Rather, it is a messy mix of foliage, branches, trunks, shrubs, dead
leaves, soil and (mostly) air filling a layer some 10 or more meters
thick.
For the purposes of our discussions to follow, however, we can
pretend that a forest is indeed a flat surface by imagining a trans-
parent horizontal plane just above the highest treetops and ignoring
the details of what happens to radiation below that plane (Fig. 5.1).
98 5. Radiative Properties of Natural Surfaces

All we really need to know, from an atmospheric perspective, is

that when radiation (e.g., sunlight), incident from above, passes
downward through that imaginary surface, some fraction of it is
never seen again. We can safely assume that that fraction was ab-
sorbed and converted to heat or some other form of energy. The
fraction of the incident radiation not absorbed reemerges (by what
detailed mechanism we need not inquire) in various directions and
contributes to “upwelling” radiation illuminating the atmosphere
from below.
The above simplified picture is particularly useful when we’re
willing to limit our attention to the average radiative effects of forests
and other natural “surfaces” over fairly large areas. Cloud lay-
ers may also sometimes be viewed in this simplified way, if we’re
willing to overlook the details of what happens inside the cloud
layer and focus only on the upwelling and downwelling radiation
through an imaginary horizontal plane just above cloud top or just
below cloud base.
I should mention one caveat before you embrace this approach
too firmly: when considering emission of radiation from a complex
medium (as we will in later chapters), it is really only safe to treat
it as an equivalent “surface” if there is also a single more-or-less
unique temperature that can be ascribed to every participating point
in the scene.

5.2 Absorptivity and Reflectivity

When radiation is incident on the earth’s surface, some fraction is
reflected and the remainder absorbed. The fraction absorbed we
call the absorptivity, while the fraction reflected is the reflectivity.
We will use the symbols a and r for absorptivity and reflectivity,
respectively. Obviously neither quantity can be less than zero nor
greater than one.
In general a and r depend on the wavelength λ. A surface that is
highly reflective at one wavelength (r ∼ 1) may be highly absorbing
at another (r ∼ 0). For example, grass and other vegetation appears
green because it reflects green, yellow, and blue wavelengths more
strongly than red and orange. Often we will make the wavelength
 Absorptivity and Reflectivity 99

dependence explicit by adding a subscript λ, unless there is no pos-

sibility of confusion.
In addition, both of these quantities usually depend on the direc-
tion Ω̂ = (θ, φ) of the incident radiation. For example, in Chapter 4
we saw that the reflectivity (in the visible band) of a smooth water
surface varies from a rather small value of 2% at normal incidence
(θ = 0) to nearly 100% at grazing angles (θ → 90◦ ).
For any given wavelength and direction of incident radiation, it
is clear that the reflectivity and absorptivity must sum to unity:

aλ (θ, φ) + rλ (θ, φ) = 1 . (5.1)

Note that many natural surfaces are azimuthally isotropic, mean-

ing they reflect about the same amount of radiation regardless of
whether the sun (for example) shines on them from the east, south,
west, or some other direction. In that case, the dependence on φ
may disappear from the above equations.
Furthermore, in the case of many rough surfaces, such as forests,
grasslands, ploughed fields, etc., even the dependence on θ may be
weak, in which case we might choose (as an approximation) to ig-
nore the directional dependence altogether. One could then relate
the reflected monochromatic flux Fλ,r to the incident flux Fλ,0 as fol-
lows:
Fλ,r = rλ Fλ,0 . (5.2)

It follows that the power per unit area (per unit wavelength) ab-
sorbed by the surface is

Fλ,0 − Fλ,r = (1 − rλ ) Fλ,0 = aλ Fλ,0 . (5.3)

5.2.1 Examples of Reflectivity Spectra

Fig. 5.2 gives examples of how reflectivity typically varies as a func-
tion of wavelength for a few natural surface types. Not surpris-
ingly, fresh snow has a very high reflectivity (85%–95%) throughout
100 5. Radiative Properties of Natural Surfaces

Yellow
Green
Violet
Blue

Red
Near Infrared
100

80 Fresh Snow
Wet Snow
Reflectivity (%)

60 Light soil
Grass
40

Alfalfa
20 Straw Dark Soil
Lake
0
0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.0
Wavelength (µm)

Fig. 5.2: Typical shortwave reflectivity of various natural surface types as a func-
tion of wavelength.

the visible band, whereas the reflectivity of fresh vegetation, soil,

and bodies of water is generally very low (< 15%). There is a pro-
nounced peak in reflectivity near 0.55 µm for fresh grass and alfalfa
which, not surprising, corresponds to the green portion of the vis-
ible spectrum and is due to chlorophyll. Dry straw, on the other
hand, exhibits no such peak and is considerably brighter overall,
though with a bias toward the orange and red end of the spectrum.
An important point to take from this diagram is that there may
be relatively little relation between the reflectivity of a surface type
in the visible part of the spectrum and that in the near-IR band. For
example, at 1 µm, grass is nearly as reflective as wet snow. And
both fresh and wet snow are far less reflective in the near-IR band —
especially beyond about 1.3 µm, than they are in the visible. These
differences can become even more extreme as one continues into
the thermal IR band (not shown) where many natural surface types,
including fresh snow, have reflectivities of less than 5%.

5.2.2 The Graybody Approximation

As we have just seen, the absorptivity and reflectivity of a surface
depends on the exact wavelength being considered. Often, how-
ever, it is useful, especially in introductory treatments of radiation
such as this one, to disregard the details of this dependence and to
pretend that the surface is “gray” over some rather broad range of
 Absorptivity and Reflectivity 101

wavelengths. That is to say, we specify a single average absorptivity

that is taken to be representative of the entire band. This is known
as the graybody approximation.
For example, if the total incident flux of radiation between two
widely separated wavelengths is denoted Fi , and the reflected flux
within the same range of wavelengths is Fr , then the effective gray-
body reflectivity is defined as

Fr
r≡ , (5.4)
Fi

and the corresponding graybody absorptivity is a = 1 − r.

A word of caution: r and a do, in general, depend on the spectral
details of the incident radiation, so it usually makes sense to use
the graybody approximation only when those details are reasonably
constant. For example, the spectrum of solar radiation reaching the
surface does not change drastically from one day to the next, so
it is reasonable to use the graybody approximation to describe the
reflection and absorption of the incident solar flux.
A common application of the graybody approximation is to as-
sign one constant absorptivity asw to the entire shortwave, or so-
lar, band, and another constant absorptivity alw to the longwave, or
thermal IR, band. For most terrestrial surfaces, alw is close to unity,
whereas asw can be highly variable, from close to zero in the case
of deep snow to close to unity in the case of forests and bodies of
water.
The complement of the shortwave absorptivity asw is of course
the shortwave reflectivity rsw = 1 − asw , also commonly known as
the shortwave albedo. Examples of albedos for common surfaces are
given in Table 5.1.
The value of this kind of partitioning between longwave and
shortwave absorptivities and reflectivites will become more appar-
ent when we consider simple radiative energy budgets for the sur-
face and atmosphere later on.
102 5. Radiative Properties of Natural Surfaces

Table 5.1: Shortwave (solar) reflectivity (in percent) of various surfaces.

Fresh, dry snow 70–90

Old, melting snow 35–65
Sand, desert 25–40
Dry vegetation 20–30
Deciduous Forest 15–25
Grass 15–25
Ocean surface (low sun) 10–70
Bare soil 10–25
Coniferous Forest 10–15
Ocean surface (high sun) <10

5.3 Angular Distribution of Reflected Radiation

Early in this chapter (Section 5.2), I introduced the concept of re-
flectivity, which describes the fraction of incident radiation that is
reflected from a surface. Among other things, I pointed out that
the reflectivity and absorptivity of an opaque surface must sum to
unity; that is, all incident radiation must be either reflected or ab-
sorbed:
aλ + rλ = 1 . (5.5)

This seems straightforward enough, and you might assume that

there’s little more to be said. But we have not yet explained what
happens to the radiation that is reflected from a surface. That is to
say, given rλ , we know how much radiation is reflected, but we don’t
yet know where it goes.

5.3.1 Specular and Lambertian Reflection

In Chapter 4 we discussed reflection from a very smooth bound-
ary between two homogeneous media. In this special case, which is
called specular reflection, the reflected angle Θr (relative to the local
normal) is just equal to the incident angle Θi . Furthermore, the re-
flectivity is then given by the Fresnel relationships (4.16) and (4.17).
Surfaces encountered in nature are not so simple. About the
only place where you sometimes find true specular reflection on
the scales important to us is from the mirror-like surface of a very
 Angular Distribution of Reflected Radiation 103

smooth body of water, such as a pond or lake on a completely calm

day. Under those conditions, the reflection of the sun and other ob-
jects is very sharp and clear.
More commonly however, the surface of open bodies of water
are at least somewhat roughened by ripples or waves generated
by the wind, so that light from the sun is not reflected in a single
direction but rather is scattered in a variety of different directions,
depending on the local slope where each light ray from the sun en-
counters the surface. For a lightly roughened water surface, most
of the radiation is scattered in a fairly narrow cone of angles sur-
rounding the specular direction. Thus, the reflection of the sun is
still identifiable to the eye (or a radiometer) as a relatively concen-
trated bright spot, but it appears blurred in comparison to the sun
itself. The rougher the surface, the greater the blurring, until the
reflected radiation is scattered almost uniformly in all directions, ir-
respective of the direction of incidence.
Apart from bodies of water, the vast majority of surfaces en-
countered outdoors, as viewed from a significant altitude, are very
rough. A forest or a corn field observed from an airplane tends to
look more or less uniformly bright when illuminated by the sun,
regardless of which direction you are looking. At least, there is no
pronounced “hot spot” in the direction where you would expect the
specular reflection of the sun to be.
As a crude approximation, one often assumes the flux of upward
reflected radiation is equally distributed over all angles, irrespective
of the direction of the source. Reflection obeying this rule is called
Lambertian . Lambertian reflection is thus the exact opposite of spec-
ular reflection, since in the latter case all reflected radiation emerges
in a single well-defined direction.

Problem 5.1: Wall paint is sold in several varieties, including

glossy, semigloss, and flat. The distinction between these lies in the
angular pattern of reflected radiation. Which of these do you think
would be best described by the Lambertian model of reflection?

For any kind of reflection, the upward reflected flux Fr is equal

104 5. Radiative Properties of Natural Surfaces

(a) specular (b) quasi-specular (c) Lambertian

(d) quasi-Lambertian (e) complex

Fig. 5.3: Examples of various types of surface reflection, presented in the form of
polar plots in which the distance of the curve to the reflection point on the surface
represents the relative intensity of reflected radiation in that direction.

to the reflectivity r times the incident flux Fi :

Fr = rFi . (5.6)

In the Lambertian case, the intensity Ir of the reflected radiation is

by definition independent of direction, so that Fr = πIr . Thus we
can express the reflected intensity I ↑ in terms of the incident flux by
simply dividing by π (see Section 2.16):

rFi
I↑ = . (5.7)
π
If the incident flux is entirely due to the direct rays of the sun posi-
tioned in the sky at an angle θi from zenith, we further have

rS0 cos θi
I↑ = , (5.8)
π
where S0 is the solar flux normal to the beam.
The angular distribution of reflection from most natural surfaces
is generally neither specular nor perfectly Lambertian. Indeed, it
generally has no simple mathematical form at all but rather must
be determined empirically from suitable measurements. Fig. 5.3
 Angular Distribution of Reflected Radiation 105

schematically depicts some reflection patterns that might be en-

countered in nature, just to give you an idea of what is possible.
It is not uncommon for observed reflection patterns to combine
specular- and Lambertian-like features, since there is often both a
strong peak in the specular direction as well as a more diffuse com-
ponent spread out more uniformly over all angles.

5.3.2 Reflection in the General Case†

In the previous section, we considered two important limiting cases:
1) specular reflection, wherein the direction of the reflected radia-
tion is uniquely determined by the incident direction, and 2) Lam-
bertian reflection, wherein the angular distribution of reflected ra-
diation is uniform and therefore completely independent of the in-
cident direction. Both cases are simple to deal with both conceptu-
ally and computationally. Radiation and remote sensing specialists
therefore tend to fall back on one or the other simplifying assump-
tion whenever they can get away with it.
Sometimes, however, they cannot, and a more sophisticated
way of describing surface reflection is needed. In general, the in-
tensity of reflected radiation is a continuously varying function of
both the incident direction Ω̂i = (θi , φi ) and the reflected direction
Ω̂r = (θr , φr ). Radiation specialists define the bidirectional reflection
function ρ(θi , φi ; θr , φr ), or BDRF,1 such that

I ↑ (θr , φr ) = ρ(θi , φi ; θr , φr )S0 cos θi , (5.9)

where S0 is the solar flux normal to the beam.

The above relationship assumes direct illumination of the sur-
face by the sun at position (θi , φi ) in the sky. Of course, it is not
uncommon for illumination to come from more than one direction.
For example, on an overcast day, the downward illumination comes
more or less equally from all directions. In this case, it is necessary
to generalize the above relationship so as to integrate the contribu-
tions from all directions:
!
↑
I (Ω̂r ) = ρ(Ω̂i ; Ω̂r ) I ↓ (Ω̂i )n̂ · Ω̂i dωi , (5.10)
2π
1 Some authors prefer bidirectional reflection distribution function,or BRDF.
106 5. Radiative Properties of Natural Surfaces

n
Ωi

Ωr
dωi θi θr

dωr

dA
φi
φr

Fig. 5.4: Geometry and relevant variables for the definition of the BDRF.

where n̂ is the vertical unit vector, so that n̂ · Ω̂i = cos θi .

Recall that the integral as written above represents a generic in-
tegration over solid angle covering the entire upper hemisphere of
2π steradians. Rewriting the above as an explicit integration over
the two polar coordinates θ and φ, we have
! 2π ! π/2
↑
I (θr , φr ) = ρ(θi , φi ; θr , φr ) I ↓ (θi , φi ) cos θi sin θi dθi dφi .
0 0
(5.11)
As a special case, consider what happens if ρ(θi , φi ; θr , φr ) = ρ L ,
where ρ L is a constant. In that case, ρ L can be taken out of the inte-
gral above, leaving
! 2π ! π/2
↑
I = ρL I ↓ (θi , φi ) cos θi sin θi dθi dφi . (5.12)
0 0

Note that the right hand side, and therefore I ↑ , no longer depends
on (θr , φr ), so we evidently have Lambertian reflection. Also, by
 Applications 107

comparing the remaining integral with (2.59), we realize that it rep-

resents the incident flux of radiation Fi , so that

I ↑ = ρ L Fi . (5.13)

If we multiply both sides of the above equation by π, we get an

expression for the reflected flux (since I ↑ is isotropic). From (5.2),
we then conclude that
r = πρ L , (5.14)
where r is the total (nondirectional) reflectivity of the Lambertian
surface.
Let us now generalize to find the reflectivity r (Ω̂i ) correspond-
ing to an arbitrary BDRF. For simplicity, we again consider the re-
lationship between the reflected flux Fr and the incident flux Fi =
S0 cos θi due to a columnated beam originating from a zenith angle
θi :
Fr = r (Ω̂i )S0 cos θi . (5.15)
The reflected flux Fr can in turn be expressed in terms of an integral
over the hemisphere of the upward reflected intensity I ↑ :
! !
↑
Fr = I (Ω̂) cos θ dω = ρ(Ω̂i ; Ω̂r )S0 cos θi cos θr dωr , (5.16)
2π 2π

where we have substituted (5.9) into the integral. Setting (5.15)

equal to (5.16) and dividing out the incident flux terms yields
!
r (Ω̂) = ρ(Ω̂; Ω̂r ) cos θr dωr . (5.17)
2π

5.4 Applications to Meteorology, Climatology,

and Remote Sensing
5.4.1 Solar Heating of Surfaces
The shortwave albedo of a surface has a large effect on the direct
heating of the surface by sunlight and, ultimately, on the heating of
air in contact with that surface. A field of freshly plowed bare soil
(albedo ∼10%) will absorb almost 30% more solar radiation than a
108 5. Radiative Properties of Natural Surfaces

field of dry wheat (albedo ∼30%) and nine times as much as a layer
of fresh dry snow (albedo ∼90%).
Whatever solar radiation is absorbed usually has the immediate
effect of heating the surface. But much of that heat is eventually
transferred to the atmosphere by way of (1) direct thermal conduc-
tion (sensible heat flux), (2) evaporation of surface moisture (latent
heat flux), and/or (3) net longwave flux. Which one of these transfer
mechanisms dominates depends on both atmospheric and surface
conditions. A warm, humid atmosphere will tend to minimize the
net loss of longwave radiation from the surface. A moist or vege-
tated surface will tend to transfer a proportionally larger fraction of
its energy to the atmosphere in the form of latent heat.
If you have two dry, relatively bare land areas side by side with
sharply differing albedos, then the darker surface, and therefore the
air immediately above that surface, will be heated more rapidly
by the absorption of solar radiation. If there is little wind, the
sharp temperature difference that develops over time will trigger
microscale circulations, as the warmer air over the darker surface
rises in the form of a thermal updraft and is replaced by relatively
cooler air flowing horizontally from the lighter surface. The latter
air, of course, must be replaced by sinking air from above. Experi-
enced glider pilots know how to exploit variations in land surface
albedo so as to find the thermal updrafts that can help keep them
aloft.

Problem 5.2: A field of snow that is just starting to melt is a spe-

cial case, in that absorbed solar radiation will contribute primarily
to a further phase change (solid to liquid) at a constant temperature
of 0◦ C. Assume that the shortwave albedo of wet snow is 60% and
that the solar flux reaching the surface on a particular sunny day
is 500 W m−2 . The latent heat of fusion of ice L f is 3.3 × 105 J/kg
and you can assume that the density ρws of wet snow stays constant
(assuming all excess meltwater drains away) at around 200 kg m−3 .
At what rate, in centimeters per hour, is the snow pack depleted by
direct solar heating?
 Applications 109

Zenith
Satellite
Sun

θs θsat

dA
φs
φsat

Fig. 5.5: Relationship between incident and reflected solar radiation, as seen by a
satellite sensor.

5.4.2 Satellite Imaging at Visible and Near-IR Wavelengths

One of the simplest and most familiar applications of remote sens-

ing is embodied in visible (“VIS”) images captured by sensors on
board both geostationary and polar orbiting meteorological satel-
lites. Conceptually (though not technologically), these images are
similar to an ordinary black and white photograph taken from the
(admittedly less than ordinary) vantage point of outer space. As
long as the satellite sensor is viewing the sunlit side of the earth,
visible images allow you to easily distinguish regions of high reflec-
tivity (clouds, snow cover, sea ice, desert sand) and low reflectivity
(vegetated ground, water, soil).
The very first meteorological satellite images (beginning in April
1960) were of this type. In fact, they were little more than televi-
sion pictures taken from space. Nevertheless, they provided the
first clear pictures of the large scale organization and evolution of
clouds and storm systems. While careful analysis of surface weather
110 5. Radiative Properties of Natural Surfaces

Fig. 5.6: A visible image of the Eastern Pacific and west coast of North America
taken by the GOES-West geostationary weather satellite.

observations had previously hinted at these structures, the network

of observing stations was (and remains) too sparse to resolve more
than grossest features, especially over sparsely populated areas.

Problem 5.3: An example of a visible image from a geostationary

satellite is given in Fig. 5.6.
(a) Based on this image, characterize the relative reflectivities of
the various geographical and meteorological features that you see.
(b) The apparent brightness of the clouds diminishes steadily to-
ward the western edge of the image. Why?

Quantitative utilization of VIS imagery – i.e., the conversion

of measured radiances to estimates of specific physical properties,
such as albedo, snow depth, etc., is also possible but requires con-
siderably more care. In particular, you need an instrument that is
 Applications 111

well calibrated, so that radiance measurements are reliable and ac-

curate in an absolute sense, not just a relative one. Also, you have
to carefully account for the variable angle and intensity of solar il-
lumination.
Recall from (5.9) that the reflected intensity I ↑ viewed by the
satellite depends on the viewing direction (θr , φr ), on the direction
of incidence (θi , φi ), and on the bidirectional reflectance function
(BDRF) ρ(θi , φi ; θr , φr ). While the various directions can be easily
calculated, given the satellite’s position, the scene location on the
earth’s surface, and the sun position (Fig. 5.5), the details of the
BDRF are highly variable and, for many natural surfaces, are poorly
known.
It is therefore not possible to reliably estimate the overall albedo
of a surface from a radiance measurement made in only one direc-
tion and with one direction of solar illumination. Thus, a high ra-
diant intensity observed by a satellite sensor at a particular point
on the surface could imply either a) a high-albedo surface, such as
snow, or b) a lower-albedo surface for which the BDRF happens to
be sending a large fraction of the total reflected radiation in the di-
rection of the satellite. Sun glint from a water surface is a good
example of the second case: the overall albedo of water is quite low,
but the reflected image of the sun can be very bright if you happen
to be looking in exactly the right direction.

Problem 5.4: A certain meteorological satellite crosses the equator

at a longitude of φsat = 50◦ W and time 1200 Greenwich Mean Time
(GMT) on March 21, at which time the sun is overhead at the in-
tersection of the equator and the Greenwich Meridian. At what ge-
ographic location is an imager on board the satellite most likely to
observe sun glint? The satellite has an altitude above the surface of
H = 1000 km, and the sun can be regarded as being infinitely far
away from the earth (i.e., rays from the sun are parallel). The radius
of the earth is R E = 6373 km. Hint: First find the trigonometric equa-
tion that must be satisfied by the longitude of the sun glint. There is
no simple closed-form solution to this equation, so you’ll need to use
successive approximation to find a numerical solution accurate to at
least the nearest degree.

Some satellite sensors have channels at two or more visible

112 5. Radiative Properties of Natural Surfaces

and/or near-IR wavelengths. Since the shape of the BDRF of most

surfaces doesn’t differ much between two nearby wavelengths, the
ratio of the observed intensities doesn’t depend too much on either
the BDRF itself or the angle of solar illumination. This makes it
possible to assess the color of a scene without too much ambigu-
ity. Common applications include 1) land use classification, 2) as-
sessment of phytoplankton concentration in coastal waters, and 3)
characterization of vegetation extent and density. Of course, all of
these techniques are subject to the adverse effects of atmospheric
haze and clouds.

Problem 5.5: A particular satellite sensor is being designed to mea-

sure the reflectivities r1 and r2 of a scene corresponding to two wave-
lengths λ1 and λ2 . Based on Fig. 5.2, choose λ1 and λ2 so as to
optimize the ability of the satellite to discriminate different scene
types. Start by drawing a diagram with r1 as the horizontal axis
and r2 as the vertical axis. Plot points corresponding to each surface
type, and label the points. Try to develop a simple mathematical
algorithm (e.g., a series of tests based on inequalities) that would al-
low you to correctly classify the surface as snow, soil, growing vegeta-
tion, dry vegetation, or water, based on observations of r1 and r2 . You
might want to consider reflectivity differences and/or reflectivity ra-
tios, as well as individual channel reflectivities, as possible variables
in your tests. Whatever criteria you develop, depict them graphically
as curves separating surface types on your plot of r1 vs r2 . Note that
if you choose your two wavelengths so that any two differing surface
types are too closely spaced on your plot, the risk of misclassification
greatly increases when you apply your algorithm to real-world data.
Hint: There is no single “correct” solution to this problem – be cre-
ative!
 CHAPTER 6

Thermal Emission

Up to this point, we have considered only the absorption, transmis-

sion, and reflection of radiation incident externally on a medium.
These interactions are sufficient for characterizing the disposition of
shortwave (solar) radiation within the atmosphere, because its prin-
cipal source, the sun, is external to the earth-atmosphere system. By
contrast, the principal source of longwave radiation is thermal emis-
sion by the earth and atmosphere itself. Emission is the process by
which some of the internal energy of a material is converted into
radiant energy. This energy may either get reabsorbed at another
location in the atmosphere or else escape to space, whereupon it is
permanently lost, at least from the perspective of earthbound ob-
servers.
Normally, we are oblivious to the fact that we are constantly
bathed in, and absorb, longwave radiation that is emitted by our
surroundings, nor do we give much thought to the fact that our
own bodies also lose heat energy via emission of radiation. One rea-
son that we don’t notice this is that the temperature of our skin and
that of our surroundings is usually not too different; hence there is
a near-balance between the heat we lose via radiation and that we
absorb from our surroundings. Also, if there is any air movement
at all (as is usually the case outdoors) then we tend to be more con-

110
 111

scious of heat exchange with the air in direct contact with our skin,
due to conduction, than that due to radiation.
The existence of thermally emitted radiation becomes more ap-
parent to us when the temperature differences are larger. A wood-
burning stove radiates heat that you can feel on your face from
across a room. A car parked under a clear sky during the fall or win-
ter can rapidly cool to the point that frost forms on the windshield,
even if the free air temperature remains a few degrees above freez-
ing. If an object becomes sufficiently hot, thermal emission may be
great enough at the shorter wavelengths so as to be visible to the
eye. The glowing embers in a fireplace or barbecue grill are one
example; the white-hot filament of an incandescent light bulb is an-
other.
It is possible to derive the relationship between temperature and
thermal emission from first principles, based on the laws of quan-
tum mechanics and statistical thermodynamics.1 In keeping with
the title of this book, we will instead go straight to the bottom line,
explaining the general characteristics of thermal emission and giv-
ing you the tools to perform radiative calculations of the type most
relevant to meteorologists.
Here, in words, are the key facts about thermal emission of ra-
diation that you should be very comfortable working with by the
time you finish this chapter. We will elaborate on these points, and
provide the relevant formulas, in the subsequent subsections.

• An object having temperature T will generally emit radia-

tion at all possible wavelengths. However, for any particular
wavelength λ, there is a hard upper bound on the amount of
that radiation. The function of T and λ that gives that upper
bound is called Planck’s function. The shape of Planck’s func-
tion for several representative temperatures in the atmosphere
is shown in Fig. 6.3. The mathematical details will be given
shortly.

• For any given (absolute) temperature, Planck’s Function has

its peak at a wavelength that is inversely proportional to that
temperature, a fact that is embodied in Wien’s Displacement
1 See for example Appendix A of L02.
112 6. Thermal Emission

Law. Thus, peak emission from a cool object, like the earth,
occurs at much longer wavelengths than that from a very hot
object, like the sun.

• By integrating Planck’s Function over all possible wave-

lengths, you get the Stefan-Boltzmann law, which states that
the theoretical maximum amount of total (broadband) radi-
ation that can be emitted by an object is proportional to the
fourth power of its absolute temperature. Thus, doubling the
temperature of an object leads to sixteen-fold increase in the
maximum amount of radiation it can theoretically emit.

• Within any given band of wavelengths, a good absorber is also

a good emitter. This fact is embodied in Kirchhoff’s law. Thus,
a perfect reflector (nonabsorber) emits no thermal radiation.
A perfect absorber emits the theoretical maximum amount of
thermal radiation, as described by Planck’s Function.

I urge you to spend some time thinking about the above four
points and committing them to memory. Once you have an intu-
itive grasp of these basic facts, your brain will be more receptive to
precise mathematical re-statements of the same facts. You will also
be less likely to use the mathematical formulas “blindly” in possibly
inappropriate ways.

6.1 Blackbody Radiation

I just made several references to the “theoretical maximum amount
of radiation” that can be emitted by an object. I also pointed out
that a good absorber is a good emitter. In order for thermal emis-
sion from an object to actually achieve the “theoretical maximum,”
it must be a perfect emitter, which is to say, a perfect absorber.
An object that absorbs radiation perfectly is called a blackbody. In
other words, it is a graybody for which the absorptivity a = 1. As
such, it is an idealization that is seldom exactly realized in nature.
Nevertheless, it is surprisingly easy to approximate a blackbody as
follows:
Take a large, empty cardboard packing box (say, two feet on a
side) and tape it closed on all sides. Then punch a small hole — one
 Blackbody Radiation 113

Fig. 6.1: Schematic depiction of the fate of a photon that enters a hollow cavity
through a small hole. With each additional reflection from the wall of the cavity,
its chance of surviving without absorption steadily decreases. A blackbody cavity
can therefore easily be constructed such that only a negligible fraction of photons
incident on the hole ever emerge again.

cm diameter will do — in one side. That hole is now a pretty good

approximation of a blackbody over a wide range of wavelengths.
If you shine a flashlight beam on it, virtually all of the light will
disappear into the hole and never reemerge. In other words, the
effective absorptivity of the hole is very close to unity. Why is this?
Consider the fate of a photon fired into the hole. First, it strikes
the opposite surface inside the box (Fig. 6.1). At that point, its
chance of getting absorbed is equal to the absorptivity of the card-
board. Let’s assume that that absorptivity is 0.5, in which case there
is a fifty-fifty chance of the photon getting absorbed. Even if it is not
absorbed, the chance is very small that it will bounce directly out the
same small hole that it came in through. Rather, it will encounter an-
other interior side of the box, with another fifty-fifty chance of get-
ting absorbed. Its total survival probability is now only one in four.
Now consider a trajectory that allows the photon to exit the hole
after reflecting off an interior wall of the box N times. The proba-
bility of the photon surviving that entire trajectory and exiting the
hole is (1 − a) N , where a is the absorptivity. If a = 0.5, as assumed
above, N could be as small as seven and still imply less than a one-
114 6. Thermal Emission

percent probability of a photon re-emerging from the hole. If the

hole is small and the box is large, the expected value of N can be
quite large. Also, if you paint the inside of the box black, then a
approaches one, pushing the value of (1 − a) N even closer to zero.
In short, it is not difficult to make the absorptivity of the hole in the
box almost arbitrarily close to unity.
Now that we have a blackbody, what can we do with it? For one
thing, we can measure the characteristics of the radiation emitted by
the blackbody. At room temperature, this radiation lies primarily in
the thermal infrared band and is therefore invisible to the eye. But if
the box were made of something heat resistant, like iron, you could
heat it to the point that visible light was emitted from the hole. If
you increased the temperature of the cavity gradually up to several
thousand degrees, you would first see a dark red glow, then more
orange and yellow, and all the while the overall brightness would
increase. At extremely high temperatures (6000 K and higher) , the
emitted light becomes white and, eventually, bluish white. Such
high temperatures are found in nature only on the surface of the
sun and stars.

6.1.1 Planck’s Function

The intensity of radiation emitted by a blackbody is given by
Planck’s Function B( T ). It takes slightly different forms depending
on whether it is written as a function of wavelength λ, frequency ν
or wavenumber ν̃. When it is given as a function of wavelength, it
is
2hc2
Bλ ( T ) = , (6.1)
λ5 (ehc/k B λT − 1)

where c = 2.998 × 108 m s−1 is the speed of light, h = 6.626 ×

10−34 J s is Planck’s constant, and k B = 1.381 × 10−23 J/K is Boltz-
mann’s constant.
The interpretation of this function is as follows:
 
total intensity of emitted radiation con-
Bλ ( T ) dλ = tributed by the wavelength interval . (6.2)
[λ, λ + dλ]
 Blackbody Radiation 115

Blackbody Emission Curves (Planck’s Function)

1e+08

1e+07

1e+06
6000 K
100000
Bλ(T) [W m-2Sr-1µm-1]

10000
Wien’s Law
1000

100 (a)
10 300 K

0.1

0.01 250 K
0.001

0.0001
0.1 0.15 0.2 0.3 0.5 1 1.5 2 3 5 10 15 20 30 50 100
λBλ(T) (normalized)

(b)
6000 K 300 K 250 K

0.1 0.15 0.2 0.3 0.5 1 1.5 2 3 5 10 15 20 30 50 100

Wavelength [µm]

Fig. 6.2: Blackbody emission curves at temperatures typical of the sun and of
the earth and atmosphere. (a) The actual value of Planck’s function, plotted on
a logarithmic vertical axis. The diagonal dashed line corresponds to Wien’s law
(6.3). (b) Normalized depictions of the same functions as in (a), so that the areas
under each curve are equal. Note that the vertical axis in this case is linear.

The physical dimensions of Bλ are thus those of intensity (power

per unit area per unit solid angle) per unit wavelength; in common
units: W m−2 µm−1 sr−1 .
Note that this interpretation is consistent with a point I made
earlier: if you allow dλ to go to zero, then the emitted radiation in
the interval [λ, λ + dλ] also goes to zero. Thus, a hypothetical (and
unrealizable) detector that only could measure radiation at exactly
one wavelength would detect no radiation at all. Consequently,
any real detector used in atmospheric remote sensing has a finite
116 6. Thermal Emission

10
300
8
Bλ [W/(m2 µm Sr)]

Blackbody Emission at Terrestrial Temperatures

4 250

2
200

0
0 10 20 30 40 50
λ (µm)

Fig. 6.3: The Planck (blackbody) function Bλ at temperatures typical of those found
in the atmosphere.

spectral “bandwidth” that describes the range of wavelengths to

which it responds. The greater the bandwidth, the more total radi-
ant power the sensor receives and the greater its overall sensitivity,
all other factors being equal.

Problem 6.1: Sometimes Planck’s function B( T ) may be expressed

as a function of frequency ν or wavenumber ν̃ instead of wavelength
λ. Given that Bλ ( T ) dλ must equal Bν ( T ) dν when dν and dλ corre-
spond to the same narrow interval of the spectrum, find the correct
expression for Bν as a function of ν only.

Examples of Bλ ( T ) are shown for various temperatures in Figs.

6.2a and 6.3. Planck’s function is seen to have its peak at a wave-
length that is inversely proportional to absolute temperature (see
Wien’s Displacement Law, below). At any given wavelength, emis-
sion increases monotonically with increasing temperature. Emis-
sion is not symmetrically distributed about this peak; rather, the
function drops off sharply at the short wavelength end of the spec-
trum while trailing off much more slowly toward long wavelengths.
 Blackbody Radiation 117

100 5

90

80 4

Max Wavelength ( µm) (for T=6000K)

Max Wavelength (µm) (for T=288K)

70

60 3

50

40 2

30

20 1

10

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Fraction of Total Blackbody Emission

Fig. 6.4: The fraction of total blackbody emission contributed by wavelengths λ

smaller than the threshold
λ value indicated on the vertical axes. The horizontal axis
gives the value of π [ 0 Bλ ( T ) dλ]/(σT 4 ). The left axis corresponds to a blackbody
with T = 288 K; the right axis corresponds to T = 6000 K. (Figure courtesy of S.
Ackerman, with modifications)

6.1.2 Wien’s Displacement Law

The wavelength λmax of the peak of the Planck function — i.e., that
of maximum emission from by a blackbody of temperature T is
given by Wien’s Displacement Law:

C
λmax = , (6.3)
T

where the constant C = 2897 µm K. Thus, peak emission from a

blackbody with a temperature of 6000 K, similar to that of the sun,
occurs at a wavelength of λmax = 0.48 µm, whereas typical atmo-
spheric temperatures in the range 200K–300K yield peak emission
in the range 9.6–14.4 µm.
118 6. Thermal Emission

Problem 6.2: Derive an approximate value of the constant C in

Wien’s Law (6.3). Start with an approximate form of (6.1) that is
valid when exp(hc/k B λT )  1. Then use standard calculus tech-
niques to find the position of the maximum. Once you have found
your expression for λmax , verify that the approximation you used is
always valid near that wavelength.

Problem 6.3: Regardless of whether an emitter is a perfect black-

body or not, one may define the color temperature of the emitter in
terms of its wavelength of maximum emission, according to Wien’s
law. The wavelength of maximum emission from the sun is approx-
imately 0.475 µm. Determine its color temperature.

Problem 6.4: The filament of an incandescent light bulb typically

has a temperature of 2,850 K. (a) Find λmax . (b) Is this wavelength
within the visible band? (c) Plot Bλ for this temperature, using linear
axes. Either graphically or numerically, estimate the fraction of the
total emission from an incandescent light bulb that falls within the
visible portion of the spectrum (i.e., between 0.4 and 0.7 µm). (d) If
the bulb consumes 60 W of electrical power, what is the maximum
amount of visible light it produces, in watts?

6.1.3 Stefan-Boltzmann Law

Planck’s function gives the monochromatic intensity emitted by a
blackbody. Of fundamental interest for energy transfer within the
atmosphere, and between the atmosphere and outer space, is the
broadband flux emitted by a blackbody. This flux is obtained by in-
tegrating Planck’s function over all wavelengths and over the 2π
steradians of solid angle of one hemisphere. For uniform intensity,
the latter integration entails nothing more than multiplication by a
factor π; hence the desired blackbody flux is

∞
FBB ( T ) = π Bλ ( T ) dλ . (6.4)
0
 Blackbody Radiation 119

When expression (6.1) for Bλ is substituted and the integral evalu-

ated analytically, the above integral reduces to the Stefan-Boltzmann
law

FBB ( T ) = σT 4 , (6.5)

where
2π 5 k4 W
σ= 2 3
≈ 5.67 × 10−8 2 4 . (6.6)
15c h m K

Problem 6.5: Given the following information, use the Stefan-

Boltzmann relationship to compute the effective emitting tempera-
ture of the Sun: Solar constant at top of Earth’s atmosphere S0 =
1370 W m−2 ; mean radius of Earth’s orbit 1.496 × 108 km; radius
of the Sun’s photosphere 6.96 × 105 km. How does this value com-
pare with the color temperature of the Sun derived in the previous
problem?

Problem 6.6: Perform the analytic integration of Bλ that yields

the Stefan-Boltzmann relationship.
∞ Hint: Use the substitution x =
hc/k B λT, and note that 0 [ x3 /(e x − 1)]dx = π 4 /15.

6.1.4 Rayleigh-Jeans Approximation

If you are concerned with microwave remote sensing of the atmo-
sphere, then the wavelengths of interest to you are quite long —
λ ∼ 1 mm or longer. In the limit of large wavelength,

2ck B
Bλ ( T ) ≈ T, (6.7)
λ4
in which case blackbody emission is seen to be proportional to abso-
lute temperature. This so-called Rayleigh-Jeans approximation signif-
icantly simplifies some kinds of radiative transfer calculations and
sensor calibration relationships in the microwave band.
120 6. Thermal Emission

Problem 6.7: Derive the Rayleigh-Jeans approximation (6.7) by ex-

panding exp(hc/k B λT ) in (6.1) as a power series and discarding
quadratic and higher order terms. State what specific condition must
be satisfied in order for the approximation to be valid. For the case
that T = 300 K, determine the minimum wavelength for which the
Rayleigh-Jeans approximation is valid to better than 1%.

6.2 Emissivity

Planck’s function Bλ ( T ) describes thermal emission from a black-

body which, as already noted, corresponds to the theoretical maxi-
mum possible emission from any real object. As such, it is an ide-
alization. In calculations, one must therefore account for the degree
to which real surfaces deviate from the ideal of a blackbody. We
therefore introduce the concept of emissivity, which is nothing more
than the ratio of what is emitted by a given surface to what would be
emitted if it were a blackbody.
There are two cases of particular interest: 1) the emissivity at a
single wavelength, and 2) emissivity over a broad range of wave-
lengths. The first case is most interesting in remote sensing applica-
tions, in which case we are primarily concerned with intensities, not
fluxes. The second is generally of greatest concern for energy trans-
fer calculations, in which case we care more about fluxes than inten-
sities. The definitions that follow reflect these biases. You should be
aware, however, that analogous, but slightly different definitions
arise in other contexts.

6.2.1 Monochromatic Emissivity

Consider a surface that emits less radiation at a given wavelength

λ and temperature T than that predicted by the Planck Function.
If the actual intensity of the emission is Iλ , then the monochromatic
 Emissivity 121

Table 6.1: Typical infrared emissivities (in percent) of various surfaces.

Water 92–96
Fresh, dry snow 82–99.5
Ice 96
Sand, dry 84–90
Soil, moist 95–98
Soil, dry plowed 90
Desert 90-91
Forest and shrubs 90
Skin, human 95
Concrete 71–88
Polished aluminum 1–5

emissivity of the surface is defined as

Iλ
ελ ≡ . (6.8)
Bλ ( T )

Note that ε λ might be a function of other variables, such as T, θ,

and/or φ. In general 0 ≤ ε λ ≤ 1. When ε λ = 1, the surface is
effectively a blackbody at that wavelength.

6.2.2 Graybody Emissivity

By analogy to the monochromatic emissivity above, one may define
a graybody emissivity ε as the ratio of the observed broadband flux
F emitted by a surface to that predicted by the Stefan-Boltzmann
relationship:

F
ε≡ . (6.9)
σT 4

Strictly speaking, no surface is truly “gray” over the full EM spec-

trum. Therefore, the use of a graybody emissivity (or absorptivity)
in calculations invariably entails an approximation. It is neverthe-
less a convenient and useful simplification, especially for the types
of problems encountered in an introductory survey of atmospheric
radiation.
122 6. Thermal Emission

Sometimes, it is useful to apply the concept of graybody emis-

sivity to a more limited range of wavelength [λ1 , λ2 ], in which case

F ( λ1 , λ2 )
ε ( λ1 , λ2 ) ≡ , (6.10)
FB (λ1 , λ2 )

where F (λ1 , λ2 ) is the actual flux emitted by the surface integrated

between λ1 and λ2 , and

λ2
FB (λ1 , λ2 ) ≡ π Bλ ( T ) dλ . (6.11)
λ1

For example, one might reasonably attempt to characterize the

emissivity of a surface in just the thermal IR band for the purpose of
simple radiative balance calculations. Examples of IR emissivities
are given in Table 6.1.

6.2.3 Kirchhoff’s Law

I already alluded to the strong connection between how well a sur-
face absorbs radiation at a given wavelength and how well it emits
at the same wavelength. It’s easy enough to confirm this experi-
mentally by taking a light colored stone and a dark colored stone
and heating them both in a furnace to the same high temperature.
Turn out the lights and observe the red glow from each stone: the
glow from the light stone will be much dimmer than that from the
dark stone. In fact, if the stone were perfectly white (nonabsorbing),
you would not be able to get it to glow at all, no matter how hot you
heated it.
In quantitative terms, the relationship between absorptivity a
and emissivity ε is embodied succinctly in Kirchhoff’s Law, which
states that

ε λ (θ, φ) = aλ (θ, φ) . (6.12)

Note that this equivalence is strictly valid only for monochromatic

radiation at a given wavelength λ and when the viewing directions
θ and φ are specified, unless a (and therefore ε) are independent of
these parameters over some range.
 Emissivity 123

I will not devote space to the derivation of Kirchhoff’s Law; suf-

fice it to say that if it did not hold, then it would be possible to con-
struct a system in which there is a spontaneous net flow of energy
from a colder object to a warmer object, in violation of the Second
Law of Thermodynamics.
It is common and convenient to apply Kirchhoff’s Law not only
to monochromatic, directional emissivities and absorptivities, but also
to graybody, flux emissivities and absorptivities specified for some
extended wavelength range. Keep in mind, however, that this is al-
ways an approximation — and sometimes a poor one — for any real
surface. This is because real surfaces generally have absorptivities
that depend on both wavelength and direction. Thus the total ab-
sorption depends on the angular and spectral distribution of the in-
cident radiation, whereas the total emission in the same wavelength
range does not.

Caveat: Local Thermodynamic Equilibrium

For all common applications in atmospheric radiation, Kirchhoff’s

Law can be taken as an absolute. It is therefore only for the sake
of completeness that I point out that Kirchhoff’s Law only applies
to systems in local thermodynamic equilibrium (LTE).2 This condition
applies, for example, when the molecules in a substance exchange
energy with each other (e.g., through collisions) much more rapidly
than they do with the radiation field or other sources of energy.
LTE, and thus Kirchhoff’s Law, breaks down at extremely high al-
titudes in the atmosphere, where collisions between molecules are
rare. LTE also breaks down in systems like lasers, fluorescent light
bulbs, gas-discharge tubes, and light-emitting diodes (LEDs), where
the average electronic energy levels of the molecules may be artifi-
cially “pumped up” by various means to levels far higher than that
expected from the thermodynamic temperature of the molecules.
The emission from such systems is therefore far greater than that
expected from a blackbody (indeed, that’s why these devices are
useful!) and therefore does not obey either Planck’s Law or Kirch-
hoff’s Law. Examples of non-LTE emission in the atmosphere in-
clude lightning discharges and the aurora (e.g. “northern lights.”).
2 See GY89 (pp. 30–32) for a more advanced treatment of this topic.
124 6. Thermal Emission

Brightness Temperature vs. Emissivity for Surface with T = 300 K

300

250
0 µm
λ=1
200
TB [K]

150
µm
100
λ=
100 m
1m
λ=

50

0
0.0 0.2 0.4 0.6 0.8 1.0
ε
Fig. 6.5: Relationship between emissivity ε and brightness temperature TB for a
surface with a physical temperature of 300 K, at three different wavelengths.

6.2.4 Brightness Temperature

Planck’s function describes a unique, one-to-one relationship be-
tween the intensity of radiation emitted by a blackbody at a given
wavelength and the blackbody’s temperature. That is to say, if you
know the intensity Iλ of radiation emitted by a blackbody, then you
can find its temperature T, and vice versa. In view of this correspon-
dence, you can convert any monochromatic intensity, regardless of
source, to an equivalent blackbody temperature, or brightness tem-
perature:

TB ≡ Bλ−1 ( Iλ ) , (6.13)

where Bλ−1 is the inverse of the Planck function applied to the ob-
served radiance.
The concept of brightness temperature is extremely useful in re-
mote sensing at infrared and microwave wavelengths. For exam-
ple, at thermal IR wavelengths, most land and water surfaces and
dense cloud layers have an emissivity ε ≈ 1. Viewed through a
transparent atmosphere, the brightness temperature of these sur-
 Emissivity 125

faces is therefore very close to their actual physical temperature.

Problem 6.8: A satellite viewing a surface location under

cloud-free conditions measures a 12 micron radiance of
6.2 W m−2 µm−1 sr−1 . (a) Compute the brightness tempera-
ture TB . (b) Compute the actual temperature, assuming that the
atmosphere is completely transparent, and that the surface in ques-
tion is known to have an emissivity of 0.9 at this wavelength. (c)
Is the ratio of the brightness temperature to the actual temperature
equal to the emissivity?

At microwave wavelengths, on the other hand, the emissivity

of some surfaces (especially water and glacial ice) is substantially
less than unity, in which case the brightness temperature may be
substantially less than the physical temperature. Nevertheless, the
Rayleigh-Jeans approximation which is valid for microwave wave-

Wavelength [µm]
25 20 18 15 14 13 12 11 10 9 8 7

140
300

120 280
Radiance [mW / m2 sr cm-1]

100
260
80
240

60
220

40 200

20

0
400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600
Wavenumber [cm-1]

Fig. 6.6: Example of an actual infrared emission spectrum observed by the Nimbus
4 satellite over a point in the tropical Pacific Ocean. Dashed curves represent
blackbody radiances at the indicated temperatures in Kelvin. (IRIS data courtesy of
the Goddard EOS Distributed Active Archive Center (DAAC) and instrument team leader
Dr. Rudolf A. Hanel.)
126 6. Thermal Emission

lengths (see Section 6.1.4) implies a direct proportionality between

intensity Iλ and brightness temperature TB . Therefore TB remains
a very convenient substitute for Iλ in radiative transfer calculations
used in microwave remote sensing.

Problem 6.9: Repeat the previous problem, only for a wavelength

of 1 cm and an intensity of 2.103 × 10−10 W m−2 µm−1 sr−1 . How
does your answer to part (c) change?

In some wavelength bands, the cloud-free atmosphere is far

from transparent. In this case, the brightness temperature observed
from space can no longer be interpreted in terms of the apparent
physical temperature of a surface but rather as a weighted average
of all of the atmospheric temperatures encountered along the line-
of-sight. The more opaque the atmosphere, the greater the altitude
of the maximum atmospheric contribution to the observed TB . This
principle is at the foundation of satellite techniques for estimating
atmospheric temperature profiles from space. We will discuss this
problem in greater detail in a later chapter.

Problem 6.10: The irregular solid curve in Fig. 6.6 depicts an ac-
tual satellite-derived spectrum of radiant intensity when the satel-
lite passed over a particular ocean location on a cloudless day. The
smooth curves represent Planck’s function curves for various tem-
peratures. (a) Estimate the brightness temperature at 11 µm. (b) Es-
timate the brightness temperature at 15 µm. (c) For which of these
two wavelengths would you guess that you are looking primarily at
emission from the surface? (d) For the wavelength that is not seeing
the surface, make a rough estimate of the approximate altitude in
the atmosphere that corresponds to the observed brightness temper-
ature, assuming a standard lapse rate of 6.5 K/km.

6.3 When Does Thermal Emission Matter?

We have seen that all surfaces reflect and absorb radiation. We
have also seen that all matter in local thermal equilibrium emits
 When Does Thermal Emission Matter? 127

thermal radiation according to its emissivity and Planck’s function.

Common experience, however, tells us that thermal emission by the
earth or atmosphere isn’t always worth worrying about. For exam-
ple, it is clear that all natural light visible to the naked eye normally
originates from extraterrestrial sources (lightning, volcanoes, and
fireflies excepted!); otherwise it would not be dark at night after the
sun goes down.
An important question, therefore, is when one can and can’t ig-
nore thermal emission from the earth and atmosphere. The sim-
plistic answer is that, for wavelengths shorter than a certain value,
the contribution of incident and reflected solar radiation to the to-
tal radiation field far exceeds the contribution due to direct ther-
mal emission. For slightly longer wavelengths, one might need to
consider both solar radiation and thermal emission from the earth-
atmosphere system itself. For even longer wavelengths, one may
often (though not always) ignore the solar component relative to
the thermally emitted component.
How short must the wavelength be in order for thermal emis-
sion to be unimportant? A common (but somewhat misleading)
way to answer that question is to compare the distributions of radi-
ation emitted by a blackbody for two temperatures: that of the sun
(approximately 6000 K) and a much cooler temperature characteris-
tic of the earth and atmosphere (200–300 K). If the Planck function is
normalized so as to have the same area under the curve (Fig. 6.2b),
we see that there is surprisingly little overlap between the two ar-
eas of significant emission. In fact, a threshold of λ ≈ 4 µm does a
very nice job of separating solar emission from terrestrial emission,
with more than 99% of solar emission taking place at wavelengths
shorter than 4 µm and more than 99% of terrestrial emission taking
place at longer wavelengths (Fig. 6.4). According to this analysis,
therefore, one might neglect the sun as a source of atmospheric ra-
diation much longer than 4 µm wavelength, and one may neglect
terrestrial emission at wavelengths much shorter than this value.
In reality, the above argument glosses over some important facts.
To begin with, while Fig. 6.2b does a nice job of showing the relative
spectral distribution of emission for terrestrial and solar sources, re-
call that we normalized the curves so that the areas would be equal!
If we had not done this, we would have seen that, for any given
128 6. Thermal Emission

wavelength, emission from a blackbody with the temperature of the

sun is far greater than that of a much cooler terrestrial source.

Problem 6.11: Use Planck’s function to compute the ratio of black-

body emission at 6000 K to that at 255 K, for a wavelength of 12 µm,
which is firmly in the thermal IR part of the spectrum.

A more convincing assessment of the relative importance of the

two sources requires us to consider the following: 1) whether it is
the flux or the intensity at a given wavelength that is of interest;
2) if the latter, then which direction is being viewed; and 3) if the
direction being viewed is not directly toward the sun, then what are
the relevant reflective properties of the surface and atmosphere.
Notwithstanding the above caution, it turns out that 4 µm is
indeed a reasonable threshold for separating thermal from solar
sources, as long as one views it only as a rough guide. A clear excep-
tion to this rule, however, occurs for satellite sensors that happen to
be looking in the direction of the reflection of the sun from a smooth
surface. Under certain conditions (e.g., over water), the intensity of
this sunglint can be quite significant relative to thermal emission,
even at microwave wavelengths, and can be an important source of
error for remote sensing methods that do not account for it.

Problem 6.12: Calculate the upward flux of reflected solar radia-

tion at 12 µm, assuming an overhead sun (temperature 6000 K, solid
angle ∆ω = 6.8 × 10−5 sr) and a surface (flux) reflectivity rλ = 0.1.
Find the ratio of the above flux to that due to terrestrial emission at
a temperature of 300 K.

6.4 Applications to Meteorology, Climatology,

and Remote Sensing
Even though we have not yet delved into the interaction of radia-
tion with the atmosphere, the relationships we defined or derived
 Applications 129

in this chapter for surfaces (either real or imaginary) already give us

powerful tools for analyzing some aspects of the radiation budget
of the earth and atmosphere, as well as for understanding simple
remote sensing applications.
Common to most of the analyses to follow is the application of
the graybody approximation separately to shortwave and longwave
fluxes. That is, we assume that a single absorptivity asw (equal to
one minus the shortwave albedo A, if a surface) applies to all short-
wave radiation and another, alw , applies to all longwave radiation.
According to Kirchhoff’s law, the longwave emissivity ε = alw , so
emitted longwave fluxes may be computed as εσT 4 . Furthermore,
it is often reasonable to assume that ε ≈ 1 in the longwave band.
Emission in the shortwave band is neglected altogether.
Given the above, the condition for radiative equilibrium is simply
that all fluxes balance at each point in the system. And if the system
is not in equilibrium, the radiative imbalance may be interpreted as
a heating or cooling rate. You may be surprised by the wide variety
of problems that can be tackled at least qualitatively using the sim-
ple approach just outlined. We will start with the simplest systems
(e.g., a single surface) and work our way up to a coupled surface-
atmosphere model that, while crude, at least hints at some impor-
tant features of the radiative interaction between the two.

6.4.1 Radiative Equilibrium in a Vacuum

The moon has no ocean or atmosphere and therefore no efficient
mechanism for the lateral redistribution of heat. Furthermore, only
a relatively shallow layer of the moon’s soil plays a role in the stor-
age and release of heat energy. To a good approximation therefore,
the equilibrium temperature at a point on the moon’s surface is deter-
mined by a balance between shortwave radiation absorbed from the
sun and longwave radiation emitted to space. Assuming the albedo
of the surface doesn’t change, the local rate of shortwave absorption
is a function only of the solar zenith angle. The emitted longwave
radiation, on the other hand, increases sharply with increasing tem-
perature according to the Stefan-Boltzmann relationship (6.5).
Therefore, if the emitted LW radiation is less than the absorbed
SW radiation, the temperature of the surface will increase. The in-
130 6. Thermal Emission

crease will continue until the LW emission has increased to the point
where it balances the SW absorption. Likewise, if the outgoing LW
emission exceeds the SW absorption, the temperature will cool until
the two are again in balance. The criterion for radiative equilibrium
is therefore
(1 − A)S0 cos θ = εσT 4 , (6.14)
where A is the shortwave albedo of the surface, θ is the zenith angle
of the sun, and ε is the longwave emissivity. Solving for T yields the
equilibrium temperature
 1/4
S0 (1 − A) cos θ
TE = . (6.15)
εσ

Problem 6.13: Assuming an albedo A = 0.10, an emissivity ε = 1,

and solar constant S0 = 1370 W m−2 , plot the equilibrium tempera-
ture TE as a function of solar zenith angle 0 ≤ θ < 90◦ .

Equation (6.15) assumes, of course, that the surface of the moon

has indefinitely long to equilibrate, a condition that is not met in
practice. It also assumes that there is no other source of shortwave
radiation. In reality, reflected moonlight from the earth, as well as
a minor contribution from starlight, ensures that the radiative equi-
librium temperature is greater than absolute zero even during the
lunar night.

Problem 6.14: (a) Derive an expression for the net flux Fnet at the
surface of the moon as a function of the actual temperature T and the
other parameters appearing in (6.15). (b) Find the numerical value of
Fnet for A = 0.10, θ = 45◦ , T = 300 K, and ε = 1.

Now consider a small, perfectly black spherical object in outer

space with radius r and exposed to direct solar radiation in the
amount of S0 (i.e., the object is at the same distance from the sun
as the earth). We can derive the equilibrium temperature TE of the
 Applications 131

Outgoing Longwave
Incoming Radiation
Shortwave
Radiation

Fig. 6.7: Relationship between incoming solar radiation and outgoing thermal in-
frared radiation for a spherical object.

object by requiring a balance between incoming solar and outgoing

longwave radiation. The first is given by the area of the shadow of
the object (i.e., its cross-sectional area) times the solar flux:

ΦSW = S0 πr2 . (6.16)

The second is given by the Stefan-Boltzmann relationship times the

surface area:
ΦLW = 4πr2 σTE4 . (6.17)
Setting the two fluxes equal to each other and simplifying yields

S0 = 4σTE4 , or (6.18)
  14
S0
TE = . (6.19)
4σ
Substituting S0 = 1370 W m−2 and σ = 5.67 × 10−8 W/(m2 K4 )
yields TE = 279 K.

Problem 6.15: Consider a spherical object like the one discussed

above, with radius r = 10 cm and heat capacity C of 1 × 104 J/K.
(a) Find an expression for the heating rate dT/dt as a function of T,
132 6. Thermal Emission

when T = TE .
(b) Assume that after the temperature reaches equilibrium, the in-
cident solar radiation is suddenly shut off – for example, the object
passes into the shadow of a planet. Use your solution to the previ-
ous part to find the temperature T of the object as a function of time
t. Ignore the effects of finite thermal conductivity.
(c) From your solution to (b), determine the time that it takes for the
object to cool to 100 K, 10 K, and 1 K.

Problem 6.16: (a) Generalize the solution for TE found above for a
black sphere to one with shortwave albedo A and longwave emis-
sivity ε.
(b) Compute the numerical value of TE for A = 0.10 and ε = 0.95.
(c) If your solution to (a) is correct, you should find that TE → ∞
when ε → 0. This would seem to imply that an object that is
highly reflective of longwave radiation (and therefore has low ε) but
strongly absorbs shortwave radiation could theoretically be heated
to an arbitrarily large temperature in a vacuum simply by exposing
it to radiation from the sun. Since the temperature of the sun itself is
finite (T ≈ 6000 K), this result would seem to violate the Second Law
of Thermodynamics. Identify the logical flaw in this analysis. Hint:
Consider the implications of Wien’s Law for this problem.

6.4.2 Top-of-the-Atmosphere Global Radiation Balance

The earth is a much more complex system than either the moon or
the spherical solid object considered in the previous system. Among
other things, the atmosphere and ocean facilitate the rapid lateral
transport of heat. Furthermore, the atmosphere itself both emits
and absorbs solar and longwave radiation. The radiation balance
at the surface of the earth cannot be considered independently of
that of the atmosphere. For both reasons, one cannot use a simple
radiative equilibrium of the type we have discussed so far as a cri-
terion for estimating local surface temperatures. Nevertheless, the
principle of balance between globally averaged absorbed solar radia-
tion and outgoing longwave radiation (OLR) must still apply at the
 Applications 133

top of the atmosphere in order for the global climate to be more or less
in a steady state over decades or centuries.
Just as we did for the sphere earlier, we compute the total ab-
sorbed solar flux (in watts) as the product of the area of the shadow
cast by the earth, the solar constant S0 , and one minus the ob-
served average planetary albedo A p of approximately 30%, which
takes into account the important contribution of clouds as well as
surface albedo.
The total flux of outgoing longwave radiation must equal the ab-
sorbed shortwave flux (Fig. 6.7). Although this OLR does not repre-
sent emission from a simple surface with a single unique tempera-
ture, it is nevertheless useful to express the average OLR in terms of
an effective emitting temperature, assuming pure blackbody emission
(i.e., ε = 1):
 
S0 (1 − A p ) 1/4
Teff = . (6.20)
4σ
Substituting the appropriate numerical values for the earth yields
Teff = 255 K.
Note that this value for Teff is about 33 K colder than the ob-
served global average surface temperature of approximately 288 K.
Recall, however, that the OLR seen from space originates not just
from the surface but also from various levels within the atmosphere
itself, most of which is substantially colder than the surface. There-
fore our calculated value of Teff is not inconsistent with reality, as
long as we recognize that our simple criterion for radiative balance
at the top of the atmosphere doesn’t give us any basis for determin-
ing the vertical distribution of temperature within the atmosphere.

Problem 6.17: If there were no meridional heat transport via the

fluid motions of the oceans and atmosphere, the mean temperature
at a given latitude would be determined by a local balance between
absorbed solar radiation and outgoing longwave radiation. Assum-
ing that the zonally averaged absorbed solar radiation remained at
the level indicated in Fig. 1.1 (this unrealistically assumes that feed-
backs due to changing cloud, ice, and vegetation cover are inopera-
tive), use the Stefan-Boltzmann relationship to estimate the effective
mean emitting temperature (as seen from the top of the atmosphere)
that would prevail at the poles and at the equator. Then add 33◦ C
134 6. Thermal Emission

Shortwave Longwave

1 3 5 7

asw Atmosphere alw Ta

2 4 6 8

A Surface ε Ts

Fig. 6.8: Schematic depiction of the radiative coupling between the surface and a
thin isothermal atmosphere.

to estimate the corresponding surface temperatures (this again un-

realistically assumes that mean lapse rate and atmospheric opacity
would not change from current values).

6.4.3 Simple Radiative Models of the Atmosphere

Single Layer, Nonreflecting Atmosphere
Let us now consider a system consisting of an opaque surface and
an overlying atmosphere that we will model as a single thin translu-
cent sheet having uniform temperature. The temperature of the sur-
face is denoted Ts and that of the atmosphere Ta . The surface has
longwave emissivity ε = 1 and shortwave albedo A. The atmo-
sphere has corresponding absorptivities alw and asw (for the time
being, we will assume that the atmosphere only absorbs but does
not scatter or reflect either shortwave or longwave radiation). Fi-
nally, at the top of the atmosphere, the globally averaged incident
shortwave flux from the sun is S = S0 /4.
 Applications 135

With the aid of the parameters given above, we can now be-
gin to construct expressions for the upwelling and downwelling
shortwave and longwave fluxes at the top of the atmosphere and
between the atmosphere and the surface. Fig. 6.8 depicts the vari-
ous components of these fluxes, and their physical interpretation is
given here:

F1 incident shortwave flux from the sun

F2 transmitted portion of F1
F3 transmitted portion of F4
F4 shortwave flux reflected by the surface
F5 longwave emission upward by atmosphere
F6 longwave emission downward by atmosphere
F7 transmitted portion of F8
F8 longwave emission by surface

Note that if we did not specify ahead of time that ε = 1, we

would have also had to include terms that describe the upward re-
flection of F6 from the surface.
We can write expressions for each of the flux terms using the
parameters specified at the beginning of this section and the basic
principles introduced earlier in this chapter. Make sure you under-
stand how each of the following expressions was obtained!

F1 = S , (6.21)

F2 = (1 − asw ) F1 = (1 − asw )S , (6.22)

F3 = (1 − asw ) F4 = A(1 − asw )2 S , (6.23)
F4 = AF2 = A(1 − asw )S . (6.24)
In the following two expressions, we are invoking Kirchhoff’s Law,
which tells us that the longwave emissivity of the layer is the same
as its absorptivity:
F5 = alw σTa4 , (6.25)
F6 = F5 = alw σTa4 , (6.26)
F7 = (1 − alw ) F8 = (1 − alw )σTs4 , (6.27)
F8 = εσTs4 = σTs4 . (6.28)
136 6. Thermal Emission

The condition for radiative equilibrium is that the net fluxes

(shortwave and longwave combined) at the top of the atmosphere
and between the surface and the atmosphere are both zero. If this
were not the case, then either the atmosphere or the surface, or both,
would experience a net gain or loss of energy over time, leading to
heating or cooling.
If we take the fluxes in the above table to be positive quanti-
ties, we thus have the following two conditions for radiative equi-
librium:

Fnet,top = F3 + F5 + F7 − F1 = 0 , (6.29)
Fnet,sfc = F4 + F8 − F2 − F6 = 0 . (6.30)
Substituting our previous expressions for each flux term yields

A(1 − asw )2 S + alw σTa4 + (1 − alw )σTs4 − S = 0 , (6.31)

A(1 − asw )S + σTs4 − (1 − asw )S − alw σTa4 = 0 , (6.32)

which we rearrange to give

(1 − alw )σTs4 + alw σTa4 = S[1 − A(1 − asw )2 ] , (6.33)

σTs4 − alw σTa4 = (1 − A)(1 − asw )S . (6.34)

Note that if we define x = σTs4 and y = σTa4 , then we have a pair of
coupled linear equations in the two unknowns x and y. We can first
solve these for x and y and then divide by σ and take the fourth root
to get Ts and Ta :
   14
S 2 − asw
Ts = [1 − (1 − asw ) A] , (6.35)
σ 2 − alw
   14
S (1 − A)(1 − asw ) alw + [1 + (1 − asw ) A] asw
Ta = . (6.36)
σ (2 − alw ) alw
These expressions look complicated, but if we focus just on Ts and
consider a couple of limiting cases, we can quickly get some useful
insight into the role of the atmosphere in controlling global surface
temperature.
 Applications 137

The simplest case is when alw = 0 and asw = 0. This is equiv-

alent to saying that there is no atmosphere at all, because it is com-
pletely invisible as far as both shortwave and longwave radiation is
concerned. In fact, if you substitute these values and S = S0 /4 into
(6.35) and simplify, you immediately recover (6.20), which is what
you would expect.
Now consider the case that the surface is completely black (i.e.,
A = 0), while alw and asw may each be nonzero. We then have
   14
S0 2 − asw
Ts = . (6.37)
4σ 2 − alw

This relationship is interesting, because it tells us that if alw > asw ,

then the surface temperature in our simple system will be warmer
than would be the case without an atmosphere (alw = asw = 0) or,
for that matter, any case where alw = asw .
In fact, our atmosphere is relatively transparent to shortwave ra-
diation from the sun, while being comparatively absorbing at ther-
mal IR wavelengths. We can approximate the earth’s atmosphere
by choosing asw = 0.1 and alw = 0.8. Thus, we expect the surface
temperature of the earth to be substantially warmed by the absorp-
tion and re-emission of longwave radiation by the atmosphere. This
warming effect due to the presence of an atmosphere is commonly
known as the Greenhouse effect.
The above analysis assumes that the atmosphere either trans-
mits or absorbs all of the solar radiation that is incident on it. It
therefore cannot be used directly to obtain a realistic estimate of
the equilibrium surface temperature of the earth. However, we can
compensate for the loss of radiation due to reflection by clouds etc.
by replacing S with S(1 − A p ), where A p = 0.30 is the observed
planetary albedo utilized earlier in our calculation from (6.20). We
are in effect pretending that all the solar radiation reflected to space
by clouds etc. experiences this reflection at the top of the atmo-
sphere, before having any opportunity to be absorbed by the sur-
face or atmosphere. With this simplifying assumption, we obtain a
new surface equilibrium temperature of 286 K, as contrasted with
255 K obtained earlier for the earth with the same planetary albedo
but no atmosphere. In view of the crudeness of our model, this is
surprisingly close to what is actually observed!
138 6. Thermal Emission

One of the chief concerns of climate scientists today is that, by

adding carbon dioxide and other “greenhouse gases” that are trans-
parent to shortwave radiation but strongly absorb longwave ra-
diation, humankind is inexorably increasing the value of alw and
thereby shifting the radiative equilibrium of the globe to warmer
temperatures.
Of course, a far more elaborate computer model of the earth and
atmosphere, one that incorporates fluid motions, cloud and water
vapor feedbacks and other complex processes, is required in order
to study this problem in any detail. The most sophisticated such
models are called general circulation models (GCMs) and are the best
tools we have today for understanding climate and climate change.
Even now, however, a fully comprehensive treatment of radiation
and other physical processes remains too complex a problem for
the most powerful computers to tackle for the entire atmosphere at
once. GCMs therefore continue to rely on grossly simplified rep-
resentations of some of these processes, with the attendant risk of
error in the models’ predictions. Finding ways to improve the ac-
curacy of radiative and other physical parameterizations within the
constraints of the available computing power is a major focus of
current research in atmospheric science.

Problem 6.18: In the example above, we set the surface albedo A

to zero and then reduced the assumed global insolation S by 30% in
order to crudely account for the solar radiation that was reflected
rather than absorbed by the earth-atmosphere system. We could
have equally well left S = S0 /4 and then chosen a larger A such
that exactly 1 − A p = 70% of the solar radiation was absorbed by
either the surface or the atmosphere.
(a) Using this revised approach, find an expression for the appropri-
ate choice of A in terms of A p and asw .
(b) Calculate the numerical value of A, given A p = 0.30 and asw =
0.1.
(c) Calculate the resulting Ts , given the above parameters and alw =
0.8.
 Applications 139

6.4.4 Nighttime Radiative Cooling

In the previous section, we discussed the criteria for radiative equi-
librium in a simple two-level model, based on an assumed balance
between the long-term globally averaged LW fluxes and the long-term
globally averaged SW flux from the sun. Of course, the solar input at
any location is not constant with time but varies from zero at night
to a maximum at around noon, as well as varying with latitude and
season. Also both SW and LW fluxes at the surface vary sharply
with temperature, humidity, and cloud cover. Although it remains
a very crude representation of reality, we can nevertheless invoke
our two-level model to learn a few things about short-term, local ra-
diative processes.
In particular, let us examine the basis for the well-known rule of
thumb that dew or frost is far more likely to occur on clear nights
than cloudy nights. Both dew and frost occur when the surface cools
to below the dewpoint or frostpoint temperature of the air, while the
air above the surface may remain at a significantly warmer temper-
ature. The latter observation rules out direct thermal conduction
(through physical contact of the air with the surface) as the reason
for the cooling of the surface, as heat always conducts from warm
to cold. The only other obvious candidate is radiation.
Because we are addressing a night-time situation, we don’t need
to worry about SW fluxes. Furthermore, we’re not interested in es-
tablishing the criteria for radiative equilibrium but rather in esti-
mating the surface cooling rate under conditions of disequilibrium.
We therefore need only consider flux components 6 and 8 in Fig. 6.8,
which we will refer to simply as F ↓ and F ↑ for the purposes of this
discussion:
F ↓ = alw σTa4 , (6.38)

F ↑ = εσTs4 = σTs4 , (6.39)

where, as before, we are taking the LW surface emissivity ε ≈ 1.
The cooling rate of the ground is proportional to the net flux

Fnet = F ↑ − F ↓ = σTs4 − alw σTa4 , (6.40)

or
Fnet = σ( Ts4 − alw Ta4 ) . (6.41)
140 6. Thermal Emission

The effective value of alw for the cloud-free atmosphere ranges from
approximately 0.7 in the wintertime arctic to approximately 0.95 in
the tropics. This variation is driven primarily by the humidity of
the atmosphere, as water vapor is a strong absorber of radiation
over much of the thermal IR band. The corresponding range of ef-
fective atmospheric temperature Ta is 235 K (arctic winter) to 290 K
(tropical), yielding typical clear-sky downwelling longwave fluxes
F ↓ ranging from a minimum near ∼120 W m−2 to a maximum of
∼380 W m−2 .
For a midlatitude winter situation, we may use alw = 0.8 and
Ta = 260 K, and take the initial surface temperature to be Ts =
275 K. Plugging these values into (6.41) yields a positive (upward)
net flux of 117 W m−2 . On level ground with no wind, very little
of this heat loss from the ground is shared with the overlying air, so
the surface temperature of the ground falls rapidly.
A crude estimate of the rate of the temperature fall may be had
by noting that heat conduction in soil is rather slow, so that only the
top few centimeters of soil experience the fluctuation of tempera-
ture associated with the diurnal (day-night) cycle. If we somewhat
arbitrarily choose an effective depth over which to average the cool-
ing as ∆Z = 5 cm, and use a typical soil heat capacity (per volume)
of C ≈ 2 × 106 J m−3 K−1 , then we have
dT − Fnet K
≈ ≈ −4.2 . (6.42)
dt C∆Z hr
You can see that it will not take long for the ground temperature to
drop below freezing and, presumably, below the frost point of the
overlying air, at which point frost will start to deposit (or sublime)
directly onto the surface.
The above calculation assumed a cloud-free atmosphere. What
happens when we introduce a low-level opaque cloud deck whose
temperature is only a few degrees below that of the surface? Taking
alw = 1 and Ta = 270 K, we now find a net surface flux of only
22 W m−2 , or less than a fifth of value for a clear sky. The cooling
rate of the surface is reduced by the same ratio and is now only
0.8 K/hr. What a difference a cloud makes!
Of course, other processes, such as surface latent and sensible
heat fluxes, while small (unless there is a significant breeze!), par-
tially offset the radiational cooling predicted by the above simple
 Applications 141

analysis. Nevertheless, you should now be persuaded that radia-

tion can have observable meteorological effects even on rather short
time scales.

6.4.5 Radiative Cooling at Cloud Top

A similar analysis can be applied to the top of a continuous cloud
layer, such as the low-level marine stratocumulus clouds that are
common within midlatitude and subtropical high pressure zones
over relatively cool ocean areas, such as the Eastern Pacific near Cal-
ifornia and Peru or in the vicinity of the Azores near North Africa
and the Iberian Peninsula.
Although the processes involved in the steady-state mainte-
nance of these persistent cloud sheets is complex and includes tur-
bulent fluxes of sensible heat and moisture, it is at least possible to
evaluate the potential role of radiative fluxes as one component in
the overall balance.
To begin with, because the cloud layer is opaque to LW radia-
tion and does not reflect appreciably at these wavelengths, we can
consider it as having two radiating “surfaces”, each with emissiv-
ity ε ≈ 1. The base height Zbase for marine stratocumulus clouds is
commonly near 300 m altitude, below which the air follows a dry
adiabatic temperature profile of 9.8 K/km. This implies a tempera-
ture difference between cloud base and the ocean surface of only
3 K. Taking the surface temperature to be Ts = 288 K, we have
Tbase = 285 K and a net LW flux at cloud base of
Fnet,base < σ( Ts4 − Tbase
4
) ≈ 16 W m−2 . (6.43)
The use of the ‘<’ sign is intended to indicate that the above is an
upper bound on the net flux at cloud base, because not all the radi-
ation reaching cloud base originates at the surface; some is emitted
at higher, slightly cooler altitudes on account of the high opacity of
the atmosphere at some wavelengths.
Within the cloud, the moist adiabatic lapse rate of approximately
6 K/km prevails. Therefore, at cloud top, typically near 1 km al-
titude, the temperature is approximately 4 K colder than at cloud
base, yielding Ttop ≈ 281 K. Because marine stratocumulus clouds
typically occur in regions of high pressure and therefore subsid-
ing air aloft, the atmosphere above cloud top is often warm but
142 6. Thermal Emission

very dry. The downward flux of radiation from the overlying at-
mosphere may therefore be estimated as alw σTa4 , where we assume
Ta ≈ 280 K and alw ≈ 0.8:

Fnet,top ≈ σ( Ttop
4
− alw Ta4 ) ≈ 211 W m−2 . (6.44)

In summary, we have substantial radiative cooling (∼211 W m−2 )

at cloud top but relatively little radiative warming (∼16 W m−2 ) at
cloud base. There is therefore net radiative cooling of ∼195 W m−2
for the cloud layer as a whole. Note, by the way, that any time you
cool a layer of air from above, you increase the lapse rate. Once the
lapse rate exceeds the applicable adiabatic lapse rate (in this case the
moist adiabatic lapse rate), you destabilize the layer, leading to ver-
tical overturning. The same applies to warming from below. Conse-
quently, the net radiative cooling experienced by the cloud ends up
being distributed pretty much throughout the entire atmospheric
boundary layer (or mixed layer), which extends from the surface to
cloud top. In the present example, this represents ∼195 W m−2 of
cooling distributed over a 1 km depth of air. We can compute the
cooling rate as

dT Fnet,base − Fnet,top
≈ ≈ −14 K/day , (6.45)
dt rad C p ρ∆Z

where C p = 1004 J kg−1 K−1 is the heat capacity at constant pressure

of air, ρ ≈ 1.2 kg m−3 is the density of air near the surface, and
∆Z = 1 km.
In reality, we know that the temperature in the marine boundary
layer is close to a steady state, implying that there must be compen-
sating inputs of energy in the form of surface latent and sensible
heat fluxes and turbulent fluxes at cloud top. In fact, the convective
overturning that results from cloud-top radiative cooling is respon-
sible for mixing between the cloud and the warmer, drier overlying
air, which plays an important role in the overall mass and energy
balance of the boundary layer. Also, it is responsible for the fact
that stratocumulus cloud layers viewed from above usually have
the appearance of rolls and “cells” rather than appearing as feature-
less, flat sheets of cloud.
 Applications 143

Fig. 6.9: A 10.7 µm image of the Eastern Pacific and west coast of North America
taken by the GOES-West geostationary weather satellite.

6.4.6 IR Imaging from Space

For some ranges of wavelength in the thermal IR and microwave
bands, the cloud-free atmosphere is rather transparent. We refer to
such a region as a spectral window. Within the thermal IR band, the
most important spectral window encompasses wavelengths in the
vicinity of 11 µm.
Almost all operational weather satellites are equipped with at
least one imaging channel in this window, typically with surface
spatial resolutions of approximately 4 km at nadir. An example of
a standard satellite 10.7 µm image from a geostationary satellite is
shown in Fig. 6.9. Similar products are also available from most
polar orbiting weather satellites.
Because the atmospheric gases themselves are nearly transpar-
ent (apart from a slight degree of absorption by water vapor), con-
ventional satellite IR imagery typically records thermal emission
144 6. Thermal Emission

from the surface and from the tops of any clouds present. It is con-
venient to express IR radiance observations at brightness tempera-
tures (see Section 6.2.4). The brightness temperature of a scene with
emissivity ε and temperature T is given by
TB = Bλ−1 [εBλ ( T )] . (6.46)
When ε ≈ 1, the above reduces to TB ≈ T. Since land and water sur-
faces, most clouds (excepting thin cirrus), and even snow-covered
land surfaces are rather black (ε ≈ 1) at thermal IR wavelengths, the
observed brightness temperature is in fact a reasonably good mea-
sure of the physical temperature.
IR imagery from satellites is most often depicted using white or
light gray for the coldest brightness temperatures and black or dark
gray for the warmest brightness temperatures. Thus, hot land sur-
faces (T ≈ 310 K or more) during the summer (especially daytime)
appear black, and cold land surfaces during the winter (T < 250 K)
appear medium gray.
The coldest scenes of all, however, are often associated not with
the surface but rather with high cirrus clouds and deep thunder-
storms, whose temperature at cloud top may occasionally be colder
than 200 K (−73◦ C). The strong brightness temperature contrast be-
tween deep thunderstorms and surrounding shallower, and there-
fore warmer cloud tops makes IR imagery particularly useful for
operational detection of developing severe weather and even for
estimating probable rainfall amounts. In this respect, satellite IR
imagery is complementary to VIS imagery, which depicts all clouds
as relatively bright features, regardless of cloud-top height. Also, IR
imagers are not constrained by the availability of sunlight but yield
high quality images of cloud systems both day and night.

Problem 6.19: An IR imager observes a cloud layer with TB =

240 K. The cloud radiates as a blackbody and the surface air tem-
perature near that location is known to be 15◦ C.
(a) Assuming a standard atmospheric lapse rate of Γ = 6.5 K/km,
estimate the height of the cloud layer top.
(b) Assuming the cloud top is a blackbody over the entire LW band,
compute the upward flux of longwave radiation (OLR) that would
be observed by an aircraft flying directly over the cloud.
 Applications 145

If you view a movie loop of GOES IR imagery covering a full

day or longer, you can see not only the evolution of cloud patterns
associated with extratropical cyclones, thunderstorms, etc., but also
the evolution in surface temperature. It is not uncommon to see
the land surface temperature vary sharply between day and night,
while ocean surfaces remain at relatively constant temperature over
long periods.

Problem 6.20: Examine Fig. 6.9 for clues as to the most likely sea-
son and/or time of day when the image was taken. Explain your
reasoning.

IR imagery is currently the primary basis for mapping global sea

surface temperatures (SSTs) at high precision (approaching 0.1 K),
allowing the depiction of the Gulf Stream and other features of im-
portance for climate and weather forecasting, as well as for fish-
eries and for physical oceanographic studies. In particular, the well-
known El Niño phenomenon, which dramatically influences large-
scale weather patterns around much of the world, is manifested in
the form of a dramatic warming of SSTs over the Eastern Tropical
Pacific ocean. In situ measurements of SST by buoys and ships are
rather sparse in this region, so IR imagery is the principal method
today for monitoring the onset and evolution of El Niño.
Keep in mind, however, that accurate IR observations of surface
temperature are only possible in cloud-free areas and with careful
attention to the correction of minor atmospheric effects, mostly due
to water vapor. One way to isolate, and correct, contamination due
to water vapor is to utilize two channels at nearby wavelengths but
with differing sensitivity to absorption by water vapor. The differ-
ence in brightness temperature between the two images is then an
indicator of how much water vapor is present. In fact, one common
form of this so-called split window technique entails the simultane-
ous retrieval of both SST and total column water vapor, also known
as precipitable water. Both variables are of great interest to meteorol-
ogists, though for different reasons.
When even thin or broken clouds are present, which is true a
majority of the time at many locations around the globe, IR imagers
146 6. Thermal Emission

can no longer accurately measure surface properties. When surface

measurements are required in the presence of clouds, we turn to the
microwave band.

6.4.7 Microwave Imaging from Space

Passive microwave remote sensing is in many respects similar to IR
remote sensing, in that a satellite sensor views naturally emitted ra-
diation from the earth and atmosphere.3 Especially at wavelengths
greater than 3 cm (frequencies less than 10 GHz), atmospheric ef-
fects are fairly small, and the primary variable observed in this case
is thermal emission from the earth’s surface. There are several im-
portant differences from IR imagery that are worth briefly noting.
For one thing, it takes a rather heavy cloud to seriously interfere
with a microwave imager’s ability to observe the surface, in con-
trast to IR imagers, which are foiled by even thin clouds. In fact, at
low microwave frequencies, only a cloud that is actually producing
rainfall presents a serious obstacle.
Also, in contrast to the IR band, the microwave emissivities of
natural surfaces are often considerably less than unity and may in
fact vary wildly from one scene to the next. In particular, ocean
surface emissivities may be as small as 0.25–0.7, whereas land sur-
face emissivities are typically in the range 0.8–0.95. This variation in
surface emissivity makes it much more difficult (though not impos-
sible) to use microwave imagery for accurately estimating surface
temperature, but greatly facilitates the determination of other sur-
face properties.
A particularly convenient property of the microwave band is the
validity of the Rayleigh-Jeans approximation introduced in Section
6.1.4. Because of the accurate proportionality between blackbody
radiance and temperature in the microwave band, the slightly cum-
bersome relationship between T, ε, and TB represented by (6.46)
simplifies down to just
TB = εT . (6.47)

3 Active remote sensing is also employed in the microwave band but entails the

measurement of backscattered radiation from an artificial source. Radar is the best-

known example.
 Applications 147

Elsewhere in this book, we tend to neglect the polarization prop-

erties of radiation. But polarization cannot be ignored for mi-
crowave remote sensing. In particular, the ocean surface emissiv-
ity is markedly higher for vertical polarization than for horizontal
polarization when the ocean is viewed at an oblique angle. In fact,
if the water surface is perfectly smooth, the reflectivity r, and thus
the emissivity ε = 1 − r, is given by the Fresnel relations discussed
in Chapter 4. Fig. 4.5b showed the Fresnel reflectivity of water for
a microwave frequency of 37 GHz. For a satellite sensor viewing
a water surface at a typical angle from vertical (i.e. nadir angle) of
∼50◦ , the horizontally polarized emissivity ε h ≈ 0.35 while the ver-
tically polarized emissivity ε v ≈ 0.65. Thus, for the same physical
temperature of T = 283 K, the brightness temperature of emission
from the ocean surface is approximately 100 K for horizontal polar-
ization and close to 185 K for vertical polarization, an 85 K differ-
ence.
Things get even more interesting when you consider that the
microwave emissivity of the ocean surface is not a constant at any
given frequency but depends (albeit fairly weakly) on the sea sur-
face temperature Ts (via changes in the complex index of refraction
N of seawater) and on the degree of roughening and foam cover-
age associated with a given near-surface wind speed U. Thus, the
observed brightness temperature can be represented as

TB,p ≈ ε p ( Ts , U ) Ts + ∆atmos , (6.48)

where p represents the polarization (V or H) and ∆atmos represents

an atmospheric correction, which is usually small (<10 K) at low
microwave frequencies.
The typical increase of TB in response to increasing ocean surface
wind speed is of the order of 1 K per m/s for horizontal polariza-
tion, which is easily measurable. The net sensitivity of TB to chang-
ing sea surface temperature, on the other hand, can be very small
or even be negative at some frequencies. This is because ε tends to
decrease with increasing Ts .
By making observations at multiple microwave frequencies
and/or polarizations, and provided a reasonably accurate model
for ε p ( Ts , U ) and ∆atmos is available, it is possible to infer sea sur-
face temperature Ts to an accuracy of a few tenths of a degree and
148 6. Thermal Emission

near-surface wind speed U to better than ±2 m/s.

Over land, it is more difficult to express the relationship between
microwave TB and surface properties in a concise way. As a general
rule, the surface emissivity is rather high over most land surfaces
but fluctuates markedly with soil type, soil wetness, and vegetation
density. All of these variables can be estimated with varying de-
grees of success using satellite microwave imagers. In addition, the
presence of snow can markedly depress brightness temperatures at
higher microwave frequencies, making it possible to use microwave
imagers to monitor snow extent and depth over remote areas of the
globe even under cloudy skies.4
A significant drawback to the microwave band for remote sens-
ing is that the longer wavelength limits the “sharpness,” or resolu-
tion, with which a satellite sensor can observe the earth. While an
IR imager can easily resolve details as small as 4 km from even a
geostationary satellite at an altitude of around 36,000 km, a current-
generation microwave imager in low earth orbit (typical altitude
800–1000 km) can resolve SST patterns down to only about 30-
50 km.

4 In addition to the surface properties just discussed,certain atmospheric prop-

erties, such as total column water vapor, total column cloud water, and surface rain
rate, can also be retrieved over ocean using microwave imaging channels at about
18 GHz and higher. For these channels, the atmospheric contribution to emission
seen from space is nonnegligible and depends strongly on the above variables.
 CHAPTER 7

Atmospheric Transmission

Up to this point, our discussion of the interaction of radiation with

matter has focused mainly on reflection, refraction, absorption, and
emission by surfaces, either real or “virtual” (note that in the last
chapter we crudely treated the atmosphere as a translucent, gray,
isothermal “surface” for the purpose of some simple radiation bud-
get calculations).
Only in Section 2.5 did we briefly consider the absorption of
a monochromatic EM wave propagating through a homogeneous
medium whose index of refraction N included a nonzero imaginary
part. I pointed out that, in this homogeneous absorbing medium,
the intensity Iλ falls off exponentially with distance:

Iλ ( x ) = Iλ,0 exp(− β a x ), (7.1)

where β a is an absorption coefficient that depends on the physical

medium and on the wavelength of the radiation.
We are now ready to generalize this result to the atmosphere.
Specifically, we will have to adjust (7.1) to allow for two minor com-
plications:

1. The fact that a beam of radiation can be attenuated not only

by absorption (i.e., conversion of the energy of the radiation to

149
150 7. Atmospheric Transmission

Fig. 7.1: Scattering of an incident wave by a particle.

heat or chemical energy) but also by scattering (redirection of

radiation out of the original direction of propagation, usually
due to interactions with particles; Fig. 7.1). This will require
us to define a more general extinction coefficient to replace the
absorption coefficient β a appearing above.

2. The likelihood that the strength of the absorption or scattering

will vary significantly along the path, so that our mathemat-
ical starting point must be the differential equivalent to (7.1),
in which we initially consider the change of intensity of the ra-
diation as it traverses an infinitesimal distance ds, over which
the extinction coefficient can be assumed constant.

In addition to the above two minor complications, there is one

more major complication that I will defer until a later chapter: the
fact that radiation can be scattered from other directions into the di-
rection of propagation, thereby enhancing its intensity. But for now
we will continue to focus strictly on line-of-sight transmission and
depletion of radiation, which can be understood entirely in terms of
local properties encountered along that line of sight.
 151

Let us start by illustrating the first of the above points with the
help of a classic classroom demonstration1 in which the instructor
places three transparent Petri dishes partly filled with water onto
an overhead projector and projects their silhouettes onto the screen.
Because the water and the dishes are transparent, nearly all the light
from the projector passes through unattenuated, so that the shad-
ows cast by the dishes are barely perceptible.
Into one of the dishes the instructor then introduces a few drops
of India ink; into another, a tablespoon or so of milk (Fig. 7.2a).
Those two dishes now cast strong shadows, indicating that the light
from the projector has been significantly attenuated or extinguished
(Fig. 7.2b). The mechanisms for the extinction are quite different,
however. In the case of the black ink, the energy carried by the
light is largely absorbed and, presumably, converted to heat. In the
case of the white milk, the light lost from the original beam is not
absorbed but merely scattered out of its original direction of propa-
gation and into all other directions.
Here are two ways to convince yourself of the truth of the above
statement:

1 C. Bohren, 1987: Clouds in a Glass of Beer. Wiley, New York

Fig. 7.2: Demonstration of radiative extinction using an overhead projector. (a)

Three dishes of liquid, as positioned on the light source of the projector. The clear
liquid is water, the black liquid is diluted ink (an absorbing medium), and the
white liquid is diluted milk (a scattering medium). (b) The projected images of the
three dishes. The dark shadows for milk and ink demonstrate that absorption and
scattering are equally effective at depleting transmitted radiation.
152 7. Atmospheric Transmission

• View the dishes from the side against a dark background: the
milk “lights up” strongly against this background due to the
illumination from the projector bulb, while the ink does so
only weakly if at all. This suggests that light from the origi-
nal beam is being redirected toward your eye in the first case
but not the second.

• Place the bulbs of two identical thermometers in the ink and

milk dishes and monitor the temperature change with time.
The temperature of the dish with the ink will rise faster than
that with the milk, indicating a greater rate of absorption of
energy from the light beam.

Despite the difference in the mechanisms by which the original

light beam is depleted, or extinguished, common sense tells us that
the rate of depletion in either case should be proportional to the in-
tensity of the incident radiation. In other words, if the illumination
is zero (the projector is shut off), the depletion is zero as well (you
can’t take away energy from a nonexistent beam!). If one doubles
the intensity of the illumination, the rate of absorption (in the case
of the ink) or scattering (in the case of the milk) is doubled as well.
Both of these expectations (which are easily confirmed experi-
mentally) are consistent with a relationship of the form of (7.1), pro-
vided that we replace the absorption coefficient β a with an extinc-
tion coefficient β e that accounts for depletion due to either scattering
or absorption or, more likely, a combination of both. We’re also as-
serting (without proof), by the way, that we can successfully adapt
this relationship to an inhomogeneous medium (such as a cloud), as
long as we restrict our attention to the extinction of the original di-
rect beam and ignore any enhancement via the scattering of radiation
from other directions into the line of sight.
Now consider what happens if you add a drop or two of India
ink to the Petri dish containing the milk solution. If any significant
fraction of the light had succeeded in passing through the dish be-
fore, it is now further reduced by the addition of the ink. Moreover,
it is now clear that some fraction of the light that is extinguished is
being absorbed while the remainder is scattered. That is, the solu-
tion is now neither white nor black, but gray.
 Extinction, Scattering and Absorption Coefficients 153

7.1 Extinction, Scattering and Absorption Coef-

ficients
One may characterize the relative contributions of scattering and
absorption to the total extinction by considering the extinction co-
efficient to be the sum of an absorption coefficient and a scattering
coefficient:
βe = βa + βs . (7.2)

In the case of the milk solution, β a ≈ 0, so that β e ≈ β s . For the ink

solution, β s ≈ 0 so that β e ≈ β a . Note, by the way, that all three of
these parameters have physical dimensions of inverse length.
To more conveniently characterize the relative importance of
scattering versus absorption in a medium, radiative transfer special-
ists have introduced an important quantity called the single scatter
albedo, which is defined as follows:

βs βs
ω̃ = = . (7.3)
βe βs + βa

The value of ω̃ ranges from zero in a purely absorbing medium to

one in a purely scattering medium. From an optical standpoint at
least, the fundamental difference between milk and India ink, there-
fore, is that ω̃ ≈ 1 for milk whereas ω̃ ≈ 0 for the ink. A mixture of
milk and ink would have a value of the single scatter albedo some-
where in between.
At this point it is worth reminding you that the quantities dis-
cussed in this chapter have values that are usually very specific to
the wavelength being considered. Milk and India ink are somewhat
unusual in that β e , ω̃, etc., are more or less constant across the entire
visible band. If that were not the case, they would have a percepti-
ble color other than white, gray, or black.

Problem 7.1: Based on what you know about their appearance, at-
tempt to characterize the apparent spectral dependence of β e and
ω̃ for the following substances: red wine, chocolate milk, a cloud,
154 7. Atmospheric Transmission

Iλ(s1) Iλ(s) Iλ(s+ds) Iλ(s2)

s1 s s+ds s2

Fig. 7.3: Depletion of radiation over an infinitesimal path ds within an extinguish-

ing medium.

a plume of exhaust from a diesel truck, the cloud-free atmosphere

(based on the color of the setting sun).

7.2 Extinction Over a Finite Path

7.2.1 Fundamental Relationships
Let us now consider how to adapt (7.1) to a radiative path over
which the extinction coefficient varies with location, since this is the
situation we encounter in the atmosphere. We do not want to limit
our attention to radiative paths aligned with the x-axis, so we gener-
alize by replacing x with the geometric distance s along a ray in any
arbitrary direction (Fig. 7.3). We further consider the attenuation of
radiation over an infinitesimal path ds, which is chosen to be small
enough so that (a) the extinction coefficient β e is effectively constant
within the interval, and (b) the incident radiation is attenuated by
an infinitesimal amount dIλ . We can then write

dIλ ≡ Iλ (s + ds) − Iλ (s) = − Iλ (s) β e (s) ds, (7.4)

which we can rewrite as

dIλ
≡ d log Iλ = − β e ds. (7.5)
Iλ
 Extinction Over a Finite Path 155

In other words, the infinitesimal decrease in intensity dIλ , ex-

pressed as a fraction of the incident intensity Iλ , is equal to the prod-
uct of the local extinction coefficient times the infinitesimal path
length ds.
In order to describe the extinction over an extended path between
points s1 and s2 , we simply integrate:

s2
log[ Iλ (s2 )] − log[ Iλ (s1 )] = − β e (s) ds, (7.6)
s1
or

s2 
Iλ (s2 ) = Iλ (s1 ) exp − β e (s) ds . (7.7)
s1

This equation gives a general form of Beer’s law. From Beer’s law
follow several extremely important definitions and associated facts
which I urge you to memorize before you continue reading:
• The integral quantity inside the brackets is called the optical
path between points s1 and s2 (also known as optical depth or
optical thickness when measured vertically in the atmosphere) :

s2
τ ( s1 , s2 ) ≡ β e (s) ds . (7.8)
s1

The optical path is dimensionless, as it must be, since it ap-

pears as the argument of the transcendental function exp(). It
may take on any nonnegative value. It is zero when s1 = s2 or
when β e = 0 between s1 and s2 ; otherwise it is positive.
• By exponentiating the optical path τ, we get the transmittance
between s1 and s2 :

t(s1 , s2 ) ≡ e−τ (s1 ,s2 ) . (7.9)

The transmittance is a dimensionless quantity ranging from

near zero (for τ → ∞) to one (for τ = 0). From (7.7) we see
that
Iλ (s2 ) = t(s1 , s2 ) Iλ (s1 ) . (7.10)
156 7. Atmospheric Transmission

Thus, t ≈ 1 implies very weak attenuation of the beam be-

tween s1 and s2 , whereas t ≈ 0 implies near total extinction of
the beam. Note that no matter how large τ, the transmittance
is never identically zero, though it is often vanishingly small.
• If β e happens to be constant between s1 and s2 , then (7.9) sim-
plifies to

τ = β e ( s2 − s1 ) . (7.11)

• Each (dimensionless) unit of optical path corresponds to a re-

duction of Iλ to e−1 ≈ 37% of its original value.
• Consider the propagation of a ray along an extended path
from s1 to s N . Break that path into several sub-paths; e.g.,
from s1 to s2 , s2 to s3 , s N −1 to s N , etc. Then from (7.8) it can
be readily shown that the total optical path can be written

τ ( s 1 , s N ) = τ ( s 1 , s 2 ) + τ ( s 2 , s 3 ) + · · · + τ ( s N −1 , s N ) ,

(7.12)
and the corresponding transmittance can be written

t ( s 1 , s N ) = t ( s 1 , s 2 ) · t ( s 2 , s 3 ) · · · · · t ( s N −1 , s N ) . (7.13)

In plain English, (1) the total optical path equals the sum of the
individual optical paths; and (2) the total transmittance equals the
product of the individual transmittances.
• Consider the propagation of a ray over an optical path
τ (s1 , s2 )  1. That is, the medium is relatively transparent
over the distance in question. This can occur if either that ge-
ometric distance ∆s is sufficiently short or the extinction coef-
ficient β e is sufficiently small. In this case, the transmittance is
approximated by

t = exp(−τ ) ≈ 1 − τ (s1 , s2 ) = 1 − β e (s2 − s1 ), (7.14)

where the equality on the right holds if we take β e to be con-

stant.
 Extinction Over a Finite Path 157

• If a medium doesn’t scatter (ω̃ = 0), then whatever is not

transmitted along a given path must be absorbed. Thus, the
path absorptance in this case is

a = 1−t . (7.15)

If the medium does scatter (ω̃ >0), then the absorptance is, in
general, less than that defined above, but it cannot be com-
puted from a simple formula.

Problem 7.2: A cloud layer has a vertical profile of β e that is

quadratic in altitude z between cloud base zbase and cloud top ztop ,
with maximum β e,m at the midpoint of the cloud. At the base and
top of the cloud, β e = 0.
(a) Find the quadratic equation that describes β e (z) within the cloud
layer.
(b) Find an expression for the total optical path τ measured vertically
through the cloud layer.
(c) Typical values for the above parameters for solar radiation inci-
dent on a thin cloud layer might be zbase = 1.0 km, ztop = 1.2 km,
and β e,m = 0.015 m−1 . Compute the total optical path for this case.
(d) Based on your answer to (c), compute the vertical transmittance
t through the cloud.

7.2.2 Mass Extinction Coefficient

We have already introduced the concept of an extinction coefficient
β e which is a measure of how strongly radiation is attenuated by
absorption and/or scattering as it traverses a given geometric dis-
tance ∆s. It is possible, and sometimes desirable, to describe the
strength of extinction not by reference to a fixed geometric distance,
but rather to a fixed mass of material or even a fixed number of par-
ticles.
For example, consider again the example of a clear glass petri
dish on an overhead project. Let’s start with the glass half-full of
clear water and then add 10 drops of India ink. Once the ink and
the water are well-mixed, the solution will noticeably attenuate the
158 7. Atmospheric Transmission

light passing through the Petri dish. One can, in principle, measure
the total transmittance t as well as the depth of the solution in the
dish and, from those two pieces of information, use (7.9) and (7.11)
to compute the extinction coefficient β e of the water-ink mixture.

Problem 7.3: Assume the measured transmission through the glass

is 70% and the depth of the ink-water solution is 10 cm. Compute β e .

Now consider what happens if you add enough water to the

mixture to fill the Petri dish to the rim. If you repeat the previous
measurements of transmittance and depth, you will find that the
transmittance remains unchanged, even though the depth of the so-
lution has doubled! Your new value for the extinction coefficient β e
will be half of what it was previously. Clearly, the total transmit-
tance through the column is conserved with respect to dilution of
the ink by clear water, as long as the solution remains confined to
the same horizontal area (i.e., the container has vertical sides).
Furthermore, if you were to leave the water out altogether and
simply spread a single drop of ink uniformly over the bottom of
the Petri dish (this is admittedly difficult to do in practice), the total
transmittance would again be conserved, although the thickness of
the layer is now very small, implying very large β e .
After a little thought, we conclude that the one constant in the
above experiment is the total mass of ink per unit area in the dish,
since dilution by water only spreads it out vertically but not hori-
zontally. Further reflection leads us to define a new quantity called
the mass extinction coefficient ke which relates our previous volume
extinction coefficient β e to the density ρ of the relevant material:

β e = ρke . (7.16)

In our experiment with the ink, the density ρ refers to that of just
the diluted mass of ink in the solution. This density in turn is given
by
M
ρ= , (7.17)
HA
 Extinction Over a Finite Path 159

where M is the mass of concentrated ink pigment added to the dish,

H is the depth of the solution, and A is the horizontal area of the
dish. Adding water increases H but does not change A or M; there-
fore ρ decreases as the solution becomes more dilute.
The total vertical transmittance through the solution is thus
given by
t = exp(−τ ) = exp(− β e H )
      
M M (7.18)
= exp −ke H = exp −ke ,
HA A
so that we discover
M
τ = ke . (7.19)
A
As you can see, H does not appear on the right hand side. This
is consistent with the observation that we can dilute the column as
much or as little as we like and still get the same total transmittance,
as long as M and A are fixed.
Note that ke has dimensions of area per unit mass. You can
think of this as an extinction cross-section per unit mass. The extinc-
tion cross-section can in turn be thought of as the equivalent area of
an opaque object blocking the same total amount of radiation (as-
suming equal illumination everywhere). That is to say, if India ink
pigment has a mass extinction coefficient ke of, say, 100 m2 kg−1 ,
then one kg of pure ink pigment poured into a large body of clear
water (and mixed thoroughly) has the capacity to block as much to-
tal sunlight from reaching the bottom as would a 100 m2 sheet of
opaque metal foil.2 Obviously, the larger the area over which the
ink is spread, the less intense the attenuation at any one point, but
the total remains the same.
The concept of mass extinction coefficient is very handy pre-
cisely because, unlike the case for the volume extinction coefficient,
it is nearly constant for most substances in the atmosphere irrespec-
tive of dilution, as long as all other variables (e.g., pressure, temper-
ature, etc.) are held constant. This is particularly convenient when
dealing with highly variable atmospheric constituents, such as wa-
ter vapor.
2 Strictly speaking, this equivalence holds only if the ink is spread out thinly

enough (i.e., over a large enough area) so that the transmission is close to 100% at
any point.
160 7. Atmospheric Transmission

7.2.3 Extinction Cross-Section

There is one further extinction variable that frequently arises in con-
texts involving known concentrations (number densities) of parti-
cles. Such particles might be molecules of an absorbing gas, droplets
of water in a cloud, or perhaps individual soot particles in a plume
of smoke.
To return to our previous thought experiment, what if, instead of
finding a proportionality between the volume extinction coefficient
β e and the mass density ρ, we instead relate the former to the number
density or concentration N of microscopic particles of pigment in the
ink? If we set up the proportionality as follows —

β e = σe N , (7.20)

we find that the constant of proportionality σe has dimensions of

area. It is thus an extinction cross-section with the same physical
interpretation as discussed earlier but now referenced to a single
particle rather than a unit mass. It follows that

σe = ke m , (7.21)

where m is the mass per particle.3

One place where the idea of an extinction cross-section σe per
particle is especially convenient and intuitive is in clouds. In the
visible and infrared part of the spectrum, a single cloud droplet has
an extinction cross-section that is similar, but not identical, to its
geometric cross-section πr2 , where r is the droplet radius. In fact,
we may define an extinction efficiency Qe according to

σe
Qe ≡ , (7.22)
A

where A is the geometric cross-sectional area of the particle. In the

case of a spherical droplet, A = πr2 , where r is the radius.
3 Equations (7.20)–(7.21) are assume that all of the particles are identical. We

will generalize later to distributions of dissimilar particles.

 Extinction Over a Finite Path 161

One might surmise (incorrectly, as it turns out) that Qe could

range only from zero to one. That is, a particle should not be able
to extinguish more radiation than its geometric cross-section would
imply — in plain English again, it shouldn’t cast a shadow bigger
than itself. Unfortunately, however, our physical intuition lets us
down in this instance. On the contrary, for cloud droplets in the
visible spectrum, Qe ≈ 2 on average, and for certain wavelengths,
it can be even larger! The reasons for this apparent paradox will be
examined more closely in Chapter 12; for now, simply ascribe it to
the wave properties of radiation.

7.2.4 Generalization to Scattering and Absorption

Throughout the last section, we have talked exclusively about ex-
tinction, and we defined a mass extinction coefficient and extinction
cross-section to go along with the previously defined volume extinc-
tion coefficient. Recalling that extinction is the sum of absorption and
scattering, it makes sense to define completely analogous mass- and
particle-normalized quantities for those two processes:

β a = ρka = Nσa β s = ρks = Nσs , (7.23)

where ka and ks are the mass absorption coefficient and mass scat-
tering coefficient, and σa and σs are the absorption cross-section and
scattering cross-section. The latter can in turn be written in terms of
absorption and scattering efficiencies as follows:

σa = Qa A σs = Qs A , (7.24)

where A is the geometric cross-sectional area of a particle.

It follows that the single scatter albedo can be written in terms
of any of these quantities as follows:

βs ks σs
ω̃ = = = . (7.25)
βe ke σe
162 7. Atmospheric Transmission

Problem 7.4: Complete the following table, using the information

in each column to determine the missing values in the same column.
(a) (b) (c) (d)
ke [m2 /kg] 3.89 × 102 ? 0.45 ?
N [m−3 ] ? ? 80 109
A [m2 ] 2.8 × 10−19 ? ? 3.14 × 10−10
Qe ? 0.2 0.6 ?
ω̃ 0 0.1 ? 0.9
m [kg] 7.3 × 10−26 1.41 × 10−17 ? 4.19 × 10−12
ρ [kg/m3 ] 4.8 × 10−4 ? 3.35 × 10−4 ?
σe [m2 ] ? ? 1.89 × 10−6 ?
β e [m−1 ] ? 1.41 × 10−4 ? 0.628
β s [m−1 ] ? 1.41 × 10−5 6.03 × 10−5 ?

7.2.5 Generalization to Arbitrary Mixtures of Components

In the above discussion, we assumed we were dealing with a single
absorbing or scattering substance; either ink or milk. In each case,
we conveniently ignored the fact that we were actually dealing with
a two-component mixture, where one of the components was water.
We were able to get away with this because, for the purposes of this
demonstration at least, pure water is completely nonabsorbing and
nonscattering at visible wavelengths.
In the atmosphere, we have to deal with a mixture of a wide
variety of gases, as well as clouds and aerosols, all of which have
the capacity to absorb and/or scatter radiation in at least some, if
not all, spectral bands. We therefore have to generalize our previ-
ous relationships to properly account for such mixtures. This is not
difficult: all you have to remember is that the total volume extinction,
scattering, and absorbing coefficients for a mixture are equal to the sums
of the corresponding coefficients for the individual components. That is,

βe = ∑ βe,i = ∑ ρi ke,i = ∑ Ni σe,i ,

i i i
βa = ∑ βa,i = ∑ ρi ka,i = ∑ Ni σa,i , (7.26)
i i i
βs = ∑ βs,i = ∑ ρi ks,i = ∑ Ni σs,i ,
i i i
 Plane Parallel Approximation 163

where ρi and Ni , respectively, are the (diluted) density and parti-

cle (e.g., molecule) number concentration of the ith constituent, and
the remaining subscripted variables are the corresponding radiative
coefficients.
The single scatter albedo of the mixture is determined by first
finding the combined β s and β e and then taking the ratio, as usual.

7.3 Plane Parallel Approximation

Now that we have introduced the fundamental concepts and vari-
ables relevant to radiative transfer in an arbitrary medium, let us
take the first step toward adapting these specifically to the atmo-
sphere. In Section 7.2, I introduced equations describing the trans-
mission of radiation over a finite distance along an arbitrary path.
The idea is that you can pick any direction from a given starting
point, integrate the extinction coefficient β e over that path up to
some ending point, and use Beer’s Law to compute the transmit-
tance between the two points. We used the variable s to denote
the geometric distance along the path, with s1 denoting the starting
point and s2 the endpoint. This approach is completely general, in
that it makes no assumptions whatsoever about how β e varies as a
function of horizontal or vertical position in the atmosphere.
As a matter of fact, the atmosphere is usually highly stratified.
That is to say, properties such as pressure, density, temperature,
composition, etc., usually vary much more rapidly in the vertical
direction z than they do in the horizontal directions x and y. For ex-
ample, it is usually necessary to move vertically only 1 km in the at-
mosphere to experience somewhere around a 7◦ C change in temper-
ature; horizontally, it might be necessary to travel a hundred kilo-
meters or more to experience a similar temperature change. Simi-
larly, horizontal pressure gradients are typically less than 1 mb per
20 km, whereas vertical pressure gradients are of the order of 1 mb
per 8 meters. Since air density is determined primarily by pressure
and (absolute) temperature via the ideal gas law
p
ρ= , (7.27)
Rd T

where Rd is the gas constant for air, it follows that density also varies
164 7. Atmospheric Transmission

much more dramatically in the vertical than in the horizontal direc-

tions.

One obvious exception to the above rule is clouds. Many clouds

do exhibit strong horizontal as well as vertical structure. Neverthe-
less, even clouds (especially stratiform clouds – e.g., stratus, stra-
tocumulus, nimbostratus, altostratus, cirrostratus, etc.) are often or-
ganized in great sheets whose horizontal dimension greatly exceeds
their vertical thickness. If you have ever spent time looking out
a window during a cross-country airline flight, you may have no-
ticed that you can sometimes traverse hundreds of kilometers over
a more or less continuous sheet of clouds below you, especially in
the vicinity of synoptic-scale weather disturbances. Thus, although
it is by no means always the case, it is at least sometimes safe (and
in any case very convenient) to treat even cloud layers as varying
much more rapidly in the vertical than the horizontal.

For all of the above reasons, it is quite common (perhaps too

common!) to treat the atmosphere for radiative purposes as plane
parallel. That is, at a given location, we ignore horizontal variations
in the structure of the atmosphere and assume instead that all rele-
vant radiative properties depend strictly on the vertical coordinate
z. Another term you will sometimes see that refers to the same ap-
proximation is slab geometry.

We also ignore the curvature of the earth in the plane parallel

approximation, since any ray of light that is not traveling at a very
shallow angle will pass through most of the mass of the atmosphere
long before the earth’s curvature comes into play. A good rule of
thumb for the validity of this approximation is that H/ cos θ  R,
where θ is the angle of the ray relative to vertical, H is the effec-
tive depth of the atmosphere (H ∼ 10 km for many purposes), and
R ≈ 6380 km is the radius of the earth. Obviously, this criterion
is no longer satisfied if cos θ becomes very small, as it does when
θ approaches 90◦ . Usually, however, we are concerned primarily
with radiation transporting energy more or less vertically through
the atmosphere, in which case we can ignore this complication.
 Plane Parallel Approximation 165

z+dz
dz

ds
z

ds = dz/µ where µ = cos θ

z=0

Fig. 7.4: Relationship between slant and vertical paths in a plane parallel atmo-
sphere.

7.3.1 Definition
Mathematically, we invoke the plane parallel assumption via the
following simplifications:

β e ( x, y, z) ≈ β e (z) , (7.28)

T ( x, y, z) ≈ T (z) , etc. (7.29)

Since everything now depends only on vertical distance z, the path
distance s we used earlier in computing transmittance etc. along a
ray (Fig. 7.4) may now be expressed as

z
s= , (7.30)
µ

where we have introduced the following new definition for conve-

nience:
µ ≡ | cos θ | . (7.31)

As before, θ is the angle of propagation of the ray relative to zenith

(i.e., straight up). Note, therefore, that 0 ≤ µ ≤ 1 regardless of
whether the ray is propagating upward or downward.
166 7. Atmospheric Transmission

Combining the above definitions with our earlier expressions for

transmittance, optical path, etc., we have the following expressions
valid for a plane parallel atmosphere:
The optical thickness between levels z1 and z2 is

z2
τ ( z1 , z2 ) = β e (z) dz , (7.32)
z1

and the transmittance for a ray propagating with direction µ is

 
1
t(z1 , z2 ) = exp − τ (z1 , z2 ) , (7.33)
µ

where z2 > z1 .
Note that the optical thickness in a plane parallel atmosphere,
as given above, is deliberately defined so as not to depend on the
direction of propagation µ. Thus, it doesn’t matter whether the sun
is directly overhead (µ = 1) or nearly setting on the horizon (µ → 0);
the optical thickness of the atmosphere between levels z1 and z2 is
the same. However, when computing the transmittance t of a ray of
the sun as it passes between the two levels, it is of course necessary
to account for µ, since the ray experiences a much longer optical
path when the sun is low in the sky.
Just in case you’ve forgotten, let me reiterate that all of the above
definitions still assume that we’re talking about monochromatic ra-
diation. That is, the parameters β e , τ, t, etc., are all implicitly func-
tions of wavelength.

Problem 7.5: At a certain wavelength in the visible band, the op-

tical thickness of the cloud-free atmosphere is τ ∗ = 0.2. Determine
the transmittance of sunlight at this wavelength when the sun is 10◦
above the horizon.
 Plane Parallel Approximation 167

Problem 7.6: A particular plane parallel cloud has liquid water

density ρw = 0.1 g m−3 and thickness ∆Z = 100 m. At a cer-
tain wavelength, the mass extinction coefficient of the cloud droplets
is ke,w = 150 m2 /kg, and the single scatter albedo is ω̃w = 1.0.
However, the air in which the droplets are suspended is itself ab-
sorbing at this wavelength, having volume absorption coefficient
β a,v = 10 km−1 and ω̃v = 0.
(a) Compute the combined β e , β a , β s , and ω̃ for the mixture.
(b) Compute the total optical thickness τ of the cloud layer. (c) If ra-
diation with intensity Iλ,top is incident on the top of the cloud from
a zenith angle θ = 60◦ , compute the directly transmitted intensity
Iλ,bot .

7.3.2 Optical Depth as Vertical Coordinate

In a plane parallel atmosphere, we may choose any level z in the at-
mosphere and evaluate the optical thickness between that level and
another, higher, level ztop , according to (7.32). Because β e is never
negative (though it may be effectively zero at some levels), the op-
tical thickness is always a nonnegative quantity. Moreover, it can
never increase as z gets closer to ztop . If we choose ztop to be effec-
tively at infinity — i.e., in outer space where there is no atmosphere
left to contribute to additional radiative extinction, then we can de-
fine an optical depth that is a function of only z:

∞
τ (z) ≡ lim τ (z, ztop ) = β e (z )dz . (7.34)
ztop →∞ z

An important property of τ (z) is that it starts at some positive

value τ ∗ ≡ τ (z = 0) at sea level and then decreases with height
until it reaches zero at some level z that is effectively above the top
of the atmosphere,4 where τ ≈ 0.
Because of this property, τ (z) may be employed as a vertical co-
ordinate for radiative transfer purposes. That is, I might tell you
4 Strictly speaking, the atmosphere has no upper limit, but by the time you reach

an altitude of a couple hundred kilometers, the residual gases found there are so
tenuous as to make no significant contribution to radiative extinction.
168 7. Atmospheric Transmission

that the top of a cloud is located at τ = 0.4 (in dimensionless units)

rather than at z = 4 km. Radiation in a plane-parallel atmosphere
doesn’t “care” about vertical altitude in geometric units, only about
how much absorbing atmosphere it must traverse. Hence, changes
in altitude expressed in units of τ are more convenient as a basis for
describing the net effect of that change on a propagating beam of
radiation.
In particular, one may adapt our earlier definitions of transmit-
tance etc. in the following ways:
The optical thickness between levels z1 and z2 (where z2 > z1 ) is
given by

τ ( z1 , z2 ) = τ ( z1 ) − τ ( z2 ) , (7.35)

where the appearance of only one altitude in the arguments of the τ

on the right hand side reminds us that we’re measuring it between
the altitude in question and the top of the atmosphere.
Likewise, the transmittance of a ray propagating with direction
µ from level z1 to z2 (or vice versa) is

 
1 t ( z1 )
t(z1 , z2 ) = exp − τ (z1 , z2 ) = , (7.36)
µ t ( z2 )

where the transmittance from level z to the top of the atmosphere is

of course
− τ (µz)
t(z) ≡ e . (7.37)

7.4 Applications to Meteorology, Climatology,

and Remote Sensing
7.4.1 The Transmission Spectrum of the Atmosphere
In the preceding sections of this chapter, we developed the ma-
chinery for relating the transmission of radiation through the atmo-
sphere to the material properties of the atmosphere at every point
 Applications 169

along the path. To at least first order, these properties are embodied
in the volume extinction coefficient β e and the single scatter albedo
ω̃, both of which are generally functions of position (in three dimen-
sions) and wavelength λ. Recall that if you know β e and ω̃, you can
easily obtain β s and β a , and vice versa, so there is no single “cor-
rect” way to describe the radiative properties of the atmosphere.
Also, there are times when it is more convenient to work with the
mass-normalized or particle-normalized quantities ke , σe , etc., but
it is straightforward to convert between these and β e , etc., if the
amount (mass density ρ or particle concentration N) of the relevant
constituents is known.
We are now ready to undertake a survey of the observed trans-
mission characteristics of the atmosphere at various wavelengths.
For now, we only want to answer the following basic questions:
• At which wavelengths is the cloud-free atmosphere reason-
ably transparent?
• At which wavelengths is the cloud-free atmosphere strongly
absorbing, and which constituents are responsible for the ab-
sorption?
• How do the extinction and scattering properties of clouds vary
with wavelength?
In later chapters, we will revisit all of the above questions in a
much more detailed way.

Absorption by Atmospheric Gases

Except at short visible and UV wavelengths (where scattering by air
molecules can be important), the overall transmittance of the cloud-
free atmosphere is controlled primarily by absorption due to vari-
ous constituent gases. Where absorption is strong, the transmittance
is small; where absorption is weak or absent, the transmittance is
close to 100%. Table 7.1 lists a number of radiatively important con-
stituents and their approximate abundances. Two points are worth
highlighting right away:
• Some constituents make up a large fraction of the total mass of
the atmosphere. In fact, nitrogen (N2 ) and oxygen (O2 ) alone
170

Table 7.1: Key constituents of air in the troposphere and stratosphere.

Constituent Fraction by Significant Remarks

volume in (or absorption bands
relative to) dry air

N2 78.1% —

O2 20.9% UV-C, MW near

60 and 118 GHz,
weak bands in
VIS and IR

H2 O (0–2%) numerous strong highly variable in

bands throughout time and space
IR; also in MW, es-
pecially near 183
GHz
Ar and other 0.936% — monoatomic
inert gases

CO2 370 ppm near 2.8, 4.3, and concentration as

15 µm of 2001;
increasing 1.6
ppm per year

CH4 1.7 ppm near 3.3 and increasing due to

7.8 µm human activities

N2 O 0.35 ppm 4.5, 7.8, and

17 µm

CO 0.07 ppm 4.7 µm (weak)

O3 ∼10−8 UV-B, 9.6 µm highly variable

concentration;
high in
stratosphere and
in polluted air

CFCl3 ,CF2 Cl2 , ∼10−10 IR industrial origin

etc.
 171

380

370

360

350
CO2 (ppmv)

340

330

320

310

300
1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005
Year

Fig. 7.5: Measurements of atmospheric carbon dioxide concentration at Mauna

Loa Observatory, Hawaii. (Source: C.D. Keeling and T.P. Whorf, Scripps Institute
of Oceanography, University of California.)
172 7. Atmospheric Transmission

are responsible for approximately 99% of the total. Many oth-

ers, however, such as carbon dioxide (CO2 ), methane (CH4 ),
and ozone (O3 ), are present only in trace amounts. As we shall
see shortly, many of the less abundant constituents have a dispro-
portionally large influence on atmospheric transmission.

• Some constituents, such as nitrogen, oxygen, argon (Ar), and

other so-called permanent gases are present in the atmosphere
in nearly unvarying concentrations. The concentration of oth-
ers, such as water vapor (H2 O) and ozone, can vary wildly
from one time and location to another. The atmospheric con-
centration of CO2 does not vary much from location to loca-
tion; however, it is experiencing a steady upward trend over
time as a result of fossil fuel use by humans (Fig. 7.5). Many of
the variable constituents are more important for atmospheric radia-
tion than the permanent gases, a fact that has profound implications
for climate variability.

Figure 7.6 depicts the vertical transmittance of the entire cloud-

and aerosol-free atmosphere as a function of wavelength from the
visible to the microwave.5 It also depicts separately the transmit-
tance spectrum for selected invididual gases as if they were the only
absorbers present. Recall that the total atmospheric transmittance due
to all constituents is the product of the transmittances due to the individ-
ual constituents. Thus, if any one constituent is a strong absorber at
a given wavelength, then the atmosphere will be effectively opaque
at that wavelength, even if all other constituents are nonabsorbing
at the same wavelength.6
Starting in the ultraviolet band, we see that the atmosphere is
almost completely opaque to radiation with wavelengths shorter
than 0.3 µm. This absorption is due to oxygen at the very short
wavelengths of the UV-C band and to ozone in the UV-B band. In
most of the UV-A band, on the other hand, the atmosphere is rather
transparent.
5 We are considering only the effects of absorption and not molecular scattering
in this depiction.
6 Note that this depiction gives a highly smoothed, “big picture” view. As we

shall see in Chapter 9, the atmospheric absorption bands are extremely complex
when viewed at fine spectral resolution.
 Applications 173

ZENITH ATMOSPHERIC TRANSMITTANCE

UV VIS Near IR Thermal IR
1

Transmittance
CO

1 0
Transmittance

N2O

0 1

Transmittance
CH4

1 0
Transmittance

O2

0 1

Transmittance
O3

1 0
Transmittance

CO2

0 1

Transmittance
H2O

1 0
Transmittance

Total

0
0.3 0.4 0.5 0.6 0.8 1 1.2 1.5 2 2.5 3 4 5 6 7 8 9 10 12 15 20 25 30 40 50
Wavelength [µm]

Fig. 7.6: Zenith transmittance of the cloud- and aerosol-free atmosphere for condi-
tions typical of a midlatitude summertime atmosphere. Each upper panel depicts
the absorption contribution due to a single atmospheric constituent; the bottom
panel depicts the combined effect of all constituents. Molecular scattering, which
becomes increasingly important for wavelengths shorter than about 0.5 µm, is not
considered in these plots.

Problem 7.7: The absorption cross-section of a single O2 molecule

is approximately 7 × 10−29 m2 at a wavelength of 0.24 µm (it in-
creases sharply for even shorter wavelengths). Given a standard
174 7. Atmospheric Transmission

sea level pressure p0 of 1.01 × 105 Pa, a molar fraction for oxy-
gen of 21%, a mean molecular mass m for air of 29 kg/kmole,
gravitational acceleration g = 9.81 m/s2 , and Avogadro’s constant
NA = 6.02 × 1026 kmole−1 , compute (a) the mass per unit area of the
atmospheric column assuming hydrostatic balance, (b) the number
of oxygen molecules per unit area in the column, and (c) the opti-
cal thickness and vertical transmittance due to molecular oxygen at
0.24 µm.

Throughout most of the visible band (0.4 µm<λ<0.7 µm), the

atmosphere is quite transparent, apart from a few weak and narrow
absorption bands due to oxygen or ozone.
In the near IR band (0.7 µm<λ<4 µm) things start to get more
interesting. Over this relatively narrow range of wavelength, the
atmosphere swings several times between being almost perfectly
transparent to being almost perfectly opaque. Most of the absorp-
tion is due to water vapor; however there are also important contri-
butions from CO2 , CH4 , and N2 O.
As already mentioned, water vapor is a highly variable compo-
nent of the atmosphere. The plot appearing in Fig. 7.6 was derived
for the typical atmospheric conditions of a station in the middle lat-
itudes during summertime. In a colder, drier environment, the total
amount of water vapor present in the column will be drastically
smaller, so that the atmosphere may be substantially more transpar-
ent in some spectral bands than indicated in this plot. In a moist
tropical environment, of course, the reverse is true.
Through the thermal IR band (4 µm<λ<50 µm) there are broad
bands of near-total absorption due to CO2 (near 4 µm), water va-
por (from 5–8 µm), ozone (near 9.6 µm), and again CO2 (λ>13 µm).
However, there is also a wide band within which the atmosphere
is rather transparent from 8–13 µm, terminated only by the ozone
band and increasing absorption by water vapor beyond 12 µm.
The far IR band (50 µm<λ<1 mm) is not terribly important in
the calculation of radiative fluxes in the atmosphere, nor is it used
much for remote sensing. For both reasons, it is hard to find plots of
atmospheric transmission in this band. As a rule of thumb, how-
ever, it can be said that water vapor is the dominant absorber
throughout most of this band.
 Applications 175

Zenith Microwave Transmittance

1

0.9 Oxygen
-2
Water Vapor (21.3 kg m )
0.8
Cloud Liquid Water (0.2 kg m-2)
0.7 Total
Transmittance

0.6

0.5

0.4

0.3

0.2

0.1

0
0 50 100 150 200 250 300
Frequency [GHz]

Fig. 7.7: Zenith atmospheric opacity in the microwave band under typical midlati-
tude conditions, including the effects of a moderately thick nonprecipitating cloud
layer.

The microwave band is of interest primarily for remote sensing,

mainly below 300 GHz. Fig. 7.7 depicts the principal features of the
absorption spectrum in this band. There is a strong absorption band
due to oxygen centered on 60 GHz; this band plays an important
role in satellite retrievals of atmospheric temperature. A second,
somewhat narrower oxygen band is also present at 118 GHz. There
is a very strong water vapor band centered on 183.3 GHz; this band
is used in microwave retrievals of atmospheric humidity profiles.
All three of the microwave absorption bands just mentioned are
strong enough to render the atmosphere completely opaque at the
centers of the bands. A difference between the oxygen bands and
the water vapor bands, however, is that the latter is only present to
the degree that water vapor is present. In an extremely dry arctic at-
mosphere, the 183.3 GHz band is substantially weaker than shown
in Fig. 7.7.
Last but not least, there is a weak water vapor absorption line
located near 22 GHz. Despite its weakness (indeed, partly because
of its weakness), microwave observations made at this frequency
have long been crucial for retrieving the total (vertically integrated)
176 7. Atmospheric Transmission

water vapor content of the atmosphere.

Apart from the major absorption bands just mentioned, Fig. 7.7
reveals a strong tendency for the atmosphere to become more
opaque with increasing frequency, with the standard midlatitude
atmosphere becoming nearly opaque by 300 GHz. This tendency is
due to the so-called continuum absorption by water vapor; that is,
absorption that is not concentrated in a few discrete lines or bands.
It is partly because of the increasing dominance of water vapor con-
tinuum absorption that frequencies higher than 300 GHz are not
terribly useful for remote sensing, except perhaps for observing cir-
rus clouds that lie well above the majority of water vapor in the
troposphere.
Throughout every major band just discussed, there exist spectral
windows for which the cloud-free atmosphere is relatively transpar-
ent. For example, the entire visible band is a spectral window, as is
the 8–13 µm region in the thermal IR band and the 80–100 GHz and
0–40 GHz regions in the microwave band. If no clouds are present,
then a satellite observing upwelling radiation within a window re-
gion has a more or less unobstructed view of the earth’s surface.
Also, radiation emitted by the earth’s surface at wavelengths falling
within one of the windows in the thermal IR band readily escapes
to space, so that these window regions play an important role in
surface cooling.
In many spectral windows, the atmosphere is not perfectly
transparent but has a moderate component of absorption due to
water vapor or some other constituent. Such regions of the spec-
trum are often referred to as “dirty windows,” because radiation
originating from the surface remains observable from space but sig-
nificant corrections for atmospheric absorption and emission must
be applied, especially when the atmosphere is humid.
One final important comment concerning the gaseous absorp-
tion features we have been discussing: The plots above give a some-
what simplified depiction of the actual dependence of transmittance
on wavelength. If you were to “zoom in” on any one of the ab-
sorption bands discussed above, you would find that it represents
a complex aggregate of many very closely spaced “lines” of ab-
sorption. There may be hundreds of such lines in a very narrow
wavelength band. In the gaps between individual lines, the at-
 Applications 177

Zenith Atmospheric Transmittance (SW)

1

0.8
Transmittance

0.6

0.4

0.2

0
0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Wavelength [µm]

Fig. 7.8: Detail of atmospheric transmission in the shortwave portion of the spec-
trum, for a summer midlatitude atmosphere. This plot includes the effects of
scattering by air molecules (dashed/smooth curve on left).

mosphere may be substantially more transparent than indicated by

Fig. 7.6. Likewise, at the exact centers of isolated narrow lines, some
of which fall within the so-called “dirty windows”, the atmosphere
may be more opaque than indicated. The reasons for the enormous
complexity of absorption line spectra, and methods for dealing with
it, are the subject of later chapters.

Scattering by Clear Air

In the previous subsection, we outlined the major absorption fea-

tures of atmospheric gases. In the absence of clouds, it is largely ab-
sorption by gases that controls the overall opacity of the atmosphere.
However, at short wavelengths (visible and shorter), molecules of
air have the capacity to significantly scatter EM radiation. Although
we will not look at the details of such scattering until Chapter 12,
it is worth mentioning now that the scattering cross-section σs of an
air molecule is approximately proportional to λ−4 . Thus, extinction
of radiation via scattering is 9.3 times stronger at 0.4 µm (the short
wavelength limit of the visible band) than it is at 0.7 µm (the long
wavelength limit). In the UV band it is stronger still, while in most
of the IR band it is negligible. The uppermost curve in Fig. 7.8 shows
178 7. Atmospheric Transmission

how vertical transmission through the atmosphere is reduced by

molecular scattering.

Extinction and Scattering by Aerosols and Clouds

In addition to the gas molecules that make up most of the atmo-

sphere, there are also countless small particles of dust, salt, water,
and other materials suspended in the air. Some, such as the wa-
ter droplets that make up a visible cloud, are comparatively large
(order 10 µm radius), while many other aerosol types are typically
much smaller than 1 µm.
Depending on the size and composition, these particles can ei-
ther scatter or absorb radiation or, more commonly, both. Clouds
in the visible band strongly scatter sunlight but absorb very little;
hence they are rather opaque but appear white. A plume of smoke
from a forest fire, on the other hand, may not appear as optically
dense as a typical water cloud, but it usually has a gray or brown
color to it even when viewed from above, suggesting that at least
some absorption is taking place.
Ignoring water and ice clouds for the moment, it is difficult to
generalize about the contribution of aerosols to the extinction of ra-
diation in the atmosphere, partly because their concentration and
composition are so highly variable. However, it is safe to say that
the contribution of “background” aerosols to the total optical thick-
ness τ ∗ is rarely more than a few tenths of a unit at visible wave-
lengths and tends to be even smaller still at infrared and longer
wavelengths (Fig. 7.8). Aerosol optical depths only reach compara-
tively large values (order unity or greater) in association with fires,
volcanoes, dust storms, and severe pollution episodes.
Clouds are another story: because the droplets they contain are
large compared with most other aerosols and because the total mass
of liquid or frozen water (integrated vertically over the atmospheric
column) can also be large (order 0.1 to 1 kg/m2 ), they routinely
achieve optical thicknesses large enough to completely block the
direct rays of the sun. In the microwave band, clouds are consider-
ably more transparent; nevertheless, their radiative effects are never
completely negligible.
Because ordinary clouds consist of nearly pure water droplets
 Applications 179

or ice particles, the degree to which clouds absorb radiation (in ad-
dition to scattering it) is a function of the imaginary part ni of the
index of refraction of water or ice, as discussed in Chapter 4. If ni is
zero at a given wavelength, then a cloud can only scatter radiation
at that wavelength; if ni is nonzero, then the cloud will also absorb
to at least some degree. We will examine this problem in far greater
detail in a later chapter; for now it is enough to point out that clouds
quickly switch over from being almost purely scattering (ω̃ ≈ 1) in
the visible band to strongly absorbing (ω̃ <1) in the IR band and
beyond. However, the details of this transition are different for liq-
uid water and ice; hence there are wavelengths (e.g., near 1.7 µm;
see Fig. 4.1) for which a water cloud is significantly less absorbing
than an ice cloud. Satellite sensors with channels at suitable wave-
lengths can exploit this difference to distinguish between water and
ice clouds.

7.4.2 Measuring Solar Intensity from the Ground

We have already seen that the solar flux S0 at the top of the atmo-
sphere is of key importance to the radiation budget of the earth and
atmosphere. Furthermore, because atmospheric absorption of solar
radiation is a function of wavelength, it is important to accurately
know how the solar intensity varies with wavelength.
Prior to the (relatively recent) advent of satellites equipped with
devices for measuring the solar spectrum, it was impossible to mea-
sure solar intensity directly. Ground-based measurements are al-
ways affected by a certain degree of attenuation of the solar beam by
atmospheric absorption and scattering. If the atmospheric transmit-
tance were known, then of course the unattenuated solar intensity
could be computed from the ground measurements, but there was
no direct method for inferring transmittance either without know-
ing the strength of the solar source. This presented a classic “Catch
22” problem to atmospheric scientists and astrophysicists. Yet rea-
sonable estimates of the solar spectrum were obtained years before
the first satellite measurements were made. How was this done?
Let us assume that the atmosphere is plane parallel and that
the properties of the atmosphere are essentially constant over the
period of a day. These assumptions are generally met only on a
180 7. Atmospheric Transmission

sun directly overhead

log(Sλ)
+++ slop
++ e=
-τ
log(Iλ)
++
+ + λ
+ ++
+

0 2 4 6 8 10
Z = sec θ

Fig. 7.9: Schematic relationship between the log-intensity of solar radiation at

wavelength λ and the secant of the solar zenith angle, for an atmospheric optical
depth τλ . The plus symbols represent individual measurements at different times
of day, from which the slope and intercept of the best-fit line may be determined.

perfectly clear day with no significant change in humidity, surface

pressure, etc. For any given wavelength λ, the two unknowns in
this problem are the solar intensity Sλ and the atmospheric optical
depth τλ .
At any particular time on the day in question, Beer’s Law states
that the intensity of solar radiation measured at sea level is
τλ
−
Iλ = Sλ e µ , (7.38)

where µ = cos θ and θ is the solar zenith angle. Taking logarithms

of both sides yields
τλ
log( Iλ ) = − + log(Sλ ) . (7.39)
µ

If we define a Z = 1/µ = sec θ, then the above is a linear equation

of the form Y = mZ + B, where Y ≡ log( Iλ ), the slope m ≡ −τλ ,
and the Y-intercept B ≡ log(Sλ ).
All that is needed now is a series of measurements of Y for differ-
ent values of Z; i.e., at different times of day as the sun rises, reaches
its highest position (smallest θ) at noon, and then sets again. By
 Applications 181

plotting a series of such measurements on a graph, one may readily

determine both the slope m and the intercept B, which may then be
converted to τλ and Sλ , respectively. Fig. 7.9 illustrates the principle.

Problem 7.8: A ground-based radiometer operating at λ = 0.45 µm

is used to measure the solar intensity Iλ (0). For a solar zenith angle
θ = 30◦ , I = 1.74 × 107 W m−2 µm−1 sr−1 . For θ = 60◦ , Iλ (0) =
1.14 × 107 W m−2 µm−1 sr−1 . From this information, determine the
top-of-the-atmosphere solar intensity Sλ and the atmospheric optical
thickness τλ .

7.4.3 Transmittance in an Exponential Atmosphere

To a fair approximation, the density ρ of the atmosphere decays ex-
ponentially with height z. That is,

ρ ( z ) ≈ ρ0 e − H ,
z
(7.40)

where ρ0 is the density at sea level and H ≈ 8 km is the scale height;

i.e., the altitude change that leads to a factor e change in density. If
a certain constituent of the atmosphere, such as carbon dioxide, is
well-mixed, then the density of just that constituent is given by

ρ 1 ( z ) ≈ w1 ρ 0 e − H ,
z
(7.41)

where w1 is the mixing ratio (mass of constituent per unit mass of

air).
Let us further assume for simplicity that the constituent in ques-
tion has a mass absorption coefficient of ka that depends on λ but
is otherwise independent of temperature, pressure, w, etc. (we will
see later that this is rarely the case for real absorbers; nevertheless,
the dependence of ka on environmental variables is usually much
weaker than it is on λ). Finally, let’s assume that the atmosphere is
nonscattering at the wavelength in question, so that ke = ka .
By combining (7.41) with (7.16), we have an expression for the
volume extinction coefficient as a function of altitude:

β e ( z ) = k a w1 ρ 0 e − H .
z
(7.42)
182 7. Atmospheric Transmission

0.01

dt
dz t(z) 0.1
z τ

1
ρ(z)
ρ0
10
0 1

Fig. 7.10: Illustration of the relationship between transmittance t between the top
of the atmosphere and altitude z and the absorption/emission weighting function
W (z) = dt/dz for an atmosphere with exponential density profile ρ(z).

Armed with this information, we can now answer the following

questions:

1. What is the relationship between altitude z and optical depth

τ as a vertical coordinate, as discussed in Section 7.3.2?

2. What is the transmittance from the top of the atmosphere to

any level z?

3. When radiation (e.g., sunlight) is incident on the top of the

atmosphere, where does it get absorbed, and at what altitude
is the rate of absorption strongest?

The answer to each of these questions depends directly on the an-

swer to the previous one. Let’s address each in turn.
 Applications 183

Optical Depth
From (7.34) and (7.42), we have

∞
∞
z

τ (z) = β e (z )dz = ka w1 ρ0 e− H dz , (7.43)
z z

or
τ (z) = ka w1 ρ0 He− H .
z
(7.44)
Recall from (7.35) that we can also use the above equation to deter-
mine the optical thickness τ (z1 , z2 ) of an atmospheric layer bounded
by two arbitrary levels z1 and z2 , since τ (z1 , z2 ) = |τ (z1 ) − τ (z2 )|.
The total atmospheric optical depth is just

τ ∗ ≡ τ ( 0 ) = k a w1 ρ 0 H . (7.45)

We see that τ starts at zero at the top of the atmosphere, and in-
creases ever more rapidly as one moves downward toward the sur-
face, where it reaches its maximum value of τ ∗ . When optical depth
is used as a vertical coordinate, therefore, a unit change of τ cor-
responds to a smaller change of z at low altitudes than it does at
higher altitudes.
Note, by the way, that (7.43) can also be written

τ (z) = ka u(z) , (7.46)

where
∞
u(z) ≡ ρ1 (z )dz (7.47)
z
is defined as the mass path of the constituent in question between
altitude z and the top of the atmosphere. This quantity has dimen-
sions of mass per unit horizontal area. The definition given by (7.47)
is valid regardless of the functional form of the constituent density
ρ1 (z). When z = 0 is substituted into (7.47), we have the total mass
path utot of the constituent.
In the atmospheric science literature, there are many equivalent
names and symbols assigned to the same quantity. For example, if
the constituent in question is water vapor, then the total mass path
of water vapor in the atmosphere is variously known as total precip-
itable water, column-integrated water vapor, the water vapor burden, and
the water vapor path, depending on the author and the context.
184 7. Atmospheric Transmission

Transmittance
From (7.37), the transmittance between altitude z and the top of the
atmosphere (regardless of whether the radiation is propagating up-
ward or downward) is determined by the optical depth τ (z) and by
the cosine of the zenith (or nadir) angle µ = cos θ:
   
τ (z) k a w1 ρ 0 H − z
t(z) = exp − = exp − e H . (7.48)
µ µ
This expression is admittedly a bit awkward-looking, in that one
rarely encounters functions that are exponentials of exponentials!
Apart from this minor peculiarity, it poses no computational prob-
lems.

Absorption
We already noted that, in a nonscattering atmosphere, the total ab-
sorption along a path is just equal to one minus the transmittance
along that path. Thus, the total absorptance between z and the top
of atmosphere is a = 1 − t(z). A much more interesting and impor-
tant question, however, is where that absorption is occurring.
Consider the absorption of solar radiation within the layer
bounded by z1 and z2 , where z2 > z1 . This must be equal to the
absorption between z1 and the top of the atmosphere minus that
occurring between z2 and the top of the atmosphere:
a(z1, z2) = [1 − t(z1 )] − [1 − t(z2 )] = t(z2 ) − t(z1 ) . (7.49)
Let us define ∆z = z2 − z1 and look at the local absorption per unit
altitude:
 
a(z, z + ∆z) t(z + ∆z) − t(z)
W (z) = lim = , (7.50)
∆z→0 ∆z ∆z
which of course reduces to

dt(z)
W (z) = . (7.51)
dz

The above relationship is extremely important, and you should spend some
time thinking about its implications before you move on. It states that,
 Applications 185

for radiation incident at the top of the atmosphere, the local rate of
absorption within the atmosphere equals the local rate of change of trans-
mittance from level z to the top of the atmosphere.
Let us take things a step further: Noting that t(z) = e−τ (z)/µ , we
have
d − τ(µz) 1 τ(z) dτ (z)
W (z) = e = − e− µ . (7.52)
dz µ dz
∞
Noting further that τ (z) = z β e (z )dz , we have

dτ (z)
= − β e (z) , (7.53)
dz
thus
β e (z) − τ(µz) β e (z)
W (z) = e = t(z) . (7.54)
µ µ
That is, W (z) equals the local extinction coefficient at level z times
the transmittance from z to the top of the atmosphere.
Note that the above relationships (7.51) and (7.54) do not depend
on any assumptions about the specific profile of absorption coeffi-
cient β a (z); that is, we haven’t yet invoked the assumption of ex-
ponentially decaying density. In fact, they can be generalized as
to apply to any path in any nonscattering, non-plane parallel atmo-
sphere:
dt(s, s )
W (s) = = β e (s)t(s, s ) , (7.55)
ds
where s is the distance along the path toward the source of the radi-
ation at s = s and t(s) is the transmittance between s and s .
Does this relationship between local absorption and the rate of
change of transmittance make physical sense? Consider a perfectly
transparent atmosphere. The transmittance t(z) is then constant and
equal to one for all z; therefore dt/dz = 0, and the local absorption
per unit distance is also zero, as you would expect. Now consider
the rate of absorption deep within a strongly absorbing atmosphere.
Except near the very top of that atmosphere, t(z) ≈ 0 for all z, so
again dt/dz ≈ 0 and there is no local absorption of radiation inci-
dent at the top of the atmosphere. That is to say, there’s essentially
nothing left to absorb, because everything already got absorbed at a
higher altitude!
186 7. Atmospheric Transmission

To summarize, it is by definition the range of z over which the

transmittance drops from near one to near zero (when moving
downward through the atmosphere) that most of the incident ra-
diation is absorbed, and it is of course where the change in trans-
mittance is most rapid that the absorption is most rapid.7
Let us conclude this discussion by working out the math for our
idealized exponential atmosphere. All that is necessary is to sub-
stitute our expressions for β a (z) and t(z), from (7.42) and (7.48) re-
spectively, into (7.54), yielding
 
1 k a w1 ρ 0 H − z
W (z) = ka w1 ρ0 e− H exp −
z
e H , (7.56)
µ µ
which can be written in simpler form by substituting (7.45):
 
τ∗ − z τ∗ − z
W (z) = e H exp − e H . (7.57)
Hµ µ
The characteristic shape of W (z) is depicted schematically by the
dotted line in Fig. 7.10. As explained above, it is nearly zero at high
altitudes where the atmospheric density is vanishingly small; it is
also zero at low levels, below the maximum depth to which radi-
ation can penetrate. Somewhere in between, of course, it has its
maximum value, which is where t(z) is changing most rapidly.
Qualitatively, we expect the altitude of peak absorption to de-
pend on the strength of absorption; i.e., on the value of ke at the
wavelength in question. If ke is small, the atmosphere is relatively
transparent, and radiation penetrates to lower altitudes, or even to
ground level, before the rate of absorption becomes strong. If ke is
large, then absorption is strongest at high altitudes where the den-
sity is still very small; at lower levels no radiation remains to be
absorbed.
We can find the altitude of the peak of W (z) using the standard
approach: Take the derivative with respect to z, set the result equal
to zero, and solve for z:
 
dW (z) d τ∗ − z τ∗ − z
= e H exp − e H ,= 0 (7.58)
dz dz Hµ µ
7 The layer of the atmosphere in which solar radiation of a particular wave-

length and incidence angle is absorbed — i.e., where W (z) > 0 — is sometimes
referred to as the Chapman layer, after the scientist who first studied it.
 Applications 187
  
−z/H τ ∗ −z/H τ ∗ −z/H
e exp e e −1 = 0 . (7.59)
µ µ
The equality is satisfied when
τ ∗ −z/H
e =1. (7.60)
µ
Using (7.44) and (7.45), we find that the above condition can be
rewritten as
τ (z)
=1. (7.61)
µ

In other words, the peak of the absorption weighting function occurs at

the altitude z for which the optical path, measured in the direction of inci-
dence from the top of the atmosphere, equals one.
Later, we will see that the above absorption weighting function
is identical to the emission weighting function describing the contri-
bution of various atmospheric levels to emitted radiation observed
from space. For now, I will just point out that this is a natural con-
sequence of Kirchhoff’s Law.

Problem 7.9: At three different wavelengths λ1 , λ2 , and λ3 , the

profile of absorption coefficient due to a certain atmospheric con-
stituent is given by

β e (z) = k n ρ0 exp[−z/H ] , (7.62)

where ρ0 = 4 g/m3 is the density of the constituent at sea level,

and H = 8 km is the scale height. The wavelength-dependent mass
extinction coefficients k1 , k2 , and k3 , respectively, are 0.05, 0.10, and
0.15 m2 kg−1 . Find the altitudes zn of the corresponding peaks of the
absorption weighting functions W (z) for radiation incident at the top
of the atmosphere with θ = 60◦ .

7.4.4 Optical Thickness and Transmittance of a Cloud

Layer
Clouds consist of a very large number of very small water droplets
and/or ice crystals suspended in air. In the case of liquid water
188 7. Atmospheric Transmission

clouds, typical droplet radii fall between 5 µm and 15 µm, though

smaller or larger droplets are also possible (droplets approaching
100 µm are large enough to fall out of the cloud as drizzle and are re-
garded as precipitation rather than cloud droplets). Typical droplet
concentrations range from 102 to 103 droplets per cubic centimeter.
Many liquid water clouds are thick enough to block virtually
all direct sunlight. That is, if a cloud passes between you and the
sun, you often cannot make out the sun’s disk. Instead, you may
see only a diffuse gray illumination contributed by solar radiation
that has been scattered numerous times in various directions before
emerging from the bottom of the cloud. Of course, much of the
radiation that is scattered within the cloud ends up emerging from
the top instead of the bottom.
When a photon of sunlight is incident on the top of a cloud layer,
it will experience one of four possible fates:

1. It may pass directly through the cloud without being scattered

or absorbed even once. The fraction of incident radiation that
passes straight through the cloud in this way is called the direct
transmittance tdir .

2. It may be scattered one or more times and then emerge from

the bottom of the cloud layer. The corresponding fraction is
called the diffuse transmittance tdiff .

3. It may be scattered one or more times and then emerge from

the top of the cloud layer. The corresponding fraction of the
incident flux is called the reflectance (or albedo) r, by analogy
to a reflecting surface, as discussed in the previous chapter.

4. Whatever is neither transmitted nor reflected by the cloud

layer must be absorbed. This fraction is called the in-cloud
absorptance a.

Obviously, the four terms must sum to one; thus

tdir + tdiff + r + a = 1 . (7.63)

Often, it is convenient to lump the diffuse and direct transmittance

 Applications 189

together, in which case we refer to the total transmittance as

t = tdir + tdiff , (7.64)

so that
t+r+a = 1 . (7.65)

The above properties of clouds are of profound importance for

the atmospheric radiation budget. In particular, the reflectivity r of
a cloud layer helps determine how much of the solar radiation inci-
dent on the top of the atmosphere gets immediately reflected back to
space, forever lost as far as the energy budget of the earth and atmo-
sphere is concerned. The total transmittance t limits the amount of
solar radiation available to directly heat the surface. And of course
the in-cloud absorptance a defines the fraction that contributes to
direct heating of the atmospheric layer in which the cloud resides.
For any given cloud, the values of the above four variables de-
pend on its optical thickness τ ∗ , the single scatter albedo ω̃, and de-
tails of how radiation is scattered by the constituent cloud droplets.
The three properties tdiff , r, and a all include contributions from mul-
tiple scattering of radiation within the cloud layer. We will not have
the tools to address multiple scattering until a later chapter.
We do, however, already have the tools to evaluate the optical
thickness τ ∗ of a cloud layer in terms of the cloud droplet radii r,
concentrations N, and extinction efficiencies Qe . The direct trans-
mittance tdir then follows from (7.33):

∗ /µ
tdir = e−τ . (7.66)

Monodisperse Cloud

Let us start by considering an ideal plane parallel cloud composed

entirely of cloud droplets of identical radius r in a concentration N
per unit volume. Such a cloud composed of equal-sized droplets
is called monodisperse , as contrasted with a more realistic cloud
190 7. Atmospheric Transmission

Monodisperse Polydisperse

Fig. 7.11: Illustration of the definitions of monodisperse and polydisperse for cloud
water droplets.

composed of polydisperse cloud droplets (Fig. 7.11). From (7.20) and

(7.22), we have the volume extinction coefficient
β e = NQe πr2 . (7.67)
Usually, it is difficult to directly measure either N or r but some-
what easier to measure or estimate the cloud water density ρc ,
which is the mass of condensed cloud water per unit volume of air.
Typical values for ρc lie in the range 0.1–1 g/m3 , where the lower
values are usually found in stratiform clouds experiencing weak up-
lift, whereas values in excess of 1 g/m3 are most often found in the
cores of vigorous convective updrafts.
For our monodisperse cloud, the cloud water density is just the
number concentration of droplets times the mass of water in each
droplet:
4
ρc = N πr3 ρl , (7.68)
3
where ρl ≈ 1000 kg/m3 is the density of pure water. Combining
with (7.16) and (7.67), we find that the volume extinction index-
Cloud!extinction coefficient indexExtinction coefficient!in cloud co-
efficient is given by
4
β e = NQe πr2 = ke ρc = ke N πr3 ρl , (7.69)
3
which allows us to solve for the mass extinction coefficient:
3Qe
ke = . (7.70)
4ρl r
 Applications 191

This result is interesting, but perhaps not surprising. It tells us that

the same mass of water broken up into droplets extinguishes more
radiation if the droplets are small and numerous than if they are
large and relatively few.
Is (7.70) consistent with everyday experience? Consider mod-
erate falling rain and thick fog. At visible wavelengths, Qe ≈ 2 for
both cases. But for rain, the drop radius r is on the order of 1 mm; for
fog it is a hundred times smaller, or about 10 µm. Substituting these
values into (7.70) yields ke ≈ 1.5 m2 /kg for rain and ke ≈ 150 m2 /kg
for fog. Yet both rain and fog can yield liquid water concentrations
ρc on the order of 0.1 g per cubic meter of air. Thus, the volume ex-
tinction coefficient β e can be on the order of 0.15 km−1 and 15 km−1
for rain and fog. In the first case, the transmittance along a 1 km
path is 86%; in the second, it is for all practical purposes zero!
Let’s return to the problem of determining the optical thickness
of the cloud layer with cloud water density profile ρc (z) between
cloud base zbot and cloud top ztop :

ztop
ztop
τ∗ = β e (z) dz = ke ρc (z) dz . (7.71)
zbot zbot

Since we’re assuming that ke is constant in this problem, we can take

it out of the integral and simply write
τ ∗ = ke L , (7.72)
where the liquid water path (vertically integrated mass of cloud water
per unit horizontal area) is defined as

ztop
L≡ ρc (z) dz . (7.73)
zbot

Combining (7.72) and (7.70) and using Qe ≈ 2, we get

3L
τ∗ ≈ . (7.74)
2ρl r
To summarize, the total optical depth of our plane parallel cloud
layer is proportional to the liquid water path L and inversely pro-
portional to radius of the constituent cloud droplets.8
8 We are neglecting any contribution by the air itself to any absorption within

the cloud layer. At some wavelengths in the solar band, absorption by water vapor
or other constituents cannot be neglected.
192 7. Atmospheric Transmission

Typical values of L range from essentially zero to as much as

0.5 kg m−2 or so in heavy stratiform cloud layers. For r = 10 µm,
this yields an optical thickness of up to 75 or so. In order for even
as much as, say, one percent of incident sunlight to be directly trans-
mitted, we have a maximum τ ∗ of

τ ∗ 1% = −µ ln(0.01) ≈ 4 , (7.75)

which corresponds to a liquid water path L on the order of only

0.03 kg m−2 . Such a value might correspond to a cloud layer that is
a mere 30 m thick and has a relatively low average cloud water den-
sity of only 0.1 g m−3 . We therefore conclude that all but the very
thinnest clouds are nearly opaque to direct solar radiation. Never-
theless, diffuse transmittance in clouds can be quite significant even
in optically thick clouds.

Cloud Condensation Nuclei and Cloud Optical Thickness

Although we have not yet shown this rigorously, common sense
tells us that, all other factors being equal, an optically thick cloud
will reflect more sunlight than an optically thin cloud. We showed
further that the optical thickness is proportional to liquid water path
L and inversely proportional to droplet radius r.
The liquid water path L is generally controlled by the cloud-
scale dynamics and thermodynamics of the air in which the cloud
forms; that is, by temperature and humidity, local ascent rate, mix-
ing, etc. For example, if a saturated parcel of air is lifted moist-
adiabatically from one temperature and pressure level to a higher
level in the atmosphere, the total mass of condensed water can
be predicted very easily using a standard thermodynamic diagram
such as a Skew-T/log-P chart.
What is not predictable from bulk thermodynamics alone is
whether the condensed water will appear in the form of relatively
few large droplets or instead as numerous smaller droplets. That
characteristic of a cloud is determined by the number of cloud con-
densation nuclei (CCN) that get activated and become cloud droplets
when the cloud first forms.
All other factors being equal, clean air masses containing very
few CCN will produce few cloud droplets, but those droplets will
 Applications 193

be relatively large. Continental and/or polluted air masses may

contain one or two orders of magnitude more CCN and therefore
produce clouds containing correspondingly higher concentrations
of much smaller cloud droplets. It can be shown that

  13
9πL2 H
τ ∗ = Qe N , (7.76)
16ρ2l

where H is the geometric thickness of the cloud layer. This relation-

ship tells us that, for a cloud layer of fixed depth and total liquid
water path, the optical thickness τ ∗ ∝ N 1/3 .
The above relationship is of more than just academic interest.
Climate scientists have come to realize that industrial pollution
contributes vast quantities of aerosols to the atmosphere, many of
which are capable of serving as CCN. It is now apparent that this
pollution may be increasing the average reflectivity of clouds and
therefore reducing the fraction of solar radiation which is absorbed
by the earth and atmosphere. The expected effect of aerosol pollu-
tion is thus a decrease in average global temperature over time, in
contrast to the increased “greenhouse warming” expected from in-
creased CO2 concentrations. Some climate change specialists now
suspect that the indirect cooling effect of aerosols may have helped
mask the expected warming due to CO2 increases, preventing the
observed warming to date from being as large as that predicted by
many climate models. Indeed, there are those who believe that the
aerosol effect may be beneficial because of its moderating influence
on greenhouse warming. That optimistic view may be premature,
however, since it glosses over important regional, seasonal, and
altitude-dependent differences in the radiative impact of aerosols
versus CO2 . A partial cancellation of the globally averaged radiative
impacts of each pollutant does not necessarily ensure that regional
climate responses to anthropogenic pollution will be negligible.

Problem 7.10: Derive (7.76).

194 7. Atmospheric Transmission

Problem 7.11: A certain cloud layer has geometric thickness H =

0.1 km and liquid water path L = 0.01 kg m−2 . Taking Qe ≈ 2
and the solar zenith angle θ = 60◦ , compute the direct transmittance
tdir for (a) N = 100 cm−3 (characteristic of clean maritime environ-
ments), and (b) N = 1000 cm−3 (characteristic of continental envi-
ronments).

Polydisperse Cloud†

In our above analysis of the effect of droplet concentration and ra-

dius on cloud optical thickness, we assumed that all cloud droplets
had the same size. This is a reasonable approach when the purpose
is just to gain qualitative insight into how the radiative properties of
clouds depend on drop size and number. It is insufficient, however,
for anything that requires accurate calculations of those properties.
In reality, of course, the droplets found in any given cloud are
distributed over a range of sizes. We may describe the drop size dis-
tribution in any given case via a function n(r ), such that
 
number of droplets (per unit volume of
n(r ) dr = . (7.77)
air) whose radii fall in the range [r, r + dr ]

It follows that n(r ) has dimensions of “per unit volume per unit
interval of r”, or length−4 . Often, the units will be written as
[m−3 µm−1 ].
With the above definition of n(r ), one may immediately derive
several related quantities. For example, the total number of droplets
of all sizes (per unit volume), is just

∞
N= n(r ) dr . (7.78)
0

The total number of droplets whose radius is smaller than some ra-
dius r is

r
N (r < r ) = n(r ) dr . (7.79)
0
 Applications 195

The total surface area (per unit volume of air) contributed by

droplets of all sizes is obtained by first multiplying n(r ) by the ex-
pression for the surface area of a single droplet of radius r and then
integrating over all sizes:

∞  
Asfc = n(r ) 4πr2 dr . (7.80)
0

An analogous approach can be used to obtain those radiative and

cloud physical quantities that we already worked with in the case
of a monodisperse cloud. For example, the local cloud water density
is given by  

∞
4π 3
ρc = n (r ) ρ l r dr . (7.81)
0 3
Finally, the local volume extinction coefficient is obtained as

∞  
βe = n(r ) Qe (r )πr2 dr , (7.82)
0

where the expression in brackets represents the extinction cross-

section of a single droplet of radius r. The mass extinction coeffi-
cient can then be written as
∞  
βe n(r ) Qe (r )πr2 dr
ke ≡ = ∞ 0
  . (7.83)
ρc 0
n(r ) ρl 4π
3 r
3 dr

If we again take Qe ≈ 2 for all r (valid when the droplets are large
compared with the wavelength of the radiation), we find that the
above simplifies to
3
ke ≈ , (7.84)
2ρl reff
where the effective drop size reff is defined as
∞
n(r )r3 dr
reff ≡ 0∞ . (7.85)
0
n(r )r2 dr

The total optical depth of the cloud layer is then approximately

3L
τ∗ ≈ , (7.86)
2ρl reff
196 7. Atmospheric Transmission

where L is the vertically integrated liquid water path defined by

(7.73) and (7.81). We see that this expression is identical to our ear-
lier expression (7.74) derived for a monodisperse cloud, except that
we have replaced the single radius r with the effective radius reff
defined by (7.85). The latter can therefore be thought of as the cloud
droplet radius a monodisperse cloud would have to have in order
to have the same optical thickness τ ∗ as a given polydisperse cloud.
Let me emphasize that the utility of the above definition of effec-
tive radius reff is compromised slightly by the assumption that Qe
is not a strong function of r. This will not always be the case, espe-
cially when the wavelength is not much larger than the radii of the
droplets contributing most of the total cloud water mass. Also, the
optical thickness τ ∗ is only one of several radiative parameters that
ultimately determine the overall reflectivity and transmittance of a
cloud layer. Thus, it is not always safe to assume radiative equiva-
lence between a polydisperse cloud with effective radius reff and a
monodisperse cloud with r = reff .
Nevertheless, it is common to use satellite observations of cloud-
top reflectivity at two or more wavelengths to try to infer reff for the
cloud in question, and considerable recent research has been de-
voted to demonstrating a link between reff and anthropogenic pol-
lution, for the reasons discussed in the previous subsection.
 CHAPTER 8

Atmospheric Emission

In the previous chapter, we began our examination of how

monochromatic radiation interacts with the atmosphere, focusing
initially on the extinction (or, conversely, transmission) of radiation by
atmospheric gases and clouds. If the depletion by the atmosphere
of radiation from an external source, such as the sun, were the only
process we had to worry about, life would be simple indeed. But we
already know from our previous discussion of simple radiative bal-
ance models that the atmosphere both absorbs and emits radiation.
And we already hinted that Kirchhoff’s Law implies a correspon-
dence between absorption and emission that is just as valid for the
atmosphere as it is for a solid surface.
We are now ready to expand our understanding of radiative
transfer to include both extinction and thermal emission. Eventu-
ally, we will also have to come to grips with the problem of scatter-
ing as another potential source of radiation along a particular line of
sight. But that is a subject for a later chapter. In this chapter, we will
restrict our attention to problems in which scattering can be safely
ignored. This is not an unreasonable approach for the majority of
real-world problems involving the thermal IR, far IR, or microwave
bands.

197
198 8. Atmospheric Emission

8.1 Schwarzschild’s Equation

Consider the passage of radiation of wavelength λ through a layer
of air with infinitesimal thickness ds, measured along the direction
of propagation. If the radiant intensity is initially I, then the reduc-
tion in I due to absorption is

dIabs = − β a I ds , (8.1)

where we are using (7.4) applied to the case that the single scatter
albedo ω̃ = 0 and therefore β e = β a .
The quantity β a ds can be thought of as the absorptivity a of the
thin slice of the medium, since it describes the fraction of the inci-
dent radiation that is lost to absorption. But Kirchhoff’s Law tells
us that the absorptivity of any quantity of matter in local thermo-
dynamic equilibrium (LTE) is equal to the emissivity of the same
matter. Therefore, we can expect the thin layer of air to emit radia-
tion in the amount
dIemit = β a B ds , (8.2)
where we are using the symbol B here as a convenient short-hand1
for the Planck function Bλ ( T ). The net change in radiant intensity is
therefore
dI = dIabs + dIemit = β a ( B − I ) ds , (8.3)
or
dI
= βa ( B − I ) . (8.4)
ds

This equation is known as Schwarzschild’s Equation and is the most

fundamental description of radiative transfer in a nonscattering
medium. Schwarzschild’s Equation tells us that the radiance along
a particular line of sight either increases or decreases with distance
traveled, depending on whether I (s) is less than or greater than
B[ T (s)], where T (s) is the temperature at point s.
Let us now consider the intensity of radiation arriving at a par-
ticular sensor looking backward along the direction of propagation.
1 Note that we have also dropped the subscript λ from I and other quantities,

as all relationships discussed in this chapter should be understood as applying to

radiative transfer at a single wavelength.
 Schwarzschild’s Equation 199

Much as we did previously for a plane parallel atmosphere, we can

introduce
S
τ (s) = β a (s ) ds (8.5)
s

as the optical path between an arbitrary point s and the sensor posi-
tioned at S. By this definition, τ (S) = 0. Differentiating yields

dτ = − β a ds , (8.6)

which, when substituted into Schwarzschild’s Equation, gives

dI
= I−B. (8.7)
dτ
We can integrate this equation by multiplying both sides by the
integrating factor e−τ and noting that

d dI
[ Ie−τ ] = e−τ − Ie−τ , (8.8)
dτ dτ
so that
dI
e−τ − Ie−τ = − Be−τ (8.9)
dτ
becomes
d
[ Ie−τ ] = − Be−τ . (8.10)
dτ
We then integrate with respect to τ between the sensor at τ = 0 and
some arbitrary point τ :

τ
τ
d
[ Ie−τ ] dτ = − Be−τ dτ , (8.11)
0 dτ 0

which simplifies to

τ
−τ
[ I (τ )e ] − I (0) = − Be−τ dτ (8.12)
0

or

τ
−τ
I (0) = I ( τ ) e + Be−τ dτ . (8.13)
0
200 8. Atmospheric Emission

Let’s stop a moment and carefully study this equation, because it

is remarkably far-reaching in its implications for radiative transfer.
On the left hand side is I (0) which is the radiant intensity observed
by the sensor stationed at τ = 0. On the right hand side are two
terms:

1. The intensity I at position τ = τ , multiplied by the trans-

mittance t(τ ) = e−τ between the sensor and τ . This term
therefore represents the attenuated contribution of any radi-
ation source on the far side of the path. For example, for
a downward-viewing satellite sensor, I (τ ) could represent

emission from the earth’s surface, in which case e−τ is the to-
tal atmospheric transmittance along the line-of-sight.

2. The integrated thermal emission contributions Bdτ from each

point τ along the line of sight between the sensor and τ .
Again, these contributions are attenuated by the respective
path transmittances between the sensor and τ , hence the ap-
pearance of t(τ ) = e−τ inside the integral.

Although we went through some minor mathematical gymnastics

to get to this point, our analysis of the resulting equation is in fact
consistent with a common-sense understanding of how radiation
“ought” to work. Among other things, (8.13) tells us that I (0) is
not influenced by emission from points for which the intervening
atmosphere is opaque (t ≈ 0). It also tells us that, in order to have
emission from a particular point along the path, β a must be nonzero
at that point, since otherwise dτ, and therefore Bdτ, is zero.
Key Point: Almost all common radiative transfer problems involv-
ing emission and absorption in the atmosphere (without scattering) can be
understood in terms of (8.13)!2
Let us try to gain more insight into (8.13) by rewriting it in a
couple of different forms. Let’s start by using the fact that t = e−τ ,
so that
dt = −e−τ dτ . (8.14)

2 The only exceptions are those in which local thermodynamic equilibrium

(LTE) doesn’t apply, as discussed in Section 6.2.3.

 Schwarzschild’s Equation 201

With these substitutions, we now have

1
I (0) = I ( τ ) t ( τ ) + Bdt . (8.15)
t(τ )

With this form, we can see that equal increments of changing trans-
mittance along the line of sight carry equal weight in determining
the total radiance seen at τ = 0.
Finally, let’s consider the relative contribution of emission as a
function of geometric distance s along the path. Noting that
dt
dt = ds , (8.16)
ds
we have
S
dt(s)
I ( S ) = I ( s0 ) t ( s0 ) + B(s) ds , (8.17)
s0 ds
where s0 is the geometric point that coincides with our chosen opti-
cal limit of integration τ . Comparing this form with (7.55), it sud-
denly dawns on us that the integral on the right hand side can be
written
S
S
dt(s)
B(s) ds = B(s)W (s)ds , (8.18)
s0 ds s0

where the weighting function W (s) for thermal emission is exactly the
same as the weighting function for absorption of radiation traveling in the
opposite direction!

A Digression on the Emission Weighting Function†

Let us take a minute to show how general the above result really
is. Although our focus in this chapter is on the atmosphere, the va-
lidity of W (s) = dt(s)/ds as a measure of where emitted radiation
is coming from is not restricted to the atmosphere. Consider, for
example, the surface of an opaque, perfectly absorbing medium po-
sitioned at s , such that the transmittance between an external point
and arbitrary s is a step function; i.e.,

1 s > s
t(s) = (8.19)
0 s < s
202 8. Atmospheric Emission

For this case dt(s)/ds ≡ W (s) = 0 for any s = s0 and infinite for
s = s . In fact, we can write

W (s) = δ(s − s ) , (8.20)

where δ( x ) is the Dirac δ-function. If you have not heard of this

function before, then let me quickly summarize its key properties:

∞ x=0
δ( x ) = (8.21)
0 x = 0

In other words, δ( x ) is a function that is zero everywhere except at

x = 0, where it is an infinitely tall, infinitely narrow spike. Despite
these unusual properties, the area under the curve is defined to be
finite and equal to one. That is,

x

0 x < x
δ( x − x ) dx = (8.22)
−∞ 1 x > x

This is of course consistent with (8.19) and (8.20).

Now if you think about it a bit, you will realize that the product
of δ( x − x ) with f ( x ) is just the same δ-function, but multiplied by
the value of f ( x ) evaluated exactly at x . Thus, (8.22) implies

x2

f (x) x1 < x < x2
δ( x − x ) f ( x ) dx = (8.23)
x1 0 otherwise

Let us now look at what this implies for (8.18), given (8.20):

S
I ( S ) = I ( s0 ) t ( s0 ) + B(s)δ(s − s )ds , (8.24)
s0

where we will take s0 to be an arbitrary point below our surface at

s , so that s0 < s < S. Since the medium in question is opaque by
assumption, the transmittance t(s0 ) = 0, and we’re left with

S
I (S) = B(s)δ(s − s )ds , (8.25)
s0

or, invoking (8.23),

I (S) = B(s ) . (8.26)
 Radiative Transfer in a Plane Parallel Atmosphere 203

We have just confirmed, in an admittedly roundabout way, what

we already knew: The emitted intensity from a opaque blackbody
is just the Planck function B evaluated at the surface of the black-
body. Why did we go to all this trouble? My purpose was simply
to demonstrate that (8.17) is quite general (for a nonscattering and
nonreflecting medium) and is valid even when the transmittance t
is a discontinuous function of distance along a path.

8.2 Radiative Transfer in a Plane Parallel Atmo-

sphere
Let us now adapt (8.17) to a plane-parallel atmosphere. We will
start by considering the case that a sensor is located at the surface
(z = 0), viewing downward emitted radiation from the atmosphere.
The appropriate form of the radiative transfer equation is then

∞
I ↓ (0) = I ↓ ( ∞ ) t ∗ + B(z)W ↓ (z)dz , (8.27)
0

where z = ∞ represents an arbitrary point beyond the top of the at-

mosphere, and t∗ ≡ exp(−τ ∗ /µ) is the transmittance from the sur-
face to the top of the atmosphere. Recall that we are using B(z) here
as a shorthand for Bλ [ T (z)], where T (z) is the atmospheric temper-
ature profile. Because the transmittance t(0, z) between the surface
and altitude z decreases with increasing z, our weighting function
W ↓ (z) in this case is given by

dt(0, z) β a (z)
W ↓ (z) = − = t(0, z) . (8.28)
dz µ

Unless the sensor is pointing at an extraterrestrial source of ra-

diation, such as the sun, I ↓ (∞) = 0. In this case, the first term on
the right vanishes, and the observed downward intensity is strictly
a function of the atmospheric temperature and absorption profiles.
204 8. Atmospheric Emission

Now let’s consider a sensor above the top of the atmosphere

looking down toward the surface. We then have

∞
I ↑ ( ∞ ) = I ↑ (0) t ∗ + B(z)W ↑ (z)dz , (8.29)
0

where
dt(z, ∞) β a (z)
W ↑ (z) = = t(z, ∞) . (8.30)
dz µ

Note the strong similarity between (8.27) and (8.29). Both state
that the radiant intensity emerging from the bottom or top of the
atmosphere is the sum of two contributions: 1) the transmitted ra-
diation entering the atmosphere from the opposite side, and 2) a
weighted sum of the contributions of emission from each level z
within the atmosphere.

8.2.1 The Emissivity of the Atmosphere

Consider the case that T (z) = Ta , where Ta is the temperature of an
isothermal atmosphere. Then B[ T (z)] = B( Ta ) = constant, so that
(8.27), together with (8.28), can be written

∞
dt(0, z)
I ↓ (0) = I ↓ (∞)t∗ + B( Ta ) − dz . (8.31)
0 dz

The integral reduces to t(0, 0) − t∗ = 1 − t∗ , yielding

I ↓ (0) = I ↓ (∞)t∗ + B( Ta )[1 − t∗ ] . (8.32)

By the same token, (8.29)and (8.30) can be manipulated to yield

I ↑ (∞) = I ↑ (0)t∗ + B( Ta )[1 − t∗ ] . (8.33)

Recall that the absorptivity of a nonscattering layer is just one minus

the transmittance, and that Kirchhoff’s Law states that absorptivity
equals emissivity. The interpretation of the above two equations is
thus remarkably simple: The total radiant intensity is just the sum of
 Radiative Transfer in a Plane Parallel Atmosphere 205

(a)the transmitted intensity of any source on the far side plus (b) Planck’s
function times the emissivity of the entire atmosphere.
In reality, of course, the atmosphere is never isothermal. Never-
theless, it is sometimes convenient to use equations similar in form
to (8.32) and (8.33) to describe the atmospheric contribution to the
observed radiant intensity:

↓
I ↓ (0) = I ↓ ( ∞ ) t ∗ + B [1 − t ∗ ] , (8.34)

↑
I ↑ ( ∞ ) = I ↑ (0) t ∗ + B [1 − t ∗ ] , (8.35)
where
∞
↓ 1
B = B(z)W ↓ (z) dz , (8.36)
1 − t∗ 0

∞
↑ 1
B = B(z)W ↑ (z) dz , (8.37)
1 − t∗ 0

give the weighted average Planck function values for the entire atmo-
sphere.

Problem 8.1: Show that when 1 − t∗  1, W ↓ (z) ≈ W ↑ (z) ≈

↓ ↑
β a (z)/µ, so that B ≈ B .

8.2.2 Monochromatic Flux †

Sometimes it is not the intensity but rather the flux that we care
about. For example, if we ignore extraterrestrial sources (appropri-
ate for the LW band), we can compute the downwelling monochro-
matic flux F ↓ at the surface simply by integrating I ↓ cos θ over one
hemisphere of solid angle, according to (2.55):

2π
π
F↓ = − I ↓ (θ, φ) cos θ sin θ dθdφ.
0 π/2

There are two simplifications we can make before we even take

the obvious step of substituting (8.27). First, in our plane parallel
206 8. Atmospheric Emission

atmosphere, nothing depends on φ; therefore we can immediately

deal with the integration over azimuth, leaving

π
↓
F = −2π I ↓ (θ ) cos θ sin θ dθ . (8.38)
π/2

Second, we can substitute µ = | cos θ | as our variable for describing

zenith angle. For downwelling radiation, cos θ < 0, so in this in-
stance µ = − cos θ and dµ = sin θdθ; using (8.27), we are now able
to write

1 
∞ 
↓ ↓
F (0) = 2π B(z)W (z, µ) dz µdµ . (8.39)
0 0

Only W ↓ depends on µ. Therefore, if we wish, we can reverse the

order of integration and write

∞
F ↓ (0) = πB(z)WF↓ (z) dz , (8.40)
0

where

1
1
∂t(0, z; µ)
WF↓ (z) ≡ 2 W ↓ (z, µ)µ dµ = −2 µ dµ (8.41)
0 0 ∂z

may be thought of as a flux weighting function. Hopefully, you will

recognize the term πB(z) appearing in (8.40) as the monochromatic
flux you would expect from an opaque blackbody having the tem-
perature found at altitude z. With a little more rearrangement, we
can write

1 
↓ ∂ ∂t F (0, z)
WF (z) = − 2 t(0, z; µ)µ dµ = − , (8.42)
∂z 0 ∂z

where the monochromatic flux transmittance t F between levels z1

and z2 is defined as

1
1 τ (z1 ,z2 )
−
t F ( z1 , z2 ) ≡ 2 t(z1 , z2 ; µ)µ dµ = 2 e µ µ dµ . (8.43)
0 0
 Radiative Transfer in a Plane Parallel Atmosphere 207

Problem 8.2: Find expressions analogous to (8.40)–(8.42) for both

the downwelling and upwelling flux at an arbitrary altitude z in the
atmosphere.

Unfortunately, there is no simple closed-form solution to the in-

tegral in (8.43), though it is easily solved numerically. However, it
turns out that the above expression for t F behaves very much like
the simple “beam” transmittance t = exp(τ/µ) — that is, it de-
creases quasi-exponentially from one to zero as τ goes from zero to
infinity. Therefore, in order to deal with the integral, it is common
to use the approximation

1
− µτ
tF = 2 e µ dµ ≈ e−τ/µ , (8.44)
0

where µ describes an effective zenith angle such that the correspond-

ing beam transmittance is approximately equal to the flux transmit-
tance between two levels.

Problem 8.3:
(a) Show that, for any specified value of τ in (8.44), there is always
a value of µ between zero and one that makes the relationship exact.
(b) Find the value of µ that makes the relationship exact for the
case that τ = 0.

Although the “perfect” value of µ is a function of τ, numerical

experiments reveal that you can often get away with choosing a sin-
gle constant value that yields a reasonable overall fit between the
approximate expression

∂t(0, z; µ)
WF↓ (z) ≈ − , (8.45)
∂z
and the exact expression (8.42). The most commonly used value is
µ = 1/r, where r = 5/3 is the so-called diffusivity factor that arises
in certain theoretical analyses of the flux transmittance.
208 8. Atmospheric Emission

8.2.3 Surface Contributions to Upward Intensity

Equation (8.29) described the intensity of monochromatic radiation
as seen looking downward from the top of the atmosphere. One of
the terms appearing in this equation is the upward radiant inten-
sity at the surface I ↑ (0). This is the only term whose value cannot
be directly computed from knowledge of β a (z) and T (z) alone (for
given wavelength λ and viewing direction µ). Therefore, in order to
have a complete, self-contained expression for I ↑ (∞), it is necessary
to find an explicit expression for I ↑ (0).
The expression we supply for I ↑ (0) depends on what we assume
about the nature of the surface. But regardless of those assumptions,
there are two contributions that must be considered: 1) emission by
the surface itself, and 2) upward reflection of atmospheric radiation
incident on the surface.

Specular Lower Boundary

Let us first consider the simplest possible case: that of a specular

surface with emissivity ε. In this case, the reflectivity r = 1 − ε, and
we have
I ↑ (0) = εB( Ts ) + (1 − ε) I ↓ (0) , (8.46)

where Ts is the temperature of the surface (not necessarily the same

as the surface air temperature!), and I ↓ is evaluated for the same µ as
I ↑ . Now recall that we have already derived an expression for I ↓ (0),
namely (8.27). Combining the latter with (8.29) and (8.46) yields:

∞

I ↑ (∞) = εB( Ts ) + (1 − ε) B(z)W ↓ (z) dz t∗
0

∞ (8.47)
↑
+ B(z)W (z) dz ,
0

where we have assumed that there is no extraterrestrial source of

downward radiation in the direction of interest (i.e., we’re not look-
ing at the sun’s reflection). Given the appropriate weighting func-
tions W ↑ and W ↓ , we have a self-contained expression for the radi-
ant intensity that would be observed by a downward-viewing satel-
lite sensor.
 Radiative Transfer in a Plane Parallel Atmosphere 209

If we wish, we can use the notation developed in the previous

subsection to hide the integrals:
 
↓ ↑
I ↑ (∞) = εB( Ts ) + (1 − ε) B [1 − t∗ ] t∗ + B [1 − t∗ ] . (8.48)

There are three limiting cases of the above that help persuade us
that our analysis makes sense. The first is that of a perfectly trans-
parent atmosphere (t∗ = 1), in which case our equation reduces to
the expected dependence on surface emission alone:
I ↑ (∞) = εB( Ts ) . (8.49)
The second is that of a perfectly opaque atmosphere (t∗ = 0), in
which case surface emission and reflection are both irrelevant, and
we have
∞
↑
I ↑ (∞) = B = B(z)W ↑ (z) dz . (8.50)
0
The third occurs when the surface is nonreflecting; i.e., ε = 1, in
which case we have
↑
I ↑ (∞) = εB( Ts )t∗ + B [1 − t∗ ] . (8.51)
It would be tempting to interpret the last of these as showing a lin-
ear relationship between I ↑ and atmospheric transmittance t∗ , but
↑
remember that B also depends on the atmospheric opacity. In gen-
↑
eral, as the atmosphere becomes more opaque, B represents emission from
higher (and therefore usually colder) levels of the atmosphere.

Lambertian Lower Boundary †

As discussed in Chapter 5, the “opposite” of specular reflection is
a Lambertian reflection, for which radiation incident on the surface
from any direction is reflected equally in all directions. When the
surface is Lambertian, the upward directed radiance at the surface
I ↑ (0) includes contributions due to the reflection of downwelling
radiation from all possible directions. Combining (8.29) with (5.12),
(5.14), and (8.41), we get

∞ 
↑ ↓
I (∞, µ) = εB( Ts ) + 2(1 − ε) πB(z)WF (z) dz t∗
0

∞ (8.52)
+ B(z)W ↑ (z, µ) dz ,
0
210 8. Atmospheric Emission

where we have made the dependence of W ↑ and W ↓ on µ explicit.

Problem 8.4: Derive (8.52).

Problem 8.5: Equation (8.52) assumes there is no extraterrestrial

source of radiation. Generalize it to include a columnated source
(e.g., the sun) of monochromatic flux S (measured normal to the
beam). The cosine of the zenith angle of the source is µ0 .

Problem 8.6: Equations (8.27)–(8.30) were derived for radiant in-

tensities observed at the bottom and top of the atmosphere, respec-
tively. Generalize these to describe the downward and upward in-
tensities I ↓ (z) and I ↑ (z) at any arbitrary level z in the atmosphere.
Include explicit expressions for the new weighting functions W ↓ (z)
and W ↑ (z) in terms of β a (z), etc.

8.3 Applications to Meteorology, Climatology,

and Remote Sensing
This chapter introduced the relationships that describe the transfer
of monochromatic radiation in an atmosphere that doesn’t scatter
appreciably but does absorb and emit radiation. When do these con-
ditions apply?
As a general rule of thumb (see Chapter 12 for details), the
longer the wavelength, the larger a particle has to be before it is ca-
pable of scattering appreciably. You can definitely neglect scattering
by air molecules for wavelengths falling anywhere in the infrared or
microwave bands. Cloud droplets and cloud ice particles continue
to scatter appreciably throughout most of the near IR band, but in
the thermal and far IR bands, water clouds (and to a lesser extent
 Applications 211

ice clouds) tend to look a lot like blackbodies. By the time you get
to the microwave band, about the only particles for which scatter-
ing can’t be neglected are precipitation particles — i.e., raindrops,
snowflakes, hailstones, etc.
In short, you’re usually safe using the relationships derived in
this chapter throughout the thermal IR and microwave bands (ex-
cepting precipitation), while you’re almost never safe using them in
the solar (shortwave) part of the spectrum.
As you already know, the thermal IR band plays an extremely
important role in the exchange of energy within the atmosphere,
and between the atmosphere, the surface, and outer space. In prin-
ciple, one might take the relationships we derived for monochro-
matic intensities and simply integrate them over both wavelength
and solid angle in order to obtain broadband radiative fluxes at any
level in the atmosphere. This is easier said than done, however, ow-
ing to the extreme complexity of the dependence of β a (z) on wave-
length, as we shall see in Chapter 9. It is necessary to develop spe-
cial methods for efficiently computing longwave (broadband) flux,
and flux divergence, in the atmosphere. Some of these methods will
be outlined in Chapter 10.
The relationships developed earlier in this chapter are therefore
most directly useful in the context of infrared and microwave re-
mote sensing. Most satellite sensors operating in these bands ob-
serve intensities rather than fluxes of radiation, and most do so for
very narrow ranges of wavelength, so that in many cases the ob-
served intensities can be thought of as quasi-monochromatic.
In the following, we will take a look at real-world atmospheric
emission spectra and also outline the principles behind temperature
and humidity profile retrieval. Since all of these topics are closely
tied to the absorption spectra of various atmospheric constituents,
you might find it helpful to go back and review the major absorption
bands depicted in Fig. 7.6.

8.3.1 The Spectrum of Atmospheric Emission

A spectrometer is a device that measures radiant intensity as a func-
tion of wavelength. Infrared spectrometers have been deployed on
satellites, which view the atmosphere looking down from above,
212 8. Atmospheric Emission

Wavelength [µm]
30 25 20 18 15 14 13 12 11 10 9 8
160

140

300 K Nauru (Tropical Western Pacific)

11/15/98
120
Radiance [mW / m2 sr cm-1]

100

Barrow, Alaska 3/10/99

80

60

245 K
40

20

0
300 400 500 600 700 800 900 1000 1100 1200 1300 1400
Wavenumber [cm-1]

Fig. 8.1: Two examples of measured atmospheric emission spectra as seen from
ground level looking up. Planck function curves corresponding to the approxi-
mate surface temperature in each case are superimposed (dashed lines). (Data cour-
tesy of Robert Knuteson, Space Science and Engineering Center, University of Wisconsin-
Madison.)

and at fixed and mobile research stations on the ground, where they
can measure the atmospheric emission spectrum reaching the sur-
face. In addition, spectrometers are occasionally flown on research
aircraft. The latter may view downward, upward, or both.
For an upward-viewing instrument at ground level, the relevant
form of the radiative transfer equation is given most generally by
(8.27), but here we will assume that there are no extraterrestrial
sources of radiation:

∞
I ↓ (0) = B(z)W ↓ (z)dz . (8.53)
0
 Applications 213

For a satellite sensor viewing downward, we have

∞
I ↑ (∞) = B( TS )t∗ + B(z)W ↑ (z)dz , (8.54)
0

where we are assuming that the surface is nonreflective and has

temperature TS , so that I ↑ (0) = B( TS ).
For the sake of the discussion to follow, we needn’t worry about
the detailed shape of the weighting functions W ↑ (z) and W ↓ (z). The
important thing to remember is that when the atmosphere is very
opaque, the measured radiation will be due principally to emission
from the nearest levels of the atmosphere; when it is less opaque,
the relevant weighting function will include emission from more
remote levels. Finally, when the atmosphere is very transparent, the
instrument will “see through” the atmosphere, measuring primar-
ily the radiance contribution (if any) from the far side — e.g., the
surface for a downward-looking instrument or “cold space” for an
upward-looking one.
Of course, no matter what the source of the measured radiance,
we can always interpret that radiance as a brightness temperature
TB , according to (6.13). If the atmosphere happens to be opaque
at a particular wavelength, then the brightness temperature in turn
gives you a reasonable estimate of the physical temperature at the
level where the weighting function peaks.
Figure 8.1 shows a pair of relatively high-resolution IR spectra
measured at ground level under two different sets of conditions.
One was taken at a location in the tropical western Pacific, where at-
mospheric temperatures are warm and humidity is high. The other
was taken at an arctic location in late winter, where temperatures
are very cold and the atmosphere contains only small amounts of
water vapor. Both were taken under cloud-free skies.
As complicated as these spectra appear to the untrained eye, the
interpretation is really rather simple. Let’s take it one step at a time.

• The two dashed curves depict the Planck function at tempera-

tures representative of the warmest atmospheric emission seen
by the instrument at each location. At the tropical site, this is
300 K; at the arctic site, it’s 245 K. Note that these curves serve
as an approximate upper bound to the radiance measured in
any wavelength band.
214 8. Atmospheric Emission

• In the tropical case, there are two spectral regions for which
the measured radiance is very close to the reference curve at
300 K. One is associated with λ > 14 µm (ν̃ < 730 cm−1 ); the
other with λ < 8 µm (ν̃ > 1270 cm−1 ). In both of these bands,
we can infer that the atmosphere must be quite opaque, be-
cause virtually all of the observed radiation is evidently being
emitted at the warmest, and therefore lowest, levels of the at-
mosphere. Referring to Fig. 7.6 in section 7.4.1, we find that
these features are consistent with (a) strong absorption by CO2
in the vicinity of 15 µm, (b) strong absorption by water vapor
at wavelengths longer than 15 µm, and (c) strong absorption
by water vapor between 5 and 8 µm.

• The two water vapor absorption bands mentioned above are

also evident in the arctic emission spectrum (for which the
measurements extend to a somewhat longer maximum wave-
length of 25 µm). However, because the atmosphere is so
dry in this case, these bands are not as uniformly opaque and
therefore the measured radiance is quite variable. In fact, be-
tween 17 and 25 µm, the general impression is of a large num-
ber of strong water vapor lines separated by what might be
termed “microwindows.” The most transparent of the latter is
found near 18 µm (560 cm−1 ).

• Between 8 and 13 µm in the tropical spectrum, observed

brightness temperatures are considerably colder than the sur-
face, and in some cases as cold as 240 K. This broad region may
be regarded as a “dirty indexSpectral window¡‘dirty” win-
dow” — the atmosphere overall is fairly transparent, but there
are a large number of individual absorption lines due to water
vapor.

• In the arctic spectrum, the above window “opens up” substan-

tially, because there is far less water vapor in the atmospheric
column to absorb and emit radiation. Therefore, observed ra-
diances between 8 and 13 µm are generally quite low indeed
— the instrument effectively has an unobstructed view of cold
space.

• A clear exception to the above statement is found between 9

 Applications 215

and 10 µm. Referring again to Fig. 7.6, we infer that the culprit
this time is the ozone absorption band centered at 9.6 µm. This
band, while not totally opaque, emits sufficiently strongly
to raise the brightness temperature to around 230 K, much
warmer than the surrounding window region. The “cold”
spike at the center of the ozone band corresponds to an iso-
lated region of relative transparency.

• Carefully examining the tropical emission spectrum, we dis-

cover that the 9.6 µm ozone band is apparent there as well.
But because the surrounding window is much less transpar-
ent in this case, the ozone band does not stand out nearly as
strongly.

• Now let’s take a closer look at the CO2 band in the arctic spec-
trum. At 15 µm, the measured brightness temperature of ap-
proximately 235 K is significantly colder than it is on the edges
of the band near 14 and 16 µm, where the brightness tempera-
ture is close to the maximum value of 245 K. Yet we know that
at the center of the band, where absorption is the strongest,
the emitted radiation should be originating at the levels clos-
est to the intrument, namely in the lowest few meters of the
atmosphere. How do we explain this apparent paradox? In
fact, there is no paradox; we are simply witnessing the effects
of a strong inversion in the surface temperature profile — i.e.,
a sharp increase in temperature with height in the lowest few
hundred meters of the atmosphere.3 At the center of the CO2
band, the spectrometer records emission from the coldest air
right at the surface. At the edges of the band, where the air is
less opaque, it sees emission from the warmer layer of air at
the top of the inversion. This example hints at the possibility
of using remote measurements of atmospheric emission to in-
fer atmospheric temperature structure. We will return to that
topic shortly.

Let’s now take a look at how the atmospheric emission spectrum

changes depending on whether you are looking down from above
3 Deep surface inversions, relatively uncommon elsewhere, are the rule in win-

tertime polar regions.

216

Wavelength [µm]
25 20 18 15 14 13 12 11 10 9 8 7 6
120

100 (a) 20 km looking down

Radiance [mW / m2 sr cm-1]

80

60 26
0K

24
40 0K
22
0K
20 200
K
180
160 K K

0
400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600 1700
-1
Wavenumber [cm ]

Wavelength [µm]
25 20 18 15 14 13 12 11 10 9 8 7 6
120

100 (b) Surface looking up

Radiance [mW / m2 sr cm-1]

80

60
26
0K
40 24
0K
200
K 220
180 K
20 K
160 K

0
400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600 1700
Wavenumber [cm-1]

Fig. 8.2: Coincident measurements of the infrared emission spectrum of the cloud-
free atmosphere at (a) 20 km looking downward over the polar ice sheet and (b) at
the surface looking upward. (Data courtesy of David Tobin, Space Science and Engi-
neering Center, University of Wisconsin-Madison.)
 Applications 217

or looking up from the surface. Fig. 8.2 gives us a rare opportunity

to compare the two perspectives for the same atmospheric condi-
tions: an aircraft flying at 20 km altitude measured the upwelling
emission spectrum at exactly the same time and location as a sur-
face instrument looking up measured the downwelling spectrum.
The measurements in this case were taken over the arctic ice pack
and are therefore comparable in some respects to the arctic spec-
trum already discussed. The following exercise asks you to provide
the physical interpretation:

Problem 8.7: Based on the measured spectra depicted in Fig. 8.2,

answer the following questions: (a) What is the approximate temper-
ature of the surface of the ice sheet, and how do you know? (b) What
is the approximate temperature of the near-surface air, and how do
you know? (c) What is the approximate temperature of the air at
the aircraft’s flight altitude of 20 km, and how do you know? (d)
Identify the feature seen between 9 and 10 µm in both spectra. (e)
In Fig. 8.1, we saw evidence of a strong inversion in the near-surface
atmospheric temperature profile. Can similar evidence be seen in
Fig. 8.2? Explain.

We’ll conclude our discussion of atmospheric emission spectra

by looking at examples of satellite observations made under diverse
conditions around the globe (Fig. 8.3). Once again, I’ll leave most of
the physical interpretation as an exercise. Only one new point re-
quires a brief explanation, because it didn’t arise in our previous
discussion of surface- and aircraft-measured spectra. That concerns
the prominent narrow spike observed at the center of the 15 µm CO2
band in all four panels. This spike occurs where absorption is by far
the most intense of any point in the thermal IR band. From the van-
tage point of a satellite sensor viewing downward, emission at this
wavelength therefore originates almost entirely in the stratosphere,
whereas most of the remaining emission spectrum is associated pri-
marily with the surface and troposphere.4
4 The troposphere is the layer of the atmosphere closest to the surface and is usu-
ally characterized by decreasing temperature with height. The troposphere is typ-
ically anywhere from a few km to 15 km deep, depending on latitude and sea-
son. The stratosphere is the deep layer above the troposphere in which temperature
218

Wavelength [µm]
25 20 18 15 14 13 12 11 10 9 8 7
200
320
(a) Sahara Desert
Radiance [mW / m2 sr cm-1]

150 300

280

100
260

240

220
50

0
400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600
-1
Wavenumber [cm ]

Wavelength [µm]
25 20 18 15 14 13 12 11 10 9 8 7
50

45
(b) Antarctic Ice Sheet
40
Radiance [mW / m2 sr cm-1]

35

30

25 220

20 200

180
15

10

0
400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600
Wavenumber [cm-1]

Fig. 8.3: Examples of moderate resolution IR spectra observed by a satellite spec-

trometer. Except for the curve labeled “thunderstorm anvil” in panel (c), all spectra
were obtained under cloud-free conditions. (Nimbus-4 IRIS data courtesy of the God-
dard EOS Distributed Active Archive Center (DAAC) and instrument team leader Dr.
Rudolf A. Hanel.)
 219

Wavelength [µm]
25 20 18 15 14 13 12 11 10 9 8 7

140
300 (c) Tropical Western Pacific
120 280
Clear
Radiance [mW / m2 sr cm-1]

100
260
80
240

60
220

40 200

20 Thunderstorm Anvil

0
400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600
Wavenumber [cm-1]

Wavelength [µm]
25 20 18 15 14 13 12 11 10 9 8 7

140 300
(d) Southern Iraq
120 280
Radiance [mW / m2 sr cm-1]

100 260

80
240

60 220

40 200

20

0
400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600
Wavenumber [cm-1]

Fig. 8.3: (cont.)

 220 8. Atmospheric Emission

To summarize, as you move from the edge of the CO2 band to-
ward the center, the general tendency for satellite-observed spec-
tra is usually: (i) decreasing brightness temperature as the emission
weighting function W ↑ peaks at higher (and colder) levels in the
troposphere, followed by (ii) a sharp reversal of this trend at the
strongly absorbing center of the band, where the weighting function
peaks at an altitude solidly within the relatively warm stratosphere.

Problem 8.8: Contrast the above explanation for the radiance spike
at the center of the 15 µm CO2 band with that given earlier for the
similar-appearing spike at the center of the 9.6 µm ozone band.

Problem 8.9: Referring to Fig. 8.3, answer each of following ques-

tions.
(a) For each of the four scenes, provide an estimate of the surface
temperature.
(b) For which scene does the surface appear to be significantly
colder than any other level in the atmosphere?
(c) Compare the apparent humidity of the atmosphere over
Southern Iraq with that over the Sahara Desert, and explain your
reasoning.
(d) Estimate the temperature of the thunderstorm cloud top in
the Tropical Western Pacific. Why is this emission spectrum so much
smoother than that for the neighboring clear atmosphere? Explain
the small variations in brightness temperature that do show up.
(e) Explain why the ozone band at 9.6 µm shows up as a relatively
warm feature in the Antarctic spectrum, but a relatively cold feature
for all other scenes.

generally increases with height. The boundary between the troposphere and the
stratosphere, called the tropopause, is often (though not always) the coldest point
in the atmospheric temperature profile below 40 km. The layer of highest ozone
concentration is found in the stratosphere between 15 and 30 km. If these facts are
new to you, then I recommend that you spend an evening with Chapter 1 of WH77
and/or Section 3.1 of L02.

Review preprint for I. Sokolik - not for distribution --

 Applications 221

8.3.2 Satellite Retrieval of Temperature Profiles

As we have seen in Sections 7.4.1 and 8.3.1, certain atmospheric con-
stituents — most notably CO2 , water vapor, ozone, and oxygen —
are associated with strong absorption lines and bands that render
the atmosphere opaque over certain ranges of wavelengths. Some
of these constituents, such as CO2 and O2 are “well mixed” through-
out the troposphere and stratosphere. That is, they are present at a
constant, accurately known mass ratio w to all other constituents
of the atmosphere. If you knew the density profile ρ(z) of the at-
mosphere, then the density of the absorbing constituent would just
be ρ (z) = wρ(z). If you also knew the mass absorption coefficient
ka (z) of the constituent, then you could calculate the optical depth
τ (z) and thus the emission weighting function W ↑ (z) that appears
in the integral in (8.29).
The relative strength of absorption at the wavelength in question
determines whether W ↑ (z) will peak at a high or low altitude in the
atmosphere. If you build a satellite sensor that is able to measure the
radiant intensities Iλ for a series of closely spaced wavelengths λi
located on the edge of a strong absorption line or band for the con-
stituent (e.g., near 15 µm for CO2 ), each channel will measure ther-
mal emission from a different layer of the atmosphere. The closer
the wavelength is to the line center, the higher in the atmosphere
yhe weighting function will peak. The intensity of the emission is
of course determined by the atmospheric temperature within that
layer.
In principle, you can find a temperature profile T (z) that is si-
multaneously consistent with each of the observed radiances Iλ for
the scene in question. Typically, you would start with a “first guess”
profile and compute the associated intensities for each channel us-
ing (8.29). You would then compare the computed intensities with
the observed intensities and adjust the profile in such a way as to
reduce the discrepancies. This process could be repeated until the
differences for all channels fell to within some tolerance, based on
the assumed precision of the instrument measurements themselves
and of the model calculations of Iλ .
The above procedure (with certain important caveats; see below)
is in fact not too far from that actually used in routine satellite-based
temperature profile retrievals. It is not my purpose here to embark
222 8. Atmospheric Emission

(a) (b) (c)

Altitude

Weight
Fig. 8.4: Idealized satellite sensor weighting functions for atmospheric tempera-
ture profile retrievals. (a) Channel weighting functions resemble δ-functions; i.e.,
all emission observed by each channel originates at a single altitude. (b) Observed
emission represents layer averages, but channel weighting functions do not over-
lap. (c) Realistic case, in which weighting functions not only represent layer aver-
ages but also overlap.

on a rigorous discussion of remote sensing theory, which is best left

for a separate course and/or textbook. It is enough for now that you
recognize the close connection between the radiative transfer princi-
ples discussed earlier in this chapter and an application of immense
practical importance to modern meteorology.
Figure 8.4 depicts the physical basis for profile retrieval at three
levels of idealization, starting with the simplest — and least realis-
tic — on the left: If weighting functions happened to be perfectly
sharp — that is, if all emission observed at each wavelength λi orig-
inated at a single altitude (Fig. 8.4a), then “inverting” the observa-
tions would be simple: in this case, the observed brightness tem-
peratures TB,i would exactly correspond to the physical tempera-
tures at the corresponding altitudes hi . Your job is then essentially
finished without even lifting a calculator. Of course, you wouldn’t
know how the temperature was varying between those levels, but
you could either interpolate between the known levels and hope for
the best or, if your budget was big enough, you could add an arbi-
 Applications 223

trary number of new channels to your sensor to fill in the vertical

gaps.
Slightly more realistically, panel (b) depicts the weighting func-
tions as having finite width, so that the observed brightness tem-
peratures correspond to an average of Bλ [ T (z)] over a substantial
depth of the atmosphere rather than a unique temperature Ti at al-
titude hi . There is now ambiguity in the retrieval, because there is
no single atmospheric level that is responsible for all of the emission
measured by any given channel. At best, you can estimate the av-
erage layer temperature associated with each channel. Nevertheless,
the profile retrieval problem itself remains straightforward, because
each channel contains completely independent information: there
is no overlap between the weighting functions.
Unfortunately, real weighting functions are constrained to obey
the laws of physics, as embodied in (8.30). This means that un-
less you have an unusually sharp change with altitude in the at-
mospheric absorption coefficient β a , your weighting functions will
be quite broad. In the worst case, the mass extinction coefficient
ka of your chosen constituent will be nearly constant with height,
so that the weighting functions will be essentially those predicted
for an exponential absorption profile as discussed in Section 7.4.3.
The situation is somewhat better for sensor channels positioned on
the edge of an absorption line (or between two lines), because pres-
sure broadening (see Chapter 9) then increases ka toward the surface,
which sharpens the weighting function. Nevertheless, the improve-
ment is not spectacular.
Therefore, given any reasonable number of channels, there is al-
ways considerable overlap between adjacent weighting functions,
as depicted schematically in Fig. 8.4c. In fact, Fig. 8.5 shows actual
weighting functions for the Advanced Microwave Sounding Unit
(AMSU), which has 11 channels on the edge of the strong O2 ab-
sorption band near 60 GHz (c.f. Fig. 7.7). Although each satellite
sounding device has its own set of channels and therefore its own
unique set of weighting functions, those for the AMSU are fairly
typical for most current-generation temperature sounders in the in-
frared and microwave bands.
To summarize: It is clear on the one hand that there is informa-
tion about vertical temperature structure in the radiant intensities
224 8. Atmospheric Emission

14

13

10
Presuure (mb) 12

11

10

9
100

8
7

6
5

4
1000
Channel Weight

Fig. 8.5: Weighting functions for channels 4–14 of the Advanced Microwave
Sounding Unit (AMSU).

observed by a sounding instrument like the AMSU. On the other

hand, one shouldn’t underestimate the technical challenge of re-
trieving temperature profiles of consistently useful quality from satel-
lite observations. In outline form, here are the main issues:

• In general, it takes far more variables to accurately describe

an arbitrary temperature profile T (z) than there are channels
on a typical satellite sounding unit. This means that you have
fewer measurements than unknowns, and the retrieval prob-
lem is underdetermined (or ill-posed). The problem is therefore
not just that of finding any temperature profile that is consis-
tent with the measurements; the real problem is of choosing
the most plausible one out of an infinity of physically admissi-
ble candidates.

• Because of the high degree of vertical overlap between adja-

cent weighting functions, the temperature information con-
tained in each channel is not completely independent from
that provided by the other channels. That is to say, if you have
 Applications 225

N channels, you don’t really have N independent pieces of

information about your profile; you have something less than
N, which makes the problem highlighted in the previous para-
graph even worse that it appears at first glance.
• Because any measurement is inherently subject to some de-
gree of random error, or noise, it is important to undertake the
retrieval in such a way that these errors don’t have an exces-
sive impact on the final retrieved profile.
• Because of the large vertical width of the individual weight-
ing functions, a satellite’s view of the atmosphere’s tempera-
ture structure is necessarily very “blurred” — that is, it is im-
possible to resolve fine-scale vertical structure in the tempera-
ture profile. As a consequence, a great many of the candidate
profile solutions that would be physically consistent with the
observed radiances Iλ,i exhibit wild oscillations that are com-
pletely inconsistent with any reasonable temperature struc-
ture of the atmosphere. Effectively weeding out these bad
solutions while retaining the (potentially) good ones requires
one to impose requirements on the “smoothness” of the re-
trieved profile or limits on the allowable magnitude of the de-
parture from the “first guess” profile.

Of course, well-established methods exist for dealing with the

above challenges, and satellite temperature profile retrievals are
successfully obtained at thousands of locations around the globe ev-
ery day. These retrievals provide indispensable information about
the current state of the atmosphere to numerical weather prediction
models. Without the availability of satellite-derived temperature
structure data, accurate medium- and long-range forecasts (three
days and beyond) would be impossible almost everywhere, and
even shorter-range forecasts would be of questionable value over
oceans and other data sparse regions.

8.3.3 Water Vapor Imagery

In the previous subsection, we looked at the case that satellite-
observed emission was associated with a constituent that was “well-
mixed” in the atmosphere. Under that assumption, the vertical
226 8. Atmospheric Emission

Fig. 8.6: An image of the Eastern Pacific and west coast of North America taken
by the GOES-West geostationary weather satellite at a wavelength of 6.7 µm.
This wavelength falls within the strong water vapor absorption band centered on
6.3 µm.

distribution of the absorber is essentially known, and variations in

brightness temperature can be attributed to variations in tempera-
ture in the atmospheric layer associated with each channel.
One can also imagine the opposite situation, in which the tem-
perature profile is reasonably well known (perhaps from satellite
measurements, as described above), but the concentration of the
absorbing/emitting constituent is unknown and highly variable in
both time and space. Such is the case for infrared images acquired at
wavelengths between about 5 and 8 µm, where atmospheric emis-
sion and absorption is very strong in connection with the water va-
por band centered at 6.3 µm.
The wavelength in this band most commonly utilized for satel-
lite imaging is 6.7 µm, where water vapor absorption is strong
 Applications 227

enough to block surface emission from reaching the satellite un-

der most conditions, but weak enough that the imager can usu-
ally “probe” well into the troposphere without being blocked by the
small amounts of water vapor found in the stratosphere and upper
troposphere.
Unlike the case for CO2 and O2 , water vapor is far from well
mixed and varies wildly in concentration, both horizontally and
vertically. Therefore, the emission weighting function W ↑ at 6.7 µm
is highly variable, peaking at low altitudes (or even the surface) in
a very dry atmosphere and in the upper troposphere in a humid
tropical atmosphere or when high altitude clouds are present.
As usual, the observed brightness temperature TB is a function
of the temperature of the atmosphere in the vicinity of the weighting
function peak, but variations in brightness temperature are a much
stronger function of height of the weighting function peak than of
the temperature at any specific altitude.
Consequently dry, cloud-free air masses typically produce rel-
atively warm brightness temperatures, because the observed emis-
sion originates at the warm lower levels of the atmosphere. Humid
air masses, on the other hand, produce cold brightness temperatures,
because emission then originates principally in the cold upper tro-
posphere. In fact, globally speaking, there is often a roughly in-
verse correlation between brightness temperature at 6.7 µm and air
mass temperature, because warm tropical air masses are capable of
supporting higher water vapor content, and may also be associated
with colder tropopause temperatures, on average, than cold extrat-
ropical air masses.
An example of a 6.7 µm image is shown in Fig. 8.6. In this in-
stance, the most dramatic features are the very dark (warm) bands
snaking across the subtropical Pacific. These bands are presum-
ably associated with regions of strong subsidence (sinking motion)
in connection with the subtropical high pressure belt. The effect of
subsidence is to bring extremely dry air from the upper troposphere
down to relatively low levels, allowing the imager to “see” warm
emission from the lowest kilometer or two. Elsewhere, deep hu-
midity and some high-level clouds connected with a weakening ex-
tratropical cyclone produce bands of relatively cold brightness tem-
peratures. Overall, one has an impression of a three-dimensional
228 8. Atmospheric Emission

“surface” representing the upper boundary of the most humid layer

of the atmosphere. This humidity structure is of course invisible in
the conventional visible and IR images shown previously for the
same time (Figs. 5.6 and 6.9).
Because the observed brightness temperature depends both on
the amount and detailed vertical distribution of water vapor present
and on the temperature structure of the atmosphere, it is generally
not possible to retrieve quantitative humidity information using this
channel alone. However, if you have several channels with varying
sensitivity to water vapor absorption, they can be used in combi-
nation with temperature sounding channels to obtain vertical pro-
files of humidity. Of course, the same practical difficulties that were
outlined for the temperature profile retrieval problem are encoun-
tered here as well. If anything, they are worse, because there are
few useful criteria for distinguishing a “realistic” from an “unrealis-
tic” humidity profile, except for the need to avoid supersaturation.
Also, because the temperature profile cannot be retrieved with per-
fect precision, errors in this profile “feed through” to the humidity
profile retrieval and constitute an additional source of error.
 CHAPTER 9

Absorption by Atmospheric Gases†

In the previous chapter, we developed the mathematical tools for

describing monochromatic radiative transfer, including both ab-
sorption and thermal emission, in a nonscattering atmosphere.
Specifically, given the absorption profile β a (z) and temperature pro-
file T (z) in a plane parallel atmosphere, we can use (8.27) through
(8.30) to compute the atmospheric contribution to the radiant inten-
sity, as seen from the surface looking up or from a satellite looking
down. With only slight modifications, they may be used to com-
pute monochromatic intensities at any level z between the surface
and the top of the atmosphere.
In these equations, it is the absorption coefficient β a that directly
links absorption and emission at any specific wavelength to the ma-
terial composition of the atmosphere. In Section 7.4.1, we saw that
a relatively small number of constituents, such as water vapor, car-
bon dioxide, and a few other trace gases, are responsible for almost
all of the important features in the observed absorption/emission
spectrum of the atmosphere. However, we have not yet examined
why certain gases strongly absorb radiation at certain wavelengths
nor how their absorption properties are influenced by temperature
and pressure.
Is it really necessary, you ask, for the nonspecialist to know

229
230 9. Absorption by Atmospheric Gases†

the whys and hows of atmospheric absorption? Even without that

insight, does anything prevent you from computing broadband
fluxes (for example), simply by integrating monochromatic inten-
sities over both solid angle and wavelength? No, of course not. All
you need is a reasonably accurate specification of β a (λ, p, T, . . . ) for
each relevant constituent. Such models are already available “off
the shelf.”

But while it is technically possible to learn to fly and land an

airplane by rote without ever actually mastering the physical prin-
ciples of flight, most of us would prefer to entrust our lives to a pilot
with a deeper understanding of his/her job. When it comes to at-
mospheric radiation, the consequences of ignorance are less grave,
but the argument is basically the same: the more you know, the less
likely you are to do something embarrassing, even when working
with someone else’s well-tested models.

Note, by the way, that even though I conceded the possibility of

blindly applying the equations cited above to the problem of com-
puting broadband fluxes (as required for general circulation mod-
els, weather forecast models and the like), this “brute force” ap-
proach is rarely attractive for routine calculations. Quite often, we
need to obtain fluxes of modest accuracy with the least possible expen-
diture of computational effort. That requires us to take advantage of
highly simplified parameterizations of radiative absorption and emis-
sion. If you’re going to stake your reputation on the results of such
methods (as many atmospheric scientists do without even realizing
it!), it’s certainly in your interest to know where they come from and
how far they can be trusted.

Atmospheric remote sensing methods typically rely on calcu-

lations of quasi-monochromatic radiant intensity. This is admit-
tedly a far simpler computational problem than that of broadband
fluxes. Nevertheless, it is the detailed absorption behavior of vari-
ous atmospheric molecules that largely determine optimal channel
wavelengths, spectral widths, and other instrument characteristics.
Those who work with remote sensing data and/or instrumentation,
as many of us can expect to do at some point in our careers, should
have at least a basic understanding of these issues.
 Basis for Molecular Absorption/Emission 231

9.1 Basis for Molecular Absorption/Emission

In Chapter 2, I pointed out that EM radiation has both wavelike and
particle-like properties. Recall that there are times when radiation
must be viewed as waves, times when it must be viewed as a shower
of quantized particles having energy E = hν, and, finally, times
when it doesn’t matter which view you take.
The absorption of radiation by gases turns out to be one of those
cases in which the quantized (particle) nature of radiation comes
to the forefront. Simply stated, interactions between radiation and
individual gas molecules — whether absorption or emission — are
possible only for photons having energies satisfying certain criteria.
Those criteria are largely determined by the arcane, and sometimes
counterintuitive, laws of quantum mechanics. But don’t despair:
the consequences of those laws aren’t difficult to grasp at the level
targeted by this book.
When a photon is absorbed by a system, the energy originally
carried by that photon must contribute to a corresponding increase
in the internal energy of the system. Likewise, when a photon is
emitted, the system must give up an equivalent amount of its inter-
nal energy. There are many different ways in which internal energy
may increase or decrease. Examples include:

• Changes in the translational kinetic energy of molecules (i.e.,

temperature).

• Changes in the rotational kinetic energy of polyatomic

molecules.

• Changes in the vibrational energy of polyatomic molecules.

• Changes in the distribution of electric charge within a

molecule, possibly including the complete separation (or re-
unification) of two components previously bound by electro-
static forces.

Collisions between molecules tend to equalize the distribution

of the total internal energy in a gas among the various “storage”
mechanisms listed above. Imagine, for example, that you were able
to put a diatomic gas like oxygen into a unusual state in which
232 9. Absorption by Atmospheric Gases†

all of its molecules had translational kinetic energy but no rota-

tional or vibrational kinetic energy. Subsequent collisions between
the molecules would quickly set many of them to spinning and vi-
brating again. Before long, the total internal energy would be dis-
tributed between all available storage modes, reestablishing a con-
dition known as local thermodynamic equilibrium (LTE; see also Sec-
tion 6.2.3).
LTE can be taken as a given for most problems in the lower and
middle atmosphere, where atmospheric density is comparatively
high and collisions are therefore quite frequent. This assures us
that knowledge of the physical temperature of the medium is suf-
ficient to accurately predict the distribution of the total internal en-
ergy among all possible modes. It also ensures that any radiative
energy absorbed or emitted by a medium will quickly give rise to a
commensurate change in the physical temperature of that medium.
Note, however, that the immediate consequence of the absorption
or emission of any particular photon is usually a change in the in-
ternal energy of a single molecule. For example, the emission by
a gas of a single photon might entail a reduction in the rotational
energy of one molecule of the gas. For another photon, the imme-
diate consequence of an absorption event might be a simultaneous
increase in the vibrational and rotational energy of the molecule.
For yet another, the response might be an increase in the electro-
static potential energy of one of the electrons in the molecule. Yet all
of these energy changes eventually get redistributed, via collisions,
between all of the molecules in the vicinity, so that there is always a
predictable distribution of the total internal energy in a gas among
all available storage modes.
Now here is the kicker: most modes of energy storage at the
molecular level and smaller are quantized. That is to say, a given
molecule cannot have just any vibrational energy, but rather only one
of a discrete set of energy levels E0 , E1 , . . . , E∞ permitted by the laws of
quantum mechanics applied to that particular molecule. The same
principle applies to other modes of energy storage, such as that as-
sociated with molecular rotation and electron excitation. Only the
translational kinetic energy of molecules and other unbound parti-
cles is unquantized.
 Absorption/Emission Lines 233

E2
∆E12
E1
a) ∆E02

Energy
∆E01
E0

Abs. Cross-section

b)

ν
ν12 ν01 ν02

Fig. 9.1: Relationship between energy level transitions and the absorp-
tion/emission line spectrum for a hypothetical molecule with only three allowed
energy levels. Panel (a) depicts the allowed transitions; panel (b) shows the corre-
sponding positions of the lines in the spectrum, where νij = ∆Eij /h.

9.2 Absorption/Emission Lines

The consequences of the quantization of energy states are far-
reaching. Basically, it implies that a photon can be absorbed or
emitted by a molecule only if the interaction leads to a transition from
the molecule’s original state to one of its other allowed states. Thus, if
a molecule is in energy state E0 , then it can absorb a photon whose
energy ∆E = En − E0 but not one whose energy is some other value.
Furthermore, a molecule in energy state E1 can only emit a photon
with energy ∆E = E1 − E0 , since there is no other allowed transition
that results in a decrease in the molecule’s energy. A molecule in its
base state E0 cannot emit anything. However, the condition of LTE
ensures that, for any given temperature above absolute zero, a pre-
dictable percentage of the molecules in a gas may be found in any
given energy state. This ensures that a predictable set of allowed
transitions (for both absorption and emission) is always available
within any macroscopic sample of the gas in question.
Consider a hypothetical molecule with only three allowed en-
ergy states E0 , E1 , and E2 (Fig. 9.1a). There are then three transi-
tions available for the absorption of an incoming photon: E0 → E1 ,
E0 → E2 , and E1 → E2 . Each of these transitions corresponds to
a specific photon energy ∆E and therefore, according to (2.41), to
234 9. Absorption by Atmospheric Gases†

a specific wavelength λ = c/ν = hc/∆E at which absorption may

occur. For all other wavelengths, the molecule is nonabsorbing.
If we plot the absorption cross-section as a function of wave-
length for our hypothetical molecule, we get an absorption line spec-
trum similar to Fig. 9.1b. The positions of the lines are determined by
energy changes associated with the allowed transitions. The relative
strengths of the lines are determined by a) the fraction of molecules
that are in the particular initial state required for the transition and
b) the intrinsic likelihood that a photon having the right energy and
encountering a molecule in the required energy state will actually
produce the relevant transition.1
It is possible for more than one transition to contribute to a single
absorption line, if they correspond to the same change of energy. For
example, if the energy levels for the above hypothetical molecule
were equally spaced, then both the E0 → E1 and E1 → E2 transitions
would correspond to the same photon wavelength. The absorption
spectrum would then consist of only two lines rather than three,
with one occurring at half the wavelength of the other one. The
latter, in this example, would be called degenerate, in that it is really
two lines collapsed into one. Multiple degeneracies occur in some
real gases; the associated absorption lines are often quite strong in
comparison to their nondegenerate neighbors.

Problem 9.1: Consider a hypothetical molecule with N allowed en-

ergy levels Ei .
(a) Determine the total number of possible transitions that could
lead to the absorption of a photon.
(b) If the energy levels happen to be equally spaced, then how
many distinct absorption lines result from the above transitions?

1 Although the existence of absorption lines is justified here by invoking the

quantized nature of radiation and of energy transitions in molecules, it turns out

that there is a close analogy as well to the absorption of energy by a classical
damped harmonic oscillator. If the frequency of the external wave coincides with
the resonant frequency of the oscillator, absorption occurs, otherwise it does not.
Some features of line absorption can be explained in classical terms using this anal-
ogy (see for example Thomas and Stamnes, 1999, Section 3.3). For this reason, you
will sometimes hear the term resonant absorption used to describe line spectra of
the type discussed above, as contrasted with nonresonant (or continuum) absorp-
tion spectra that do not exhibit line structure.
 Absorption/Emission Lines 235

The energies carried by the photons of interest to atmospheric

scientists cover a wide range — five orders of magnitude from
∼ 10−23 J for microwaves to ∼ 10−18 J for the far UV band. Be-
cause the quantization of energy states in atoms and molecules is
quite different for the different modes of energy storage, different
wavelength bands tend to be associated with different types of tran-
sitions.
For example, the allowed rotational energy states of most
molecules are rather closely spaced; hence transitions between ro-
tational states tend to involve low-energy photons, such as those in
the far IR and microwave bands. On the other extreme, the tightly
bound electrons in the inner shell of an atom require large amounts
of energy in order to be “kicked up” to a higher level of excitation;
consequently, this kind of transition is important primarily for the
absorption and emission of X-rays. Intermediate energies are asso-
ciated with transitions in the outer electronic shells, and with vibra-
tional transitions. The following table gives the dominant transition
type within the wavelength bands of interest to us:

Wavelengths Band Dominant transition

>20 µm Far IR, microwave Rotation
1 µm– 20 µm Near IR, thermal IR Vibration
<1 µm Visible, UV Electronic

As we shall see later, low-energy transitions, such as those asso-

ciated with rotation, often occur simultaneously with high-energy
transitions due to vibration and/or electronic excitation. The total
energy change is the sum of the individual changes. The effect of
such combinations is to add fine-scale structure to the absorption
line spectrum that would result from “pure” vibrational or elec-
tronic transitions alone.
This book assumes that the reader has had little if any prior in-
struction in quantum physics. The following discussion of quan-
tized energy transitions is therefore intentionally simplistic and
largely descriptive, though mathematical relationships are derived
for a few simple cases.2 My purpose here is to give you a qualitative
understanding of why atmospheric absorption spectra look the way
2 More detailed treatments may be found in S94 (Section 3.2), TS02 (Section 4.5),

and GY89 (Chapter 3).

236 9. Absorption by Atmospheric Gases†

they do, and why certain trace constituents, like CO2 , are far more
important for radiation than are other far more prevalent species,
like nitrogen. We will begin by examining the physical basis for the
absorption line spectra associated with rotational, vibrational, and
electronic transitions separately and in combination. This will be
followed by a discussion of absorption line shape and width and,
finally, by a survey of principal atmospheric absorbers.

9.2.1 Rotational Transitions

Moments of Inertia of Molecules
Let’s first begin with a quick review of some familiar relationships
from freshman physics. All physical objects, including molecules,
have mass. According to Newton’s law the mass m is a measure of
the object’s resistance to linear acceleration, i.e.,

F = ma (9.1)

where F is the applied force and a is the resulting acceleration. Also,

the mass figures prominently in the object’s translational kinetic en-
ergy product, which is given by

1 2
Ekt = mv , (9.2)
2
where v is the speed. In the case of molecules in a gas, their average
translational kinetic energy is proportional to the absolute tempera-
ture of the gas. Also, the linear momentum is given by

p = mv . (9.3)

The above relationships all apply to linear motion by a mass;

completely analogous relationships also exist for rotational motion.
In particular, all physical objects that are not point masses have mo-
ments of inertia. Analogous to the role of mass in linear acceleration,
the moment of inertia I is a measure of the object’s resistance to ro-
tational acceleration when subjected to a torque T:

dω
T=I . (9.4)
dt
 Absorption/Emission Lines 237

Similarly, the rotational kinetic energy is given by

1 2
Ekr = Iω . (9.5)
2
Finally, the angular momentum is given by

L = Iω . (9.6)

The moment of inertia I depends not only on the mass m but also
on how that mass is distributed in space about the object’s center
of gravity. Specifically, it is the sum of the products of the masses
δmi of each volume element in the object with the squares of their
respective distance ri from the axis of rotation:

I= ∑ ri2 δmi . (9.7)

i

Thus, for two objects of equal mass, the more compact object
will have the smaller moment of inertia. A hula hoop has about the
same mass as a softball, but a hula hoop spinning about its axis at
the rate of one revolution per second stores much more energy and
angular momentum then does a softball spinning at the same rate.
In one sense, the analogy between rotational and translational
measures of energy and momentum is very close — indeed, the
respective equations for linear and rotational acceleration, kinetic
energy, and momentum all have similar algebraic forms. In two
respects, however, the picture is slightly more complicated for rota-
tional motion than it is for translational motion.
First of all, translational motion is not quantized — a molecule
is allowed to have any speed (how could this not be the case, since
there is no absolute frame of reference for translational motion?).
Rotational motion at the molecular level, on the other hand, is quan-
tized according to the laws of quantum mechanics, and this dis-
cretization of energy/angular momentum states is responsible for
the existence absorption/emission lines in connection with rota-
tional transitions.
Secondly, any object has only one mass m, but it has three prin-
cipal moments of inertia I1 , I2 , I3 . Each of these corresponds to one
of three perpendicular axes of rotation whose overall orientation is
determined by the particular mass distribution (i.e., shape) of the
238 9. Absorption by Atmospheric Gases†

object. Although the actual axis of rotation of an object is not con-

strained to coincide with one of the principal axes, the overall an-
gular momentum and/or rotational kinetic energy can nevertheless
be analyzed in terms of the projection of the angular velocity vector
on each of the three principal axes.
For objects having simple symmetry, it is often easy to guess the
alignment of the principal axes and to judge the relative size of the
corresponding moments of inertia. A rectangular brick, for exam-
ple, has principal axes passing at right angles through each face and
through the center of mass. The corresponding moments of iner-
tia will be different for each axis: smallest for rotation about the
longest axis of the brick (because the mass is most compactly dis-
tributed about that axis) and largest for rotation about the axis pass-
ing through the shortest dimension of the brick (because more of the
mass is found at a greater distance from that axis).
For many common shapes, one or more of the moments of iner-
tia have the same value. For example, a pencil has one very small
moment of inertia corresponding to rotation about its lengthwise
axis and two much larger, identical moments of inertia correspond-
ing to rotation about any pair of axes at right angles to the first.
A flat circular disk has one large moment of inertia about its radial
axis and two smaller and equal moments of inertia about any pair of
perpendicular axes that pass edgewise through the disk. A uniform
sphere or a cube has three identical moments of inertia; moreover,
it is irrelevant in these cases how you define the principal axes, be-
cause rotation about any axis is equivalent.
We are now ready to see how the above concepts apply to
molecules. Let us start by considering an isolated atom. Virtually
the entire mass of any atom is confined to its nucleus which, on
a molecular scale, has a vanishingly small radius. For all practi-
cal purposes relevant to atmospheric radiation, a single atom has
effectively zero moment of inertia for any axis of rotation; i.e.,
I1 = I2 = I3 ≈ 0. It follows that isolated atoms exhibit no rotational
transitions of interest to us.
A diatomic molecule, on the other hand, consists of two or more
bound atoms, each with finite mass, separated by a finite distance.
Moreover, the distance separating the two atoms is quite large com-
pared with the radius of the individual nuclei. One of the axes of
 Absorption/Emission Lines 239

Permanent Electirc
Molecule Structure Dipole Moment?

linear No
Oxygen O O
(magnetic
dipole)

Nitrogen N N linear No

Carbon Monoxide O C linear Yes

Carbon Dioxide O C O linear No

O N N
Nitrous Oxide linear Yes

O
Water asymmetric top Yes
H H

O
Ozone asymmetric top Yes
O O
H
H
Methane C spherical top Yes
H
H

Fig. 9.2: Molecular structure of several key atmospheric constituents. Also indi-
cated is whether the molecule has a permanent electric dipole moment (note that
oxygen has a permanent magnetic dipole moment).

rotation can be taken to pass through the two nuclei. For this axis,
the moment of inertia I1 = 0. The remaining two axes are perpen-
dicular to each other and to the first; for them the moments of inertia
are nonzero and I2 = I3 . These properties also apply more generally
to a linear polyatomic molecule — i.e., any molecule with all of its
constituent atoms lying in a straight line.
Finally, we come to nonlinear polyatomic molecules, which have
three nonzero moments of inertia. Depending on the symmetry of
240 9. Absorption by Atmospheric Gases†

the molecule, there are three possible cases: 1) all moments of inertia
equal, corresponding to a spherical symmetric top molecule, 2) two
equal and one different in the case of a symmetric top molecule, and
3) all three different — the so-called asymmetric top.
The significance of the above distinctions lies in the number
of mechanically dissimilar modes of rotation available to a given
molecule. There can be no rotational energy, and therefore no ab-
sorption or emission of radiation, in connection with a principal
axis for which I = 0. Furthermore, axes of rotation for which the
moments of inertia are equal are energetically indistinguishable and
therefore give rise to identical absorption spectra. Thus, linear and
spherical symmetric top molecules have the fewest distinct modes
of rotation, and therefore the simplest absorption spectra due to ro-
tational transitions, while asymmetric top molecules have the rich-
est set of possible transitions, and therefore the most complex rota-
tion spectra.
The following table summarizes the different types of rotational
symmetry, in order of increasing complexity, and lists the most im-
portant atmospheric molecules in each category (see also Fig. 9.2).

Description Moments of Inertia Examples

Monoatomic I1 = I2 = I3 = 0 Ar
Linear I1 = 0; I2 = I3 > 0 N2 ,O2 ,CO2 ,N2 O
Spherical symmetric top I1 = I2 = I3 > 0 CH4
Symmetric top I1 = 0; I2 = I3 > 0 NH3 , CH3 Cl, CF3 Cl
Asymmetric top I1 = I2 = I3 H2 O, O3

Quantization of Angular Momentum

A diatomic molecule may be regarded as a two-body rigid rotator

consisting of two masses m1 and m2 separated by distance r = r1 +
r2 , where ri is the distance of each mass mi from the axis of rotation.
For any freely rotating body, the axis of rotation concides with the
center of mass which, in this example, satisfies m1 r1 = m2 r2 . It can be
shown from (9.7) that the moment of inertia I2 = I3 = I (recall that
I1 = 0 for a linear molecule) is given by

I = m r2 , (9.8)
 Absorption/Emission Lines 241

where we introduce the so-called reduced mass

m1 m2
m ≡ . (9.9)
m1 + m2
.
If the molecule is rotating with angular velocity ω, then its ki-
netic energy of rotation is

1 2 L2
E= Iω = , (9.10)
2 2I
where
L = Iω (9.11)
is the angular momentum. Quantum mechanics tells us (via
Schrödinger’s equation), that the angular momentum of our rigid
molecule is restricted to discrete values given by

h
L= J ( J + 1) , (9.12)
2π
where J = 0, 1, 2, . . . is the rotational quantum number. Combining
the above with (9.10) yields a set of discrete energy levels

1 2 J ( J + 1) h2
EJ = Iω = . (9.13)
2 8π 2 I

Problem 9.2: Equations (9.11) and (9.12) imply that the angular ve-
locity ω of an object can only take on discrete (quantized) values.
Explain why this quantization is not observable for macroscopic ob-
jects like frisbees and hula hoops.

Rotational Absorption Spectrum

If we now consider transitions between adjacent rotational states
J and J + 1 (it turns out that none other are allowed by quantum
mechanics), we find that the associated energy change is

h2
∆E = E J +1 − E J = [( J + 1)( J + 2) − J ( J + 1)] , (9.14)
8π 2 I
242 9. Absorption by Atmospheric Gases†

or
h2
∆E = ( J + 1) . (9.15)
4π 2 I
The corresponding photon frequency ν is then
h
ν = ∆E/h = ( J + 1) , (9.16)
4π 2 I
or
ν = 2B( J + 1) , (9.17)
where B is the rotational constant defined as
h
B= . (9.18)
8π 2 I
From (9.17), we see that the rotational absorption line spec-
trum of our simple diatomic molecule consists of a series of equally
spaced lines separated by ∆ν = 2B.

Problem 9.3: Oxygen (O2 ; molecular mass 32.0 kg/kmole) is a di-

atomic molecule whose lowest-frequency rotational absorption band
is found at a frequency of 60 GHz. From the relationships given in
this section, find the distance r separating the two oxygen atoms in
the molecule.

We have just discussed the quantization of rotational energy,

and the associated line spectrum, for the simplest case: that of a di-
atomic molecule with only one nonzero moment of inertia I and ro-
tational quantum number J. Analogous relationships exist for non-
linear molecules possessing up to three different moments of inertia
In and associated rotational quantum numbers Jn . Each quantum
number Jn has its own distinct set of energy levels, the spacing of
which is determined by the value of In . In this case, however, the ab-
sorption or emission of a photon often entails a simultaneous change
in two or more of the rotational quantum numbers. Because of the
likelihood of simultaneous transitions, the rotation line spectrum of
a nonlinear molecule is considerably more complex and irregular
than that of a linear molecule.
There are additional complications that I will mention but not
dwell on. For example, both vibrational and rotational motions can
 Absorption/Emission Lines 243

perturb the moments of inertia In from their static values — e.g., by

slightly stretching the mean interatomic distance r. These interac-
tions can give rise to slight shifts in the positions of lines relative to
the “pure” case. Since the perturbing influence is itself quantized,
the usual result is the splitting of what would otherwise be a single
line into family of closely spaced lines.

Dipole Moments
The above discussion hopefully gives you some insight into the ba-
sic nature of rotational line spectra, but there is one final issue that
we have not yet considered. In order for a molecule to interact with
an electromagnetic wave via rotational transitions, it must possess
either a magnetic or electric dipole moment. That is to say, an exter-
nally applied magnetic or electric field, respectively, must have the
capacity to exert a torque on the molecule.
For example, the needle in a compass has a magnetic dipole
moment. The earth’s magnetic field exerts a torque on the needle
that causes it to rotate until it is aligned with the field. Likewise, a
molecule with no net electric charge but with an asymmetric inter-
nal distribution of positive and negative charge can have an electric
dipole moment that allows it respond to an external electric field by
changing its orientation.
A diatomic molecule that is also homonuclear — i.e., consisting
of two identical atoms (e.g, N2 or O2 ) — has no permanent electric
dipole moment because of the symmetry of the distribution of pos-
itive and negative charge within the molecule. The same applies to
a symmetric linear triatomic molecule at rest, such as CO2 . A het-
eronuclear diatomic molecule, such as carbon monoxide (CO), will
generally have a permanent electric dipole. Also, with the excep-
tion of CO2 , all triatomic (and larger) molecules of interest to us for
atmospheric radiation, such as H2 O, N2 O, O3 , and CH4 , have per-
manent electric dipole moments due to various asymmetries in their
structure.
Here is the “bottom line” for rotational absorption by atmo-
spheric constituents:

• As noted previously, monoatomic constituents such as argon

(Ar) and other noble gases have effectively zero moment of
244 9. Absorption by Atmospheric Gases†

inertia I and therefore no rotational transitions.

• Molecular nitrogen (N2 ), the most abundant atmospheric con-

stituent, has neither electric nor magnetic dipole moment and
therefore has no rotational absorption spectrum.

• Oxygen (O2 ) also has no electric dipole moment but, unlike

most other diatomic gases, it does have a permanent magnetic
dipole moment. This property is what permits it to have rota-
tional absorption bands at 60 and 118 GHz.

• Carbon dioxide (CO2 ) has no permanent electric or magnetic

dipole moment and is therefore radiatively inactive with re-
spect to pure rotational transitions. However, bending vibra-
tional motions can break the linear symmetry of the molecule
and introduce an oscillating dipole moment whose presence
permits combined vibration-rotation transitions at shorter
wavelengths (see Section 9.2.2).

• All other major molecules found in the atmosphere exhibit

permanent electric dipole moments and therefore also major
rotational absorption bands.

9.2.2 Vibrational Transitions

The covalent bonds between two atoms in a molecule arise from
a balance of attractive and repulsive electrostatic forces. The for-
mer dominates when the two atoms are relatively widely separated;
the latter, due to mutual repulsion of the positively charged nuclei,
takes over when the atoms get pushed too close together. The ar-
rangement of atoms in a molecule at rest corresponds to the posi-
tions for which all attractive and repulsive forces exactly cancel.

Diatomic Molecules
The molecular bond is thus not rigid but behaves like a spring. In
particular, for sufficiently small displacements, the force between
two atoms in a diatomic molecule is given by

F = − k (r − r ) , (9.19)
 Absorption/Emission Lines 245

where k is analogous to a spring constant, and F is thus the restor-

ing force when the separation r is smaller or greater than the equi-
librium separation r . Because the atoms have mass, the molecule
would, in the classical limit, behave like a simple harmonic oscilla-
tor with resonant frequency

1 k
ν = , (9.20)
2π m
where m is the reduced mass defined by (9.9).
In the quantum mechanical limit, however, the actual vibra-
tional frequency ν of the the oscillator is quantized according to
 
1
ν = v+ ν , (9.21)
2
where v (letter ’V’, not Greek ’nu’) is the vibrational quantum number
and, like the rotational quantum number J, can only take on non-
negative integer values.
The energy associated with each frequency ν is the same as it is
for a photon with the same frequency:
 
1
Ev = hν = v + hν . (9.22)
2
A vibrational transition ∆v = ± N therefore entails a change of en-
ergy
|∆Ev | = Nhν , (9.23)
and we see, somewhat to our amazement, that the photon frequency
associated with the transition is just an integer multiple of the clas-
sical resonant frequency of the harmonic oscillator.

Vibration/Rotation Spectra
There are two important points to note:
• Vibrational transitions tend to be associated with consider-
ably larger energies than rotational transitions. Therefore vi-
brational transitions give rise to absorption/emission lines at
much shorter wavelengths — e.g., in the thermal and near IR
bands — than those due to pure rotational transitions, which
are generally associated with the far IR and microwave bands.
246 9. Absorption by Atmospheric Gases†

J'

{
5

4
v' 3
2
1
0
Energy

J"

{
5

4
v"
3
2
1
0

P Q R Branch
ν
ν0

Fig. 9.3: Vibrational-rotational transitions for ∆v = ±1, ∆J = [−1, 0, +1], showing

the relative positions of the transitions in the spectrum. The P-branch corresponds
to transitions involving ∆J = −1, while the R-branch corresponds to ∆J = +1.
The Q-branch, when present, represents a superposition of all possible transitions
involving ∆J = 0 and occurs close to frequency ν0 = ∆E/h, where ∆E is the energy
associated with pure vibrational transitions.

• Vibrational and rotational transitions may, and usually do,

occur simultaneously. It follows from the previous point
that the energy (and photon wavelength) of combined vibra-
tion/rotation transition is slightly greater or less than that of
a pure vibrational transition, depending on whether the rota-
tional quantum number J increased or decreased during the
transition. The effect of rotational transitions is therefore to
split up vibrational absorption lines into a series of rather
closely spaced separate lines.

The second of the above points is illustrated by Fig. 9.3, which

depicts the positions of lines associated with various combinations
of ∆v = ±1 and ∆J = [−1, 0, 1]. The nominal frequency of a pure
vibrational transition (∆J = 0) is indicated in this figure as ν0 and is
approximately the same regardless of the value of J in force when
 Absorption/Emission Lines 247

the transition occurs. The associated absorption/emission lines,

when present (see below), belong to what spectroscopists refer to
as the Q branch of the vibration/rotation spectrum.3
For transitions involving ∆J = −1 and ∆J = +1, respectively,
the absorption lines associated with ∆v = 1 are spread out slightly
below or above ν0 . These lines do not overlap because of the uneven
spacing of the rotational energy levels. The lines associated with
∆J = −1 belong to the so-called P branch; the ∆J = +1 transitions
are known as the R branch.
Note that for certain molecular configurations, quantum me-
chanics requires a vibrational transition of ∆v = ±1 to be accompa-
nied by a nonzero rotational transition (typically ∆J = ±1). In such
cases, the central Q branch is absent and the absorption spectrum
consists of the P and R branches alone.

Polyatomic Molecules

When an molecule consists of more than two atoms, the variety of

possible vibrational motions increases substantially and, in general,
consist of combinations of bending and stretching of interatomic
bonds. It turns out that, for a given molecular structure, any phys-
ically admissible vibrational motion can be expressed as the super-
position of a finite set of normal modes, each of which can be consid-
ered independently of the others, and each of which is associated
with its own set of vibrational energy levels. The situation is analo-
gous to the existence of three distinct rotational quantum numbers
for aspherical top molecules, but the number of possible vibrational
modes is a function of the structure of the molecule.
For the diatomic molecule considered earlier, the sole possible
vibrational mode consisted of stretching and compression along the
axis of the molecule. Its vibrational energy levels can therefore be
expressed in terms of only one vibrational quantum number v.
For triatomic molecules, there are three energetically distinct
modes: (1) symmetric stretching, (2) asymmetric stretching, and (3)

3 In actuality, the rate of rotation of the molecule has a weak influence on vi-

brational energy levels, so that the Q branch is very finely split into a number of
nearly overlapping lines (not pictured in Fig. 9.3).
248 9. Absorption by Atmospheric Gases†

Diatomic (N2, O2, CO)

Linear triatomic (CO2, N2O)

Symmetric stretch Bending Asymmetric stretch

ν1 ν2 ν3

Nonlinear Triatomic (H2O, O3)

Symmetric stretch Bending Asymmetric stretch

ν1 ν2 ν3

Fig. 9.4: Illustration of normal modes of vibration for simple molecules.

bending.4 The nature of these vibrational motions is illustrated in

Fig. 9.4. For this case, three vibrational quantum numbers v1 , v2 , v3
are required to describe all possible vibrational energy levels. Dur-
ing the absorption or emission of a photon, simultaneous transitions
may occur in one, two, or three of these quantum numbers. It fol-
lows that the vibrational absorption spectra of triatomic molecules
is considerably more complex than those of diatomic molecules.
For molecules consisting of four or more atoms — e.g., methane
(CH4 ), chlorofluorocarbons, etc., the number of normal modes mul-
tiplies very rapidly, and the associated absorption spectrum be-
comes very rich indeed. In general, a molecule consisting of n > 1

4 Fora linear molecule such as CO2 , there are actually two normal modes as-
sociated with bending - one lying in each of two orthogonal planes. However,
these two modes are energetically equivalent and therefore needn’t be considered
separately.
 Absorption/Emission Lines 249

atoms has N = 3n − 6 normal modes if it is nonlinear, or N = 3n − 5

if it is linear.

Problem 9.4: Chlorofluorocarbons (CFCs) are synthetic com-

pounds that have strong absorption bands in the IR band. One of
the simplest CFCs has the formula CFCl3 . Given that it is a nonlin-
ear molecule, how many vibrational normal modes does it have?

Problem 9.5: Assume that a molecule has N distinct vibrational

quantum numbers vi . How many possible simultaneous (nonzero)
energy transitions of the form ∆vi = {−1, 0, 1} exist?

9.2.3 Electronic Transitions

Having discussed rotational and vibrational transitions, we now
briefly consider a third transition type, which involves the energy
levels of the electrons orbiting an atomic nucleus. Just as energy has
to be added to a satellite in order to raise it to a higher orbit above
the earth’s surface, the energy associated with an electron increases
with its mean distance from the nucleus. As before, the allowed en-
ergy levels of electronic orbits are quantized. Also, the more tightly
bound an electron is to a nucleus, the larger the magnitude of the
energy steps ∆E between levels.
In general, a photon can be absorbed when its energy corre-
sponds to the excitation of an electron to a more energetic state; a
photon of an appropriate wavelength can be emitted when an ex-
cited electron drops back to a lower state.
For the purposes of this book, it is principally the outermost
electrons that are of interest, since transitions between their ground
state and the first excited state are associated with photon wave-
lengths in the near IR, visible and ultraviolet bands. Inner electrons,
which are more strongly bound by attractive forces to the nucleus,
tend to have energy transitions corresponding to very short wave-
lengths.
250 9. Absorption by Atmospheric Gases†

v
2

1st Excited
State 0
Energy

J
v 4
2
3

2
1
1
0
Ground
State
0

Electronic + Vibration + Rotation

Fig. 9.5: Schematic depiction of the superposition of electronic, vibrational and

rotational energy levels. The absorption spectrum of a molecule is determined by
all nonforbidden transitions between pairs of levels in the righthand column.

At ordinary atmospheric temperatures, collisions between

molecules are very rarely energetic enough to kick electrons into
excited states. Those few that do briefly find themselves in an ex-
cited state, for whatever reason, usually give up their energy again
in the course of subsequent collisions without emitting a photon.
There are therefore extremely few electrons found naturally in ex-
cited states and even fewer opportunities for the spontaneous emis-
sion of photons due to electronic transitions back to the ground
state. This fact is of course consistent with the sharp fall-off in ther-
mal emission at the short wavelength end of the Planck function.
When electron orbits in atmospheric molecules are found in an
excited state, it is usually because of the absorption of an incoming
photon with the right energy, usually one of solar origin. As for
rotational and vibrational transitions, we therefore expect to find
discrete absorption lines associated with each allowed transition to
higher electronic orbital states.
 Line Shapes 251

9.2.4 Combined Energy Transitions and Associated Spec-

tra
As noted in the previous subsections, each atmospheric molecule
has associated with it a number of discrete energy levels associated
with electronic, vibrational, and rotational quantum states. Elec-
tronic energy states are the most widely separated and therefore
correspond to photon absorption/emission at the shortest wave-
lengths; vibrational and rotational transitions, respectively, are as-
sociated with intermediate- and low-energy transitions.
The combination of all three modes of excitations leads to a set of
discrete energy levels that are depicted schematically in Fig. 9.5. The
total absorption line spectrum of the molecule is determined by all
allowed transitions between pairs of energy levels (recall that not all
transitions are allowed by quantum mechanics) and by whether the
molecule exhibits a sufficiently strong electric or magnetic dipole
moment (either permanent or otherwise) to be able to interact with
the radiation field.

9.3 Line Shapes

The above introduction of absorption line spectra might have
understandably left you with the impression that absorption by
molecules can occur only at the precise wavelengths determined by
the permitted energy transitions. If this were indeed the case, line
absorption would be completely unimportant. Why? Recall that,
for natural radiation, the energy (or flux) associated with exactly
one wavelength is zero. In order to account for a finite percentage of
the total radiation field, it is necessary to consider a finite interval of
wavelength. That being the case, if absorption lines had zero width,
then no matter how strong they were and no matter how many of
them were scattered through the spectrum, their practical impact
on atmospheric radiative transfer would be zilch. The observed
broadening of absorption lines is therefore of tremendous practical
importance for radiative transfer, as it permits a given atmospheric
constituent to absorb radiation not only at the nominal wavelength
associated with a transition but also at nearby wavelengths as well
(Fig. 9.6). Where absorption lines are closely spaced, as they are in
252 9. Absorption by Atmospheric Gases†

(a) without broadening

Absorption
Wavelength

(b) with broadening

Absorption

Wavelength

Fig. 9.6: Schematic depiction of the effect of broadening on a hypothetical line

spectrum.

much of the thermal IR band, line broadening can completely close

the gaps between neighboring lines so that the atmosphere becomes
effectively opaque over a continuous range of wavelength.
There are three distinct processes responsible for line broaden-
ing. Their relative importance depends on local environmental con-
ditions.

Natural broadening: The Heisenberg uncertainty principle, which

is one of the most fundamental and far-reaching early find-
ings of quantum physics, ensures that an absorption line must
have finite (though very small) width, even in the absence
of other factors. Throughout most of the atmosphere, natu-
ral broadening is entirely negligible compared with the other
two broadening mechanisms. We will not consider it further
in this book.

Doppler broadening: Because of the random translational motions

of individual molecules in any gas, absorption and emission
 Line Shapes 253

occurs at wavelengths that are Doppler-shifted relative to the

natural line position. This is the primary broadening mecha-
nism in the mesosphere and above.
Pressure broadening: Collisions between molecules randomly dis-
rupt natural transitions between energy states, so that emis-
sion and absorption occurs at wavelengths that deviate from
the natural line position. This is the primary broadening
mechanism in the troposphere and lower stratosphere.

9.3.1 Generic Description of Lines

Before going into the details of the line shapes associated with
the above broadening mechanisms, it is convenient to introduce a
generic framework for describing any individual absorption line.
Specifically, there are three pieces of information that one needs to
know about a line under a given set of environmental conditions:
Line position: Where does the line fall in the EM spectrum?
Line strength: How much total absorption is associated with the
line?
Line shape: How is the above absorption distributed about the cen-
ter of the line?
All of the above characteristics are captured succinctly in the follow-
ing formulation:
σν = S f (ν − ν0 ) , (9.24)
where σν is the absorption cross-section per molecule (or some other
equivalent description, such as cross-section per unit mass) at fre-
quency ν. ν0 is the center frequency of the absorption line, S is the
line strength, and f (ν − ν0 ) is the line shape function. In order that
the strength and the shape be completely decoupled in the above
description, the shape function is normalized to unit area:

∞
f (ν − ν0 ) dν = 1 , (9.25)
0
so that the line strength is given by

∞
∞
σν dν = S f (ν − ν0 ) dν = S . (9.26)
0 0
254 9. Absorption by Atmospheric Gases†

Doppler

Lorentz

-4.0 -3.0 -2.0 -1.0 0.0 1.0 2.0 3.0 4.0

(ν−ν0)/α

Fig. 9.7: Comparison of the Doppler and Lorentz line shapes for equal widths and
strengths.

Although the shape of f (ν − ν0 ) depends on the broadening

mechanism, one feature common to all shape functions is that the
maximum occurs at the line center ν = ν0 , and the function falls
sharply and monotonically with increasing |ν − ν0 |. Moreover, the
line shape is usually symmetric about ν0 (except in the microwave
band); therefore it is common to succinctly characterize the overall
width of the line via a parameter α1/2 known as the half width at
half-maximum. That is to say, α1/2 represents the value of |ν − ν0 |
for which the absorption cross-section falls to half of its maximum
value at the line center; i.e. f (α1/2 ) = f (0)/2.
Let us now consider the two major line broadening mechanisms
and their implications for the shape function f (ν − ν0 ).

9.3.2 Doppler Broadening

In any gas, individual molecules are in constant motion, follow-
ing random ballistic trajectories interrupted only by collisions with
other molecules. The average kinetic energy of all of the molecules
is proportional to the temperature. From the perspective of a sta-
tionary observer, each molecule has a random velocity component
vs toward or away from the observer. The statistical probability
of any particular velocity component vs along the line-of-sight s is
 Line Shapes 255

given by the Maxwell-Boltzmann distribution

1
√ e−(vs /v0 ) ,
2
p(vs ) = (9.27)
v0 π
√
where v0 = 2k B T/m represents the standard deviation (or root-
mean-squared value) of vs , m is the mass per molecule (= M/N0 ,
where M is the molar mass and N0 = 6.02 × 1023 ), T is the temper-
ature, and k B is the Boltzmann constant.
The motion of each molecule along a particular line-of-sight in-
troduces a Doppler shift into the frequency of the photons it emits
and absorbs, as measured from the perspective of a stationary ob-
server. In Problem 2.2, you were asked to show that, in the case of
an EM wave traveling at the speed of light c, the Doppler-shifted
frequency is given by
ν = ν(1 − v/c) , (9.28)
where ν is the frequency as measured by an observer who is station-
ary relative to the source, and ν is the frequency measured by an
observer moving with a velocity component v away from the source.
Because of the Doppler effect, EM frequencies that would ap-
pear not to coincide with the nominal position ν0 of an absorption
line can be absorbed, nevertheless, by any molecule having the right
relative velocity. Conversely, radiation whose frequency is equal to
ν0 can only be absorbed by those relatively few molecules whose rel-
ative velocity is close to zero. The net effect, therefore, is a decrease
in the likelihood of absorption by a molecule at ν0 and an increase
at nearby frequencies.
Combining (9.27) with (9.28) yields the line shape for Doppler
broadening:
 
1 (ν − ν0 )2
f D (ν − ν0 ) = √ exp − , (9.29)
αD π α2D
where 
2k B T
α D = ν0 . (9.30)
mc2
We can find the line halfwidth at half max α1/2 by setting
f D (α1/2 ) 1
= (9.31)
f D (0) 2
256 9. Absorption by Atmospheric Gases†

and solving for α1/2 . We find that

√
α1/2 = α D ln 2 . (9.32)

The interpretation of (9.29) is straightforward. This Maxwell-

Boltzmann distribution of line-of-sight velocities (9.27) is Gaussian;
therefore the Doppler-broadened shape of an initially narrow line is
also Gaussian. The mean speed of the molecules, and therefore the
line width, increases with temperature and decreases with molecu-
lar mass. The shape of the Doppler profile is depicted in Fig. 9.7.

9.3.3 Pressure Broadening

In Section 9.2, the occurrence and position of absorption lines was

justified under the implicit assumption that molecules were free to
make transitions between their various energy states without exter-
nal interference. For gases in the tenuous upper atmosphere, this is
a reasonable assumption. In the denser portions of the atmosphere
(i.e., in the stratosphere and troposphere), collisions occur between
molecules with very high frequency. The effect of a collision is to
“shock” the molecule at a time when it might just be in the process
of emitting or absorbing a photon. Not surprisingly, this adds a sig-
nificant new level of complexity to the problem of predicting which
wavelengths might be most readily absorbed or emitted during a
particular transition. In fact, no exact theory has yet been developed
to describe the so-called pressure broadening of absorption/emission
lines that results.
It is beyond the scope of this book to go through the heuristic ar-
guments that have been used to derive an approximate expression
for the shape of pressure broadened lines. Suffice it to say that one
common model invokes collision-induced random phase shifts im-
posed on an otherwise “pure” sinusoidal oscillation to explain the
resulting “smearing” of emitted (and absorbed) frequencies about
the nominal frequency ν0 . The interested student is referred to Sec-
tion 3.3.1 of S94 and Section 3.3 of GY89 for details.
The bottom line, for our purposes, is that pressure broadening is
usually described adequately, though by no means perfectly, by the
 Line Shapes 257

Lorentz line shape

α L /π
f (ν − ν0 ) = , (9.33)
(ν − ν0 )2 + α2L

where α L is the Lorentz halfwidth at half max and is roughly pro-

portional to the number of collisions per unit time. Thus, to first
order
α L ∝ pT −1/2 , (9.34)
where p is the pressure and T is the temperature. As a practical
matter, α L for any particular line is usually modeled as
  n
p T0
α L = α0 , (9.35)
p0 T

where α0 is the laboratory-measured line width at reference pres-

sure p0 and temperature T0 , and n is an empirically determined
exponent. When measured in units of wavenumber, typical val-
ues of α0 lie in the range 0.01–0.1 cm−1 , for T0 = 273 K and
p0 = 1000 mb. The corresponding range in units of frequency is
3 × 108 to 3 × 109 Hz.
The Lorentz line shape has two notable deficiencies. One is that
the far wings of actual absorption lines have been found to be poorly
represented by this line shape. It might strike you as unimportant
whether a line shape model does well in the far wings, where ab-
sorption due to a single line is very small anyway. However, recall
that the absorption coefficient at any particular frequency ν is due
to the combined contributions of all lines, near and far. And there
are of course many more distant lines than nearby lines! In partic-
ular, at any frequency located in a spectral window where there are
no nearby lines, all of the absorption is due to the accumulated far
wings of countless thousands of distant lines. In general, it has been
found that the Lorentz model underestimates this contribution.
Another limitation of the Lorentz model is that it is only valid
when α L  ν0 ; that is, when the line width is very small compared
to the center frequency. When this condition doesn’t hold, then the
Lorentz model is invalid, because it fails to account for the fact that
line wings cannot extend down to zero frequency. This is primar-
ily a problem in the microwave band, where ν0 is quite small. In
258 9. Absorption by Atmospheric Gases†

a) O2 at 100 mb pressure b) O2 at 1000 mb pressure

0.8 3
0.7
2.5
Absorption Coeff.

Absorption Coeff.
0.6
2
0.5
0.4 1.5
0.3
1
0.2
0.5
0.1
0 0
50 60 70 50 60 70
Freq. (GHz) Freq. (GHz)

Fig. 9.8: Absorption coefficient of oxygen in the microwave band near 60 GHz, (a)
at 100 mb pressure, revealing the individual lines making up the absorption band;
(b) at 1000 mb, for which pressure broadening obliterates the line structure.

this case, the preferred line shape model is the van Vleck-Weisskopf
function
   
1 ν 2 αL αL
f VW (ν − ν0 ) = + .
π ν0 (ν − ν0 )2 + α2L (ν + ν0 )2 + α2L
(9.36)
Note that the van Vleck-Weisskopf line shape is asymmetric about
ν0 and, more importantly, is constrained to go obediently to zero as
ν → 0.
An example of the how pressure broadening affects absorption
line spectra is shown in Fig. 9.8. At high altitudes, individual lines
are narrow enough to be clearly separated. At lower altitudes, pres-
sure broadening often completely obliterates the individual lines,
leading to the appearance of one single broad absorption feature.
Note that the careful positioning of a satellite sounding channel be-
tween lines can lead to a sharpening of the absorption/emission
weighting function W (z), because the mass extinction coefficient at
that wavelength then sharply increases toward the surface, rather
than being constant, as we assumed in the previous chapter.

9.3.4 Comparing Doppler and Pressure Broadening

The shape of the Lorentz profile is compared with the Doppler pro-
file in Fig. 9.7 for equal values of the width parameters α L and α D .
The most important difference is that the Lorentz shape puts much
 Line Shapes 259

O2 CO2
λ=2.5 mm λ=15 µm λ=4.3 µm
100

Doppler Broadening
80

60
Z (km)

Pr
40 es
su
re
B ro
a de
20 nin
g

0
104 105 106 107 108 109 1010
α1/2 (Hz)

Fig. 9.9: Approximate linewidth as a function of altitude for three prominent ab-
sorption bands.

more absorption out in the far wings of the line.

Both Doppler and pressure broadening occur at all levels of the
atmosphere. However, if the degree of broadening by one mech-
anism is substantially greater than that due to the other, then the
latter may be neglected. The relative importance of each may there-
fore be assessed by way of the ratio of α D to α L , using (9.30) and
(9.35):

  
αD p0 2k B Tν0 −13 −1 ν
≈ √ ∼ [5 × 10 mb Hz ] 0 , (9.37)
αL α0 c T0 p m p

where we have assumed that n ≈ 1/2 and, in the far righthand

terms, used “typical” values for m, T, α0 . Because pressure varies
by several orders of magnitude through the atmosphere, whereas
all other variables (except ν0 ) typically vary by less than a factor of
two, we conclude that p is the principal factor determining whether
Doppler broadening or pressure broadening prevails in any given
case. If we choose ν0 = 2 × 1013 Hz , corresponding to a wavelength
260 9. Absorption by Atmospheric Gases†

of 15 µm, we find that α D ≈ α L at p ∼ 10 mb. At this altitude,

the two broadening processes are of roughly equal importance. For
lines in the microwave band, we have ν0 ∼ 1011 Hz, so that the pres-
sure level at which parity occurs is in the vicinity of 0.1 mb or less.
Fig. 9.9 depicts the transition from predominantly pressure broad-
ening at low altitudes to primarily Doppler broadening at higher
altitudes for the two cases discussed above.
When α D is of the same order of magnitude as α L , both Doppler
and pressure broadening must be considered simultaneously. In
this case, it is necessary to use the hybrid Voigt line shape, which
accounts for both mechanisms. Near the center of the line, it be-
haves like a Doppler profile; in the wings, it has the characteristics
of the Lorentz profile. The interested student should consult p. 112
of GY89 for further details.

9.4 Continuum Absorption

The most important absorption features in the IR and microwave
bands are due to bands of discrete absorption lines, as discussed
above. However, outside the major resonant absorption bands, one
generally finds some level of atmospheric absorption that does not
exhibit line-like structure. This is known as continuum (or nonreso-
nant) absorption, because it tends to vary smoothly with frequency.
There are at least three causes of continuum absorption, two of
which, photoionization and photodissociation are well-understood and
affect primarily the very short wavelength end of the solar spec-
trum. The third, affecting spectral windows throughout the infrared
and microwave bands, is significant both for remote sensing and for
thermal radiative transfer in the atmosphere but is, unfortunately,
less well understood.

9.4.1 Photoionization
Photoionization occurs when a photon has enough energy not only
to excite an electron in an atom to a higher level but to actually
strip it completely from the atom, creating a positively charged ion
and a free electron. We saw earlier that ordinary electron excita-
tion was constrained to discrete energy levels, leading to absorp-
 Continuum Absorption 261

tion/emission lines in the visible and UV bands. A fixed amount

of energy is also required to ionize an atom. However, any photon
exceeding that ionization energy can be absorbed in the course of an
ionization event; the excess energy appears in the form of kinetic
energy of the ion and/or the free electron. Recall that translational
kinetic energy is not quantized; hence there is no constraint on the
exact energy level of the photon.
The ionization of atoms requires very energetic photons. It is
observed primarily in connection with X-ray and gamma radiation
from extraterrestrial sources. It is therefore a phenomenon that is
of relatively little interest from an atmospheric radiation point of
view. However, it is responsible for the existence of the ionosphere,
a highly conductive region of the atmosphere that has important
implications for the propagation of radio waves.

9.4.2 Photodissociation

A second cause of continuum absorption is photodissociation. Just

as photoionization is electronic excitation carried to the extreme
that the electron completely separates from the atom, photodissoci-
ation is molecular vibration carried to the extreme that the molecule
breaks into two pieces. Thus, any photon whose energy E ex-
ceeds the chemical binding energy Ebond between two components
of a molecule can induce photodissociation, with the excess energy
∆E = E − Ebond appearing as unquantized kinetic (thermal) en-
ergy.5 At the molecular level, of course, a gain in kinetic energy
is synonymous with an increase in temperature.
The binding energies in the diatomic molecules of gases like O2 ,
N2 , etc., tend to be large. Therefore, only very short wavelengths
can dissociate them (see Section 3.4.1).

5 Again, to return to the refrigerator magnet analogy: imagine that only 0.01 J of
energy is necessary to separate a 20 g magnet from the door, but that, by whatever
means, you supply 1 J of mechanical energy to the magnet (e.g., by snapping a wet
dish towel at it). The magnet will acquire 0.99 J of kinetic energy, corresponding to
a speed of almost 10 m/sec.
262 9. Absorption by Atmospheric Gases†

9.4.3 Continuum Absorption by Water Vapor

In addition to the above well-understood modes of continuum ab-
sorption, which are primarily active at UV wavelengths, there is also
significant continuum absorption by water vapor between the major
absorption bands in the infrared and microwave bands. The physi-
cal mechanism behind the water vapor continuum remains a matter
of some controversy. There are basically two competing theories:

Far wings of lines - As noted earlier in this chapter, the Lorentz

line shape doesn’t do a good job of describing the intensity
of absorption in the far wings of collision-broadened absorp-
tion lines. Therefore, one possible explanation of continuum
absorption is that it is simply that portion of the accumulated
contribution of the far wings of countless distant lines that is
not accounted for by the Lorentz model.

H2 O clusters - Water molecules exert a significant attractive force

on one another (it is for this reason that water condenses read-
ily at relatively low temperature and pressure combinations).
Therefore, molecules that encounter one another at low rel-
ative speeds tend to stick to each other, forming temporary
clusters of two or more molecules that are subsequently bro-
ken up again by other collisions. Such dimers (two molecules),
trimers (three) and polymers (many molecules) would naturally
be expected to exhibit far more complex vibrational and ro-
tational transitions than an isolated molecule. Furthermore,
the range of structural configurations that these loose clusters
might exhibit at various points in their formation and disin-
tegration is almost infinitely variable. There is therefore no
reason to expect absorption to be limited to a finite set of dis-
crete photon energies. Instead, we expect absorption to be
smoothly distributed over a broad range of wavelengths, as
is in fact observed with the water vapor continuum.

It is possible that the ultimate explanation of water vapor contin-

uum will be found to combine elements of both theories. Regardless
of the explanation, there are two points that are worth noting: (1) the
strength of the continuum tends to increase steadily with frequency
through the microwave and far IR bands but is weak again in the
 Continuum Absorption 263

a) H20 b) CO2
>20 µm 15 µm 4.3 µm
6.3 µm

Wavenumber (cm-1) Wavenumber (cm-1)

c) O3 d) CO
14.3 µm
9.6 µm

Wavenumber (cm-1) Wavenumber (cm-1)

Fig. 9.10: Spectra of absorption cross-sections for several molecules encountered in

the atmosphere. Absorption bands of particular significance for thermal radiative
transfer in the atmosphere are labeled with the center wavelength. (a) Water vapor.
(b) Carbon dioxide. (c) Ozone. (d) Carbon monoxide.

thermal and near IR bands, and (2) the volume extinction coefficient
β a is not proportional to the water vapor density ρv (as would be
expected if the mass absorption coefficient ka were approximately
constant) but rather something closer to the square of that density.
Therefore, continuum absorption is generally significant mainly in
the lower troposphere, where water vapor density is the highest.
264 9. Absorption by Atmospheric Gases†

9.5 Applications to Meteorology, Climatology,

and Remote Sensing
The main focus in this chapter has been on the following questions:
(1) how does absorption by gases in the atmosphere occur, (2) what
physical properties of molecules determine where in the EM spec-
trum absorption lines or bands will occur, and (3) by what processes
are absorption lines broadened about their nominal wavelengths,
and what is the shape of the resulting broadened line? I will now
put some of the above ideas into a practical context with a survey of
those constituents responsible for most absorption and emission in
the cloud-free atmosphere.

9.5.1 Atmospheric Absorbers in the IR Band

From the perspective of most mainstream atmospheric scientists,
absorption by gases in the atmosphere is mainly a feature (or defect,
depending on your point of view) of the IR bands. While not dis-
counting the importance of absorption in other bands to a handful
of overachievers (e.g., atmospheric chemists, stratospheric dynam-
icists, and microwave remote sensing specialists), it is safe to say
that those individuals are not relying on this book alone for their
education in atmospheric radiation.
Within the IR band (including near-, thermal-, and far-IR), the
most important absorbers are carbon dioxide (CO2 ), water vapor
(H2 O), ozone (O3 ), methane (CH4 ), and nitrous oxide (N2 O). The
major absorption features associated with each of these constituents
was depicted in Fig. 7.6, which you may wish to refer back to occa-
sionally while reading this section. It is perhaps no surprise that
all of the above absorbers are (at least) triatomic, owing to the
wealth of both rotational and vibrational transitions available in
such molecules.
A broad-spectrum view of absorption by several representa-
tive molecules is depicted in Fig. 9.10. Included in this figure is a
molecule of lesser importance, carbon monoxide (CO), which (for-
tunately for our health) is not usually present in the atmosphere
in concentrations high enough to contribute much to atmospheric
opacity. The reason for including CO here is to illustrate how much
 Applications 265

simpler the absorption spectrum is for a diatomic molecule, which

(as discussed earlier) has just one vibration quantum number v and
one rotational quantum number J.
The most obvious feature of Fig. 9.10d is the evenly spaced series
of major absorption bands of CO. Although it is less apparent in this
figure, each band is actually a cluster of numerous closely spaced
lines, most of which overlap strongly due to pressure broadening.
The first band on the far left (near zero wavenumber) is a pure
rotation band with ∆v = 0, each line corresponding to a transi-
tion from a particular initial J to the next higher value. Subsequent
bands, moving from left to right, are associated with vibrational
transitions ∆v = 1, 2, . . . ; their central positions are predicted by
(9.23). The occurrence of simultaneous J transitions (both positive
and negative) again splits each band into a cluster of closely spaced
lines. In fact, it is not too much of a stretch to consider each major
vibration-rotation band a near-clone of the one before it, since the
fine-scale structure imposed by rotational transitions is similar in
each case.
Turning our attention to the three triatomic molecules in
Fig. 9.10(a–c), we see that the possibility of simultaneous transitions
between multiple rotational and vibrational quantum numbers (up
to three each) adds a fair amount of complexity to the absorption
spectra. In particular, the “windows” between adjacent vibration
bands are not nearly as broad and clear as they were for CO.
Note that CO2 is the only molecule in Fig. 9.10 exhibiting no
pure rotation band on the far left edge of the plot. As pointed out
earlier, this is because CO2 lacks a permanent electric dipole mo-
ment and can interact with radiation only via combined rotation-
vibration transitions.
Let’s now take a closer look at absorption by selected atmo-
spheric molecules.

Water Vapor

Globally speaking, water vapor (H2 O) is the single most important

atmospheric absorber in the IR band. Fig. 9.11 gives a moderate-
resolution depiction of the zenith transmittance due to water vapor
alone under the conditions of temperature and humidity found in
266 9. Absorption by Atmospheric Gases†

Zenith Transmittance due to H2O

Wavelength (µm)
2
1
Transmittance

Fig. 9.11: Zenith transmittance of the atmosphere in a portion of the thermal IR

band due to water vapor in a U.S. Standard atmosphere.

the U.S. Standard Atmosphere. The enormous complexity, and ap-

parent randomness, of the line structure is readily apparent. This
randomness is characteristic of a nonlinear molecule, owing to the
wide range of rotational transitions that are available to it.
The most transparent region of the H2 O spectrum in the thermal
IR band lies between about 8 and 12 µm. At longer wavelengths
(smaller wavenumbers), the spectrum is increasingly dominated by
the numerous lines due to pure rotational transitions. In fact, be-
yond about 25 µm, the atmosphere is effectively opaque and doesn’t
begin to open up again until well into the microwave band (not
shown).
Not shown in Fig. 9.11 but visible in Fig. 9.10 are vibration-
rotation bands centered around 6.3 µm and 2.7 µm, the latter falling
solidly in the near-IR band. The 6.3 µm band is due to the v2 (bend-
ing mode) fundamental (∆v = 1), while both the v1 (symmetric
stretch) and v3 (asymmetric stretch) fundamental modes contribute
to absorption in the vicinity of 2.7 µm. Higher-order vibrational
transitions (∆v > 1) contribute weak absorption bands at shorter
wavelengths throughout the solar part of the spectrum. Also, as
discussed earlier, not all of H2 O absorption is due to lines; there is
also a significant continuum contribution throughout the IR band,
though its overall intensity varies somewhat with wavelength.
An additional factor contributing to the complexity of the H2 O
absorption band is the existence in the atmosphere of multiple iso-
 Applications 267

Zenith Transmittance CO2

20 19 18 17 16 15 14 13 µm

Transmission

Wavenumber (cm-1)

Fig. 9.12: Zenith transmittance of the atmosphere due to CO2 in the vicinity of
15 µm.

topes of the constituent elements. In addition to the common hydro-

gen isotope 1 H, a small fraction (about 0.03%) of water molecules
include at least one 2 H, or deuterium (D) atom. Also, oxygen ex-
ists both as the 16 O isotope and also the less common 18 O isotope,
found in about 0.2% of water vapor molecules. Naturally, chang-
ing the nuclear masses of one or more of the constituent atoms in
a molecule alters both the rotational and vibrational spectra of the
molecule, adding numerous new lines to those already contributed
by the most common form. However, because isotopic variations of
the water molecule comprise such a small fraction of the total, most
of these added lines are relatively weak.

An important property of water vapor that is not shared with

other radiatively important atmospheric gases is of course its enor-
mous variability in both time and space. There are significant
ranges of the IR spectrum for which the atmosphere may be virtu-
ally opaque under conditions of high humidity but relatively trans-
parent (or at least translucent) in a dry arctic atmosphere.
268

CO Transmission in air over 1 m path

2
a) p = 1000 mb

b) p = 1000 mb

c) p = 100 mb

Fig. 9.13: (a) High resolution depiction of the transmission spectrum of a one-
meter path through air with typical CO2 concentration, at 1000 mb pressure. (b)
Same as (a), but “zoomed in” on the central Q-branch. (c) Same as (b), but at
reduced atmospheric pressure of 100 mb, revealing that what previously appeared
as a single absorption line is actually a cluster of closely spaced but distinct lines.
 Applications 269

Carbon Dioxide
No other atmospheric constituent is better known to the general
public as a “greenhouse gas” than carbon dioxide (CO2 ). In actual-
ity, water vapor has a larger overall impact on the radiative energy
budget of the atmosphere. But CO2 is of particular concern from a
public policy point of view owing to its steadily increasing abun-
dance in the atmosphere due to fossil fuel use (see Fig. 7.5).
CO2 possesses two very strong rotation-vibration bands in the
IR band, one centered on 4.3 µm (v3 fundamental), the other at
15 µm (v2 fundamental). The 4.3 µm band is actually the stronger
of the two, but because it’s located out on the edge of both the so-
lar and longwave bands, it isn’t terribly important for broadband
radiative fluxes in either band.
The 15 µm band is very important for longwave radiative trans-
fer in the atmosphere, because it is positioned near the peak of
the Planck emission function for terrestrial temperatures and ren-
ders the atmosphere completely opaque between 14 and 16 µm
(Fig. 9.12). Moreover, the atmosphere is at least partly absorbing for
several micrometers more to either side of the above wavelengths.

Problem 9.6: The effect of the CO2 15 µm band on atmospheric

transmission to space of surface emission can be crudely approxi-
mated by assuming total opacity between 13.5 and 17 µm, and total
transparency outside these limits. Use Fig. 6.4 to estimate the frac-
tion of surface longwave emission that is reabsorbed by the CO2 in
the atmosphere (ignoring other absorbers). Assume a surface tem-
perature of 288 K.

Examining the center of the 15 µm band closely (Fig. 9.13), we

find that it looks much like the idealized vibration-rotation band
associated with a diatomic molecule — that is, it exhibits a strong
central Q branch associated with pure vibration transitions (∆v = 1;
∆J = 0), as well as the usual P branch (∆v = 1; ∆J = −1) and
R branch (∆v = 1; ∆J = +1). Unlike the random-appearing H2 O
rotation spectrum, the P and R branches for CO2 consist of very
regularly spaced lines, again as predicted for a diatomic molecule
(Fig. 9.3).
270 9. Absorption by Atmospheric Gases†

Why does CO2 , which is triatomic, yield a spectrum so unlike

that of another triatomic molecule, H2 O? The reason is that CO2 is
a linear molecule with effectively only one rotation mode (similar
to a diatomic molecule), as compared with three for H2 O, which is
an asymmetric top. This seemingly minor difference eliminates the
whole universe of lines that would arise from transitions involving
J2 or J3 .
As noted also for water vapor, isotopic variations of the con-
stituent atoms multiply the total number of lines found in the CO2
absorption spectrum. Here we have not only the two most common
oxygen isotopes 16 O and 18 O, but also two major carbon isotopes
12 C and 13 C, the latter comprising about 1% of total carbon in the

atmosphere.
We conclude our discussion of CO2 by noting that the central
lines of the 15 µm band are really quite strong. Despite the di-
luteness of CO2 in air (about 370 parts per million by volume), the
strongest part of the central Q branch (Fig. 9.13b) absorbs all but 5%
of radiation over just a 1-meter path in air at 1000 mb pressure! Also,
we see that, at reduced pressure (Fig. 9.13c), the Q branch resolves
into a series of very closely spaced but separate lines. Each of these
is associated with a (∆v = 1; ∆J = 0) transition, but depending on
the rate of rotation of the molecule (i.e., value of J), there are slight
differences in the energy associated with a pure ∆v = 1 transition.

Ozone

Ozone (O3 ), which (except in urban pollution) is confined primar-

ily to the stratosphere, is another nonlinear triatomic molecule, like
water vapor. Therefore, it possesses many of the same general ab-
sorption features as water vapor, such as a relatively strong rotation
spectrum consisting of numerous quasi-randomly spaced lines, as
well as three fundamental vibrational bands v1 , v2 , and v3 . These
bands, respectively, are found at 9.066, 14.27, and 9.597 µm. In
the atmosphere, the 14.3 µm band is essentially masked by CO2 ’s
15 µm band — that is, the presence or absence of ozone doesn’t
change much in this band, because absorption by CO2 is already
overwhelmingly strong.
The moderately strong v1 and very strong v3 bands are close
 Applications 271

enough to each other that they are commonly viewed as a single

entity known as the 9.6 µm band of ozone. Note that this band sits
squarely in the middle of the 8–12 µm “window” of H2 O as well
as falling not far from the peak of the Planck function for terrestrial
emission. For both reasons, ozone is an important player in the at-
mospheric radiation budget, especially in the stratosphere.
Ozone also has a strong band at 4.7 µm, but this is at the edge of
the Planck functions for both solar and terrestrial emission (as was
also the case for the CO2 4.3 µm band) and therefore is of very lim-
ited relevance for broadband radiative fluxes. Additional vibration-
rotation absorption bands can be found in the near-IR regions of the
spectrum down to about 2.5 µm (see Fig. 9.10c), but these are weak
and therefore of only minor importance.
Also, as is well-known to the many specialists and laypeo-
ple alike concerned about recent observed declines in stratospheric
ozone concentrations, ozone exhibits very strong absorption in the
UV band below 0.28 µm due to electronic rather than vibrational
excitations. As was discussed in Section 3.4.1, this absorption is
exceedingly important for life on earth, because of the biologically
damaging effects of UV-B radiation. A near-steady-state balance be-
tween absorption of solar UV-B and re-emission of longwave IR in
the 9.6 µm band is largely responsible for the temperature structure
of the stratosphere and lower mesosphere.

Methane

Methane (CH4 ) is a spherical top molecule with five atoms and

therefore 3(5) − 6 = 9 fundamental modes of vibration. However,
because of the symmetry of the molecule (four hydrogen atoms all
bound to one central carbon atom), five of the modes are equiva-
lent, leaving only four distinct modes v1 , v2 , v3 , and v4 . The most
important of these are the v3 and v4 fundamentals at approximately
3.3 µm and 7.6 µm, respectively. In addition, there are a few weak
overtone (∆vi > 1) and combination bands scattered through the
near-IR.
Although methane is present in the atmosphere in relatively low
concentrations, its 7.6 µm absorption band is strong enough, and
is placed in an otherwise relatively transparent part of the atmo-
272 9. Absorption by Atmospheric Gases†

spheric spectrum (especially in a dry atmosphere) to have a mea-

surable impact on longwave fluxes. In addition, it is now known
that methane concentrations are increasing as a direct or indirect
consequence of human activities.6

Nitrous Oxide
Nitrous oxide (N2 O) is another minor atmospheric constituent that
nevertheless has a measurable radiative impact, primarily owing to
its absorption band at 7.8 µm, which acts to broaden and strengthen
the absorption band already present at 7.6 µm due to methane. In
addition, there is a strong 4.5 µm band which, however, is less sig-
nificant for the same reasons as given for the CO2 4.3 µm band and
the ozone 4.7 µm band.

6 To give just one example, widespread clearing of tropical forests has led to

a marked increase in the number of methane-producing termite colonies, which

thrive in the resulting open spaces!
Powered by TCPDF ([Link])
 Line-by-line Calculations 281

niques are also applicable to atmospheric absorption of solar radi-

ation (especially in the near IR band), we will keep things simple
by limiting our attention initially to the emission and absorption of
radiation in the thermal IR band.

10.1 Line-by-line Calculations

For monochromatic radiation, we can easily generalize (8.27)–(8.30)
to describe the upward and downward intensity at any level z in a
plane-parallel, nonscattering atmosphere with a black lower bound-
ary at temperature Ts and no extraterrestrial sources:
Z •
Iñ# (z) = Bñ [ T (z0 )]Wñ (z0 , z) dz0 , (10.1)
z

Z z
Iñ" (z) = Bñ ( Ts )tñ (0, z) + Bñ [ T (z0 )]Wñ (z0 , z) dz0 , (10.2)
0

where 
tñ (z1 , z2 )
tñ (z1 , z2 ) = exp , (10.3)
µ
Z z2
tñ (z1 , z2 ) = b a ñ (z) dz , (10.4)
z1

and
0 ∂tñ (z0 , z)
Wñ (z , z) = . (10.5)
∂z0

Note that we have introduced the subscript ñ to make explicit the

dependence on wavenumber,2 and we have introduced absolute
value operators to permit a more compact set of expressions valid
for both upwelling and downwelling radiation.
In view of what we learned from the previous chapter about ab-
sorption spectra, it is evident that the absorption coefficient b a ñ (z)

2 Traditionally,
spectroscopists working in the thermal IR band are accustomed
to using wavenumber rather than wavelength to describe where they are in the
spectrum.
282 10. Broadband Fluxes and Heating Rates†

can be expanded as
N
b a ñ (z) = Â ri (z)ka,i (z)
i =1
N
"
Mi
# (10.6)
= Â ri ( z ) kcont,i (ñ; z) + Â Sij (z) f ij (ñ ñij ; z) ,
i =1 j =1

where ri is the local density of each of N atmospheric constituents,

Mi is the number of significant absorption lines associated with the
ith constituent, Sij , f ij and ñij are the respective line strengths, shapes
and positions, and kcont,i represents the continuum component of
absorption for that constituent, if applicable. The z-dependence in
many of the above parameters arises from the influence of local tem-
perature, pressure and constituent partial pressure on line strength
and width.
In other words, in order to compute the monochromatic inten-
sity at altitude z and zenith angle µ = | cos q |, you have to eval-
uate the sum of the contributions of all relevant absorption lines
to the absorption coefficient b a at your chosen wavenumber ñ, and
you have to repeat this for all z0 . Then you can numerically evalu-
ate (10.1) and/or (10.2) to get the corresponding radiant intensity at
your chosen level z. The above procedure is the essence of so-called
line-by-line (LBL) calculations of radiative transfer — that is, there
is an explicit summation of the individual contributions of all lines
in the vicinity to the emission and absorption at each wavenum-
ber of interest. Note that “in the vicinity” means all lines whose
wings contribute nonnegligible absorption at the wavenumber in
question.3
A couple of comments are in order. If the intended applica-
tion for your radiance calculation is remote sensing, then a full LBL
treatment is reasonably manageable, since you’re often dealing with
only an extremely narrow range of wavenumber for each sensor
channel and therefore don’t have to repeat the calculation many
times. In fact, if the wavenumber of the calculation is far enough
3 ActualLBL computer codes in wide use include FASCODE, GENLN2, and
LBLRTM. All of these use various strategies to maximize computational efficiency
without sacrificing precision, but the essence of the calculation remains as outlined
above.
 Line-by-line Calculations 283

from any absorption lines, then you can probably get away with
using just a single wavenumber to represent the response of your
sensor channel to the atmospheric profile of temperature, humidity,
etc. If, on the other hand, your channel is positioned very close to a
strong absorption band (as is the case for satellite sounders), then
the monochromatic intensity Iñ will probably vary strongly even
over the narrow bandwidth of the sensor channel, and a numeri-
cal integration over wavenumber is required. Nevertheless, this is
still not an excessive computational burden in most instances.
Far more problematic is the computation of longwave fluxes and
heating rates. Recall that, in contrast to remote sensing applications,
which commonly require quasi-monochromatic radiances at only one or
two levels (e.g., the top and/or bottom of the atmosphere), the calcu-
lation of radiative heating profiles within the atmosphere requires
broadband fluxes at every level within the atmosphere.
Getting from a monochromatic intensity to a monochromatic flux
is straightforward — in fact, we already solved that problem for a
plane-parallel nonscattering atmosphere in Section 8.2.2. To refresh
your memory, the key result from that section is that the flux can
be computed in a manner analogous to the intensity simply by sub-
stituting the flux transmittance t F (z1 , z2 ) for the beam transmittance
t(z1 , z2 ) and pBñ for Bñ , everywhere where these terms appear in
equations (10.1)–(10.5). Furthermore, I pointed out that, to a pretty
good approximation,
t (z1 ,z2 )/µ
t F ( z1 , z2 ) ⇡ e , (10.7)

where the effective zenith angle µ is usually taken to be constant

and equal to 0.6.
With the above adaptations, we have
Z •
Fñ# (z) = pBñ [ T (z0 )]WF,ñ (z0 , z) dz0 , (10.8)
z
Z z
Fñ" (z) = pBñ ( Ts )t F,ñ (0, z) + pBñ [ T (z0 )]WF,ñ (z0 , z) dz0 , (10.9)
0
where

0 ∂t F,ñ (z0 , z) ∂tñ (z0 , z; µ)

WF,ñ (z , z) = ⇡ . (10.10)
∂z0 ∂z0
284 10. Broadband Fluxes and Heating Rates†

So much for the easy part. Getting radiative heating profiles

(for example) requires us to compute the broadband net flux at each
level z in the atmosphere. If we use the above LBL approach, we ef-
fectively have to repeat our monochromatic calculations for [a very
large number of discrete wavenumbers]⇥[a moderate number of
discrete altitudes]. The wavenumber spacing of our calculations
must be fine enough to accurately resolve the shapes of individ-
ual lines, a hurdle which is especially imposing at high altitudes,
where lines are extremely narrow. In short, the radiative heating pro-
file problem is potentially many millions of times more expensive than the
monochromatic radiance problem.
If we only need to do this once or twice, a LBL flux calculation
might still be a manageable proposition – simply let the program
run on an unused computer in a back room somewhere and forget
about it until word arrives that the program has completed. Unfor-
tunately, some of the most important applications of radiative trans-
fer calculations are found in the general circulation models (GCMs)
used in climate studies and in numerical weather prediction (NWP)
models. In GCMs, for example, radiative fluxes and heating profiles
must be computed at every point on the entire globe and updated on
regular time intervals as the model atmosphere evolves. It’s safe to
assume that, using the LBL approach and current computer tech-
nology, it would take a GCM far longer than a decade in computer
time to simulate a decade of real time! Clearly this is unacceptable
for graduate students trying to finish their dissertations on global
warming.
Hopefully I have persuaded you that we need some way to
obtain reasonably accurate broadband fluxes with greatly reduced
computational effort. The goal is to find techniques for greatly
streamlining the otherwise painstaking integration over complex
line spectra. In this book, I will give you a fairly cursory overview
of two basic strategies:

• band models, and

• the k-distribution method.

Band models have been in use in various forms for many

decades and remain of both practical and historical interest. They
 Line-by-line Calculations 285

come in two basic flavors: wide-band emission models and narrow-

band transmission models. The former represent the most radical
simplification of the radiative transfer equation and are widely used
in applications (e.g., climate models) in which computational ef-
ficiency is of paramount importance. Narrow-band transmission
models, which we will discuss in this book, are not as efficient but
are capable of yielding more accurate results while still being sub-
stantially faster to execute than LBL codes.
The k-distribution method is a relatively recent innovation and is
rapidly gaining favor in view of its ability to achieve fairly accurate
results with two or three orders of magnitude less computer time
than LBL methods.
For both narrow-band transmission models and the k-distribution
method, the longwave spectrum is subdivided into N intervals Dñi
that are each

1. large enough to encompass a significant number of absorption

lines associated with a particular atmospheric constituent, and
yet

2. small enough so that the Planck function Bñ ( T ) can be consid-

ered approximately constant and equal to Bi over the range.

Thus, the downwelling broadband longwave flux at level z can be

written
Z • N Z N
#
F (z) = Fñ# (z) dñ = Â Fñ# (z) dñ = Â Fi# (z) , (10.11)
0 i =1 Dñi i =1

where the flux contribution from each spectral interval can be ex-
panded as
Z Z Z •
∂tñ (z0 , z; µ) 0
Fi# (z) = Fñ# (z) dñ = pBñ [ T (z0 )]
dz .
Dñi Dñi z ∂z0
(10.12)
Making use of condition 2 above, and reordering the differentiation
and integration operators, we can write
Z •
∂Ti (z0 , z; µ) 0
Fi# (z) = pDñi Bi [ T (z0 )] dz , (10.13)
z ∂z0
286 10. Broadband Fluxes and Heating Rates†

where the band-averaged transmittance is

Z Z
0 1 0 1 tñ (z0 ,z)/µ
Ti (z , z; µ) = tñ (z , z; µ) dñ = e dñ .
Dñi Dñi Dñi Dñi
(10.14)
Analogous equations can of course be found for the upwelling flux
Fi" (z).
The efficient computation of broadband fluxes evidently boils
down to the problem of finding good approximations to Ti between
two levels for the particular spectral interval Dñi . In practice, this
problem is solved in two steps:

1. Develop a method for efficiently estimating Ti over an arbi-

trary homogeneous path — that is, one over which line shapes
and strengths can be considered constant, and

2. generalize the above method to inhomogeneous (e.g. vertical)

paths, over which linewidths can be expected to vary substan-
tially owing to pressure broadening, etc.

Band transmission models and the k-distribution method offer two

rather different frameworks for tackling both steps. We will con-
sider each in turn.

10.2 Band Transmission Models

Previously, when we examined the extinction of radiation (Chap-
ter 7), we were always concerned with the fate of monochromatic
radiation along a finite path between points s1 and s2 . The trans-
mittance t(s1 , s2 ) in this case is very nicely described by Beer’s Law
(7.7), which is both easy to remember and easy to apply. In particu-
lar, you’ll remember that one consequence of Beer’s Law is that the
transmittance over an extended path is equal to the product of the
transmittances over a series of shorter paths.
You can forget about Beer’s law in the case of spectrally aver-
aged transmission, except in the rare case that the medium happens
to be gray over the relevant interval Dñ – that is, when tñ is inde-
pendent of ñ. Qualitatively, this is because radiation with differ-
ent wavenumbers within the band are depleted at different rates.
 Band Transmission Models 287

At some wavenumbers for which the medium is transparent, there

is very little depletion even over long paths, and these wavenum-
bers ensure that the average band transmittance remains nonzero
even at a great distance from the source. At other wavenumbers
(e.g., at line centers), radiation is depleted rapidly, contributing to a
sharp decrease in band transmission over a short distance. But once
those wavenumbers are gone, they can’t be depleted further, and so
the transmission ceases to decrease as rapidly with further distance
along the beam.

Problem 10.1: Imagine that between wavenumbers ñ1 and ñ2 , the
atmosphere is perfectly transparent, whereas between ñ2 and ñ3 , the
absorption coefficient b a is constant and nonzero.
(a) Find an expression for the average transmittance T over a
pathlength s for the spectral band Dñ = ñ3 ñ1 .
(b) Sketch a graph of T (s) for the case that ñ2 is the midpoint
of the interval, being sure to show the asymptotic behavior of T for
large s.
(c) Assuming that the radiation incident at point s=0 is “white”,
compute the fraction that is absorbed by the time it reaches s = 1 km,
if b a = 3 km 1 .
(d) Assume that the radiation that survives the transit in (c) goes
on to traverse the path from s = 1 km to s = 2 km. What fraction
of the radiation that begins this second leg is absorbed by the time it
reaches the end?
(e) Explain why the fraction absorbed over the first kilometer is
substantially different than the fraction absorbed over the second
kilometer.

We will now consider the problem of band transmission in a

more quantitative fashion. To start with, it is instructive to consider
the band-averaged transmission and absorption properties in the
idealized case of a single isolated absorption line. We will use this
example to gain some basic insight and to introduce some defini-
tions that can later be applied to more complex cases.
288 10. Broadband Fluxes and Heating Rates†

10.2.1 Absorption by an Isolated Line

Recall that the band-averaged path transmission in spectral interval
Dñ is defined as Z
1
T = e tñ dñ (10.15)
Dñ Dñi
from which we can also define the band-averaged absorption

A=1 T . (10.16)

For the special case of a single isolated absorption line and a homo-
geneous path
tñ = S f (ñ)u , (10.17)
where S is the line strength, f (ñ) is the line shape function, and u is
the mass path: Z s2
u= r(s) ds , (10.18)
s1

where s is the distance along the line-of-sight.

Equivalent Width
Since we’re only considering a single line embedded somewhere in-
side the spectral interval Dñ, the band transmission is obviously a
function of how wide an interval we choose; the more clear spec-
trum we include on either side of the line, the higher T . It is there-
fore convenient to define a more fundamental property (for a given
mass path u) called the line’s equivalent width W:
Z
tñ
W⌘ 1 e dñ , (10.19)
Dñi

which doesn’t depend on Dñ as long as it’s large enough to encom-

pass all significant contributions from the line’s wings.
The physical interpretation of W is straightforward: it is the
spectral width of a hypothetical “square,” completely opaque line
that absorbs the same total amount of radiation in the band as the
actual (non-square, non-opaque) line. Therefore, the band-averaged
absorption can be conveniently written
W
A= . (10.20)
Dñ
 Band Transmission Models 289

Of fundamental interest is the manner in which W increases

with increasing mass path u under different assumptions. We will
look at a few important limiting cases.

Weak Line (Linear) Limit

For the special case that tñ ⌧ 1 for all ñ (even at the line center), we
can use the approximation exp( x ) ⇡ 1 x to evaluate (10.19):
Z Z Z
tñ
W= 1 e dñ ⇡ tñ dñ = S f (ñ)u dñ , (10.21)
Dñi Dñi Dñi

or, taking S and u out of the integral and using the normalization
property of f (ñ),
W = Su . (10.22)
In other words, in the so-called weak line limit, also known as the lin-
ear regime, total path absorption is proportional to the line strength
and the mass path u, irrespective of line shape.
The corresponding band absorption and transmission are then
given by
Su
A=1 T = . (10.23)
Dñ

Simple Case: Ideal Square Line

Let’s now consider an ideal square line with the property that
8 ✓ ◆ ✓ ◆
>
> 1 W 0 W 0
>
< W0 for ñ0 < ñ < ñ0 +
2 2
f (ñ) =
>
>
>
:
0 otherwise.
(10.24)
In this case, Z
tñ
W= 1 e dñ
Dñi
⇣ ⌘ (10.25)
Su/W0
= W0 1 e .
The corresponding band absorption and transmission are
W0 ⇣ Su/W0
⌘
A=1 T = 1 e . (10.26)
Dñ
290 10. Broadband Fluxes and Heating Rates†

The square line is of course an extreme and unrealistic case, as

real lines have wings that extend indefinitely outward from the line
center. Nevertheless, this example is useful as a simple illustration
of the effects of non-gray absorption within a band. More impor-
tantly, the square line model is as far from being gray as you can
get, so it defines an upper bound on the radiative effects of non-
grayness.
Specifically, we find that the band absorption for large mass path
has a limiting value of W0 /Dñ. Since our square line has no wings,
it can never absorb a larger fraction of the incident (white) radiation
than the ratio of the line width to the width of the spectral band.
For any realistic line shape — i.e., for one with tails extending to
infinity, we expect that the reduction in transmission with increas-
ing mass path will be
1. slower than that predicted by Beer’s Law for a gray medium,
but

2. faster and more complete than that expected from a square

line of the same width.

Lorentz Line
Let’s now look at the absorption behavior of a realistic line, as de-
scribed by the Lorentz shape
aL
f (ñ) = , (10.27)
p [(ñ ñ0 )2 + a2L ]
for which the equivalent width when integrated over the entire
spectrum is
Z •  ✓ ◆
Sua L
W= 1 exp dñ . (10.28)
• p [(ñ ñ0 )2 + a2L ]
We have extended the lower limit of integration in the above from
zero to • for the sake of being able to obtain an analytic solution.
To simplify the integral into a standard form that can be looked up
in a table, we further define the nondimensional mass path
Su
ũ ⌘ , (10.29)
2pa L
 Band Transmission Models 291

Transmittance over an extended path for an isolated Lorentz line

1.0
1
2
0.8
4

Transmittance 0.6 8
16
32
0.4 128
64

256
0.2 512
1024
0.0
-10 -8 -6 -4 -2 0 2 4 6 8 10
(ν−ν0) / α

Fig. 10.1: The evolution of transmittance with increasing mass path for an isolated
Lorentz line. For each curve, the line center optical path ũ is equal to the label
value divided by 2p.

which is nothing more than one-half of the optical path t at the line
center. The above integral can then be written
Z •  ✓ ◆
2ũa2L
W= 1 exp dñ , (10.30)
• (ñ ñ0 )2 + a2L
which has the solution

W = 2pa L L(ũ) , (10.31)

where the Ladenberg-Reiche function L(ũ) can be expressed in terms

of modified Bessel functions of the first kind of order 0 and 1:
ũ
L(ũ) = ũe [ I0 (ũ) + I1 (ũ)] . (10.32)

Admittedly, you won’t find the modified Bessel functions on your

pocket calculator, but apart from that minor inconvenience, they are
standard mathematical functions. Here we will confine our atten-
tion to the behavior of W (u) in the limit of very small and very
large mass path. But wait: we already know what happens in the
first case from (10.22), which is valid for any line shape — absorp-
tion in the weak line limit is simply proportional to u.
In the limit of large mass path (or, more precisely, large ũ), it can
be shown that p p
W ⇡ 2a L 2p ũ = 2 Sa L u . (10.33)
292 10. Broadband Fluxes and Heating Rates†

In other words, in the strong line limit, an isolated Lorentz line ab-
sorbs in proportion to the square root of the mass path.
The average transmission for a spectral band Dñ is, as usual,
W
T =1 , (10.34)
Dñ
where W is given by (10.31), provided that Dñ is large enough to en-
compass all significant absorption by the wings of the Lorentz line.
If this condition is not satisfied, then the limits of integration em-
ployed in (10.28) are not applicable, and no closed-form solution
exists. However, it is easy to convince yourself that the use of a
narrower band Dñ will have the following consequences:
• The equivalent width W will be reduced relative to (10.31),
because contributions from the far wings of the line are lost;

• The band-averaged transmittance T will be reduced, because

the clearest portions of the spectrum furthest away from the
line center are excluded.
The behavior of an isolated Lorentz line with varying ũ is de-
picted graphically in Fig. 10.1. For the spectral interval depicted,
the band transmittance is equal to the area under any given curve
(labeled with values proportional to ũ), expressed as a fraction of
the total area of the plot.
For small values of ũ, the transmission is close to unity every-
where, even at the line center. This represents the linear or weak
line regime, for which W = Su. For large values (ũ 10), the line
becomes saturated in the center, so that those wavelengths can no
longer contribute to further increases in A with increasing ũ. In-
stead, the width of the saturated zone increases, though at a slower
rate than ũ itself; this represents the strong line limit or square root
regime (as noted above, this description is strictly valid only when
the edges of the spectral window are far enough from the line cen-
ter).
Unfortunately, we rarely have the luxury of dealing with iso-
lated lines; rather, in the usual scenario we have dozens, if not hun-
dreds of lines in our spectral interval. If the lines don’t overlap (as
might be the case at high altitudes), then their contributions to the
band absorption A can be determined individually as for isolated
 Band Transmission Models 293

lines, above, and then summed to give the total absorption. More
commonly, we have to deal with lines that at least partly overlap,
which is the subject of the next subsection.

10.2.2 Defining a Band Model

When setting out to characterize the transmission T of a band con-
taining numerous lines, we don’t want to have to perform a com-
plete line-by-line calculation over all of the actual lines present —
that would defeat the purpose of a band model. Rather, we want to
be able to specify, in fairly generic terms, some key characteristics of
the lines present in the band and then derive a suitable analytic ap-
proximation for T (u). The characteristics that matter most include
the following:
• The average spacing d between the lines, defined as d = Dñ/N,
where N is the total number of lines in the spectral interval
Dñ.
• The manner in which the lines are distributed throughout the spec-
tral interval. There are two common choices: (1) random, and
(2) regular (or periodic). A random model (a.k.a. Goody
model) might be an appropriate description of a portion of
a water vapor band in which there is no apparent relationship
beween the location of one line and those of its fellow lines
(Fig. 9.11). A regular (periodic) model might be more appro-
priate in the P or R branch of a vibration/rotation band for a
linear molecule, such as CO2 (Fig. 9.13a). The difference mat-
ters because it determines how many of the lines present are
likely to overlap other lines, all other factors being equal. In a
regular distribution, the occurrence of overlap is minimized.
• The line widths a. Typically, band transmission models are de-
rived under the assumption that a is the same for all lines
present, as this simplifies the problem of finding closed-form
approximations.
• The statistical distribution of line strengths S. Do they all have the
same strength, or is there a wide range of different strengths
present? If the latter, what is the relative proportion of lines of
different strengths?
294 10. Broadband Fluxes and Heating Rates†

Line Strength Distributions

The last item in the previous list is dealt with by specifying a dis-
tribution function p(S) that characterizes the relative proportion of
lines with strength S. By definition, p(S) is nonnegative and satis-
fies the normalization condition
Z •
p(S) dS = 1 . (10.35)
0

The mean line strength is given by

Z •
S= Sp(S) dS . (10.36)
0

Common models for p(S) include the following:

1) d-distribution p(S) = d(S S) (equal strength)

2) Exponential distribution p(S) = (1/S) exp( S/S)
3) Godson distribution p(S) = S/(Smax S) for S < Smax
4) Malkmus distribution p(S) = (1/S) exp( S/S)

Although all of the above line strength distribution models have

their uses, we will consider only two: the d-distribution in conjunc-
tion with a regular line model, and the Malkmus distribution in
conjunction with the random line model. The first of these com-
binations is known as the Elsasser band model, after the scientist who
first studied its characteristics.

10.2.3 The Elsasser Band Model

A few absorption bands reveal a fairly regular repeating pattern of
lines of similar strength. As discussed in Chapter 9, this property is
what you would expect for the rotation spectrum (including P- and
R-branches of a vibration/rotation band) of a simple linear molecule
like CO2 . Therefore, in the Elsasser model of a simple absorption
band, we imagine the structure to be represented by a periodic array
of identical lines of strength S separated by spacing d:
n=+•
k(ñ) = Â S f (ñ nd) . (10.37)
n= •
 Band Transmission Models 295

Elsasser Band Model

y=0.1

y=0.2
y=0.5

−2δ −δ 0 δ 2δ

Fig. 10.2: Schematic depiction of the absorption coefficient k in the Elsasser (regu-
lar) band model, for three different values of the grayness parameter y ⌘ a L /d.

With substitution of the Lorentz line shape, we have

n=+•
S aL
k (ñ) = Â nd)2 + a2L ]
. (10.38)
n= • p [( ñ

Elsasser showed that this is mathematically equivalent to

S sinh(2py)
k(ñ) = , (10.39)
d cosh(2py) cos(2px )
where
aL ñ
, y⌘ x⌘ . (10.40)
d d
y can be regarded as a “grayness parameter”: if y is large, then adja-
cent lines strongly overlap and blur together, so that the line struc-
ture is increasingly obscured; for small y, the lines are well sepa-
rated and resemble isolated lines for small to moderate mass paths
(Fig. 10.2).
Because the pattern of lines is periodic, the band transmission
T over a large spectral range can be obtained by integrating the
monochromatic transmission over just a single interval d, which is
equivalent to one unit of the nondimensional wavenumber param-
eter x: Z 1/2
T = exp[ k( x )u] dx , (10.41)
1/2
or Z 1/2 
2p ũy sinh(2py)
T = exp dx , (10.42)
1/2 cosh(2py) cos(2px )
296 10. Broadband Fluxes and Heating Rates†

Band Transmission Models

1

0.8
Band Transmittance
0.0001

0.6 0.001 0.0001

0.01
0.4 0.001
Gray limit
0.1
0.01
Random- 1
0.2 Malkmus 0.1
Elsasser
0
0.1 1 10 100 1000
(S/δ)u

Fig. 10.3: Comparison of the Elsasser (solid curves) and random-Malkmus (dashed
curves) band models for varying values of the grayness parameter y (labels on
curves). For both models, curves approach the gray limit (Beer’s Law) when y
1.

where the nondimensional mass path ũ is as defined in (10.29).

Unfortunately, the above integral cannot be solved in closed
form. However, there are important limiting cases. The first is when
y is large (order 10 or greater), in which case the line width is much
greater than the spacing between lines. In this case, the medium is
effectively gray, and the band transmission reduces to
T = exp( 2pyũ) = exp( Su/d) . (10.43)
In other words, we’re back to Beer’s law, with mass absorption co-
efficient k = S/d.
The second limiting case of interest is for ũ 1. The band trans-
mission then has the asymptotic form
h p i
T ⇡ 1 erf py 2ũ , (10.44)

where erf( x ) is the so-called error function, defined as

Z x
2 t2
erf( x ) ⌘ p e dt . (10.45)
p 0
 Band Transmission Models 297

Random Line Model

Absorption Coefficient

1.0

0.8
Transmittance

0.6

0.4

0.2

0.0
∆ν

Fig. 10.4: Example of the distribution of line locations and strengths according to a
random/Malkmus model. Upper panel shows the absorption coefficient (arbitrary
units); lower panel depicts the spectrum of transmittance for four different mass
paths.

The behavior of Elsasser band transmission T (u) is depicted as

solid curves in Fig. 10.3 for various values of y. The main thing to
take away from this plot is that when the spacing between lines is
substantial (y ⌧ 1), the transmission decreases in a sub-exponential
fashion.

10.2.4 The Random/Malkmus Band Model

Line spectra for important nonlinear molecules like water vapor,
ozone, methane, etc. do not exhibit any of the regularity assumed by
the Elsasser model. Even the P- and R-branches of a linear molecule
like CO2 band are more complicated in reality than might be ex-
pected based on our simplified discussion in Chapter 9. Conse-
298 10. Broadband Fluxes and Heating Rates†

quently, the Elsasser band model is not widely used in actual band
transmission calculations.
Examination of Fig. 9.11 reveals that the line spectrum for water
vapor, one of the most important atmospheric absorbers, is highly
irregular in nature. Line positions would be better described as ran-
dom rather than regular. Also, the line strengths are quasi-random
— in any given spectral interval there is a mix of relatively strong
and weak lines.
A better match to reality in many cases is therefore offered by
the random model, in which line positions are assumed to be scat-
tered perfectly randomly throughout the chosen interval Dñ. Fur-
thermore, the line strengths are allowed to vary according to a pre-
scribed distribution p(S), as discussed earlier. One of the best gen-
eral purpose models for p(s) is the Malkmus distribution p(S) =
(1/S) exp( S/S).
As before, the average line spacing is given by d ⌘ Dñ/N, where
N is the number of lines in spectral interval Dñ. The definitions of y
and ũ are the same as for the Elsasser model, except that we use S
in place of S. It can then be shown (see L02 4.4.3) that
2 8s 93
pa L < 4Su =
T = exp 4 1+ 1 5
2d : pa L ; (10.46)
h py np oi
= exp 1 + 8ũ 1 .
2
The band transmission for this model is plotted as dashed curves
in Fig. 10.3, where they can be compared with the results for the
Elsasser model. For both models, the limiting behavior for large
y is equivalent to Beer’s law transmittance (‘gray limit’). But be-
cause of the unevenness of the spacing of the lines, the transmission
is greater for any given value of y than is the case for the Elsasser
model.

10.2.5 The HCG Approximation

We now need to generalize from a homogeneous to inhomogeneous
(vertical) path, allowing for the fact that pressure, and therefore
linewidth a L varies substantially over any significant vertical path
in the atmosphere.
 The k-Distribution Method 299

One appealing approach is to try to find an expression for an

“effective” mass path u and pressure p such that the homogeneous
path transmission with those values is equal to the correct inhomo-
geneous path transmission between two pressure levels p1 and p2 :

Tinhom (u) = Thom (u, p) . (10.47)

The van de Hulst/Curtis/Godson (HCG) approximation, which

was derived under the assumption of uniform temperature and con-
stituent mixing ratio over the path, offers such a solution: let
1
p= ( p1 + p2 ), u=u. (10.48)
2
That is, the effective mass path is the same as the actual mass path,
and the effective pressure is just the average of the pressures at the
endpoints of the path. It seems almost too simple, and yet this is the
accepted method for using band transmission models in a vertically
inhomogeneous atmosphere.

10.3 The k-Distribution Method

The narrow-band transmission models discussed in the previous
section are based on finding analytic approximations to the band-
averaged transmittance over a finite path, assuming a reasonable
statistical model for the distribution of line positions and strengths
in the spectral interval.
More recently, the so-called k-distribution method has emerged
as a very effective and flexible way of dealing with the integration
over a spectral interval. It is based on the idea that the integration
over frequency (or wavenumber) of a complex line spectrum can
be replaced with an equivalent integration over a much smoother
function. In the latter case, much larger discretization steps (and
therefore greatly reduced computational effort) can be used in the
numerical integration without sacrificing accuracy. An extremely
important property of the k-distribution method, and one which
distinguishes it from band transmission models, is that it is well-
suited to problems involving scattering as well as absorption.
This section gives a brief conceptual overview of the k-
distribution method without delving too much into the practical
300 10. Broadband Fluxes and Heating Rates†

details, which are best left to an advanced course in radiation (see

TS02 sections 10.4 and 10.5.4).

10.3.1 Homogeneous Path

Let’s start with an absorption spectrum k (ñ) like that depicted in
Fig. 10.5a. In this plot, we see approximately 30 Lorentz lines with
random positions and strengths, and subject to a modest degree of
pressure broadening. If we undertake a “brute force” line-by-line
calculation of the average transmittance of the band over a finite
mass path u, we have to evaluate the following integral:
Z ñ2
1
T (u) = exp[ k (ñ)u] dñ . (10.49)
ñ2 ñ1 ñ1

To numerically evaluate such an integral, we typically approximate

it as a sum of the following form
N
T (u) ⇡ Â ai exp[ k (ñi )u] , (10.50)
i =1

where N is the number of wavenumbers ñi at which k (ñ) is evalu-

ated within the interval, and the coefficients ai are weights which
depend on the specific quadrature method used (e.g., the trape-
zoidal rule, Simpson’s rule, etc.). In the simplest possible method
(which we will assume for the sake of the present illustration)
the frequencies ñi are equally spaced and separated by an interval
dñ = (ñ2 ñ1 )/N, and ai = 1/N, so that T is simply the arithmetic
average of exp[ k (ñi )u]. In order to accurately resolve the shapes of
individual lines, dñ must generally be at least an order of magnitude
smaller than the linewidths, so N may be rather large.
Note, however, that the end result of the summation in (10.50)
does not in any way depend on the order in which the terms are
summed. Furthermore, if you draw any straight horizontal line
across your absorption spectrum (e.g., the dotted line in Fig. 10.5a),
you’ll find that the same value of k occurs for 2M values of ñi , where
M is the number absorption lines encountered along the way. This
observation suggests that you could kill 2M birds with one stone
in the evaluation of exp[ k (ñi )u], if you just knew what M was for
each different value of k.
 301

100 100
(a) low pressure (b)
10 10

1 1
k

k
0.1 0.1

0.01 0.01
100 100

(c) high pressure (d)

10 10

1 1
k

k
0.1 0.1

0.01 0.01
ν1 ν2 0 1
ν g(k)

100 100
(e)
(f )
10
10
k (high pressure)

1
k

1
0.1
0.1
0.01
ν1 ν2
ν 0.01
0.01 0.1 1 10 100
k (low pressure)

Fig. 10.5: Illustration of the k-distribution method and its extension, the correlated-
k method. (a) A hypothetical spectrum of absorption coefficient k at relatively low
pressure. (b) By sampling the spectrum at fine intervals and then sorting the results
so that k increases monotonically, we define the function 0  g(k)  1 (horizontal
axis). Panels (c) and (d) are the same as (a) and (b) except with stronger pressure
broadening. (e) Comparison of the actual spectrum for low pressure [from panel
(a)] (solid curve) with one estimated from the spectrum at higher pressure [panel
(c)] (dotted curve), using the mapping in panel (f). (f) The mapping between k
values at the two pressure levels, based on equal values of g.
302 10. Broadband Fluxes and Heating Rates†

The essence of the k-distribution method is that you can take all
N of your values of k (ñi ) and sort them so that they are arranged
in ascending order. The result can be treated as a function of a new
variable g that ranges from g = 0 for the smallest value of k present
to g = 1 for the largest value. Fig. 10.5b shows the results of this
process for the absorption spectrum in Fig. 10.5a. In striking con-
trast to the original k (ñ), the new k( g) is a rather smooth, monotonically
increasing function on the interval [0, 1]. By working with k ( g) rather
than the much more complicated k (ñ), we are able to replace the
nasty integral (10.49) with the much nicer one
Z 1
T (u) = exp[ k ( g)u] dg . (10.51)
0
Now, if you are only going to perform the calculation once, then
nothing is gained by this procedure, because you still need to first
evaluate k(ñ) for each of your N frequencies ñi . But if your purpose
is set things up so that you can efficiently recompute T for a variety
of different mass paths u (though all still at the same constant pres-
sure, for now), then the ability to work with the simple function k ( g)
is a decided advantage: whereas at least 1000 data points are needed
to adequately characterize the shape of k (ñ) in Fig. 10.5a, a mere 5–
10 points or so would probably suffice for the k ( g) in Fig. 10.5b. Not
only storage requirements but also the computational effort needed
to evaluate (10.51) are reduced by up to a couple of orders of mag-
nitude.
We can summarize the main difference between the band mod-
els covered in the previous section and the k-distribution method as
follows:
• Band transmission models are based on specific simple sta-
tistical models of the line positions and strengths, such as the
random/Malkmus model. Since real line spectra almost never
obey such a model exactly, the convenient analytic expressions
derived for band models are almost guaranteed to be a bit
rough in terms of how accurately they describe T (u).
• In contrast to band models, the k-distribution method makes
no prior assumption about the distribution of the line
strengths or positions in the band; you simply take what na-
ture gives you in a particular spectral interval and evaluate
 The k-Distribution Method 303

k (ñ) and, from there, k ( g). In this regard, the k-distribution

may be thought of as a data compression scheme rather than
as a transmission model: you are keeping the information you
really need for computing T , namely, the relative frequency
of different values of k, and throwing out the information that
doesn’t matter, which is how those different values of k are
ordered with respect to ñ.

10.3.2 Inhomogeneous Path: Correlated-k

The above discussion of the k-distribution method assumes a homo-
geneous path — i.e., one over which the pressure and temperature
is essentially constant so that k (ñ) is the same at every point along
the path. In practical terms, this implies that we are restricted to
relatively short horizontal paths in the atmosphere. We have to find
a way to generalize the k-distribution method so that it can be used
for extended vertical atmospheric paths over which pressure, and
therefore linewidths, can vary drastically. In the case of the narrow
band models discussed earlier, the HCG approximation gave us a
basis for treating an inhomogeneous path as an equivalent homo-
geneous path. Can we come up with something analogous for the
k-distribution method?
The answer is of course “yes,” and the applicable approxima-
tion is known as the correlated-k method. First, let’s simply write
down the applicable equation, and then take a moment to under-
stand what it’s letting us get away with:
Z 1  Z u
T (u) = exp k ( g, u0 ) du0 dg . (10.52)
0 0

In one sense at least, the above equation makes sense. For each
value of g, we are simply computing the transmittance over the path
between our location u0 = 0 and the remote location u0 = u. We are
then averaging this transmittance over the interval 0 < g < 1 to get
the band transmittance. What could be more straightforward?
But there is a serious flaw in the above reasoning: The above
equation makes physical sense only if a given value of g corresponds
to the same wavenumber ñ at every point along the path. After all,
a vertically propagating photon can’t adjust its frequency along the
304 10. Broadband Fluxes and Heating Rates†

way so as to preserve its original value of g! Recall that our def-

inition of k ( g) at a particular pressure (in the homogeneous case)
entailed a scrambling of the frequency information in the absorp-
tion spectrum. Indeed, any given value k is likely to be associated
with not just one single frequency but rather with many different
frequencies. For example, in Fig. 10.5a, the horizontal dotted line,
corresponding to a single value of k intersects the plot of k (ñ) at
24 distinct frequencies! The absorption coefficient at all of these
frequencies is associated with a single value of g, as indicated by
the intersection of the horizontal dotted line with the k ( g) curve in
Fig. 10.5b.
Note that as you move vertically through the atmosphere, pres-
sure broadening substantially alters the spectrum k (ñ). In Fig. 10.5c,
we see the absorption spectrum at a pressure that is three times
greater than that assumed in Fig. 10.5a. Broadening of the indi-
vidual lines leads to an absorption spectrum that is smoother and
for which the extreme values of k are moderated – the peaks are re-
duced and the valleys are filled. The corresponding plot of k ( g) is
shown in Fig. 10.5d. Although the shape is fairly similar to that in
Fig. 10.5b, the curves are in fact different. There is no a priori reason
to expect the “scrambling” of frequency information to be identi-
cal for both pressures. Therefore, picking a single value of g and
integrating along a path that experiences changing pressure is not
guaranteed to give you a meaningful estimate of the vertical trans-
mittance at any particular frequency.
It turns out that for an isolated Lorentz line, there is, in fact, a
unique mapping between g and ñ that is independent of pressure.
For this special case, the correlated-k method represented by (10.52)
can be used with complete confidence. Unfortunately, we are rarely
dealing with isolated lines, and when we are, we don’t need a band
model! We must face the fact that the correlated-k method conceals
a rather daring assumption for any realistic line spectrum.
As is often the case, heated arguments over whether a given
method is theoretically defensible eventually give way to the sim-
ple practical question: does it appear to work well enough in most
common situations? In fact, it does. We can demonstrate this in two
ways:
Consider again the horizontal dotted line in Fig. 10.5a. This rep-
 The k-Distribution Method 305

resents a fixed value of k for our lower pressure spectrum. Follow

the line to where it intersects the curve in Fig. 10.5b. This inter-
section determines the value of g that corresponds to our chosen k.
Now follow the vertical dotted line down to where it intersects the
curve in Fig. 10.5c. This intersection determines a new value of k
at the higher pressure level for which g(k ) has the same value as for
our original k at low pressure. Now follow the horizontal dotted
line back to the left and note where it crosses k (ñ). These intersec-
tions define the new set of frequencies at high pressure that were
determined by our particular choice of k at low pressure. For the
most part, these frequencies are very close to those determined by
the intersections in Fig. 10.5a. The correspondence is not perfect —
for example, the fifth line from the left has two intersections in the
top panel that have no counterparts in the lower panel.

More generally, we can determine the complete mapping be-

tween k at the two pressure levels, based on the requirement of
equal values of g. The results for this example are shown in
Fig. 10.5f. Using this mapping, we can test how well we are able
to predict k (ñ) at one pressure based on k (ñ) at the other pressure.
Fig. 10.5e depicts both the original k (ñ) at low pressure (solid curve)
and the predicted k (ñ), based on applying the mapping in panel (f)
to the high pressure k (ñ) in panel (c). Over most of the spectral in-
terval, the agreement is nearly perfect. Only in the vicinity of a few
line centers (e.g., that fifth line from the left) does the agreement
break down noticably. However, these isolated errors occupy such
a small fraction of the spectral band that they don’t introduce much
error into the calculation of T (u).

The bottom line is that the correlated-k method, while not per-
fect, typically allows fluxes and heating rates to be calculated with
overall errors of less than 1%. Most importantly, it does so with
at least three orders of magnitude less computational effort than
a brute-force LBL calculation. If you ever have occasion to work
with a modern model for computing band-averaged IR radiances
or fluxes in the atmosphere, there’s a good chance that you can find
a correlated-k procedure lurking somewhere inside.
306 10. Broadband Fluxes and Heating Rates†

10.4 Applications to Meteorology, Climatology,

and Remote Sensing
The previous section outlined the methods used to compute broad-
band upwelling and downwelling fluxes in a cloud-free atmo-
sphere. Of course hardly anyone computes fluxes just for the fun
of it; rather, they do it because it is an essential part of modeling the
energy budget of the atmosphere.

10.4.1 Fluxes and Radiative Heating/Cooling

Radiative Heating Equations
In Section 2.7, we defined the net flux as

Fnet (z) ⌘ F " (z) F# (z) . (10.53)

Depending on the context, we might be interested in the net flux

computed (or measured) for all wavelengths, for just the shortwave
or longwave band, or even for a narrow spectral interval. In any of
these cases, Fnet represents the corresponding net upward flow of
radiative energy, measured in watts per meter squared, through a
unit horizontal area at level z.
Now consider a thin layer of the atmosphere with its base at al-
titude z and its upper boundary at z + Dz. The net flux Fnet (z) rep-
resents the rate at which radiative energy enters the bottom of this
layer; likewise, Fnet (z + Dz) gives the rate at which energy leaves
at the top of the layer. If the two fluxes are equal, then there is no
net change over time in the internal energy of the layer. If they are
not equal, then the layer must be experiencing a net gain or loss of
energy.
It follows that the radiative heating rate at level z is given by

1 ∂Fnet
H⌘ (z) , (10.54)
r(z)C p ∂z

where r(z) is the air density at level z and C p = 1005 J/(kg K) is the
specific heat capacity of air at constant pressure. The minus sign is
 Applications 307

needed because an increase in Fnet with height implies a net loss of

energy from level z. Traditionally, H is expressed in units of C/day.
When the value of H is negative (as it is more often than not), one
might prefer to speak instead of positive radiative cooling rate.
In order to utilize the band model machinery we developed ear-
lier, let’s confine our attention to the heating/cooling rate associated
with a particular spectral interval Dñi . The complete expressions
for upwelling and downwelling flux, including boundary contribu-
tions, are then
Z z
∂ Ti ( z0 , z ) 0
Fi" (z) = Fi" (0)Ti (0, z) + Dñi pBi (z ) 0
dz , (10.55)
0 ∂z0
Z •
∂Ti (z, z0 ) 0
Fi# (z)= Fi# (•)Ti (z, •)Dñi pBi (z ) 0
dz , (10.56)
z ∂z0
where, as usual, Ti (z, z0 ) is the band-averaged flux transmittance
between levels z and z0 . Note that we are using Bi (z) as a shorthand
notation for the average value of B[ T (z)] in the ith spectral interval.
Let’s use the above expressions to evaluate Fnet and, from there,
∂Fnet /∂z, as required for the heating rate in (10.54):

Finet (z) = Fi" (0)Ti (0, z) Fi# (•)Ti (z, •)

Z z
∂ Ti ( z 0 , z ) 0
0
+ Dñi pBi (z ) dz (10.57)
0 ∂z0
Z •
∂Ti (z, z0 ) 0
+ Dñi pBi (z0 ) dz ,
z ∂z0
∂Finet (z) ∂T (0, z) ∂Ti (z, •)
= Fi" (0) i Fi# (•)
∂z Z ∂z ∂z
∂ z 0
∂ Ti ( z , z ) 0
+ Dñi pBi (z0 ) dz (10.58)
∂z 0 ∂z0
Z •
∂Ti (z, z0 ) 0
+ pBi (z0 ) dz .
z ∂z0
To evaluate the partial derivatives of the integral terms, in which z
appears both as one of the limits of integration and as an argument
to the integrand itself, we invoke the following mathematical iden-
tity:
Z x Z x
∂ ∂ f ( x, y)
f ( x, y) dy ⌘ dy + f ( x, x ) , (10.59)
∂x x0 x0 ∂x
308 10. Broadband Fluxes and Heating Rates†

so that the heating rate H at level z is proportional to

∂Ti (0, z) ∂T (z, •)

H(z)r(z)C p = Fi" (0) + Fi# (•) i
" Z ∂z ∂z
Z •
z
0∂2 T i ( z 0 , z ) 0 ∂2 Ti (z, z0 ) 0
0
+ Dñi pBi (z ) dz pBi (z ) dz
0 ∂z0 ∂z z ∂z0 ∂z
#
0
∂ Ti ( z , z ) 0
∂T (z, z )
pBi (z) 0
+ pBi (z) i 0 .
∂z z0 =z ∂z z0 =z
(10.60)
Before we make further adjustments, let’s take a second to interpret
the terms in the above equation.
The first term is the contribution of the upward flux at the sur-
face to heating at level z. Note that ∂T (0, z)/∂z (which is nega-
tive) plays exactly the same role as the “absorption weighting func-
tion” we derived in (7.51), except that it now applies to the band-
integrated flux rather than the monochromatic intensity. The sec-
ond term is of course the same as the first, except that it represents
the absorption of radiation incident at the top of the atmosphere.
Both the first and the second terms are always greater than zero and
represent positive heating terms.
The two integrals in the second line of the equation represent the
contributions of emission at all other levels z0 to heating at level z.
These are also positive terms.
The final two terms,4 respectively, represent losses of energy due
to upward and downward emission of radiation by the atmosphere
at level z. Note that both terms are equal and negative, because

∂ Ti ( z0 , z ) ∂Ti (z, z0 )
= . (10.61)
∂z0 z0 =z ∂z0 z0 =z

Although we could combine the last two terms in (10.60) using the
above identity, we will leave them separate in order to make the
next steps clearer.
Although (10.60) is certainly an accurate and complete statement
of how local heating depends on the profile of transmission and
temperature (via B(z)), we’re less interested here in computational
4 These terms are omitted in Equation (11.29) of TS02.
 Applications 309

validity than we are in physical insight. Watch carefully while we

rearrange our equation into a new form that clarifies some aspects
of the relationship between local heating and the surrounding envi-
ronment.
We will start with an expression that has the form we are seek-
ing and show that it is mathematically equivalent to, and can be
substituted for, part of (10.60).

Z z ⇥ ⇤ ∂2 T i ( z 0 , z ) 0
0
p Bi ( z ) Bi ( z ) dz
0 ∂z0 ∂z
Z z 2 Z z
0
0 ∂ Ti ( z , z ) 0 ∂2 T i ( z 0 , z ) 0
= pBi (z ) dz pBi (z) dz
0 ∂z0 ∂z 0 ∂z0 ∂z
Z z 2 0 Z z 2
0 ∂ Ti ( z , z ) 0 ∂ Ti ( z0 , z ) 0
= pBi (z ) dz pB i ( z ) dz
0 ∂z0 ∂z 0 ∂z0 ∂z
Z z 2 0  z0 =z
0 ∂ Ti ( z , z ) 0 ∂ Ti ( z0 , z )
= pBi (z ) dz pBi (z)
0 ∂z0 ∂z ∂z z 0 =0
Z z
∂2 T i ( z 0 , z ) 0 ∂ Ti ( z0 , z ) ∂Ti (0, z)
= pBi (z0 ) dz pBi (z) + pBi (z)
0 ∂z0 ∂z ∂z z0 =z ∂z
Z z
∂2 T i (z0 , z) ∂ Ti (z0 , z) ∂Ti (0, z)
= pBi (z0 ) dz0 + pBi (z) + pBi (z)
0 ∂z0 ∂z ∂z0 z0 =z ∂z
, (10.62)

or, if we subtract the right side of the above equality from the left
side,
Z z ⇤ ∂2 T i ( z 0 , z ) 0 Z z 2 0
0 ∂ Ti ( z , z )
⇥ 0
p Bi ( z ) Bi ( z ) dz pB i ( z ) dz0
0 ∂z0 ∂z 0 ∂z 0 ∂z
0
∂ Ti ( z , z ) ∂Ti (0, z)
pBi (z) 0
pBi (z) = 0. (10.63)
∂z z0 =z ∂z

Using similar reasoning, we can show that

Z • ⇤ ∂2 Ti (z, z0 ) 0 Z • 2 0
0 ∂ Ti ( z, z )
⇥ 0
p Bi ( z ) Bi ( z ) dz pB i ( z ) dz0
z ∂z0 ∂z z ∂z 0 ∂z

∂Ti (z, •) ∂Ti (z, z0 )

+ pBi (z) + pBi (z) = 0. (10.64)
∂z ∂z0 z0 =z
310 10. Broadband Fluxes and Heating Rates†

Since the expressions on the left-hand sides of both (10.63) and

(10.64) are equal to zero, we can subtract both expressions from in-
side the square brackets of (10.60) without changing the latter’s va-
lidity. After considerable manipulation and cancellation of terms,
we get

(
1
H(z) =
r(z)C p
h i ∂T (0, z)
" i
Fi (0) Dñi pBi (z) ( A)
∂z
h i ∂T (z, •)
i
+ Fi# (•) Dñi pBi (z) ( B)
∂z
Z •⇥ ⇤ ∂2 Ti (z, z0 ) 0 (10.65)
0
Dñi p Bi ( z ) Bi ( z ) dz (C )
z ∂z0 ∂z
Z z⇥ ⇤ ∂2 T i ( z 0 , z ) 0
Dñi p Bi ( z 0 ) Bi ( z ) dz (D)
0 ∂z0 ∂z
)
.

Equation (10.65) certainly looks different than the mathemati-

cally equivalent (10.60), and it tells a different kind of story as well.
Each of the partial derivatives represents the degree of radiative cou-
pling between level z and some other part of the column. It charac-
terizes the degree to which radiation emitted by one component is
absorbed by the the atmosphere at z and vice versa.
Thus, for each of the lines (A)–(D), the difference inside the
square brackets represents the imbalance between (a) radiation
emitted from the remote location and reabsorbed at level z, and (b)
energy lost by emission from level z and reabsorbed at the remote
location. If the difference is positive, then the contribution to heat-
ing at level z is positive; negative implies cooling. I will emphasize
the word exchange wherever it appears below in order to reinforce
this idea of a two-way process.
Let’s now interpret each line in turn:
Term (A) represents net heating/cooling through radiative ex-
change with the lower boundary. In the thermal IR band, we
 Applications 311

can usually take the surface to be black, in which case we can

replace Fi" (0) with Dñi pBi ( Ts ), where Ts is the surface temper-
ature. Since Ts is usually greater than T (z), Term A usually
represents a heating term.

Term (B) represents heating/cooling through radiative exchange

with the top of the atmosphere (TOA). If we’re working in the
thermal IR band, then Fi# (•) is usually taken to be zero, in
which case the exchange is strictly one-way. In the LW band,
therefore, term (B) describes cooling to space. If, on the other
hand, we’re working with solar radiation, then Fi# (•) rep-
resents the incident flux of solar radiation at the TOA, and
B(z) = 0, so that term (B) describes heating through the direct
absorption of solar radiation.

Terms (C) and (D) collectively represent radiative exchanges be-

tween level z and all other levels in the atmosphere z0 . In
order for the net effect of this exchange to be significant, there
must be a large temperature difference between the two levels,
and the radiative coupling (represented by the second deriva-
tives of T ) must be strong. The coupling is strongest when the
transmission T changes rapidly in response to changes in both
z and z0 . In regions of the spectrum for which the atmosphere
is strongly absorbing (e.g., the middle of the CO2 15 µm band),
the coupling is strongest with levels z0 that are very close to z
and where the temperature difference is small; therefore the
contribution of these wavelengths to local heating at z is neg-
ligible.

Note further that (C) represents exchanges between z and higher

levels of the atmosphere; (D) represents exchanges with lower alti-
tudes. In the middle of the troposphere, the temperature generally
decreases in a quasi-linear fashion with increasing z. This implies
that the cooling contribution by (C) will usually be offset to a large
degree by heating from (D). The reverse applies in the stratosphere,
where temperature tends to increase with height. It follows, there-
fore, that heating/cooling due to exhanges with other levels will
be strongest where there is a minimum or maximum in T (z). In
particular, the tropopause level “sees” radiation arriving from the
312 10. Broadband Fluxes and Heating Rates†

warmer stratosphere (term C) as well as from warmer levels of the

troposphere (term D); hence, the tropopause experiences positive
heating contributions from both (C) and (D).
In the longwave band, every term except (B) represents radiative
exchanges between levels having temperatures falling somewhere
in the range 190–310 K. The typical temperature difference between
two levels that are strongly coupled is on the order of 10s of K or
less. Furthermore, as noted above, terms (C) and (D) tend to partly
cancel each other, except in the vicinity of the tropopause.
Term (B), by contrast, represents a direct loss of radiation to
space, with no compensating return radiation. This term is therefore
always negative, and it is quite often the largest term overall in the
longwave radiation budget, especially at altitudes for which the at-
mosphere rapidly becomes more transparent with increasing height
(e.g., due to the rapid narrowing of absorption lines). To a very good
approximation in many cases, radiative cooling profiles in the atmosphere
can be estimated from term (B) alone. This is called the cooling-to-space
approximation.

Model Atmospheres

It’s nice to have an equation like (10.65) to tell you how heating or
cooling at a particular level is physically related to profiles of tem-
perature and band transmittance. But simply staring at it won’t tell
you much about the actual magnitude of the heating or cooling to
expect at any particular level z. You have to write a program to nu-
merically evaluate the various terms in the equation using suitable
band transmittance models, and you then have to apply your pro-
gram to a particular profile of atmospheric temperature, humidity,
and trace gas composition.
Although you could, in principle, pull any old radiosonde
sounding off the Internet and run your program on it (although
some assumptions would still be required concerning ozone pro-
files and the like), atmospheric scientists usually like to start by
running their radiative transfer codes on idealized profiles called
model atmospheres. These don’t represent actual observations but are
designed to reflect typical atmospheric conditions for a particular
location and season. The use of standard model atmospheres has
 Applications 313

°
30 Tropical Conditions, solar zenith of 30

25
Altitude (km)
20 H2O
CO2
15 O3

10

0
0 0.5 1 1.5 2 2.5 3
Shortwave Radiative Heating ( °C per Day)

Fig. 10.6: Typical heating profiles due to solar absorption in a cloud-free tropi-
cal atmosphere, segregated according to the responsible atmospheric constituents
ozone (O3 ), carbon dioxide (CO2 ), and water vapor (H2 O). (Figure courtesy of S.
Ackerman.)

two advantages:

• It facilitates the comparison of different radiative transfer

codes. If results differ significantly for two methods using the
exact same model atmosphere, then at least one of them (per-
haps both) must be doing something wrong!

• The results of a radiative transfer calculation for a model at-

mosphere can be regarded as typical, in some sense, for the
region and season and question it represents, even if it is un-
derstood that actual conditions will vary from day to day and
location to location. Thus a model atmosphere will usually
be employed in a calculation any time an actual sounding is
not available but you need representative radiative transfer
results.

There are at least seven standard model atmospheres in circula-

tion that radiation and remote sensing scientists tend to rely on for
314 10. Broadband Fluxes and Heating Rates†

the above purposes: (1) tropical, (2) midlatitude summer, (3) mid-
latitude winter, (4) subarctic summer, (5) subarctic winter, (6) arctic
summer, and (7) arctic winter. Although the associated temperature
and humidity profiles are highly idealized, they vary in the ways
you’d expect: the tropical model atmosphere is warm and humid
with a high, cold tropopause; the arctic winter atmosphere is ex-
tremely cold with low humidity and a very low tropopause, and the
other five models fall somewhere in between these two extremes.
It’s worth keeping in mind that variations in water vapor mixing
ratio are often even more important for radiative heating profiles
than are variations in the temperature profiles themselves. This is
because of water vapor’s relative opacity over much of the SW and
LW bands.

Shortwave Heating
Figure 10.6 depicts heating profiles due to the absorption of solar
radiation in a cloud-free tropical atmosphere. The heating rate is
given in degrees per day. Separate profiles are shown for the three
major contributors to atmospheric absorption in the solar band: wa-
ter vapor, ozone, and carbon dioxide. The first two of these con-
stituents are clearly the dominant absorbers in the solar band, with
water vapor being responsible for the bulk of the absorption (up to
1.3 K/day) in the troposphere (z < 15 km in this model atmosphere)
and ozone dominating in the stratosphere, with heating rates in ex-
cess of 2 K/day.5 Carbon dioxide plays a minor role, being respon-
sible for around 0.05 K/day of heating at all levels.
To first order, the differences between the shapes of the heat-
ing profiles for different constituents mirrors the difference in con-
stituent mixing ratios:

• Water vapor is most abundant at the warm altitudes in the

lower troposphere, because it condenses to liquid water or ice
and precipitates out when it is transported to higher, colder
altitudes.

• Ozone is concentrated primarily in the stratosphere, where it

5 The heating rate continues to increase with altitude above 30 km, peaking at
an altitude high in the upper stratosphere.
 Applications 315

Tropical conditions with varying solar zenith angle

30

25

Altitude (km)
20 15°
30°
15 60°

10

0
0 0.5 1 1.5 2 2.5 3
Shortwave Radiative Heating ( °C per Day)

Fig. 10.7: Typical heating profiles due to solar absorption in a cloud-free tropical
atmosphere at three different solar zenith angles. The heating profile for qs =
30 corresponds to the sum of the three profiles in Fig. 10.6. (Figure courtesy of S.
Ackerman.)

is created by the action of UV-C radiation on molecular oxy-

gen. In fact, the stratosphere itself, which is defined by the
general increase of temperature with height starting around
10–15 km owes its very existence to the absorption of solar ra-
diation by ozone. If there were no molecular oxygen in the atmo-
sphere, there would be no ozone layer. Without ozone, there would
be very little solar heating of the middle atmosphere and therefore no
stratosphere!

• Carbon dioxide is evenly mixed throughout the atmosphere,

because atmospheric sources and removal mechanisms for
CO2 operate very slowly in comparison to those for water va-
por and ozone.

Problem 10.2: Actual water vapor mixing ratios in the standard

tropical atmosphere continue to increase all the way to the surface.
316 10. Broadband Fluxes and Heating Rates†

Yet the heating rate associated with water vapor peaks at an altitude
near 5 km and then decreases sharply below that level. Explain why.

Total shortwave heating rates are of course the sum of those

contributed by the individual constituents. Profiles of total heat-
ing due to the absorption of solar radiation in a tropical atmosphere
are shown in Fig. 10.7. The different profiles correspond to different
solar zenith angles. Not surprisingly, when the sun is lower in the
sky, overall heating rates are reduced.
Of course, the heating profiles described above include only con-
tributions due to direct absorption of shortwave radiation. In the
absence of clouds, solar radiation not absorbed by the atmosphere
reaches the surface, where a fraction equal to one minus the surface
albedo gets absorbed. Much of that energy indirectly heats the atmo-
sphere by way of three mechanisms: (1) emission and reabsorption
of longwave radiation (discussed below), (2) direct conduction of
heat from the surface to the overlying air, followed by convective
mixing, and (3) evaporation of water from the surface, followed by
latent heat release in clouds.

Longwave Cooling

In the longwave band, each part of the atmosphere is simultane-

ously emitting and absorbing radiation. Where absorption domi-
nates, there is net heating; where emission dominates, there is net
cooling. Profiles of heating/cooling in the longwave band are more
difficult to interpret because all four of the terms in (10.65) are po-
tentially significant.
Fig. 10.8 shows longwave heating profiles for a tropical atmo-
sphere, again segregated according to the responsible constituent.
Water vapor is actually represented here by two curves, one for the
heating contribution due to its rotation/vibration bands near 6.3 µm
and beyond 15 µm; the other due to the relatively weak but perva-
sive continuum component that dirties up the spectral “windows”
between conventional absorption bands.
Here is a brief rundown of the major features seen in this plot:
 Applications 317

Longwave Radiative Heating for Tropical Conditions

30

25
Altitude (km) H2O
20 CO2
O3
15 Continuum

10

0
-6 -5 -4 -3 -2 -1 0 1
Longwave Radiative Heating ( °C ) per Day

Fig. 10.8: Typical heating rate profiles due to longwave (thermal IR) radiative
transfer in a cloud-free tropical atmosphere, segregated according to the responsi-
ble atmospheric constituents ozone (O3 ), carbon dioxide (CO2 ), water vapor res-
onant absorption (H2 O), and water vapor continuum. Negative values represent
cooling. (Figure courtesy of S. Ackerman, with modifications.)

CO2 Because of the high opacity of the pressure broadened 15 µm

band, carbon dioxide contributes rather little to net radiative
heating in the troposphere. Radiation emitted at one level is
reabsorbed at nearby level having almost the same tempera-
ture. Only at the tropopause (near 15 km), where the tem-
perature profile has a minimum, is there a small amount of
net heating. At higher altitudes, pressure broadening is much
weaker and the band “opens up,” allowing emitted radiation
to escape to space with little compensating radiation down-
ward from higher levels. This is of course the cooling to space
previously discussed in connection with term (B) in (10.65).

H2 O Because water vapor is concentrated at low altitudes, the

cooling-to-space effect kicks in strongly between 3 and 10
km altitude, with maximum cooling rates between 2 and
3.5 K/day. The two peaks in the profiles are associated with
318 10. Broadband Fluxes and Heating Rates†

different absorption bands, the stronger of these being associ-

ated with the higher altitude peak. Above the tropopause near
15 km, there is very little water vapor present and so its con-
tribution to cooling at those altitudes is very modest – about
0.2 K/day.
H2 O continuum One of the unique characteristics of continuum
absorption is that it is very sensitive to pressure. The mass
absorption coefficient of water vapor falls off rapidly with
height. Consequently, the atmosphere in the upper tropo-
sphere and above is effectively transparent in this spectral
band, while lower altitudes see fairly strong absorption. Once
again, cooling-to-space is the dominant radiative exchange
term at the rather sharp transition between the two zones,
with peak cooling of 5 K/day occurring at the relatively low
altitude of 3 km. Above 10 km, the continuum contribution to
cooling is essentially zero.
O3 Ozone is responsible for the sole instance of significant LW
warming in the atmosphere below 30 km, with the peak warm-
ing of 0.5 K/day occurring between 20 and 30 km. This heat-
ing is due to the absorption at the base of the ozone layer of ra-
diation emitted by the ground in the 9.6 µm band — i.e., term
(A) in (10.65). If the plotted profiles in Fig. 10.8 were extended
to higher altitude, you would see the heating due to term (A)
disappear and get replaced with fairly strong cooling-to-space
at the top of the ozone layer.
The above discussion applies to a tropical atmosphere. Different
model atmospheres have different profiles of temperature and LW
absorption and therefore different profiles of LW heating/cooling.
Fig. 10.9 shows total LW heating profiles for a tropical, midlatitude
summer, and subarctic summer atmosphere. Qualitatively they are
similar; only the details differ, mainly due to differences in the wa-
ter vapor contribution and in the gross features of the temperature
profile.
Note that the LW radiative exchanges in the cloud-free atmo-
sphere have a net cooling effect at almost all levels of the atmo-
sphere, whereas the effect of SW absorption was exclusively warm-
ing. In fact, in the stratosphere, SW warming and LW cooling nearly
 Applications 319

Longwave Radiative Heating

30

25
Altitude (km)
20 Tropical
Midlatitude Summer
Subarctic Summer
15

10

0
-4 -3.5 -3 -2.5 -2 -1.5 -1 -0.5 0 0.5 1
Longwave Radiative Heating ( °C per Day)

Fig. 10.9: Total longwave heating rate profiles for three different model atmo-
spheres. (Figure courtesy of S. Ackerman, with modifications.)

cancel. This is because there are few other effective energy exchange
mechanisms at work in the stratosphere; therefore the temperature
of the stratosphere must necessarily approach that required for ra-
diative equilibrium. In the troposphere, a variety of other processes
are available to compensate for any radiative imbalance; therefore
we don’t expect the SW and LW heating profiles to be mirror images
of one another.
 CHAPTER 11

The Radiative Transfer Equation With Scattering

Throughout this book so far, we have discussed scattering primar-

ily as one of two mechanisms for extinguishing radiation, the other
mechanism being absorption. Thus, the extinction coefficient β e
could be decomposed into the sum of the absorption coefficient β a
and scattering coefficient β s . The single scatter albedo ω̃, defined
as β s /β e , was introduced as a convenient parameter describing the
relative importance of absorption and scattering: when ω̃ = 0, ex-
tinction of radiation is entirely by way of absorption; when ω̃ = 1,
then there is no absorption, only scattering.
When radiation is extinguished via scattering, its energy is not
converted to another form; rather, the radiation is merely redi-
rected. The loss of radiation along one line-of-sight due to scattering
is therefore always associated with a gain in radiation along other
lines-of-sight passing through the same volume.
You can easily observe the above phenomenon with the help of
a powerful, narrow beam of light, such as that from a searchlight, an
automobile headlight or a laser pointer. When the air is very clear
— i.e., free of smoke, dust, haze, or fog — then the beam can pass di-
rectly in front of you and you will not see it, because essentially none
of the radiation is scattered out of the original path into the direction
toward your eyes. But if the air contains suspended particles, then

313
314 11. RTE With Scattering

these particles will scatter some fraction of the beam into all direc-
tions, including toward your eyes, and the path of the beam will be
clearly apparent, especially against a dark background. In this case,
scattering clearly serves as a source of radiation, as seen from your
vantage point. Of course the original beam is depleted by the same
process. For example, if a fog is thick enough, the headlights of an
oncoming car can’t be seen at all until it is relatively close to you.
This chapter introduces the terminology and mathematical no-
tation required to account for scattering as a source of radiation in
the radiative transfer equation.

11.1 When Does Scattering Matter?

When scattering is important as a source of radiation along a partic-
ular line-of-sight, then the complexity of calculations of radiative
transfer along that line-of-sight greatly increases compared with
the nonscattering case. This is because one must, in the worst
case, solve for the intensity field not just in one direction along a
one-dimensional path but for all directions simultaneously in three-
dimensional space! You would therefore like to be able to neglect
scattering (as a source, at least) whenever you can get away with
it.
In fact, you can safely ignore scattering as a source whenever
gains in intensity due to scattering along a line-of-sight are negligi-
ble compared with (a) losses due to extinction and (b) gains due to
thermal emission. In the atmosphere, these conditions are usually
satisfied for radiation in the thermal IR band and for microwave ra-
diation when no precipitation (e.g., rain, snow, etc.) is present. In
addition, if one is concerned only with the depletion of direct radi-
ation from an isolated, point-like source, such as the sun, then the
above conditions are usually satisfied to reasonable accuracy.
For virtually any problem involving the interaction of short-
wave (ultraviolet, visible, and near-IR) radiation with the atmo-
sphere, scattering is the dominant atmospheric source of radiation
along any line-of-sight other than that looking directly at the sun.
The blue sky, white or gray clouds, the atmospheric haze that re-
duces the visual contrast of distant objects — all of these make their
presence known primarily by way of scattered radiation.
 Radiative Transfer Equation with Scattering 315

11.2 Radiative Transfer Equation with Scattering

11.2.1 Differential Form
Previously, we derived Schwarzschild’s Equation (8.4) under the as-
sumption that scattering was unimportant and that therefore β e =
β a . Under that assumption, we found that the change in intensity
dI along an infinitesimal path ds could be written as

dI = dIabs + dIemit , (11.1)

where the depletion due to absorption is given by

dIabs = − β a I ds , (11.2)

and the source due to emission is

dIemit = β a B( T ) ds . (11.3)

In order to generalize the equation to include scattering, we must

recognize that depletion occurs due to both absorption and scatter-
ing, so that β e rather than β a must appear in the depletion term.
Moroever, we must now add a source term that describes the con-
tribution of radiation scattered into the beam from other directions,
so that
dI = dIext + dIemit + dIscat , (11.4)
where
dIext = − β e I ds . (11.5)
The term dIscat requires more thought. First, we know it must be
proportional to the scattering coefficient β s , since without scatter-
ing there can be no contribution from this term. Second, we recog-
nize that radiation passing through our infinitesimal volume from
any direction Ω̂ can potentially contribute scattered radiation in the
direction of interest Ω̂. Moreover, these contributions from all di-
rections will sum in a linear fashion — that is, the path taken by a
photon arriving from one direction is not influenced by the presence
of, or paths taken by, other photons.
Mathematically, these ideas are expressed as follows:


βs
dIscat = p(Ω̂ , Ω̂) I (Ω̂ ) dω ds , (11.6)
4π 4π
316 11. RTE With Scattering

where the integral is over all 4π steradians of solid angle, and the
scattering phase function p(Ω̂ , Ω̂) is required to satisfy the normal-
ization condition

1
p(Ω̂ , Ω̂) dω = 1 . (11.7)
4π 4π

The complete differential form of the radiative transfer equation can

thus be written


βs
dI = − β e Ids + β a Bds + p(Ω̂ , Ω̂) I (Ω̂ ) dω ds . (11.8)
4π 4π

Dividing through by dτ = β e ds, we can write

dI (Ω̂) ω̃
= − I (Ω̂) + (1 − ω̃ ) B + p(Ω̂ , Ω̂) I (Ω̂ ) dω ,
dτ 4π 4π

(11.9)
where, in the interest of clarity, we make the dependence of I on
direction Ω̂ explicit. This is the most general and complete form of the
radiative transfer equation that we will normally have to deal with in this
book.1
Note that it is often convenient lump all sources of radiation into
a single term, so that (11.9) may be written in shorthand form as

dI (Ω̂)
= − I (Ω̂) + J (Ω̂) , (11.10)
dτ

where the source function is given by

ω̃
J (Ω̂) = (1 − ω̃ ) B + p(Ω̂ , Ω̂) I (Ω̂ ) dω . (11.11)
4π 4π

1 We have defined dτ here to be positive for translation in the same direction

as the propagation of the beam. Some textbooks use the opposite convention, in
which case the signs of all terms on the right hand side are reversed.
 Radiative Transfer Equation with Scattering 317

We see that the total source is a weighted sum of thermal emission

and scattering from other directions, with the single scatter albedo
controlling the weight given to each. If ω̃ = 0, the scattering term
vanishes; if ω̃ = 1, the thermal emission component vanishes.

11.2.2 Polarized Scattering†

Throughout most of this book, we have ignored the role of polariza-

tion in atmospheric radiative transfer and considered the effects of
transmission, absorption and scattering only on the scalar intensity
I. Although this is almost always an approximation, it is often a
very good one. There are times, however, when it is necessary to re-
vert to a more accurate fully polarized treatment of radiative transfer,
which requires us to consider changes not only in I but in all ele-
ments of the four-parameter Stokes vector
I = ( I, Q, U, V ) that was
introduced in (2.48). The fully polarized version of the differential
radiative transfer equation (11.9) can be written

d
I(Ω̂) ω̃
= −
I(Ω̂) + (1 − ω̃ ) B
U + P(Ω̂ , Ω̂)
I(Ω̂ ) dω ,
dτ 4π 4π
(11.12)
where P(Ω̂ , Ω̂) is a 4 × 4 scattering phase matrix, and
U ≡ (1, 0, 0, 0)
when ω̃ is considered to be independent of polarization. The latter
assumption is not guaranteed to be valid; indeed, for some prob-
lems involving preferentially oriented particles, such as might be
encountered in ice clouds or snowfall, even ω̃ and the extinction
coefficient β e (implicit in τ) may each depend on both polarization
and direction.
You are most likely to encounter the fully polarized RTE in the
context of certain remote sensing problems. A more comprehen-
sive discussion of polarized radiative transfer (though still with
some simplifications, such as polarization-independent extinction
and optical path) is given by L02 (Section 6.6). For the remainder
of this book, we will continue to rely on the scalar form of the RTE
given by (11.9) unless otherwise noted.
318 11. RTE With Scattering

11.2.3 Plane Parallel Atmosphere

Although we know that the atmosphere is far from horizontally
homogeneous, especially where clouds are concerned, most ana-
lytic solutions and approximations to the radiative transfer equa-
tion with scattering have been derived for the plane parallel case.
Why? There are three basic reasons:

• Plane-parallel geometry is really the only semi-realistic case

that lends itself to straightforward analysis and/or numerical
solution (e.g., in climate and weather forecast models).

• There are indeed problems (e.g., the cloud-free atmosphere,

horizontally extensive and homogeneous stratiform cloud
sheets) for which the plane-parallel assumption usually seems
quite reasonable as an approximation to reality.

• Even where it is not reasonable, there remains considerable

doubt about the best way(s) to handle three-dimensional inho-
mogeneity, especially when computational efficiency is essen-
tial. Therefore investigators tend to fall back on plane-parallel
geometry (with minor embellishments, such as the so-called
independent pixel approximation), knowing that it is not perfect
but believing it to be better than nothing at all (this is fine, as
long as the potential for large errors is understood by all con-
cerned!).

To adapt (11.10) to a plane-parallel atmosphere, we reintroduce

the optical depth τ, measured from the top of the atmosphere, as
our vertical coordinate, and we will henceforth use µ ≡ cos θ to
specify the direction of propagation of the radiation measured from
zenith.2 We then have

dI (µ, φ)
µ = I (µ, φ) − J (µ, φ) , (11.13)
dτ
2 Some textbooks, such as L02 and S94, specify that µ ≡ cos θ. Others, such as
TS02, instead define µ ≡ | cos θ |, as I also did in an earlier chapter of this book.
When writing the equations of radiative transfer with scattering, each convention
has its own advantages and disadvantages. Here I have chosen the definition that
permits the same equation to be used for both upward and downward radiation.
 The Scattering Phase Function 319

where the source function for both emission and scattering is

2π
1
ω̃
J (µ, φ) = (1 − ω̃ ) B + p(µ, φ; µ , φ ) I (µ , φ ) dµ dφ .
4π 0 −1
(11.14)
There is only a relatively small class of applications in which it
is necessary to consider both scattering and emission at the same
time. Two examples include (1) microwave remote sensing of pre-
cipitation, and (2) remote sensing of clouds near 4 µm wavelength,
for which scattered solar radiation may be of comparable impor-
tance to thermal emission. Except where noted, the rest of this book
will focus on problems involving scattering of solar radiation only,
without the additional minor complication of thermal emission.

11.3 The Scattering Phase Function

One way to give physical meaning to the scattering phase function
is to regard 4π 1
p(Ω̂ , Ω̂) as a probability density: Given that a pho-
ton arrives from direction Ω̂ and is scattered, what is the proba-
bility that its new direction falls within an infinitesimal element dω
of solid angle centered on direction Ω̂? The normalization condi-
tion (11.7) simply ensures that energy is conserved when there is no
absorption (ω̃ = 1); i.e., the new direction of a scattered photon is
guaranteed to fall somewhere within the available 4π steradians of
solid angle, and you can’t get more (or fewer) photons out than you
put in.
The functional dependence of the phase function on Ω̂ and Ω̂
can be quite complicated, depending on the sizes and shapes of the
particles responsible for the scattering. Nevertheless, an important
simplification can be made when particles suspended in the atmo-
sphere are either spherical or else randomly oriented. For exam-
ple, cloud droplets are spherical, and small aerosol particles and air
molecules, while generally not spherical, have no preferred orienta-
tion.3 In such cases, the scattering phase function for a volume of
air depends only on the angle Θ between the original direction Ω̂
3 Falling ice crystals, snowflakes, and raindrops generally do have a preferred

orientation due to aerodynamic forces, and this directional anisotropy must some-
times be considered in radiative transfer calculations.
320 11. RTE With Scattering

and the scattered direction Ω̂ , where

cos Θ ≡ Ω̂ · Ω̂ . (11.15)

The ability to replace p(Ω̂ , Ω̂) with p(Ω̂ · Ω̂) ≡ p(cos Θ) is very
helpful, inasmuch as the number of independent directional vari-
ables needed to fully characterize p is reduced from four (two each
for Ω̂ and Ω̂ ) to only one. The normalization condition (11.7) then
reduces to

2π
π
1
p(cos Θ) sin ΘdΘdφ = 1 , (11.16)
4π 0 0

or

1
1
p(cos Θ) d cos Θ = 1 . (11.17)
2 −1

Except where noted, this simplified notation for the phase function
will be utilized throughout the remainder of this book.4

11.3.1 Isotropic Scattering

The simplest possible scattering phase function is one that is con-
stant; i.e.
p(cos Θ) = 1 . (11.18)
Scattering under this condition is known as isotropic. It describes
the case that all directions Ω̂ are equally likely for a photon that has
just been scattered. Thus, the new direction the photon takes is in
no way predictable from the direction it was traveling prior to being
scattered; in other words, the photon “forgets” everything about its
past.
An example of the random path of a single photon experiencing
isotropic scattering is shown in Fig. 11.1a. Note that once the pho-
ton passes into the interior of the cloud layer it wanders aimlessly,
4 It sometimes necessary, however, to recast a phase function that is inherently

of the form p(cos Θ) in terms of the absolute directions (Ω̂ , Ω̂) in order to facilitate
integration over zenith and/or azimuth angles θ and φ.
 The Scattering Phase Function 321

a) 1 photon b) 3 photons
g=0 g = 0.85
τ=0

τ∗=10
Fig. 11.1: Examples of the random paths of photons in a plane-parallel scattering
layer with optical thickness τ ∗ = 10. Photons are incident from above with θ =
30◦ . Heavy diagonal lines indicate the path an unscattered photon would take. (a)
The trajectory of a single photon when scattering is isotropic. (b) The trajectories
of three photons when asymmetry parameter g = 0.85, which is typical for clouds
in the solar band.

often changing directions quite sharply with each scattering. Even-

tually, its “drunkard’s walk” takes it back to the cloud top, where
it emerges and, in this case, contributes to the albedo of the cloud.
A different random turn at any point in its path could have instead
taken it to the cloud base, where it would have then contributed
to the diffuse transmittance. Note that because of the large opti-
cal depth, the direct transmittance of the layer is vanishingly small;
therefore there is virtually no chance that the photon could have
passed all the way through to cloud base without first being scat-
tered numerous times.
For isotropic scattering, the scattering source term in the radia-
tive transfer equation simplifies to



ω̃ ω̃
p(Ω̂ , Ω̂) I (Ω̂ ) dω → I (Ω̂ ) dω . (11.19)
4π 4π 4π 4π

That is, the source is independent of both Ω̂ and Ω̂ and is simply

equal to the single scatter albedo times the spherically averaged in-
tensity.
Scattering by real particles in the atmosphere is never even ap-
proximately isotropic. Nevertheless, because the assumption of
322 11. RTE With Scattering

isotropic scattering leads to important simplifications in the analytic

solution of the radiative transfer equation, it is frequently employed
in theoretical studies in order to gain at least qualitative insight into
the behavior of radiation in a scattering medium.
Furthermore, for some kinds of radiative transfer calculations,
it is possible to find approximate solutions to a problem involv-
ing nonisotropic scattering by recasting it as an equivalent isotropic
scattering problem, for which analytic solutions are easily obtained.
Such so-called similarity transformations will be discussed in a later
chapter.

11.3.2 The Asymmetry Parameter

In order compute scattered intensities to a high degree of accuracy,
it is necessary to specify the functional form of the phase function
p(cos Θ). As will be seen in Chapter 12, the phase functions of real
atmospheric particles can be complex and don’t lend themselves to
simple mathematical descriptions. Often, however, we don’t care
about intensities at all but only fluxes. In such cases, it is not neces-
sary to get bogged down with details of the phase function; rather,
it is sufficient to know the relative proportion of photons that are
scattered in the forward versus backward directions. The scattering
asymmetry parameter g contains this information and is defined as

1
g≡ p(cos Θ) cos Θ dω . (11.20)
4π 4π

The asymmetry parameter may be interpreted as the average value

of cos Θ for a large number of scattered photons. Thus

−1 ≤ g ≤ 1 . (11.21)

If g > 0, photons are preferentially scattered into the forward hemi-

sphere (relative to the original direction of travel), while g < 0 im-
plies preferential scattering into the backward hemisphere. If g = 1,
this is the same as scattering into exactly the same direction as the
photon was already traveling, in which case it might as well not
have been scattered at all! A value of −1, on the other hand, implies
 The Scattering Phase Function 323

an exact reversal of direction with every scattering event, a special

case that is imaginable but physically unlikely.
For isotropic scattering, as discussed in the previous subsection,
we expect g = 0, since scattering into the forward and backward
hemispheres is equally likely. This can be shown explicitly by sub-
stituting p = 1 into (11.20), expanding dω in spherical polar coordi-
nates as sin θdθdφ, and choosing Ω̂ = ẑ so that the scattering angle
Θ is the same as the zenith angle θ. Thus,

2π π/2
1
g= cos θ sin θ dθ dφ
4π 0 −π/2


1 π/2
= cos θ sin θ dθ
2 −π/2 (11.22)


1 1
= µ dµ
2 −1
=0.

Note that while g = 0 for isotropic scattering, other phase func-

tions can also have g = 0 and not be isotropic. The best example is
the Rayleigh phase function derived in section 12.2, which describes
the scattering of radiation by particles much smaller than the wave-
length.
For many problems of interest, such as scattering of solar radi-
ation in clouds, the asymmetry parameter g falls in the range 0.8–
0.9. In other words, cloud droplets are strongly forward scattering
at solar wavelengths. Fig. 11.1b shows examples of photon paths
for g = 0.85. Although the average distance traveled by a pho-
ton between scattering events is the same as for isotropic scatter-
ing (Fig. 11.1a), the photon is now far more likely to be scattered
into a direction that is not too different from its previous direction
of travel. As a result, the photon’s path, while still random, is far
less chaotic than the isotropic case. Statistically, the photon trav-
els a much greater distance before experiencing a sharp reversal in
course. It is therefore also more likely to reach the cloud base and
less likely to exit at cloud top. In other words, we expect the dif-
fuse transmittance to increase and the cloud-top albedo to decrease
when the asymmetry is large.
324 11. RTE With Scattering

Henyey-Greenstein Phase Function

100
g=0.8
g=0.6
10 g=0.4
g=0.2
g=0
P(cosΘ)

0.1

0.01
-1 -0.5 0 0.5 1
cosΘ

Fig. 11.2: The Henyey-Greenstein phase function plotted versus cos(Θ) (left) and
as a log-scaled polar plot (right).

11.3.3 The Henyey-Greenstein Phase Function

The scattering phase functions of particles are often rather compli-
cated (we will return to this subject in Chapter 12). As already
pointed out, it is not always necessary to use a complete and ac-
curate description of p(cos Θ) in a radiative transfer calculation, as
long as we know the asymmetry parameter g. For some types of
calculations, we might want to employ a “stand-in” phase function
that satisfies the following criteria:
• It should have a convenient mathematical form, ideally one
that is an explicit function of the desired asymmetry parame-
ter g.
• It should bear at least some resemblance to the shape of real
phase functions, even if it doesn’t have details like the rain-
bow, corona, etc. (See Chapter 12.)
• In order to be physically meaningful, the value of the phase
function should be nonnegative for all values of Θ.
The Henyey-Greenstein phase function is the most widely used
“model” phase function that satisfies all of the above criteria. It is
given by
1 − g2
pHG (cos Θ) = . (11.23)
(1 + g2 − 2g cos Θ)3/2
 Single vs. Multiple Scattering 325

As you can see from Fig. 11.2, the HG phase function is isotropic for
g = 0. For positive g, the function peaks increasingly in the forward
direction but remains quite smooth. In other words, it captures the
asymmetry of real phase functions rather well but not the higher
order details.

Problem 11.1: Show that the parameter g appearing in (11.23)

equals the asymmetry parameter as defined by (11.20).

Although the HG phase function with g > 0 does a good job

of reproducing the observed forward peak in the phase functions
of real particles, there is often also a pronounced (but somewhat
smaller) backward peak which is not captured. Therefore, you will
sometimes see the use of a double HG function, with one of the two
terms serving to represent the backward peak:

pHG2 (cos Θ) = bpHG (cos Θ; g1 ) + (1 − b) pHG (cos Θ; g2 ) , (11.24)

where g1 > 0, g2 < 0, and 0 < b < 1.

Problem 11.2: (a) Given g1 , g2 , and b, find the asymmetry parame-

ter g of the double Henyey-Greenstein phase function.
(b) For marine haze particles in the visible band, it has been
found that good values for the above parameters are b = 0.9724,
g1 = 0.824, and g2 = −0.55. Find g.
(c) Plot the phase function p(cos Θ) described in part (b), using a
logarithmic vertical axis.

11.4 Single vs. Multiple Scattering

When a solar photon enters the atmosphere or a cloud layer from
the top, it will eventually either exit again (top or bottom) or else
get absorbed. There are no other possibilities. Before either one
happens, however, the photon may experience anywhere from zero
to a very large number of scatterings from atmospheric particles.
326 11. RTE With Scattering

Recall that if the photon passes entirely through the layer with-
out getting either scattered or absorbed, then it is said to be directly
transmitted. The probability of this happening to any particular pho-
ton is given by the direct transmittance tdir , which we already know
how to compute from Beer’s Law. If, on the other hand, the photon
exits the layer after having been scattered at least once, then it con-
tributes to either the diffuse transmittance or the albedo of the layer,
depending on whether it exits at the bottom or the top, respectively.
It is helpful now to distinguish between two general classes of
problems: those in which single scattering dominates and those in
which multiple scattering is the rule. In the first case, almost all of
the photons contributing to the albedo and/or diffuse transmittance
were scattered exactly once. Single scattering prevails whenever the
layer is optically thin — i.e., τ ∗  1, because each photon that
is scattered in the interior of the layer then has a high probability
of exiting the cloud before getting scattered a second time. Single
scattering is also favored whenever the layer is strongly absorbing
(ω̃  1), since a photon is then much more likely to get absorbed
than to get scattered a second time.
If, on the other hand, the layer is both optically thick (τ ∗ > 1) and
strongly scattering (1 − ω̃  1), then many or most of the photons
that enter the layer will be scattered more than once, perhaps even
hundreds of times, before reemerging at the base or top of the layer.
It takes a fair amount of sophistication to solve multiple scattering
problems accurately. In fact, whole textbooks have been devoted to
just this subject. In this book, we will defer until Chapter 13 our
own fairly rudimentary treatment of radiative transfer with multi-
ple scattering.

RTE for Single Scattering

For now, let’s focus on the much simpler single scattering problem.
In the absence of thermal emission, (11.13) and (11.14) can be com-
bined to give

2π
1
dI (µ, φ) ω̃
µ = I (µ, φ) − p(µ, φ; µ , φ ) I (µ , φ ) dµ dφ .
dτ 4π 0 −1
(11.25)
 Single vs. Multiple Scattering 327

What makes the single scattering problem simple is that the inten-
sity I (µ , φ ) inside the integral is then, by definition, the attenuated
intensity from the direct source (e.g., the sun) with no significant
contribution from radiation that has already been scattered. If we
assume a parallel beam of incident radiation from a point source
above the cloud (a good approximation for direct sunlight), then
we can write
τ
I (µ , φ ) = F0 δ(µ − µ0 )δ(φ − φ0 )e µ0 , (11.26)

where µ0 < 0 and φ0 give the directions of the incident beam, F0

is the solar flux normal to the beam, and the exponential term is
just the direct transmittance from the layer top τ = 0 to level τ
within the cloud. The Dirac δ-function δ( x ) is defined to be zero
 ∞ all x = 0 and infinite
for  ∞ for x = 0, and
it is normalized so that
−∞ δ ( x ) dx = 1, and −∞ f ( x ) δ ( x ) dx = f ( x ).
With the above substitutions, (11.25) reduces to

dI F0 ω̃
µ = I− p(cos Θ)eτ/µ0 , (11.27)
dτ 4π
where the dependence of I on µ, φ, and τ is understood, and where
cos Θ ≡ Ω̂ · Ω̂0 is the cosine of the angle between the incident sun-
light and the direction of the scattered radiation.
Let’s rearrange the above equation and multiply through by
e−τ/µ :
dI −τ/µ F0 ω̃
e − Ie−τ/µ = − p(cos Θ)eτ/µ0 e−τ/µ . (11.28)
dτ 4πµ

The point of doing the above is so that the left hand side can then
be rewritten as a simple derivative of a single expression:

d  −τ/µ  F0 ω̃ τ µ1 − µ1
Ie =− p(cos Θ)e 0 . (11.29)
dτ 4πµ

In order to compute the scattered intensity emerging from the top

or bottom of the atmosphere, we just have to integrate the above
equation from τ = 0 to τ = τ ∗ . For simplicity, we will assume here
that that ω̃ and the phase function are independent of height, so that
328 11. RTE With Scattering

we can take them outside the integral. We get

  
∗
∗
− τµ − F0 ω̃ τ ∗ µ1 − µ1
I (τ )e − I (0) =  p(cos Θ) e 0 −1 .
4πµ µ10 + µ1
(11.30)
It may surprise you to learn that the above equation is valid both
for downwelling radiation at the bottom of the atmosphere and for
upwelling radiation at the top of the atmosphere. In the first case,
we’re interested in I (0) for the case that µ > 0:
  
−τ
∗
F ω̃ τ∗ 1 − 1
I (0) = I ( τ ∗ ) e µ + 0  p(cos Θ) e µ0 µ − 1 .
4πµ µ10 + µ1
(11.31)
∗
In the second case, we want I (τ ) for µ < 0, which requires only a
slight rearrangement:
 ∗ 
∗
τ∗ F 0 ω̃ τ τ∗
I ( τ ) = I (0) e µ +  p(cos Θ) e 0 − e µ . (11.32)
µ

4πµ µ10 − µ1

To summarize, the above equations give scattered radiances at the

top and bottom of the atmosphere (or a thin cloud layer) for the spe-
cial case that all of the above are satisfied: (a) multiple scattering is
negligible, (b) ω̃ and p(cos Θ) are constant, and (c) the sole external
illumination is a parallel beam source such as the sun. It’s important
to recall the requirement that either ω̃  1 and/or τ ∗  1 in order for the
first of these requirements to be satisfied.
Let’s take things a step further. First, we’ll focus only on the scat-
tered atmospheric contribution to the radiance and drop the term
that describes the direct transmission of radiation from the opposite
side of the atmosphere (we can always add it back, if we want it).
Second, we’ll assume that the reason why we can neglect multiple
scattering is that τ ∗  1, and we’ll further assume that µ0 and µ are
not much smaller than one. Taking advantage of the fact that, for
small x, e x ≈ 1 + x, we can then simplify our equations to

*
For µ > 0, I (0) F0 ω̃τ ∗
= p(cos Θ) . (11.33)
For µ < 0, I (τ ∗ ) 4πµ
 Applications 329

The interpretation of the above equation is straightforward — so

straightforward in fact, that we probably could have guessed it
without going through all the previous steps. First, the quantity
F0 τ ∗ tells us the magnitude of the extinguished solar flux (recall
that this is valid only in the limit of small τ ∗ ). Second, the quan-
tity (ω̃/4π ) p(cos Θ) tells us how much of the intercepted flux con-
tributes to the scattering source term in a given new direction Ω̂.
Finally, the factor 1/µ accounts (to first order) for the fact that you
are looking through less atmosphere if you view it vertically than if
you look toward the horizon; consequently the path-integrated con-
tribution of scattering to the observed intensity increases toward the
horizon.

Problem 11.3: If you have access to a decent plotting program, set

things up so that you can conveniently plot I (τ ∗ ) versus −1 < µ < 0
using both (11.33) and (11.32) on the same graph. Assume no ex-
traterrestrial source of radiation from direction µ = µ0 . Assume
isotropic scattering. Determine the range of µ, µ0 , and τ ∗ for which
the second equation is a good approximation to the first. When the
two disagree significantly, describe the nature of the disagreement.
Focus on values of τ ∗ ≤ 0.1, since we know that neither equation is
valid unless the atmosphere is optically thin. Note also that (11.32)
cannot be directly evaluated when µ0 = µ, though it gives physically
reasonable values in the limit as µ → µ0 .

11.5 Applications to Meteorology, Climatology,

and Remote Sensing
11.5.1 Intensity of Skylight
We imposed several seemingly drastic restrictions in deriving
(11.33): τ ∗  1, ω̃ and p(cos Θ) independent of τ, µ and µ0 not too
small. In fact, these assumptions are reasonably well justified for
molecular scattering of visible and near-IR sunlight in the cloud-
and haze-free atmosphere, as long as (a) you stay away from the
blue and violet end of the spectrum, and (b) you don’t get too close
to the horizon.
330 11. RTE With Scattering

Therefore, to evaluate the radiant intensity of the sky (apart from

the direct rays of the sun itself) you need only specify the optical
depth τ ∗ of the cloud-free atmosphere at the wavelength in ques-
tion, supply a suitable phase function p(Θ), and substitute these
into (11.33) for arbitrary µ and µ0 .
As will be shown in Chapter 12, the scattering phase function of
air molecules in the visible band is

3
p(Θ) = (1 + cos2 Θ) . (11.34)
4

This so-called Rayleigh phase function is quite smooth and is per-

fectly symmetric with respect to forward and backward scattering
(g = 0). The factor-of-two variation in intensity implied by the
above phase function is relatively minor and is unlikely to be ob-
vious to the eye, especially since it is such a smooth function of the
scattering angle Θ. Consequently, we expect the radiant intensity
of the sky to appear rather uniform, punctuated only by the narrow
spike of high intensity associated with the directly transmitted light
of the sun.
Although p(Θ) has the same shape for molecular scattering at
all visible wavelengths, the optical depth τ ∗ of the cloud free atmo-
sphere is a strong function of wavelength. In fact, it is shown in the
next chapter that τ ∗ ∝ λ−4 . Thus, (11.33) implies that the intensity
of skylight due to molecular scattering should also be proportional
to λ−4 and, indeed, it is precisely this dependence that gives us the
blue sky. It is also because τ ∗ stops being “small” at shorter wave-
lengths that we can’t trust (11.33) to give us accurate sky intensities
in the blue and ultraviolet part of the spectrum.
Of course, even the cleanest air found in nature contains not only
molecules, but other kinds of particles called aerosols. There are
typically many thousands of aerosol particles in every cubic cen-
timeter of air. Those of interest to us here have sizes ranging from
10−2 µm to ∼1 µm or larger. The scattering of visible light by such
comparatively large particles (compared to molecules, that is!) is
not as strongly dependent on wavelength as is molecular scatter-
ing; furthermore the scattering phase function for aerosols is not
symmetric like the Rayleigh phase function but rather exhibits fairly
strong forward scattering.
 Applications 331

We can summarize the comparative scattering behavior of air

molecules and aerosols in the solar band as follows:
Molecules Aerosol

Wavelength
λ −4 weak
dependence:

p ( Θ ): smooth, symmetric strongly asymmetric

Time/location
nearly constant highly variable
dependence:

Problem 11.4: Based on the above information, explain how the

presence of scattering aerosols (e.g., haze) would be expected to visi-
bly affect both (a) the color of the sky and (b) the angular dependence
of the intensity of scattered sunlight. Is your analysis consistent with
everyday experience?

11.5.2 Horizontal Visibility

Every hour, at tens of thousands of locations around the globe, de-
tailed weather observations are made by trained observers or auto-
mated weather instruments. It is no coincidence that a large major-
ity of these stations are associated with airports. It was the need for
timely local weather observations in support of aviation, more than
any other single factor, that led to the emergence of a dense global
weather observing network during the twentieth century.
Although pilots care about a lot of weather variables, the two
that are most often of critical concern are (a) cloud ceiling height and
(b) horizontal visibility. Both affect pilots’ ability to safely land at
airports and to see and avoid other air traffic. It is the latter variable
we will address here, since it is closely tied to the subject of this
chapter.
Visibility is defined as the maximum horizontal distance over
which the eye can clearly discern features like runways, obstacles,
navigation lights, etc. On a clear day in the desert, visibility often
332 11. RTE With Scattering

exceeds 100 km. But in a pea-soup fog on the California coast, visi-
bility may be measured in meters rather than kilometers.
On first confronting this problem, your initial assumption might
be that visibility is controlled entirely by the extinction coefficient β e
along the line-of-sight. After all, the transmittance over a distance s
is just
t = e− βe s . (11.35)
One might argue, therefore, that there is some minimum transmit-
tance tmin associated with the limit of human perception, so that the
visibility V should be related to β e as follows:

1
V=− log(tmin ) . (11.36)
βe

But such an analysis is too simple. Consider the following exam-

ples:

• Translucent (“two-way”) mirrors are often used in department

stores to facilitate the detection of shoplifters by security per-
sonnel. The transmittance is the same for light traveling in ei-
ther direction through the mirror, but a shopper in a brightly
lit room viewing the mirror from the reflective side can’t nor-
mally see what’s on the other side and probably doesn’t even
realize that it transmits at all. The person (or camera) viewing
from the nonreflective side, however, can see through easily,
especially if they are situated in a darkened room.

• If you let the windshield on your car get moderately dusty, its
transmittance is somewhat reduced, but normally this reduc-
tion is fairly minor. In fact, when driving away from the sun
during daytime, it may be scarcely noticeable. But if you turn
in the direction of the setting sun, you may suddenly find it al-
most impossible to see! What has changed? Not the transmit-
tance, but rather the glare of light scattered in your direction
by the coating of dust particles.

From the above examples, we can perhaps begin to appreciate that

it’s not transmittance but rather visual contrast that determines what
we can and can’t see. We will define the contrast here as the frac-
tional difference between the apparent brightness (radiant intensity)
 Applications 333

I of an object and the brightness I of its surroundings:

I − I
C≡ . (11.37)
I
In a purely absorbing atmosphere, a mere reduction in trans-
mittance along a line-of-sight has no impact on the visual contrast
between two objects at the same distance, and therefore relatively
little on visibility (up to the limit imposed by your eyes’ sensitivity
to light), as long as the fractional reduction in brightness is the same
for both.
Atmospheric scattering reduces contrast by adding a source of
radiation to the line-of-sight that is independent of the brightness of
whatever is at the far end of the path. Since this source is integrated
along the line-of-sight, a long path produces a greater reduction in
contrast than a short path. The distance at which the contrast of an
object is reduced to the minimum level required for visual detection
defines the visibility.
Let’s analyze the visibility problem quantitatively, by consid-
ering the contribution of single-scattered radiation to the radiance
along a finite horizontal path s. Because of the latter condition, we
can’t use the plane-parallel form of the RTE but must start with an
adaptation of (11.9):
dI
= −I + J , (11.38)
d( β e s)
where I is the intensity measured horizontally in azimuthal direc-
tion φ, J is the scattering source function given by


ω̃
J= p(µ0 , φ0 ; 0, φ) I (Ω̂ ) dω , (11.39)
4π 4π

and s is the distance in the direction toward the observer.

For this problem, we can assume a horizontally homogeneous
atmosphere, so that both the extinction coefficient β e and the scat-
tering source function J are constant along the line-of-sight. With
these assumptions, we can integrate (11.38) to get

I ( S ) = I (0) e − β e S + 1 − e − β e S J , (11.40)

where I (0) is the “intrinsic” radiance of the remote scene as seen

without any intervening atmosphere, and I (S) is the brightness of
334 11. RTE With Scattering

the same scene at the observer’s distance S. We see that the ob-
served intensity is just a weighted average of the intrinsic bright-
ness of the distant object and the scattering source function, with
the weight being the path transmittance t = e− βe S for the first term
and 1 − t for the second. Obviously, if t = 0, we see only the atmo-
spheric scattering and no trace of the object at s = 0.

Problem 11.5: Fill in the steps of the derivation of (11.40) from

(11.38). Hint: Multiplying both sides by an integrating factor e βe s
will allow you to recast the differential equation into a form that can
be directly integrated.

Now let’s use the above equation to compute the contrast of a

black object with I (0) = 0 viewed against a white background with
intensity I (0):

I (S) − I (S) I (0) t

C= = . (11.41)
I (S) I (0) t + (1 − t ) J
We are interested in the distance S corresponding to the minimum
contrast that still permits the human eye to distinguish the object
from its background, so we invert the above equation to get
 
1 I (0)(1 − C )
S= ln +1 . (11.42)
βe CJ

Now all that is left is to make reasonable assumptions about I (0),

C, and J.
For the background, we assume an intensity I (0) = αF0 , where
α depends on the reflective properties of the background for the par-
ticular viewing geometry and direction of the incident sunlight. For
example, if the background is a nonabsorbing Lambertian reflector,
then α ≤ 1/π, with the equality applying in the case of normal solar
incidence.
As before, we’ll assume that the atmosphere is optically thin in
the vertical and that the sun is high in the sky, so the scattering
source function can be approximated as
F0 ω̃
J≈ p(µ0 , φ0 ; 0, φ) , (11.43)
4π
 Applications 335

where µ0 is the cosine of the solar zenith angle. We’ll assume that
the phase function can be expressed in terms of the cosine of the
scattering angle alone, with

cos Θ = Ω̂0 · Ω̂
 
= 1 − µ20 cos ∆φ, 1 − µ20 sin ∆φ , µ0 · (1, 0, 0) (11.44)

= 1 − µ20 cos ∆φ

where ∆φ = φ − φ0 is the angle between the viewing azimuth and

the solar azimuth.
We can now substitute the above expressions for J and I (0),
with α ≈ µ0 /π and C ≈ 0.02, to get
 
1 200µ0
S≈ ln +1 . (11.45)
βe ω̃ p(cos Θ)

Problem 11.6: Use (11.45) together with the phase function for ma-
rine haze given in Problem 11.2 to plot the visibility in km versus
azimuth ∆φ relative to the sun’s direction for two cases: µ0 = 1 and
µ0 = 0.5. For both cases assume β e = 1.0 km−1 . Explain the differ-
ences between the two curves. Are your results consistent with your
experience?
 CHAPTER 12

Scattering and Absorption By Particles

In the previous chapter, we introduced the mathematical frame-

work and terminology needed to account for radiative scattering
in the atmosphere. It is safe to say that whenever you find yourself
struggling with a thorny problem involving radiative scattering at
microwave and shorter wavelengths, some kind of particles are to
blame, whether they be molecules or hailstones.1
Formally, the scattering component of the radiative transfer
equation (11.9) depends on the local extinction coefficient β e (since
dτ = β e ds), single scatter albedo ω̃ and the scattering phase func-
tion p(cos Θ). These in turn depend both on wavelength and on the
size, composition, shape and number of suspended particles, in ad-
dition to any absorption contributions by atmospheric gases. The
purpose of this chapter is to examine some basic aspects of the rela-
tionship between a particle’s physical and geometric properties and
its absorption and scattering properties.

1 Weak scattering can also occur at radio wavelengths due solely to turbulent

fluctuations in the index of refraction of air and/or due to the presence of electri-
cally conducting ionized gases.

336
 Atmospheric Particles 337

Table 12.1: Examples of atmospheric particle types, with representative dimen-

sions and number concentrations. Note that actual values can vary far more widely
than indicated here.

Type Size Number

Gas molecule ∼10−4 µm < 3 × 1019 cm−3

Aerosol, Aitkin < 0.1 µm ∼104 cm−3
Aerosol, Large 0.1–1 µm ∼102 cm−3
Aerosol, Giant >1 µm ∼10−1 cm−3
Cloud droplet 5–50 µm 102 –103 cm−3
Drizzle drop ∼100 µm ∼103 m−3
Ice crystal 10–102 µm 103 –105 m−3
Rain drop 0.1–3 mm 10–103 m−3
Graupel 0.1–3 mm 1–102 m−3
Hailstone ∼1 cm 10−2 –1 m−3
Insect ∼1 cm <1 m−3
Bird ∼10 cm <10−4 m−3
Airplane ∼10 m <1 km−3

12.1 Atmospheric Particles

12.1.1 Overview
The variety of particles encountered in the atmosphere is enor-
mous. Examples include individual gas molecules, haze, smoke,
dust and pollen particles, cloud droplets and ice crystals, rain drops,
snowflakes, hailstones, insects, birds, and airplanes. Every one of
these examples has at least some practical significance as a scatterer
of EM radiation in the atmosphere.2 Table 12.1 gives representa-
tive dimensions and number concentrations for some common at-
mospheric particles.
For the scattering of radiation by particles, size matters. The size
of a particle is in fact its most important defining characteristic. In
general, particles that are far smaller than the wavelength will scat-
ter only very weakly, though they may still absorb radiation (e.g., the
gas molecules discussed in Chapter 9). We will revisit the question
of what “far smaller” means in a moment.
2 The last three of these are significant mainly for radar.
338 12. Scattering and Absorption By Particles

At the other extreme, if the particle is very large compared to the

wavelength of the radiation, then the laws of reflection, refraction,
and absorption presented for homogeneous media in Chapter 4 can
be used to evaluate σe , ω̃, and p(Θ) for the particle via the approxi-
mate technique known as ray-tracing or geometric optics.3
Unfortunately, many particles in the atmosphere fall in between
the two extremes cited above. For these particles, more complex
methods are needed in order to compute their scattering and ab-
sorption properties. Such methods generally have to consider the
effects of diffraction, constructive and destructive interference and
other wave-related phenomena.
In this book, we will discuss only those methods applicable
to very small randomly oriented particles (Rayleigh theory) or to
spheres of arbitrary size (Mie theory). Fortunately, a great many
atmospheric particles, from molecules to haze droplets to cloud
droplets to rain drops to hailstones, are reasonable (though not al-
ways perfect) candidates for one or both of these methods, so we
can cover a fair amount of ground.

12.1.2 Relevant Properties

As already mentioned, the relationship between the size of a parti-
cle and the wavelength of the radiation of interest is of crucial im-
portance to particle’s optical properties as well as to the choice of a
suitable method for calculating those properties. We therefore de-
fine the nondimensional size parameter as

2πr
x≡ , (12.1)
λ

where r is the radius of a spherical particle. In the case of nonspher-

ical particles, r might represent the radius of a sphere having the
same volume or surface area, depending on the context.
3 Even for large particles, geometric optics gives results that are seemingly at

odds with exact theories. The discrepancy is due to the inability of ray tracing
alone to account for subtle bending of light waves passing near the particle. How-
ever, because the bending is slight, it is often acceptable to treat this radiation as if
it had never been scattered at all, in which case the geometric optics approximation
yields perfectly acceptable results.
 Atmospheric Particles 339

Microwave
Thermal IR
Near IR
Visible
UV

s
1 cm p tic Hail
ic O
etr 00
1 mm om 20 Raindrops
Ge x=
100 µm ri ng Drizzle
tte
Particle Radius

ca
eS
10 µm Mi Cloud droplets

1 µm

0.1 µm
Ra
x=

yle
0.2

igh
Sc

x=0
a

.0
tte
ring

02
ring
} Dust,
Smoke,
Haze

Aitken Nuclei
tte
ca
10 nm
leS
ligib
1 nm N eg
Air Molecules

0.1 µm 1 µm 10 µm 100 µm 1 mm 1 cm 10 cm
Wavelength

Fig. 12.1: Relationship between particle size, radiation wavelength and scattering
behavior for atmospheric particles. Diagonal dashed lines represent rough bound-
aries between scattering regimes.

Given the value of x, one can immediately determine whether

scattering by the particle is likely to be significant and, if so, which
broad scattering regime — Rayleigh, Mie, or geometric optics — is
most applicable. Figure 12.1 shows how various combinations of
particle type and EM wavelength relate to these regimes.

Another key property is the relative index of refraction m, which

340 12. Scattering and Absorption By Particles

was defined in (4.19) as

N2
m≡ ,
N1

where N1 and N2 are complex refractive indices of the particle and

the surrounding medium, respectively. At the risk of slightly over-
simplifying, the real part nr = ( N ) governs the phase speed of
propagation of a wave within the material and the imaginary part
ni = ( N ) governs absorption. N1 is usually taken to be equal to
one for particles suspended in air, so that m ≈ N2 . N2 depends on
both the composition of the particle and on the wavelength. The
dependence of the refractive index of water and ice on wavelength
was shown in Fig. 4.1.
Finally, the shape of a particle may potentially play a large role
in determining its radiative properties. It is convenient, and there-
fore common, to assume that particles are spheres for radiative
purposes, even when this assumption is not entirely appropriate.
Ice crystals, snowflakes, and solid-phase aerosols (e.g., soot) are
good examples of particles that are far from spherical and there-
fore should really not be treated as such, at least not without promi-
nently posted disclaimers. Unfortunately, computational methods
appropriate for nonspherical particles are far more difficult to work
with and have only recently come into common use with the advent
of fast computers. We will not consider them here.

12.2 Scattering by Small Particles

12.2.1 Dipole Radiation
When a particle is sufficiently small relative to the wavelength —
i.e., |m| x  1 — every part of the particle simultaneously expe-
riences the same externally imposed oscillating electric field. The
response of the particle to the electric field is to become partially
polarized. That is, there is a small displacement of positive charge
within the particle in the direction of the electric field vector, while
there is a displacement of negative charge in the opposite direction.
In short, it becomes an electric dipole, with induced dipole moment
 Scattering by Small Particles 341

p. The physical dimensions of
p are charge times distance, which

can be interpreted as the net amount of charge Q displaced times an
effective displacement
x.
For most particles of interest to us, the dipole moment of a small
spherical particle is proportional to the strength of the external elec-
tric field:

p = α
E0 exp(iωt) , (12.2)
where α is called the polarizability of the particle. It depends on the
composition and the size of the particle, as well as on the frequency
ω = 2πν of the incident wave. Note that α may be complex. Any
nonzero imaginary part implies a phase difference between the real
part of
p and the real part of
E.
In summary, we have an oscillating dipole whose strength and
orientation fluctuates in lockstep with the electric field due to the
incident wave. But an oscillating dipole produces its own oscillating
electric field, and these oscillations propagate outward at the speed
of light. This is of course the origin of the scattered radiation.
Now imagine that the incident wave is traveling in direction Ω̂
and you are positioned at a large distance R  r from the dipole, in
direction Ω̂ . There are several facts we can jot down that will aid
us in visualizing the relationship between the scattered wave at our
location and the incident wave:

1. We know that in any EM wave, the electric field vector is per-

pendicular to the direction of propagation Ω̂.

2. We are assuming here that
p is aligned with the electric field

E0 of the incident wave,4 so
p is also perpendicular to Ω̂.

3. Because of the symmetry of the charge distribution in the

dipole, the electric field vector
Escat of the scattered wave at
any location must lie in the plane that contains both
p and Ω̂ .

4. The strength of the electric field at your location is proportional

to the projection of
p as seen from your direction. Specifically,
4 Inother words, we are assuming that the polarizability α is a scalar rather than
a 3 × 3 tensor which would alter the direction of
p relative to
E0 . This is always
valid for spherical particles composed of an electrically isotropic substance like
water.
342 12. Scattering and Absorption By Particles

Escat is zero if you are viewing the dipole “end on” and it is
a maximum (for a given distance) when you are viewing it at
right angles. We can put this in mathematical terms by saying
that
Escat ∝ sin γ, where γ is the angle between
E0 (using Fact
2, above) and the scattered direction Ω̂ .

5. Less obvious, but equally important, is that the power radi-

ated by the dipole is proportional to the acceleration of the elec-
tric charge in the dipole. That is to say, a stationary dipole will
create a static electric field but no propagating EM wave, and
it will therefore radiate no energy. A vibrating dipole, on the
other hand, induces a vibrating electric field (and therefore an
outward propagating EM wave) whose amplitude is propor-
tional to the square of the frequency of the vibration.

Facts 4 and 5 together, combined with (12.2), give us the follow-

ing proportionality:

∂2
p
|
Escat | ∝ 2 sin γ ∝ ω 2 sin γ . (12.3)
∂t

As discussed in section 2.5, the power per unit area, and therefore the
intensity I, is proportional to the square of the electric field ampli-
tude. Therefore, the scattered intensity is given by the following
proportionality:
I ∝ ω 4 sin2 γ . (12.4)

We now want to recast the above proportionality in terms of the

scattering angles Θ and Φ, where Θ is the angle between Ω̂ and
Ω̂ , and Φ is the polar angle about Ω̂ measured from an arbitrary
starting point.
For convenience, we let the direction of incidence Ω̂ coincide
with the x-axis, and the incident electric field vector
E0 be aligned
with the z-axis, consistent with Fact 2, above. We can then expand
Ω̂ and Ω̂ in Cartesian coordinates as follows:

Ω̂ = (1, 0, 0) , (12.5)

Ω̂ = (cos Θ, sin Θ sin Φ, sin Θ cos Φ) . (12.6)

 Scattering by Small Particles 343

90o

180o 0o

o
270

Fig. 12.2: Polar plot of the phase function for scattering by small particles (Rayleigh
scattering). The outermost curve (dashed) represents the scattered intensity for
directions Ω̂ lying in a plane perpendicular to the electric field vector of the incident
wave. The innermost curve (dot-dashed) corresponds to directions lying in a plane
parallel to the electric field vector. The solid curve represents the scattered intensity
for unpolarized incident radiation, as given by (12.10).

This allows us to write

cos γ = ẑ · Ω̂
= (0, 0, 1) · Ω̂ (12.7)
= sin Θ cos Φ ,
and
sin2 γ = 1 − cos2 γ = 1 − sin2 Θ cos2 Φ . (12.8)
Substituting into (12.4) gives

I ∝ ω 4 (1 − sin2 Θ cos2 Φ) . (12.9)

The above equation contains all of the essential features of what we

will henceforth refer to as Rayleigh scattering. Before we continue,
let’s take a moment to interpret this result:
344 12. Scattering and Absorption By Particles

• The intensity of scattered radiation is proportional to the

fourth power of the frequency of the incident radiation, assum-
ing that the polarizability α is not a strong function of fre-
quency (this may or may not be true for any given particle).
You should mentally file this piece of information; we will re-
turn to it later.

• For Φ equal to either 90◦ or 270◦ — in other words, for any

scattered ray lying in the plane perpendicular to
E0 , the scat-
tered intensity is both constant and at its maximum value, ir-
respective of Θ (the outermost curve in Fig. 12.2).

• For Φ equal to either 0◦ or 180◦ and Θ = 90◦ — in other words,

for either of the two directions along the axis of the dipole
— the scattered intensity is zero (see the innermost curve in
Fig. 12.2).

12.2.2 The Rayleigh Phase Function

The complete shape of the Rayleigh phase function for polarized
incident radiation is shown in the top two panels of Fig. 12.3. For
unpolarized incident radiation, the phase function p(Θ) is obtained
by averaging (12.9) over Φ and normalizing according to (11.7) to
get

3
p(Θ) = (1 + cos2 Θ) . (12.10)
4

The above expression is the one that we normally regard as de-

scribing the scattering phase function of very small particles. It is
depicted as the solid curve in Fig. 12.2 and in the bottom panel of
Fig. 12.3.

Problem 12.1: Verify the derivation of (12.10) from (12.9).

 345

Fig. 12.3: Three-dimensional rendering of the Rayleigh phase function. The vector
Ω indicates the direction of the incident radiation. The vector E indicates the ori-
entation of the electric field vector in the incident wave. Top: Incident radiation is
vertically polarized. Middle: Incident radiation is horizontally polarized. Bottom:
Incident radiation is unpolarized.
346 12. Scattering and Absorption By Particles

Problem 12.2: From (12.10), show that the asymmetry parameter g

for Rayleigh scattering is zero.

12.2.3 Polarization
Equation (12.10) assumes that the incident radiation is unpolarized.
But note that even if unpolarized light is incident on the particle, the
scattered light will, in general, be polarized. You can convince yourself
of this by imagining yourself positioned in the direction indicated
by the ray labeled E in the middle panel of Fig. 12.3, which is to say,
at a 90◦ angle to a horizontal beam of incident radiation (e.g., from
the setting sun). At this position, you will observe no scattering
of the horizontally polarized component of the incident radiation.
You will, however, observe the maximum amount of scattering of
the vertically polarized component of the incident radiation. More-
over, the scattered radiation in the latter case will itself be vertically
polarized.
In short, viewing in any direction at 90◦ to the incident beam,
you will see scattered radiation that is completely polarized. At
most other angles, the scattered radiation is partially polarized, be-
cause neither component is zero. Only in the forward direction
(Θ = 0◦ ) or the backward direction (Θ = 180◦ ), is the degree of
scattering the same regardless of the polarization of the incident ra-
diation; hence, unpolarized incident radiation gives rise to unpo-
larized scattered radiation in these two directions. In general, the
degree of polarization of Rayleigh-scattered radiation is given by

1 − cos2 Θ
P= . (12.11)
1 + cos2 Θ

Radiation from the cloud- and haze-free sky is dominated by

Rayleigh scattering by air molecules. According to the above equa-
tion, skylight will be unpolarized when looking directly toward or
away from the sun, and 100% polarized when viewing the sky at a
90◦ angle from the sun.
 Scattering by Small Particles 347

In reality, the inevitable presence of aerosols, which are much

larger than molecules and don’t satisfy the Rayleigh criterion, re-
duce the polarization somewhat. Also, multiple scattering, which is
weak but not negligible in this instance, further reduces the polar-
ization slightly.
Nevertheless, wearing a pair of polarized sunglasses, you can
easily verify the above effect by viewing a portion of the blue sky
at right sangles from the sun and rotating the sunglasses (or your
head) about the line-of-sight. The sky will appear darker or lighter,
depending on whether the sunglasses transmit or block the po-
larized radiation. The bluer the sky (and therefore the less haze
present) the more pronounced the effect will be.

12.2.4 Scattering and Absorption Efficiencies

We were able to infer the scattering phase function for small parti-
cles based on relatively simple handwaving arguments. Let’s now
turn to the question of how much radiation a small particle scatters
and/or absorbs. While it is possible to obtain this information di-
rectly based on the dipole model we developed above (see BH83,
section 5.2), the complete derivation requires more space and ex-
planation than seems warranted at this introductory level. Among
other things, it would be necessary to explain the relationship be-
tween the (relative) complex index of refraction m of the particle
and its polarizability α, as well as to show how the imaginary part
of α bears on absorption of the incident electromagnetic wave by the
particle.
An alternate way of getting at the same information is to take
the general Mie solutions for spheres of arbitrary size, which I will
briefly discuss in section 12.3, and find limiting expressions for
x  1. Specifically, you rewrite the solutions as power series in
x and discard all but the first few terms. Here, I will give you the
essential results without going through the derivations (see BH83,
section 5.1).
348 12. Scattering and Absorption By Particles

General Relationships

To terms of order x4 , the extinction and scattering efficiencies, re-

spectively, of a small spherical particle are
    
m2 − 1 x2 m2 − 1 m4 + 27m2 + 38
Qe = 4x  1 +
m2 + 2 15 m2 + 2 2m2 + 3
 2 *
8 4 m2 − 1
+ x  , (12.12)
3 m2 + 2

and
 2
8 4  m2 − 1 
Qs = x  2 . (12.13)
3 m + 2

The absorption efficiency is then Qa = Qe − Qs . For sufficiently

small x (see BH83, p. 136 for details), Qa simplifies to

 
m2 − 1
Qa = 4x  . (12.14)
m2 + 2

We see that the absorption efficiency Qa is proportional to x, while

the scattering efficiency Qs is proportional to x4 . It follows that, for
sufficiently small x, and assuming that m has a nonzero imaginary
part,

Qs  Qa ≈ Qe , (12.15)

and the single scatter albedo

Qs
ω̃ ≡ ∝ x3 . (12.16)
Qe

The above relationships have a number of important practical im-

plications for atmospheric radiation and remote sensing. We will
highlight a few of these here.
 Scattering by Small Particles 349

Scattering Cross-Section

First of all, if we assume that we’re in a part of the spectrum where

m for our particle varies slowly with wavelength, then according
to (12.13) the scattering efficiency Qs of a particle in the Rayleigh
limit is proportional to x4 , which is in turn proportional to (r/λ)4
or, equivalently, to (rν)4 . [Recall that we already saw this propor-
tionality in (12.9), since ω ≡ 2πν.] The scattering cross-section σs ,
which is what actually determines how much radiation is scattered,
is of course the product of Qs with the particle cross-sectional area
πr2 , so that

r6
σs ∝ . (12.17)
λ4

This proportionality is well worth memorizing, as long you also remem-

ber that it’s only valid in the Rayleigh regime — i.e., for x  1.

Single Scatter Albedo

The second relationship worth commenting on is (12.16), which tells

us that the single scatter albedo for small particles goes with x3 . Of
course this is only true for particles that are at least slightly absorb-
ing; if the imaginary part of m is zero, then ω̃ = 1 no matter how
small x.
What this means in practice is that for sufficiently small x you
can pretty much forget about scattering and focus instead on just
the absorption properties of the particles. This limiting behavior
arises in at least two important cases: (1) molecular absorption (but
not scattering) of thermal infrared radiation by atmsospheric gases
(see Chapter 9), and (2) absorption (but not scattering) of microwave
radiation by cloud droplets.

Mass Absorption Coefficient

Not only can you forget about scattering in the limit of small x, but a
surprisingly convenient fact emerges concerning absorption by par-
ticles in this limit. Recall that the mass absorption coefficient ka of
a substance is defined as its absorption cross-section per unit mass.
350 12. Scattering and Absorption By Particles

For a spherical particle of radius r and density ρ, we can write

Qa πr2 3Qa
ka = = . (12.18)
ρ(4/3)πr3 4ρr

Substituting (12.14) and (12.1), we have

 
6π m2 − 1
ka =  . (12.19)
ρλ m2 + 2

Note that there is no dependence here on the particle radius r!

Imagine a volume V of air containing a number of spherical par-
ticles (e.g., cloud droplets) which may be of various sizes but which
are in any case all much smaller than the wavelength of interest.
We can write the volume absorption (≈ extinction) coefficient (di-
mensions of inverse length) in terms of the sum of the particles’
individual absorption cross-sections σ as follows:

1
βa =
V ∑ σi . (12.20)
i

But σi = ka Mi , where Mi is the mass of the droplet, so we have

1 1
βa =
V ∑ k a Mi = k a V ∑ Mi (12.21)
i i

or, quite simply,

βa = ka ρ , (12.22)

where ρ is just the combined mass of the substance (e.g., cloud wa-
ter) per unit volume of air. We therefore conclude that for radiation
passing through a cloud of sufficiently small absorbing particles, the total
absorption is equal to ka [as given by (12.19)] times the total mass path,
regardless of the exact sizes of the constituent particles.

Summary
Let us conclude this section by summarizing some key facts about
scattering and absorption in the Rayleigh regime:
 Scattering by Spheres — Mie Theory 351

1. If you have a particle of fixed size and expose it to radiation

with two different wavelengths λ1 < λ2 , then it will scatter the
shorter wavelength λ1 more strongly by a factor of (λ2 /λ1 )4 .

2. If you have radiation of a fixed wavelength λ and use it to

illuminate two particles of radius r1 < r2 , the larger particle
will scatter the radiation more strongly by a factor (r2 /r1 )6 .

3. For sufficiently small particles with complex (not pure real)

refractive index m, scattering is negligible and absorption is
proportional to mass path only, irrespective of particle size. In
this limit, a cloud behaves radiatively like an absorbing gas
rather than a collection of discrete scatterers.

The first of these facts is directly responsible for the blue sky and
the reddish setting sun. The second fact is of central importance to
weather radar. The third is relevant to microwave remote sensing of
cloud water. We will revisit each of these topics in the Applications
section at the end of this chapter.

12.3 Scattering by Spheres — Mie Theory

A brief outline of Mie theory for scattering and absorption by ho-
mogeneous spheres of arbitrary size parameter x and relative in-
dex of refraction m is given by S94 (pp. 235–243). Full derivations
are given by BH83 (pp. 82–107) and L02 Section 5.2. In a nut-
shell, the Maxwell equations are used to derive a wave equation
for electromagnetic radiation in three dimensional space, and these
are expressed in spherical polar coordinates (r, φ, Θ), with appro-
priate boundary conditions at the surface of the sphere. The result
is a separable partial differential equation, the solution of which is
expressed as an infinite series of products of orthogonal basis func-
tions, including sines and cosines (for the dependence on φ), spher-
ical Bessel functions (for the dependence on r), and associated Leg-
endre polynomials (for the dependence on cos Θ).
If you have already had a course in partial differential equations,
then that last sentence will be at least vaguely intelligible. The bot-
tom line is that the extinction and scattering efficiencies of a sphere
352 12. Scattering and Absorption By Particles

may be written as
∞
2
Qe =
x2 ∑ (2n + 1)(an + bn ) , (12.23)
n =1
∞
2
Qs =
x2 ∑ (2n + 1)(|an |2 + |bn |2 ) , (12.24)
n =1
where the coefficients an and bn are referred to as Mie scattering
coefficients and are functions of x and m. The mathematical form
of these coefficients is not particularly informative to the untrained
eye, so they will not be reproduced here.
Similar summations are used to describe the wave scattering
amplitudes as a function of scattering angle Θ. These are used to
obtain expressions for the elements of the 4 × 4 scattering phase ma-
trix Pij (Θ) (see 11.2.2). The P11 element of this matrix is the same as
our scalar phase function p(Θ) for unpolarized incident light.
As a practical matter, one cannot actually compute an infinite
sum; therefore it is always necessary to truncate the series and keep
only enough terms to yield a sufficiently accurate approximation.
Generally speaking, the required number of terms N is a little larger
than x; the criterion developed by BH83 based on extensive testing
is that N should be the integer closest to x + 4x1/3 + 2. For a typical
cloud droplet of 10 µm radius and a visible wavelength of 0.5 µm,
the size parameter x ≈ 120; thus the number of terms required in
the summation is 127.
For much larger particles (e.g. raindrops in the visible band,
with x ∼ 104 or more) the number of terms that must be retained is
rather large. Although the computer time required to evaluate these
terms is no longer a huge issue for most applications, numerical
precision begins to suffer due to the accumulation of roundoff error.
There are therefore practical limits to the size of a sphere whose
properties can be evaluated using Mie theory. Geometric optics (or
ray tracing; see Section 4.3.1) becomes the preferred method in such
cases.

12.3.1 Extinction Efficiency for Nonabsorbing Sphere

Figure 12.4 depicts the extinction efficiency Qe as a function of x for
a sphere with m = 1.33. This is a representative value for water in
 Scattering by Spheres — Mie Theory 353

Extinction Efficiency (m=1.33)

4
3.5
(a)
3
2.5
Qext

2
1.5
1
0.5
0
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100
Size parameter x=2πr/λ
4

3.5 (b)
3

2.5
Qext

1.5
Reddening Blueing
1

0.5

0
0 5 10 15 20
Size parameter x=2πr/λ
0.05
0.045
0.04
0.035
0.03 Exact Mie Theory
Qext

0.025 Rayleigh Theory

0.02
0.015
0.01
(c)
0.005
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Size parameter x=2πr/λ

Fig. 12.4: The extinction efficiency Qe as a function of size parameter x for a non-
absorbing sphere with m = 1.33, for various ranges of x. (a) “Big picture” view,
showing that Qe → 2 as x → ∞. (b) Detail for x < 20, with examples of subranges
for which extinction increases with x (reddening) or decreases with x (blueing).
(c) Detail for x < 0.8, comparing the Rayleigh (small particle) approximation and
exact Mie theory.

the visible band. Note that no imaginary part is assumed here, so

the droplet is nonabsorbing (ω̃ = 1) for all x.
The top panel (Fig. 12.4a) shows the typical behavior of Qe over
a wide range of x. It starts out at zero for x = 0 and rises mono-
tonically up to about x = 6, where Qe achieves a maximum value
354 12. Scattering and Absorption By Particles

of about 4. In other words, for this value of x, the droplet scatters

four times as much radiation as one might surmise from its cross-
sectional area alone! Thereafter, it exhibits an ever-dampening os-
cillation about a mean value of 2, which is the limiting value of Qe
for large x (recall that we already exploited this behavior in sec-
tion 7.2.3).
At the other end of the range, we have the opportunity to com-
pare Qe (= Qs ) computed using the exact Mie theory with that ob-
tained for the small-particle (Rayleigh) limit (Fig. 12.4c). We can see
that the agreement is quite good up to about x = 0.6. Beyond that
point, Qe increases less rapidly than the x4 dependence predicted
by (12.13).

Reddening/Blueing

Let’s now zoom in on the first couple of big wiggles in the curve
(Fig. 12.4b). Let’s further assume for the moment that r is fixed, so
that variations in x are due to variations in the wavelength λ, not
particle size — that is, increasing x implies decreasing λ, and vice
versa. Despite allowing the wavelength to vary, we will pretend
(somewhat unrealistically) that m also remains approximately con-
stant, so that our Qe curves are still valid.
With the above assumptions in mind, we find that in the region
0 < x < 6, Qe increases with x and therefore decreases with wave-
length. This means that for radiation passing through a cloud of our
fixed-size particles, the shorter wavelengths will be attenuated more
strongly than the longer wavelengths. This phenomenon is known
as reddening, and is responsible for the reddish color of the setting
sun. In fact, we already found a similar phenomenon in connection
with Rayleigh scattering in the previous section; those findings ap-
plied to particles with x  1, which is of course a small portion of
the range of x we are looking at right now.
For x between 6 and 11, on the other hand, extinction is stronger
for longer wavelengths than for shorter, giving rise to blueing of
the radiation passing through our particles. Blueing of sunlight or
moonlight is only very rarely observed; it would require an unusual
distribution of aerosol sizes in order for the blueing effect to dom-
inate over the usual reddening by both air molecules and smaller
 Scattering by Spheres — Mie Theory 355

Extinction Efficiency
4

3.5
r = 1 µm
3

2.5
r = 10 µm
Qext

1.5
r = 0.3 µm
1

0.5 r = 0.1 µm

0
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Wavelength λ [µm]

Fig. 12.5: The extinction efficiency as a function of wavelength for water droplets
of the indicated sizes.

aerosols.5

Problem 12.3: Visible radiation spans the wavelength range from

0.4 µm to 0.7 µm. Assuming that atmospheric aerosols have approxi-
mately the same refractive index m as that used to produce Fig. 12.4b,
determine the range of aerosol radii that would give rise to blueing.

We can look at the above phenonomenon from a slightly dif-

ferent (and more realistic) perspective by plotting the extinction ef-
5 Large volcanic eruptions occasionally inject matter into the stratosphere that
coalesces into aerosols of fairly uniform size; on rare occasions these may have a
size that leads to blueing of visible light. Some authors have suggested that such
rare events gave rise to the phrase “once in a blue moon.”
356 12. Scattering and Absorption By Particles

ficiency versus wavelength for selected water droplet radii, as in

Fig. 12.5. For droplet radii of 0.1 and 0.3 µm, which are character-
istic of small haze droplets, extinction is a strong function of wave-
length, with short wavelengths (e.g., UV-A, violet, blue) being extin-
guished far more strongly than longer wavelengths (red, near IR).
This is again the classic reddening behavior that we observe on a
hazy day, especially when the sun is low in the sky.
At an intermediate radius of 1 µm, the extinction behavior is
more complex. Near infrared wavelengths are fairly strongly at-
tenuated, violet light (near 0.4 µm) and red light (near 0.7 µm) is
attenuated slightly less strongly, and there is a pronounced mini-
mum in extinction between 0.5 and 0.6 µm. We can conclude that
if the aerosol population of the atmosphere consisted primarily of
droplets of 1 µm radius, the setting sun would take on a rather un-
natural greenish hue!6
The largest radius for which Qe is plotted in Fig. 12.5 is 10 µm,
which is a typical radius for ordinary cloud droplets. Over the en-
tire range of wavelength plotted, Qe ≈ 2. The lack of strong wave-
length dependence is why the color of sunlight passing through thin
clouds is not noticeably altered by the encounter. Even the obvious
wiggles seen on this curve are actually irrelevant in practice, be-
cause cloud droplets never have exactly one size but rather are dis-
tributed over a fairly broad range of sizes. Even a fairly small 10%
variability in droplet size would be enough to average away most
of the wiggle structure in the Qe curve.

12.3.2 Extinction and Scattering by Absorbing Spheres

Let’s now broaden our perspective in the following two ways: (1)
we will allow the imaginary part of m to be nonzero, and (2) we
will look at not only Qe but also the absorption efficiency Qa , the
single scatter albedo ω̃, and the scattering asymmetry parameter g.
Representative results are shown in Fig. 12.6. Based on these plots
6 It is tempting, though probably futile, to speculate on a possible role for 1 µm

haze or cloud droplets in the sickly greenish light that is observed to accompany
some severe thunderstorms. Various other physical mechanisms have been pro-
posed; as of this writing none has been widely embraced as the definitive explana-
tion for green thunderstorms, partly because of the scarcity of direct measurements
that could be used to test the various theories.
 Scattering by Spheres — Mie Theory 357

Extinction Efficiency Absorption Efficiency

4 2
m = 1.33 m = 1.33 + 0.001i
3.5 m = 1.33 + 0.1i m = 1.33 + 0.01i
m = 1.33 + 1.0i m = 1.33 + 0.1i
3 1.5 m = 1.33 + 1.0i

2.5

Qabs
Qext

2 1

1.5

1 0.5

0.5

0 0
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
Size parameter x=2πr/λ Size parameter x=2πr/λ
Single Scatter Albedo Scattering Asymmetry Parameter
1 1

0.8 0.8

0.6 0.6
ω0

0.4 0.4

m = 1.33 + 0.001i m = 1.33

0.2 m = 1.33 + 0.01i 0.2 m = 1.33 + 0.01i
m = 1.33 + 0.1i m = 1.33 + 0.1i
m = 1.33 + 1.0i m = 1.33 + 1.0i
0 0
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
Size parameter x=2πr/λ Size parameter x=2πr/λ

Fig. 12.6: Key optical properties of spheres as functions of x, for varying values of
the imaginary part of m.

we can make the following general statements:

• Increasing absorption by the particle material (by increasing

the imaginary part of m) has the effect of suppressing the wig-
gles in the curve of Qe . Apart from that change, the curves
are similar, all having a limiting value of approximately 2 for
large x.

• In the limit as x goes to zero, the single scatter albedo also

goes to zero, as predicted by (12.16). The sole exception is if
(m) = 0 (not shown), in which case ω̃ = 1 regardless of x.

• For x > 10, there is no completely predictable relationship

358 12. Scattering and Absorption By Particles

between (m) and either Qa or ω̃ . The absorption tends to

increase with small increases in (m), as you might expect,
but the trend reverses when (m) is of order one — parti-
cles with large (m) scatter more effectively than those with
smaller (m).

• For (m) = 0, there is considerable fine ripple structure in

the curves for both Qe and g. The presence of even slight ab-
sorption (e.g., the case m = 1.33 + 0.001i) pretty much elimi-
nates these ripples. But even for nonabsorbing particles, these
fine ripples tend to be unimportant. This is because you never
have particles all having the same exact value of x but rather a
mix of particles of various sizes. When combining the contri-
butions from various sizes, the small ripples quickly average
away.

• For x = 0, the asymmetry parameter g is also zero, as expected

for Rayleigh scattering. As x increases, g very rapidly in-
creases as well, plateauing somewhere in the range from about
0.8 and 0.95.

Forward Scattering
The last item is worthy of particular attention. It indicates that parti-
cles comparable to or larger than the wavelength tend to strongly forward
scatter, as contrasted with the case that x  1, for which the back-
ward and forward scattered components are about equal. It turns
out that this observation is generally applicable not only to spheres but to
particles of all types and shapes. This behavior is due to constructive
interference in the forward direction by waves scattered by different
parts of the particle, as discussed for example by S94 (Section 5.2.1).

12.3.3 Scattering Phase Function

The forward-scattering properties of larger particles becomes even
more apparent as we turn our attention to the scattering phase func-
tion p(Θ) of our spheres, as depicted for example in Fig. 12.7.
For x = 0.1 (bottom), we have the classical Rayleigh phase func-
tion, which is symmetric in the forward and backward directions.
 359

Scattering Phase Function

1e+09

1e+08
x=10,000

1e+07
x=3,000

1e+06

x=1,000

100000

x=300

10000

x=100
p(Θ)

1000

x=30
100

x=10

10

x=3

x=1

0.1

x=0.3
0.01

x=0.1
0.001
0 20 40 60 80 100 120 140 160 180
Θ [deg.]

Fig. 12.7: Plots of Mie-derived phase functions p(Θ) for various values of x, assum-
ing m = 1.33 (fine-scale oscillations in the curves for large x have been smoothed
out by allowing x to vary over a narrow range). The vertical scale is logarithmic
but otherwise arbitrary; each curve has been displaced upward from the previ-
ous one for clarity. Note increasing asymmetry and complexity of phase functions
with increasing x. The topmost curve (x = 10, 000) is very similar to that predicted
by geometric optics except for the narrow forward and backward peaks at 0◦ and
180◦ . See Figs. 12.8 and 12.9 for polar plots of some of these same curves.
360 12. Scattering and Absorption By Particles

For slightly larger x, there is a tendency for the phase function in

the forward direction Θ < 90◦ to have larger amplitude than in the
backward direction.
By the time we get to x = 3, we have a broad lobe of enhanced
scattering for Θ between about 0◦ and 40◦ . Within that range, the
amplitude of p(Θ) is around a factor of 100 larger than it is for Θ
between 120◦ and 180◦ . Now watch this forward-scattering lobe
carefully as we move upward on the figure — it becomes both nar-
rower and more intense with increasing x. In fact, for very large x, this
so-called forward diffraction peak, starts to resemble a δ-function and
falls right on y-axis in Fig. 12.7, so that it can no longer be distin-
guished.
At the same time as the forward scattering peak gets narrower
with increasing x, the rest of the phase function becomes more com-
plex, exhibiting an ever greater number of ripples. By the time
we get to x = 100, we start to see unmistakable signs of an en-
hanced scattering feature near Θ = 140◦ . This feature also sharpens
and intensifies dramatically with increasing x, until there is nearly
a hundred-fold difference between the amplitude of the peak at
Θ = 137◦ and the “floor” of the valley at just a slightly smaller

x=10

x=3

x=1

x=0.1

Fig. 12.8: Polar plots of the Mie-derived scattering phase function p(Θ) for selected
values of x.
 Scattering by Spheres — Mie Theory 361

angle! This feature is the primary rainbow whose existence was

previously explained in section 4.3.1 using the ray tracing method
(Fig. 4.8) and assuming a single internal reflection of the ray. The
slightly weaker peak at Θ = 130◦ , just to the left of the primary
rainbow, is the secondary rainbow, which is associated with rays
undergoing two internal reflections in the sphere.
In summary, the Mie solution, which is based on an infinite se-
ries solution of the EM wave equation with suitable boundary con-
ditions, yields results which basically converge to the geometric op-
tics results, once we let x get large enough. In fact, for x > 2000
or so, we have crossed out of the range of x for which Mie theory
is traditionally applied (see Fig. 12.1). Even for very large x, how-
ever, there are aspects of scattering by particles that geometric op-
tics alone can never explain, such as the forward diffraction peak
as well as the intensified scattering near 180◦ known as the glory.
The polar plots in Fig. 12.8 and Fig. 12.9 provide an alternative
way of visualizing the evolution of the phase function with increas-
ing x. In the first of these, the amplitude of the phase function is pro-
portional to the distance along a particular radial at angle Θ, with
Θ = 0 pointing horizontally to the right. For x = 0.1 we again have
the symmetric Rayleigh phase function; you’ll probably recognize
the shape from Figs. 12.2 and Figs. 12.3. For even modest increases
in x, the asymmetry quickly becomes very pronounced. By the time
we reach x = 10, the forward scattering lobe is already so intense
that it no longer fits on the page!
For larger x, we can tame the extreme features of the phase func-
tion by making the radial amplitude proportional to the logarithm
of p(cos Θ) (Fig. 12.9). Among other things, these plots allow us
to clearly see, for the first time, what’s happening at Θ = 0◦ and
Θ = 180◦ . There are a few features that deserve special mention,
because they are associated with commonly observed optical phe-
nomena:

Forward Diffraction Peak

We have already mentioned the strong forward scattering that oc-

curs in connection with larger particles, spherical or not. This phe-
nomenon is readily observable in daily life. It is much harder to
362

Glory Fogbow x=30

Corona

Forward Diffraction
Peak

Secondary Rainbow

Primary Rainbow x=100

x=10,000

Fig. 12.9: Similar to Fig. 12.8, but plotted as log[ p(Θ)] so as to better accommodate
the extreme variations in the amplitude of the phase function for large x. Com-
monly observed optical features associated with the phase function are indicated.
Note the δ-function-like characteristic of the forward and backward peaks for the
largest values of x.
 Scattering by Spheres — Mie Theory 363

see through the glare of a dirty windshield when driving toward

the sun than away. Dust particles settling through a shaft of sun-
light in a room are easiest to spot when looking generally toward
the source of the light. The rays of light from the setting sun emerg-
ing from a break in the clouds (so-called crepuscular rays) are most
evident when viewed in the general direction of the sun.
Fig. 12.9 makes much clearer the profound narrowing of the
diffraction peak with increasing x. For x = 10, 000, the peak is so
narrow that the scattered radiation it represents might as well be
considered as never having been scattered at all. In the geometric
optics approximation, this feature doesn’t even exist! In fact, while
Mie theory predicts Qe ≈ 2 in the limit of large x, geometric op-
tics always predicts Qe = 1. The forward diffraction peak is largely
responsible for the discrepancy.

Corona

For intermediate values of the size parameter x, the forward diffrac-

tion peak is accompanied by number of weaker sidelobes. If you
were to view the sun through a very thin cloud made up of iden-
tical spherical droplets with x of order 100 or less, you would see
a series of closely spaced rings immediately surrounding the light
source. Moreover, because the precise angular position of the rings
depends on wavelength, the rings would be brightly colored. This
optical feature is known as a corona.
Coronas observed in real clouds are more diffuse, and less
brightly colored, than the corona you would expect from a cloud
composed of identical droplets. In fact, one reason colored coronas
are rarely observed at all is because few clouds have a sufficiently
narrow distribution of drop sizes.7
Far more commonly you will just see a diffuse circular bright
patch surrounding the sun with little if any coloration. This feature
represents a blending of both the forward diffraction peaks and the
sidelobes contributed by a variety of different drop sizes.

7 Another reason, of course, is that only true enthusiasts of optical phenomena

take time each day — and risk their eyesight — in order to stare almost directly at
the sun in the hope of spotting a spectacular corona!
364 12. Scattering and Absorption By Particles

Glory

The glory is in many respects analogous to the forward diffraction

peak and the corona, except that it is exactly on the opposite end of
the phase function. It is called a “glory” because if you stand on a
hill overlooking a fog bank with the sun at your back, you will see
a bright patch or ring surrounding the shadow of your head in a
manner reminiscent of medieval paintings of saints.
A much more predictable setting for observing glories is as a
passenger in an airplane flying above a cloud layer, from whence
a bright ring is often seen immediately surrounding the shadow of
the airplane. If the airplane is high enough above the cloud layer,
the shadow will be too indistinct to see but the glory will be visible
nonetheless.
As was also the case for the corona, glories may involve multiple
rings and vivid colors, provided only that the range of drop sizes is
sufficiently narrow. More commonly, the glory is seen as a fairly
indistinct white ring or circular bright patch.
For very large x, the glory narrows to a δ-function-like spike
in the exact backscattering direction Θ = 180◦ . Like the forward
diffraction peak, the glory is a feature not predicted by geometric
optics, at least for spheres with the index of refraction of water.8

Fogbow/Rainbow

We already mentioned the occcurence of sharp spikes in the scat-

tering phase function corresponding to the primary and secondary
rainbows. The positions of these features are noted on the polar
plot for x = 10, 000 in Fig. 12.9. For smaller x, the primary rainbow
8 You have probably noticed that a lot of street signs, license plates, reflective
leg straps for bicycle riders, etc., are intensely reflective when the light source is
very close to being in line with the object and the observer. For example, when ap-
proaching a stop sign at night from a couple of blocks away, your own headlights
cause the sign to light up brightly, whereas illumination by light sources in other
directions doesn’t have nearly as intense an effect. Close examination reveals the
presence of retroreflective beads, which are simply small spheres with an index of
refraction falling between approximately 1.5 and 2.0. For this range of index of
refraction, geometric optics is able to explain the unusually strong backscatter as
the result of rays that pass into the sphere and experience total internal reflection
on the far side.
 Distributions of Particles 365

feature is still present but not nearly as sharp. Because the peak
is much more diffuse, the separation of colors (due to varying nr )
will not be nearly as vivid as you find in a “normal” rainbow, and
you will instead observe a more or less whitish ring centered on the
point opposite the sun (i.e., centered on your own shadow). In this
case, a better name for the feature is fogbow, because it arises (for
visible light) when the water droplets have a size characteristic of
fog and clouds rather than rain.

Problem 12.4: Assuming a wavelength λ = 0.5 µm, which is near

the middle of the visible band, determine the water droplet radii cor-
responding to each of the three phase functions depicted in Fig. 12.9.

12.4 Distributions of Particles

The atmosphere never contains particles of just one size. In per-
forming radiation transfer calculations for clouds and aerosols, it is
invariably necessary to start out by determining the combined op-
tical properties of a distribution of particles of varying sizes and,
possibly, shapes and compositions. We will limit our attention to
the case of varying size only, though the formal extension to shape
and composition is straightforward.
In Section 7.4.4, I already introduced the concept of a size distri-
bution function n(r ) for cloud droplets. To refresh your memory,
 
number of droplets (per unit volume of
n(r ) dr = . (12.25)
air) whose radii fall in the range [r, r + dr ]

The same concept is applicable to any particle type.

We already saw that the volume extinction coefficient for the dis-
tribution of particles described by n(r ) is

∞
βe = n(r ) Qe (r )πr2 dr . (12.26)
0

In other words, the total extinction β e is equal to the extinction cross-

section contribution from a single particle of radius r multiplied by
366 12. Scattering and Absorption By Particles

the number of particles (per unit volume) having that radius and
then integrated over all possible radii.
A completely analogous relationship gives the scattering coeffi-
cient
∞
βs = n(r ) Qs (r )πr2 dr . (12.27)
0
From there, we immediately have the single-scatter albedo of the
distribution as ω̃ = β s /β e .
The combined scattering phase function is the scattering cross-
section weighted average of the individual phase functions:

∞
1
p(cos Θ) = n(r ) Qs (r )πr2 p(cos Θ; r ) dr , (12.28)
βs 0

which also implies a combined asymmetry parameter of

∞
1
g= n(r ) Qs (r )πr2 g(r ) dr . (12.29)
βs 0

12.5 Applications to Meteorology, Climatology,

and Remote Sensing
12.5.1 The Scattering Properties of Clouds
The radiative properties of clouds, including their ability to reflect
and absorb both solar and thermal radiation, depend on their opti-
cal depth τ ∗ , their single scatter albedo ω̃ and the scattering phase
function p(cos Θ). These properties in turn depend on the size pa-
rameter x and on the complex index of refraction m for the cloud’s
constituent particles. Both x and m depend on wavelength λ, and
x also depends on the droplet radius r. The index of refraction m
depends on composition and material phase as well, but for most
clouds, there are only two possibilities: liquid water or ice.
For reasonably large x, we already saw that Qe ≈ 2, so that the
optical thickness τ ∗ can often be taken to be almost independent of
wavelength. The phase function p(Θ) is adequately characterized
for many purposes by the asymmetry parameter g, which we saw
tends to hover in the fairly narrow range 0.8–0.9 for x greater than
around 10.
 Applications 367

(a) Single Scatter Albedo - Water vs Ice (b) Single Scatter Co-Albedo - Water vs Ice
1 1
0.9
0.8 0.1
0.7
0.6 0.01

1-ω0
ω0

0.5
0.4 0.001
0.3
0.2 0.0001
0.1 Water, r = 20 µm Water, r = 20 µm
Ice, r = 20 µm Ice r = 20 µm
0 1e-05
1 10 0 0.5 1 1.5 2 2.5 3
Wavelength λ [µm] Wavelength λ [µm]
(c) Single Scatter Albedo - Cloud Droplets (d) Single Scatter Co-Albedo - Cloud Droplets
1 1

0.9

0.8 0.1

0.7
0.01
0.6
1-ω0
ω0

0.5
0.001
0.4

0.3 0.0001
r = 5 µm r = 5 µm
0.2 r = 10 µm r = 10 µm
r = 20 µm r = 20 µm
0.1 1e-05
1 10 0 0.5 1 1.5 2 2.5 3
Wavelength λ [µm] Wavelength λ [µm]

Fig. 12.10: Single scatter albedo (or co-albedo) as a function of wavelength for
water and ice spheres of various sizes. The left column depicts ω̃ over the entire
visible, near IR and thermal IR range. The right column depicts the scattering co-
albedo (defined as 1 − ω̃) for just the solar band. The top row compares water and
ice for particle radius r = 20 µm; the bottom row compares water droplets of three
different radii (5, 10, and 20 µm).

This leaves the single scatter albedo ω̃ as the one variable that
could potentially have a large influence on how cloud reflectiv-
ity/absorptivity varies with λ. And indeed this conjecture is val-
idated by the plots of ω̃ vs. λ shown in Fig. 12.10. This information
is represented in two different ways. The first is by simply plotting
ω̃ on a linear vertical axis, as is done in the two panels in the left
column. This is fine for showing the coarse variations of ω̃ with
wavelength but tends to obscure subtle deviations of ω̃ from ex-
actly one (pure scattering, no absorption), which can nevertheless
368 12. Scattering and Absorption By Particles

be significant for absorption by clouds. Therefore, for the shorter

wavelengths where absorption is comparatively weak, we plot the
scattering co-albedo, defined as 1 − ω̃, on a logarithmic vertical axis.
Here are the basic points you should take away from these plots:

• The visible band (0.4 µm < λ < 0.7 µm) coincides almost ex-
actly with a surprisingly narrow portion of the EM spectrum
for which absorption by cloud droplets is, for all practical pur-
poses, zero. You can think of it as an astonishing coincidence
that clouds (when viewed from the sunlit side) appear to our
eyes as white rather than gray, black, or some other color!
As soon as you move into either the UV or near-IR bands, ω̃
quickly decreases to well below 1, settling into the range 0.5–
0.8 for most of the IR band. For even ω̃ = 0.8, the albedo of a
thick cloud is only around 15%.
• At several wavelengths, there is a significant difference be-
tween the single scatter albedo of a spherical ice particle and
that of a water droplet of the same size (top row). For some
of these wavelengths, ice particles are less absorptive than the
water droplets; for others, the reverse is true. These differ-
ences can be exploited by satellite sensors to distinguish ice
phase clouds (cirrus) from water clouds9 .
• For most wavelengths, there is a significant dependence of
the single scatter albedo on the droplet radius in liquid wa-
ter clouds (bottom row). As a general rule (although there
are exceptions), a larger droplet has lower ω̃ (i.e., is more
absorptive) than a smaller droplet at the same wavelength.
Once again, satellite remote sensing techniques can exploit
this property to estimate the effective droplet radius reff in wa-
ter clouds.

12.5.2 Radar Observations of Precipitation

Radar has become one of the most important observational tools of
operational meteorologists as well as hydrologists. Weather radar
9 The fact that ice particles in clouds are generally not spheres complicates the

problem somewhat, but the principle is still valid.

 Applications 369

allows severe weather systems to be tracked in real-time. It also

allows the monitoring of rainfall with far more detail in both time
and space than is possible with conventional rain gauges.
The basic principle of operation of a radar system is simple.
A transmitter sends out a continuous series of short pulses of mi-
crowave radiation. A sensitive receiver then measures the intensity
of the backscattered radiation as a function of the time elapsed fol-
lowing each transmitted pulse. The time elapsed ∆t is of course just
the round trip distance divided by the speed of light c, so that the
one-way distance d to the target is given by

c∆t
d= . (12.30)
2
The backscattered power Pr received by the radar antenna is given
by the following proportionality:
η
Pr ∝ , (12.31)
d2
where η (Greek letter eta) is the backscatter cross-section per unit vol-
ume of air. It is just the sum of the backscatter cross-sections σb of all
particles in the sampled volume of air V, divided by V:

1
η=
V ∑ σb,i , (12.32)
i

The backscatter cross-section σb is closely related to to the more fa-

miliar scattering cross-section σs , except that it only accounts for ra-
diation scattered exactly backward toward the radar antenna, rather
than radiation scattered into all directions. As a matter of fact,

σb ≡ σs p(Θ)|Θ=π . (12.33)

Now let’s assume that we’re dealing with particles that all have
the same composition (e.g., liquid water), are spherical in shape,
and are distributed in size according to a size distribution function
n( D ), where D is the droplet diameter. The assumption of spherical
shape, while approximate, is reasonable for raindrops that are “not
too large.”
370 12. Scattering and Absorption By Particles

Radar Backscatter from Sphere, λ= 10.71 cm

1
0.1 Water
Ice
0.01
0.001
0.0001
1e-05
Qback

1e-06
1e-07
1e-08
1e-09
1e-10
1e-11
1e-12
0.1 mm 1 mm 1 cm 10 cm
Diameter

Fig. 12.11: Radar backscatter efficiency Qb for water and ice spheres at the wave-
length of the WSR-88D operational weather radar.

We can then replace the summation in (12.32) with an integral

involving n( D ) and σb ( D ):

∞
η= σb ( D )n( D ) dD , (12.34)
0
or
∞ π 
η= Qb ( D ) D2 n( D ) dD , (12.35)
0 4
where the term in square brackets is just the cross-sectional area of
a sphere with diameter D, and Qb is the backscatter efficiency.
If our spherical particles happen to have size parameters x  1,
then we’re in the Rayleigh regime. This means that (a) the Rayleigh
formula (12.13) for σs applies, and (b) the phase function is given by
(12.10). Substituting these into (12.33) gives
 2 2

4 m − 1

Qb = 4x  2 . (12.36)
m + 2
If our particles are too large, then Rayleigh theory no longer ap-
plies, and we have to calculate σb using Mie theory. Fig. 12.11 shows
 Applications 371

accurate calculations of Qb for a wide range of sizes of water and

ice spheres. The wavelength λ = 10.71 cm chosen for these calcula-
tions corresponds to that used by the current-generation operational
weather radar network in the United States.
You can see that for liquid water spheres up to a diameter of
about 6 mm (solid curve), the Rayleigh relationship (12.36) holds to
a high degree of accuracy: each decade (factor ten) increase in D cor-
responds to a four decade (factor 104 ) increase in Qb . In fact, 6 mm
corresponds to a rough upper limit on the observed sizes of rain-
drops in heavy rain; beyond this size, raindrops tend to be broken
up by aerodynamic forces as they fall.
Hailstones can of course become considerably larger than rain-
drops. It is therefore convenient that the Rayleigh approximation
apparently holds up to a diameter of around 3 cm for a pure ice
sphere (dashed curve). Note that for any given D in the Rayleigh
regime part of the curve, Qb for pure ice is only 20% of that for liq-
uid water. This difference is due to the substantially smaller value
of m for ice in the microwave band, as compared to liquid water10 .

Problem 12.5: Based on the above information, compute the di-

ameter of a spherical hailstone that has the same radar backscatter
cross-section σb (not Qb !) as a spherical raindrop with a diameter of
2 mm.

Let’s assume that the hydrometeors (e.g. raindrops, hailstones,

etc.) that are observed by a 10-cm weather radar all fall in the
Rayleigh regime. We can then substitute (12.36) into (12.35) to get
 2
∞
π 5  m2 − 1 
η= 4 2 n( D ) D6 dD . (12.37)
λ m + 2 0
Substituting this expression back into (12.31), we find that the
backscattered power measured by the radar receiver is
 2 
 m − 1 2 Z

Pr ∝  2  , (12.38)
m + 2  d2
10 It is worth keeping in mind, however, that growing hailstones often have a

coating of liquid water. Even a thin coating of water can drastically alter the radar
backscattering properties of an ice particle.
372 12. Scattering and Absorption By Particles

where Z is the reflectivity factor, defined as

∞
Z= n( D ) D6 dD . (12.39)
0

In other words, the reflectivity factor is numerically equal to the sum

of the sixth powers of the diameters of all of the drops in a unit volume of
air. The standard units of Z used by meteorologists are [mm6 m−3 ].
An estimate of the reflectivity factor Z at each range d along the beam is
what most weather radars record and display.
Because observed values of Z span an enormous range, mete-
orologists prefer to work with a logarithmic representation of Z,
defining a nondimensional unit dBZ, which means “decibels with
respect to one standard unit of Z.” You convert the reflectivity factor
from standard units to units of dBZ as follows:

Z [dBZ ] = 10 log10 ( Z ) , (12.40)

where Z on the right hand side is the numerical value of Z ex-

pressed in standard (dimensional) units of reflectivity. Thus, an in-
crease in reflectivity by 10 dBZ corresponds to a factor ten increase
in Z expressed in standard units. An increase of 30 dBZ implies a
thousand-fold increase in reflectivity.

Problem 12.6: Depending on range, a typical weather radar can

measure reflectivities from as low as −20 dBZ to as high as 70 dBZ.
In terms of physical units, what is the ratio of the two reflectivity
factors?

In converting the received power Pr to an estimate of the reflec-

tivity factor Z, the radar processing software assumes a value of
m appropriate to liquid water in (12.38). The displayed quantity is
therefore actually better regarded as an equivalent reflectivity factor Ze
which may or may not be equal to the true reflectivity factor Z de-
fined by (12.39), depending on whether the targets are liquid water
or something else, like ice. If the particles are in fact ice, then

Ze ≈ 0.20Z . (12.41)
 Applications 373

Problem 12.7: During a particular (and peculiar) rainstorm, each

cubic meter of air contains 1000 falling drops, each of identical diam-
eter D. (a) Compute the reflectivity factor Z, assuming D = 1 mm.
(b) Repeat for D = 2 mm. (c) By what factor did Z increase on ac-
count of a mere two-fold increase in D? (d) Express your answers to
(a)–(c) in units of dBZ. (e) If you replace the liquid raindrops with ice
spheres of the same size, by how many dBZ will the radar-estimated
effective reflectivity Ze be reduced? (f) Notwithstanding Eq. (12.41),
hailstorms are often recognized on radar displays by virtue of their
anomalously high Ze . Why?

In actual rainfall, drops do not all have one size but rather are
distributed over a wide range of sizes. Because of the D6 depen-
dence in Z, observed reflectivities are heavily influenced by the few
largest drops in the volume of air. A single drop with a diame-
ter 5 mm reflects more microwave radiation than 15,000 drops of
1 mm diameter! And clouds, with their typical droplet diameters of
around 20 µm, are completely invisible to all but the most sensitive
radars, despite typical droplet concentrations in excess of 108 m−3 .

Problem 12.8: From the information given above concerning cloud

droplets, find a typical reflectivity factor Z for clouds, expressed in
dBZ.

Radar Rainfall Estimation

Raindrops passing through the air eventually reach the surface, and
the rate at which water is deposited (depth per unit time) is known
as the rainfall rate R. One of the most important applications of radar
is the operational estimation of accumulated rainfall for agricultural
and hydrological purposes.
Unfortunately, there is no unique relationship between the rain-
fall rate and the relative number of larger and smaller droplets
present in the column of air. Consequently, there can be no unique
relationship between radar reflectivity Z and rain rate R. However,
we know from experience that heavier rainfall tends to be associated
374 12. Scattering and Absorption By Particles

with a greater number of large raindrops, whereas light rain is usu-

ally characterized by smaller drops. On average, therefore, we expect
heavy rain rates to be associated with large Z and light rain rates to
give rise to correspondingly weaker radar echoes.
Field observations of raindrops have revealed that the dropsize
distribution n( D ) for rain is often reasonably well approximated by

n( D ) = N0 exp(−ΛD ) , (12.42)

where N0 and Λ are parameters that are functions of the rain rate R.
In fact, the most widely used model of the above form is known as
the Marshall-Palmer size distribution, after the researchers who devel-
oped it. In the Marshall-Palmer distribution, N0 is a constant and
Λ = aRb , where the parameters a and b were chosen so as to max-
imize the agreement between the above size distribution function
and a large number of actual observations of drop sizes at various
rain rates.
It is beyond the scope of this text to discuss the M.-P. distribution
in detail, except to note that, when it is substituted into (12.39) and
combined with suitable assumptions about raindrop fall speed as a
function of D, it is possible to obtain the following Z–R relationship:

Z = 200R1.6 , (12.43)

where R is assumed to be given in mm hr−1 , and Z is in standard

units of mm6 m−3 . Other assumed (or measured) drop size distri-
butions usually lead to Z–R relationships having a similar form, but
with different values for the two numerical coefficients.

Problem 12.9: Use the Marshall-Palmer Z–R relationship above to

estimate the rain rates R associated with displayed radar reflectivi-
ties of (a) 10 dBZ, (b) 30 dBZ, and (c) 50 dBZ.

12.5.3 Microwave Remote Sensing and Clouds

Microwave radiometers operating at various frequencies from 3 to
183 GHz are assuming an increasingly prominent role in the satel-
lite remote sensing of the atmosphere. One of the main attractions of
 Applications 375

the microwave band is the relative transparency of clouds at these

wavelengths, so that some properties of the surface and of the at-
mospheric column can be estimated under nearly all-weather con-
ditions.
At even the high end of the frequency range given above, the
wavelength λ is a relatively long 3 mm which, for typical 10 µm
radius cloud droplets, gives size parameter x ≈ 0.02. This is so
small that we can ignore scattering, and the mass extinction (ab-
sorption) coefficient k L of cloud liquid water is accurately given by
(12.19). Figure 12.12 shows how ka varies with frequency over the
microwave band.
Consider a microwave radiometer at ground level viewing ver-
tically incident radiation emitted by the atmosphere. In the mi-
crowave band, the Rayleigh-Jeans approximation allows us to work
with brightness temperature TB as a convenient stand-in for radi-
ant intensity, with TB = εT, where ε is the emissivity of a surface
or atmospheric layer, and T is its physical temperature (see section
6.1.4).

Mass absorption coefficient for cloud water

0.1
k (m2/ kg)

0.01

0.001

0.0001
0 50 100 150 200 250 300
Frequency (GHz)

Fig. 12.12: The mass absorption coefficient for cloud water at microwave frequen-
cies.
376 12. Scattering and Absorption By Particles

Zenith Microwave Transmittance

1

0.9 O2

0.8

0.7 H2O
Transmittance

H2O O2
0.6
2
0.5 Dry (0 kg/m )
Polar (3.1 kg/m22)
0.4 Midlatitude (21.3 kg/m2)
Tropical (53.6 kg/m )
0.3

0.2

0.1

0
0 50 100 150 200 250 300
Frequency [GHz]

Fig. 12.13: Zenith microwave transmittance of the cloud-free atmosphere for dif-
ferent models of atmospheric temperature and humidity. The vertically integrated
water vapor content associated with each model is given in parentheses.

If we assume for the moment that the cloud-free atmosphere is

perfectly transparent (it is not) and that there is a single cloud layer
with average temperature T and total vertically integrated cloud liq-
uid water L, then the measured brightness temperature is given ap-
proximately by

TB = εT = [1 − t( L)] T = [1 − exp(−k L L)] T . (12.44)

You could then use your upward-looking microwave radiometer to

estimate the cloud water path by simply (i) solving the above equa-
tion for L, (ii) assuming something reasonable for T, and (iii) plug-
ging in the observed brightness temperature TB .
The reality is of course slightly more complicated. In particu-
lar, there are two other atmospheric constituents that always con-
tribute additional absorption and emission in the microwave band:
water vapor, and oxygen (Fig. 12.13). If we stay well away from
the 60 GHz and 118 GHz absorption bands due to oxygen, then
the reduction in transmittance due to the dry atmosphere alone is
only a few percent. Furthermore, since the surface air pressure at
any given location, and therefore the total column oxygen content,
 Applications 377

varies only by about 5%, we can get away with assuming a fixed
optical depth τO due to oxygen.
Water vapor is a bigger problem, because atmospheric column
vapor content V varies from very low (∼1 kg m−2 ) in dry polar
air masses to rather high (up to 60 kg m−2 ) in humid tropical air
masses. In order to limit the total optical depth due to water vapor,
let’s confine our attention to the spectrum below about 40 GHz, so
that even in the worst case, we still have a zenith transmittance of
at least 60% or so. That way, the atmosphere will never become so
opaque due to water vapor that it becomes hard to see changes in
opacity due to cloud water.
If we assume that the mean emitting temperature of the atmo-
spheric water vapor and oxygen isn’t too different from that of the
cloud layer, then we can write

TB ≈ [1 − exp(−τ )] T , (12.45)

where the total atmospheric optical depth is approximated as

τ ≈ τO + k L L + kV V , (12.46)

and kV is the column-averaged mass absorption coefficient of water

vapor. We can divide through by T, rearrange, and take the loga-
rithm of both sides to get
 
T − TB
y ≡ log ≈ −k L L − kV V − τO . (12.47)
T

Given a reasonable value for T, the new variable y is a known func-

tion of the observed TB . The definition of y is convenient because
it turns out to be a simple linear function of our two unknowns V
and L. Unfortunately, we have one equation in two unknowns, and
so a measurement of TB at a single wavelength is not sufficient to
uniquely determine both variables.
Let’s therefore design our radiometer to measure TB at two dif-
ferent frequencies ν1 and ν2 . We can then write our equation in ma-
trix form as
      
y1 k L,1 kV,1 L τO,1
=− − . (12.48)
y2 k L,2 kV,2 V τO,2
378 12. Scattering and Absorption By Particles

We now have two linear equations in two unknowns. In principle,

we can solve for L and V as follows:
    −1  
L k L,1 kV,1 y1 + τO,1
=− , (12.49)
V k L,2 kV,2 y2 + τO,2

assuming that the inverse of the matrix of absorption coefficients K

exists.
Mathematically speaking, the inverse exists if the determinant
K = 0, a condition that is almost guaranteed to be satisfied for
any pair of distinct microwave frequencies. Practically speaking,
however, the mere existence of an inverse is not enough! Why not?
Recall that (12.47) was presented as an approximate model of the
dependence of y on L and V. This suggests that we should modify
(12.49) to allow for the likelihood of errors i in the relationship:
   −1    
L k L,1 kV,1 y1 + τO,1 1
=− − , (12.50)
V k L,2 kV,2 y2 + τO,2 2

where L and V are now estimates of the true L and V. The estimation
error can then be written
    −1  
L −L k L,1 kV,1 1
= . (12.51)
V − V k L,2 kV,2 2

The goal of a remote sensing technique, of course, is to make sure

that the estimation errors are as small as possible. In this instance, it
means ensuring that the matrix K−1 does not excessively “amplify”
the model and/or instrument errors i . Since the magnitude of K−1
is proportional to 1/K, it follows that we require not only that
K be nonzero (the strict mathematical requirement for invertibil-
ity) but that it be as large as possible!
In plain English, we want our two sensor channels to respond in
substantially different ways to L and V, so that the opacity contri-
butions of each constituent can be separated with as little ambiguity
as possible. Now, regardless of frequency, k > 0 for both L and V,
so the most we can hope for is that one channel should have a pro-
portionally larger response to V than to L, while for other channel
the reverse should be true. Examination of Figs. 12.12 and 12.13
suggests the way to achieve this: Choose one channel to fall on or
 Applications 379

near the center of the water vapor absorption line at 22.235 GHz;
choose the second channel to fall between 30 and 40 GHz, where
water vapor absorption is much weaker than for the first channel
but liquid water absorption is significantly stronger. This is in fact
what is done in the design of commercial two-channel microwave
radiometers, though the precise choice of frequencies depends on a
more sophisticated analysis than we have offered here.
Several other subtleties have also been glossed over here with
respect to the optimal inversion of radiometric measurements; these
are best left to a course in remote sensing. Suffice it to say that an
explicit matrix-inverse method, as described above, is usually less
satisfactory than a statistical or semi-statistical method that is cho-
sen so as to minimize the average squared error over the widest
possible range of atmospheric conditions.

Problem 12.10: Assume that you have an upward-looking mi-

crowave radiometer with channels at 23.8 GHz and 31.4 GHz. For
the first frequency, take k L,1 = 0.087, kV,1 = 0.0052, and τO,1 = 0.02.
For the second frequency, take k L,2 = 0.15, kV,2 = 0.0021, and
τO,2 = 0.03. All values for k are in units of m2 /kg. Assume a mean
atmospheric temperature T = 280 K.
(a) Approximately what brightness temperature TB should each
channel observe in a perfectly dry atmosphere, with V = L = 0?
(b) Repeat your calculation for V = 60 kg m−2 , which is typical
of a humid tropical atmosphere. Leave L = 0 (i.e., no cloud).
(c) Repeat your calculation again, this time assuming L =
0.3 kg m−2 , which is typical of a fairly thick nonprecipitating stra-
tocumulus layer.
(d) Derive a retrieval algorithm for the estimation of V. The final
algorithm should have the form V = a1 log( T − TB,1 ) + a2 log( T −
TB,2 ) + a3 . Provide the coefficients ai to four significant figures.
(e) Test your algorithm by applying it to your results from parts
(a), (b), and (c). Do you recover the correct values of V in each case?
 CHAPTER 13

Radiative Transfer with Multiple Scattering

We are now nearing the end of our introductory survey of radiative

transfer in the atmosphere, and it is perhaps no surprise that we
have saved the best — or at least the most challenging — for last.
Every problem we have dealt with so far entailed either absorption
and emission with no scattering, or else at most single scattering.
These restrictions enabled us to solve the radiative transfer equation
along a single line of sight without worrying about what was going
on in other locations and directions.
Such simplifications are utterly useless for solar radiative trans-
fer in clouds. Because most water clouds are both optically thick
(τ  1) and are only weakly absorbing (ω̃ ≈ 1), multiple scattering
cannot be neglected. That is to say, at most points in the interior
of the cloud, the majority of radiation incident on a cloud particle
will have already been scattered at least once by some other particle.
Photons incident at cloud top will typically be scattered numerous
times before re-emerging from the cloud, either at the top or base in
the plane parallel case, or even from the sides in the case of three-
dimensional clouds. What this means in practice is that you cannot
consider what is happening to the radiant intensity along one line-
of-sight without simultaneously considering what it is doing every-
where else.

380
 381

The full radiative transfer equation for a plane parallel atmo-

sphere was given earlier as (11.13) and (11.14). If we neglect thermal
emission, the two can be combined to give

2π
1
dI (µ, φ) ω̃
µ = I (µ, φ) − p(µ, φ; µ , φ ) I (µ , φ ) dµ dφ .
dτ 4π 0 −1
(13.1)
This integrodifferential equation tells us that in order to determine
I (µ, φ) at a particular level τ in a cloud, we must simultaneously
determine I (µ , φ ) for all values of µ and φ and for all other values
of τ.
In general, (13.1) cannot be solved exactly except under ex-
tremely restrictive (and inevitably unrealistic) assumptions about
the scattering phase function, among other things. Therefore, radi-
ation specialists have put much effort into

• studying closed form solutions to (13.1) for highly idealized

cases (e.g., isotropic scattering, infinite cloud optical depth)
with an eye toward gaining qualitative insight into radiative
transfer in clouds, and

• developing computational techniques for obtaining reason-

ably accurate numerical solutions for real-world problems.

We will not delve deeply into such methods here, as they are
mainly of interest to those who perform radiative transfer calcula-
tions for a living. If this book has piqued your own interest in atmo-
spheric radiation, then you should plan to continue your education
with advanced textbooks, such as L02 and TS02, that devote consid-
erable space, and literally hundreds of equations, to computational
methods.
Here, we will begin by trying to convey some insight into how
multiple scattering “works” in a plane-parallel cloud layer. We will
then walk through one of the simplest possible analytic solutions to
(13.1) known as the two-stream method. The two-stream approxima-
tion is not terribly useful for computing accurate radiant intensities
as a function of µ and φ, but it’s not bad for estimating hemispheri-
cally averaged fluxes in plane parallel cloud layers. In fact, because
it is relatively undemanding of computer resources, some variation
382 13. Radiative Transfer with Multiple Scattering

on the two-stream method (or the slightly more sophisticated four-

stream method) is usually at the heart of the shortwave radiative
parameterization schemes found in virtually all climate models.

13.1 Visualizing Multiple Scattering

When radiation is incident on a plane-parallel scattering layer of
the atmosphere (e.g., a cloud), we can visualize the radiative trans-
fer process in terms of the fates of a large number of individual pho-
tons. When the radiation is due to a quasi-point source, like the sun,
then each photon enters at cloud top (τ = 0) traveling in the same
initial direction Ω̂0 . The subsequent fate of each photon, however,
is determined by its chance encounters with the scattering and/or
absorbing particles in the layer, as summarized by the following se-
quence of events:
1. First, each photon travels a random distance before it encoun-
ters its first extinction event. Although the distance traveled
by each photon individually is random, the average distance is
dictated by Beer’s Law. In fact, if the total cloud optical depth
is τ ∗ , then we know that a fraction tdir = exp(τ ∗ /µ) of the
incident photons will reach the cloud base without being ex-
tinguished at all.
2. The fraction 1 − tdir of the photons that do not make it all the
way through the cloud on their first attempt get extinguished
(i.e., scattered or absorbed). Of those photons, the fraction that
gets absorbed is simply 1 − ω̃. For those photons, their long
journey from the sun has come to an abrupt end; their energy
hν goes into warming the cloud.
3. Those photons that get scattered instead of being absorbed
now have a new direction of propagation µ . Along that di-
rection, Beer’s Law (applied to the new optical distance to the
boundary) again determines the probability that a particular
photon will reach the upper or lower boundary and exit the
cloud.
4. For each photon, steps 2 and 3 are repeated until the photon
has either been absorbed or exited the cloud layer. The albedo
 383

a) ω=0.3 b) ω=0.5
τ=0

τ∗=10

c) ω=0.7 d) ω=0.9
τ=0

τ∗=10

e) ω=0.99 f ) ω=1.00
τ=0

τ∗=10

Fig. 13.1: Examples of the random paths of 100 photons in a plane-parallel, isotrop-
ically scattering layer (g = 0) with optical thickness τ ∗ = 10 and variable single
scatter albedo ω̃. Photons are incident from above with θ = 30◦ . Heavy diagonal
lines indicate the path an unscattered photon would take.
384

a) g=0.0 b) g=0.4
τ=0

τ∗=10

c) g=0.6 d) g=0.8
τ=0

τ∗=10

e) g=0.9 f ) g=0.99
τ=0

τ∗=10

Fig. 13.2: Same as Fig. 13.1, but for conservative scattering (ω̃ = 1) and varying
values of the asymmetry parameter g.
 The Two-Stream Method 385

and diffuse transmittance of the cloud, respectively, describe

the fraction of incident photons that exit the top or the bottom
of the layer.

The above process is depicted graphically for a variety of com-

binations of ω̃ and g in Figs. 13.1 and 13.2. Each photon trajectory is
followed until either the photon gets absorbed or else it exits from
the cloud. Note that even a slight chance of absorption in any single
extinction event translates into a rather high probability of absorp-
tion by the time the photon has been extinguished numerous times
on its passage through an optically thick cloud. Note also the degree
to which the asymmetry parameter g influences the relative likeli-
hood of the photon emerging from the base or the top of the cloud
in the conservative scattering case (Fig. 13.2).
It is possible to numerically simulate radiative transfer in clouds
using exactly the procedure described above. Such a method is
called a Monte Carlo model, because the fate of each photon on each
leg of its journey is simulated as random process, much like the spin
of a roulette wheel or the roll of a pair of dice. Monte Carlo models
are easy to understand and easy to implement. They are especially
useful for handling non-plane parallel cloud geometries. The only
real drawback to Monte Carlo methods is that a very large number
of photon trajectories might have to be simulated in order to obtain
statistically accurate results. For some problems, especially those in-
volving very deep clouds with little or no absorption, Monte Carlo
methods are inefficient, because each photon might undergo thou-
sands of scattering events inside the cloud before finally emerging
to be counted toward either the albedo or diffuse transmittance.
Other methods for solving multiple scattering problems do not
simulate individual photon trajectories but rather seek mathemat-
ical or numerical solutions to the radiative transfer equation. We
now turn our attention to the simplest such method.

13.2 The Two-Stream Method

13.2.1 Azimuthally Averaged RTE
Recall first of all that the contribution a beam of radiation makes to
the horizontal flux doesn’t depend on its azimuthal angle φ but only
386 13. Radiative Transfer with Multiple Scattering

on the angle from vertical, as embodied in µ = cos θ. This means

that we needn’t worry about the azimuthal dependence of I, as long
as we get its dependence on µ and τ right.
Therefore, let’s define the azimuthally averaged intensity as

2π
1
I (µ) ≡ I (µ, φ) dφ . (13.2)
2π 0

Also, if we assume an azimuthally isotropic scattering medium, so

that p(µ, φ; µ , φ ) is equivalent to p(µ, µ , ∆φ), where ∆φ ≡ φ − φ ,
then we can also define the azimuthally averaged phase function as

2π
1
p(µ, µ ) ≡ p(µ, µ , ∆φ) d(∆φ) . (13.3)
2π 0

With these definitions, we can eliminate φ from (13.1) by taking

the azimuthal average of all terms in (13.1) and simplifying to get

1
dI (µ) ω̃
µ = I (µ) − p(µ, µ ) I (µ ) dµ , (13.4)
dτ 2 −1

This is the azimuthally averaged radiative transfer equation.1

Problem 13.1: Show that (13.4) indeed follows from (13.1)–(13.3).

13.2.2 The Two-Stream Approximation

Equation (13.4) is a reasonably general restatement of the full RTE
and contains no indefensible assumptions or approximations. It
therefore can be (and is) used as the starting point for other meth-
ods of solution as well. The specific “leap of faith” that is unique
to the so-called two-stream method, which we will now derive, is
1 The question of how you actually specify p(µ, µ ) in terms of the (usually)
more fundamental p(cos Θ) is addressed in section 6.1.2 of L02. The conversion
depends on the expansion of the phase function as a sum of Legendre polynomials,
as discussed in Appendix A.
 The Two-Stream Method 387

Fig. 13.3: Assumed angular distribution of intensity in the two-stream approxima-

tion.

the assumption that the intensity I (µ) is approximately constant in

each hemisphere. That is,


I↑ µ>0
I (µ) = ↓
(13.5)
I µ<0

where I ↑ and I ↓ are constants. Schematically, this assumption is

depicted in Fig. 13.3.
The assumption of constant intensity through each hemisphere
sounds quite radical, but in fact, it’s not as bad as you might think.
Imagine, for example, that you are hovering in a balloon between
a uniform, flat surface (e.g., the ocean, cornfield-covered Iowa, or
snow-covered Greenland) and the base of heavy stratiform cloud
layer. If you look at any point below the horizon (µ = 0), you see
the more or less uniform color and brightness associated with the
surface. Everywhere above the horizon, you see the more or less
uniform grayness of the cloud layer. The horizon itself presents a
rather sharp discontinuity in intensity between the ground and the
overlying cloud layer.
Within the cloud layer, the contrast between the lower and up-
per hemisphere is less dramatic, but the assumption of constant I for
each hemisphere is still not patently ridiculous. And even if there is
in fact some dependence of I on µ within each hemisphere, it’s safe
to say that this dependence will usually be monotonic (e.g., steady
brightening from the horizon to zenith) and that you could always
find an intermediate value of µ for which the intensity is close to the
average for the entire hemisphere.
388 13. Radiative Transfer with Multiple Scattering

The Two-Stream Equations

So much for the rationalization; now for the math. We will set up
separate equations for the upwelling and downwelling streams of
radiation. The mechanics of the derivation are the same for each
stream, so we’ll save space by going through the details for only the
upwelling stream I ↑ . We start by substituting the constants I ↑ and
I ↓ for I (µ) in the appropriate locations in (13.4):

1
0
dI ↑ ω̃ ω̃
µ = I↑ − ↑
p(µ, µ ) I dµ − p(µ, µ ) I ↓ dµ . (13.6)
dτ 2 0 2 −1

Note that the integration of p(µ, µ ) I (µ ) is split into two pieces to

allow us to separately treat the constant I in each hemisphere. I ↑
and I ↓ can be taken outside of the integrals, yielding

1  
0 
dI ↑ ↑ ω̃ ↑ ω̃
µ =I − p(µ, µ ) dµ I − p(µ, µ ) dµ I ↓ .

dτ 2 0 2 −1
(13.7)
It is convenient at this point to define a quantity called the
backscattered fraction b, which will be used as a shorthand for inte-
grals like those appearing above:

0

 1 1 1

 p(µ, µ ) dµ = 1 − p(µ, µ ) dµ , µ>0,
2 −1 2 0
b(µ) ≡



 1 1 1 0
 p(µ, µ ) dµ = 1 − p(µ, µ ) dµ , µ<0.
2 0 2 −1
(13.8)
b(µ) represents the fraction of radiation that is scattered into the
opposite hemisphere, with new direction of propagation µ. In the
case of (13.7), it characterizes the degree to which backscattering
enhances I ↑ at the expense of I ↓ . We can now rewrite (13.7) as
dI ↑
µ = I ↑ − ω̃ [1 − b(µ)] I ↑ − ω̃b(µ) I ↓ . (13.9)
dτ
This equation still contains an explicit dependence on µ, which just
muddies the water, because we’ve already stipulated that I is inde-
pendent of µ in each hemisphere. So let’s eliminate µ by averaging
(13.9) over the entire hemisphere:

1 
dI ↑ ↑ ↑ ↓
µ = I − ω̃ [1 − b(µ)] I − ω̃b(µ) I dµ . (13.10)
0 dτ
 The Two-Stream Method 389

The result can be written as

1 dI ↑
= I ↑ − ω̃ (1 − b) I ↑ − ω̃bI ↓ , (13.11)
2 dτ
or
1 dI ↑
= (1 − ω̃ ) I ↑ + ω̃b( I ↑ − I ↓ ) , (13.12)
2 dτ

where
1
b≡ b(µ) dµ . (13.13)
0
Repeating the above steps for the downward stream yields the anal-
ogous equation

1 dI ↓
− = (1 − ω̃ ) I ↓ − ω̃b( I ↑ − I ↓ ) . (13.14)
2 dτ

Equations (13.12) and (13.14) are the so-called two-stream equations

for diffuse incidence.2 Since I ↑ (τ ) and I ↓ (τ ) are unknown and appear
in both equations, we’re evidently dealing with a coupled pair of
ordinary linear differential equations. The usual way to solve such
a system is to combine them into a single second-order differen-
tial equation, apply boundary conditions, and solve for the specific
boundary conditions of interest. But before we do, let’s take a closer
look at the mean backscatter fraction b.

The Backscatter Fraction and g

The mean backscatter fraction b is explicitly related to the scattering
phase function p(µ, µ ) via (13.8) and (13.13). The properties of the
phase function can in turn be partly characterized via the asymme-
try parameter g, which was defined by (11.20). It follows that there
could be some kind of systematic relationship between b and g that
would allow us to replace b in (13.12) and (13.14) with a suitable
2 With relatively little effort, one can generalize the above equations to accom-

modate illumination at the top of the atmosphere by a direct beam of radiation

from the sun — see for example TS02, Ch. 6.
390 13. Radiative Transfer with Multiple Scattering

function of g. This possibility can be made clearer by considering

three special cases.
If scattering is perfectly isotropic [p(cos Θ) = p(µ, µ ) = 1], then
g = 0. In this case, regardless of the direction from which the radi-
ation comes originally, it is equally likely to be scattered into either
hemisphere, so clearly b = 1/2.
If g = 1, this implies that all radiation is scattered in exactly
the same direction as it was traveling before being scattered. Thus,
absolutely no radiation can ever be scattered back into the opposite
hemisphere; therefore b = 0 for this case. Likewise, if g = −1, then
all radiation is scattered into the opposite hemisphere, and b = 1.
To summarize, we have the following known mappings be-
tween g and b:
g = −1 → b=1

1
g=0 → b=
2

g=1 → b=0

If we now assume that the relationship between g and b is linear,3

then we can write
1−g
b= . (13.15)
2
Making this substitution in (13.12) and (13.14), we have

1 dI ↑ ω̃ (1 − g) ↑
= (1 − ω̃ ) I ↑ + (I − I↓) , (13.16)
2 dτ 2

1 dI ↓ ω̃ (1 − g) ↑
− = (1 − ω̃ ) I ↓ − (I − I↓) . (13.17)
2 dτ 2

3 This is another approximation; see TS02 Section 7.5 for a full discussion of the

relationship between b and g.

 The Two-Stream Method 391

13.2.3 Solution
We start by adding and subtracting (13.16) and (13.17) to obtain

1 d ↑
( I − I ↓ ) = (1 − ω̃ )( I ↑ + I ↓ ) , (13.18)
2 dτ

1 d ↑
( I + I ↓ ) = (1 − ω̃g)( I ↑ − I ↓ ) . (13.19)
2 dτ
We then differentiate (13.19) to get

d2 ↑ d
( I + I ↓ ) = 2(1 − ω̃g) ( I ↑ − I ↓ ) . (13.20)
dτ 2 dτ
But note now that the derivative on the right hand side can be re-
placed with an expression obtained from (13.18), yielding

d2 ↑
( I + I ↓ ) = 4(1 − ω̃g)(1 − ω̃ )( I ↑ + I ↓ ) . (13.21)
dτ 2
Applying the same procedure as above to (13.18) gives

d2 ↑
( I − I ↓ ) = 4(1 − ω̃g)(1 − ω̃ )( I ↑ − I ↓ ) . (13.22)
dτ 2
These two equations are the same, except that in the first one the in-
dependent variable is I ↑ + I ↓ while in the second it’s I ↑ − I ↓ . We can
therefore kill two birds with one stone by solving the single equa-
tion
d2 y
= Γ2 y , (13.23)
dτ 2
where
y ≡ (I↑ + I↓) or y ≡ (I↑ − I↓) , (13.24)
and
√
Γ ≡ 2 1 − ω̃ 1 − ω̃g . (13.25)

The general solution is

y = αeΓτ + βe−Γτ . (13.26)

392 13. Radiative Transfer with Multiple Scattering

It follows that the solutions for I ↑ and I ↓ are likewise sums of expo-
nentials:
I ↑ (τ ) = AeΓτ + Be−Γτ , (13.27)
I ↓ (τ ) = CeΓτ + De−Γτ , (13.28)
where the coefficients A–D remain to be determined. Although
there are four coefficients, they are not all independent. This can
be seen by substituting (13.27) and (13.28) into (13.16), which yields

dI ↑ 
= 2(1 − ω̃ ) AeΓτ + Be−Γτ
dτ   (13.29)
+ ω̃ (1 − g) ( A − C )eΓτ + ( B − D )e−Γτ .

But differentiating (13.27) gives

dI ↑
= AΓeΓτ − BΓe−Γτ , (13.30)
dτ
so we can equate (13.29) and (13.30) to get

[2(1 − ω̃ ) + ω̃ (1 − g)( A − C ) − AΓ] eΓτ =

(13.31)
[− BΓ − 2(1 − ω̃ ) B − ω̃ (1 − g)( B − D )] e−Γτ .

The above equation must be valid for all τ, which is only possible if
the terms multiplying the exponentials are each zero. Solving for A
and C in the first case and B and D in the second, we find
√
C B 2 − ω̃ − ω̃g − Γ 1 − ω̃g − 1 − ω̃
= = = √ ≡ r∞ . (13.32)
A D ω̃ (1 − g) 1 − ω̃g + 1 − ω̃

The reason for using r∞ to denote this ratio will become apparent
later. With this definition, we can write

C = r∞ A ; B = r∞ D , (13.33)

which allows us to rewrite (13.27) and (13.28) as

I ↑ (τ ) = AeΓτ + r∞ De−Γτ , (13.34)

I ↓ (τ ) = r∞ AeΓτ + De−Γτ . (13.35)

 Semi-Infinite Cloud 393

Boundary Conditions
At this point we are nearly finished — only two coefficients in our
solution remain undetermined. To find these coefficients, we need
to supply two boundary conditions appropriate to the problem we
wish to solve. Let’s choose the following:

I ↑ (τ ∗ ) = 0 ; I ↓ (0) = I0 , (13.36)

which state that the lower boundary is black (no upward reflected
radiation at τ = τ ∗ and a known hemispherically averaged intensity
I0 is incident on the top of the atmosphere. With these boundary
conditions, we have
∗ ∗
0 = AeΓτ + r∞ De−Γτ , (13.37)

and
I0 = r∞ A + D . (13.38)
These are solved for A and D, which are then substituted back into
(13.34) and (13.35) to get our final solutions

r∞ I0  
Γ(τ ∗ −τ ) −Γ(τ ∗ −τ )
I ↑ (τ ) = Γτ ∗ 2 e−Γτ ∗ e − e , (13.39)
e − r∞

I0  
Γ(τ ∗ −τ ) 2 −Γ(τ ∗ −τ )
I ↓ (τ ) = e − r ∞ e . (13.40)
eΓτ ∗ − r∞2 e−Γτ ∗

Equations (13.39) and (13.40) give the general two-stream solu-

tion for the case of known uniform (diffuse) illumination I0 at the
top of a cloud layer and a completely absorbing lower boundary.
We now look at the implications of this solution for a few specific
examples.

13.3 Semi-Infinite Cloud

If you want to gain insight into how single scatter albedo ω̃ and
asymmetry parameter g influence the reflection and absorption
394 13. Radiative Transfer with Multiple Scattering

properties of clouds, the interpretation of your results is easiest if

there are no other variables to consider. If a cloud is optically thin,
what you see from above the cloud layer is at least as sensitive to
the radiative properties of the surface below the cloud as it is to the
cloud properties themselves.
To exclude such influences, we will first consider the case of a
semi-infinite cloud; i.e., a cloud layer with an upper boundary at
τ = 0 but which is effectively infinite in depth below that level. A
cloud need not be semi-infinite in a literal sense in order to behave
radiatively like a semi-infinite cloud; all that’s necessary is for it be
so thick that a photon incident on the cloud top has essentially zero
chance of emerging from the bottom before either getting absorbed
or else getting scattered back up through the cloud top.
We adapt (13.39) and (13.40) to the case of a semi-infinite cloud
simply by letting τ ∗ → ∞, which gives us

I ↑ (τ ) = I0 r∞ e−Γτ , (13.41)

I ↓ (τ ) = I0 e−Γτ . (13.42)

13.3.1 Albedo

Armed with (13.41) and (13.42), we can look at several interesting

radiative properties of our cloud. Let’s start by finding the albedo
at cloud top, which is defined as the ratio of the reflected to incident
radiation flux:

π I ↑ (0) I0 r∞ e−Γτ
Albedo = = , (13.43)
π I ↓ (0) I0 e−Γτ

which simplifies to

Albedo = r∞ . (13.44)

We discover that the albedo of a semi-infinite cloud is just r∞ , which

explains (retroactively) why that particular function of ω̃ and g was
singled out for its own special symbol.
 Semi-Infinite Cloud 395

With that fact in mind, let’s study the properties of r∞ more

closely. For convenience, its definition is repeated here:

√
1 − ω̃g −
1 − ω̃
r∞ ≡ √ . (13.45)
1 − ω̃g + 1 − ω̃

For starters, if ω̃ = 1, then r∞ = 1, regardless of the value of g (as

long as g < 1). This makes sense, because if there is zero absorption,
then any photon incident on the top of a semi-infinite cloud must
eventually emerge from the top again, no matter how many times
it gets scattered first. It cannot get permanently lost deep inside
the cloud, because no matter where it is, the photon is still closer to
the top than it is to the (infinitely distant) bottom and therefore its
random wanderings are statistically guaranteed to take it to the top
eventually.
You can also see that if g = 1, then r∞ = 0, regardless of the
value of ω̃ (as long as ω̃ < 1). Again, this makes sense, because
in this case, every photon that is “scattered” continues traveling in
exactly the same direction as before and can never change direction
to return to the surface. However, this case is unrealistic for two
reasons: (i) g is always less than one for any real scattering medium,
and (ii) even if g were equal to one, then you might as well say that
the medium doesn’t scatter, since all “scattered” radiation continues
traveling in its original direction as if it had never been scattered in
the first place.
Having dealt with those two limiting cases, let’s consider the
more realistic situation in which g < 1 and 0 < ω̃ < 1. Fig. 13.4
shows how r∞ varies with ω̃ for two different values of g, the larger
value (g = 0.85) being typical of real clouds in the solar band. The
albedo is zero for ω̃ = 0 and goes to one for ω̃ → 1, as expected. The
slighly less obvious point to note is that the overall absorptivity of
the cloud, which is given in this case by one minus the albedo (since
transmittance is zero for a semi-infinite cloud), is quite significant
even for ω̃ fairly close to one.
For example, ω̃ = 0.999 and g = 0.85 yields r∞ = 0.85, cor-
responding to a cloud absorptivity of 15%. In other words, even
though there is only a very slight chance for a photon to get ab-
sorbed in any single extinction event (in this example, that chance is
396 13. Radiative Transfer with Multiple Scattering

Albedo of a Semi-Infinite Cloud

1
0.9
0.8
0.7
0.6
r 0.5 g=0
8

0.4
0.3
g=0.85
0.2
0.1
0
0.0

0.9

0.99

0.999

0.9999

0.99999

0.999999
∼
ω
Fig. 13.4: The albedo of a semi-infinite cloud, as computed from (13.45).

only 0.1%), there is a much greater probability (15%) that a photon

incident on the top of the cloud will get absorbed at some point in
its wanderings before emerging from the cloud top again. The rea-
son, of course, is that the photon’s probability of survival is equal to
ω̃ n , where n is the number of scattering events it experiences inside
the cloud. For a deep cloud with small scattering co-albedo (i.e.,
1 − ω̃  1), n can be a fairly large number.

Problem 13.2: Assume that r∞ = ω̃ n , where n is the effective mean

number of scatterings that photons incident on a semi-infinite cloud
undergo inside the cloud before reemerging from the cloud top. (a)
For the case that ω̃ = 0.9999 and g = 0.85, compute r∞ and n. (b)
Repeat the above calculation, but with ω̃ = 0.9. (c) Explain why n is
much different for the above two cases.
 Semi-Infinite Cloud 397

13.3.2 Flux and Heating Rate Profile

We can also look at the profile of the net flux Fnet = F ↑ − F ↓
within our semi-infinite cloud. Recall first of all that the two-stream
method assumes uniform (isotropic) intensity within each hemi-
sphere; hence the flux in each direction is just π times the relevant
intensity:
Fnet = π ( I ↑ − I ↓ ) . (13.46)
Substituting (13.41) and (13.42), we have 4

Fnet = −π I0 (1 − r∞ )e−Γτ . (13.47)

Fnet in this case decays in exponential fashion, just as it would

in a purely absorbing medium following Beer’s Law. The rate of
decay is proportional to Γ, which is equal to 2 for a purely absorbing
medium (ω̃ = 0) and decreases with increasing scattering (ω̃ → 1).
In fact, for the purely absorbing case, we can write

Fnet = −π I0 e−2τ = −π I0 e−τ/µ , (13.48)

where µ = 1/2. This is equivalent to Beer’s Law for the case that a
parallel beam of radiation is incident on the top of a purely absorb-
ing layer at an angle of 60◦ from zenith.
For partially scattering cases (0 < ω̃ < 1), the parameter Γ is
less than 2. In this case, scattering effectively reduces the optical
depth of a cloud as compared to the purely absorbing case. If you
measure the net flux at a position τ in the cloud, you will find that it
is equivalent to the flux you would expect to find at a “flux optical
depth” of τflux = Γτ/2 in a purely absorbing cloud.
Recall from (10.54) that the heating rate at a level z in the atmo-
sphere is given by
1 ∂Fnet
H=− (z) , (13.49)
ρC p ∂z
4 Recall that I in this chapter is the spectral (i.e., monochromatic) radiance.

Therefore, Fnet , as used here, is the spectral net flux (units of W m−2 µm−1 ). To
obtain the broadband net flux, you would need to integrate Fnet over a suitable
range of wavelengths, with due allowance for the dependence of I0 , r∞ , Γ, and
τ on λ. It follows that H, as used in this section, is also a spectral rather than
broadband heating rate.
398 13. Radiative Transfer with Multiple Scattering

where C p is the heat capacity of air at constant pressure, and ρ is the

air density at level z. Assuming that the volume extinction coeffi-
cient β e is constant in our cloud, we have β e dz = −dτ, and we can
write
β e ∂Fnet
H= (τ ) . (13.50)
ρC p ∂τ
Substituting (13.47) and letting F0 ≡ π I0 be the incident flux at the
top of the cloud, we get

β e F0 (1 − r∞ )Γ −Γτ
H= e . (13.51)
ρC p

The maximum heating occurs at cloud top (τ = 0) and decreases ex-

ponentially downward, exactly like the net flux itself. Not surpris-
ingly, the rate is also proportional to β e , since this parameter deter-
mines how much physical air mass is associated with a unit change
in τ. If β e is large, the same absorbed radiation heats a smaller mass
of air and the temperature increase is therefore greater.

Problem 13.3: Assume that in the wavelength interval 0.5 µm <

λ < 1.5 µm, the scattering co-albedo of a water cloud is given ap-
proximately by 1 − ω̃ ≈ (1.8 × 10−8 ) exp[10.5λ], with λ given in
units of µm (see Fig. 12.10). Also assume that, in the same wave-
length range, the spectral flux (W m−2 µm−1 ) from the sun inci-
dent normally on the top of a semi-infinite cloud is given by F0 =
(6.1 × 10−5 ) Bλ ( T ), where T = 6000 K.
(a) For intervals of 0.2 µm or less, evaluate and graph the spec-
tral heating rate H at cloud top (τ = 0), assuming the following
parameter values: g = 0.85, β e = 0.2 m−1 , C p = 1004 J/(kg K), and
ρ = 1.0 kg/m3 . Give your results in units of K/(day µm). Hint: It
might save time to write a short computer program to perform the
above calculations.
(b) Explain why the maximum heating rate is found near the mid-
dle of the wavelength range rather than at either extreme.
(c) By estimating the average height of your plotted curve in (a)
and multiplying that height by the appropriate interval ∆λ, deter-
mine the total heating rate [K/day] contributed by absorbed solar
radiation between 0.5 µm and 1.5 µm. If your results seem unrea-
sonably large or small, check your units.
 Nonabsorbing Cloud 399

13.4 Nonabsorbing Cloud

Let’s now abandon our semi-infinite cloud, and turn to the more
realistic case of cloud layer with finite optical depth τ ∗ . We will
initially consider the case that scattering is conservative; i.e., ω̃ = 1.
This assumption sounds drastic, but in fact the single scatter albedo
of cloud droplets is very close to one over most of the visible band
(see Fig. 12.10), and absorption by clouds is indeed negligible for
most purposes within that band.
Note that if we simply set ω̃ = 1 and then try to evaluate (13.39)
and (13.40) we run into the problem that Γ = 0. Each equation
then collapses to the ratio 0/0, which is undefined. The obvious
workaround is to take the limit of each equation as ω̃ → 0. But
there is an easier way: let’s go back to an earlier step in our original
derivation, namely (13.18) and (13.19). We can now re-solve these
from scratch, using ω̃ = 1. Equation (13.18) then becomes
1 d ↑
(I − I↓) = 0 , (13.52)
2 dτ
which implies

I ↑ − I ↓ = constant → π ( I ↑ − I ↓ ) = Fnet = constant . (13.53)

or
Fnet
I↑ − I↓ = = constant . (13.54)
π
In other words, the net flux does not change with depth in the cloud.
This is what you would expect, because a change of Fnet would im-
ply absorption (and heating), and there can be no absorption when
ω̃ = 1. Similarly, (13.19) becomes

d ↑ 2Fnet
( I + I ↓ ) = 2(1 − g)( I ↑ − I ↓ ) = (1 − g ) , (13.55)
dτ π
which integrates to

2Fnet τ
I↑ + I↓ = (1 − g ) + K , (13.56)
π
Fnet and K are constants of integration whose values will be deter-
mined by the boundary conditions. Solving (13.54) and (13.56) for
400 13. Radiative Transfer with Multiple Scattering

I ↑ and I ↓ gives

Fnet K
I↑ = [1 + 2τ (1 − g)] + , (13.57)
2π 2
and
Fnet K
I↓ = − [1 − 2τ (1 − g)] + . (13.58)
2π 2
We now apply the same boundary conditions as before:
I ↑ (τ ∗ ) = 0, and I ↓ (0) = I0 , giving us

K Fnet
= I0 + , (13.59)
2 2π
and
−π I0
Fnet = . (13.60)
1 + (1 − g ) τ ∗
The general solution of the two-stream equations for the case of
conservative scattering is then

I0 (1 − g)(τ ∗ − τ )
I ↑ (τ ) = , (13.61)
1 + (1 − g ) τ ∗

I0 [1 + (1 − g)(τ ∗ − τ )]
I ↓ (τ ) = . (13.62)
1 + (1 − g ) τ ∗

From these, we can immediately find the cloud-top albedo

I ↑ (0) (1 − g ) τ ∗
r= = , ω̃ = 1 , (13.63)
I ↓ (0) 1 + (1 − g ) τ ∗

and the transmittance

I ↓ (τ ∗ ) 1
t= = , ω̃ = 1 . (13.64)
I ↓ (0) 1 + (1 − g ) τ ∗
 General Case 401

The above expressions for r and t sum to one, as they must, since
there is no absorption.
Not surprisingly, in the limit τ ∗ → ∞, the cloud-top albedo r →
1, which also implies t → 0. A bit more surprising, perhaps, is how
large τ ∗ can become while still permitting significant transmission
of radiation through a nonabsorbing cloud. For example, with τ ∗ =
100, the transmittance t is still about 6%. All of this transmittance
is associated with photons that have been scattered many hundreds
of times on their journey through the cloud layer. It follows that if
ω̃ were even slightly less than one, the fraction of incident photons
that would survive this journey would be substantially reduced.

Problem 13.4: A typical heavy stratocumulus cloud layer has an

optical thickness τ ∗ = 50, ω̃ = 1, and g = 0.85 in the visible band.
(a) Compute its albedo and total transmittance.
(b) If the cloud were perfectly absorbing rather than perfectly
scattering, what optical thickness would yield the same transmit-
tance as in (a), assuming µ = 0.5?

Problem 13.5: Repeat problem 7.11 but this time, for each case,
compute the albedo. What is the difference in albedo for the two
cases, and what does this difference suggest about the potential role
of aerosol pollution in the global energy budget?

13.5 General Case

We previously considered the limiting cases of (i) a semi-infinite
cloud (τ ∗ = ∞) with arbitrary ω̃, and (ii) a nonabsorbing cloud
(ω̃ = 1) with arbitrary optical thickness τ ∗ . Last but not least, we
may look at the more general case of (iii) arbitrary τ ∗ in combination
with arbitrary ω̃ < 1.
402 13. Radiative Transfer with Multiple Scattering

13.5.1 Albedo, Transmittance, and Absorptance

Starting from our two-stream solutions (13.39) and (13.40), we find
that the general expression for the albedo for the case that ω̃ < 1 is

 ∗ ∗
r∞ eΓτ − e−Γτ
r = Γτ ∗ 2 e−Γτ ∗
, ω̃ < 1 , (13.65)
e − r∞

and the total transmittance is

1 − r∞
2
t= ∗ 2 e−Γτ ∗
, ω̃ < 1 . (13.66)
eΓτ − r∞

Problem 13.6: Show how (13.65) and (13.66) were obtained.

The dependence of t and r on τ ∗ is depicted in in Figs. 13.5 and

13.6 for selected values of ω̃, assuming a fairly typical value of g
for cloud droplets in the shortwave part of the spectrum. The cloud
absorptance a = 1 − r − t is plotted in Fig. 13.7 .
These figures speak for themselves. For any given value of τ ∗ ,
even slight decreases in ω̃ can potentially lead to pronounced de-
creases in both t and r and thus an increase in a. We see also that
each of these variables tends to have an asymptotic limit with in-
creasing τ ∗ . The asymptotic limit for r is of course r∞ . Decreasing ω̃
causes the limit to be reached at smaller optical depth.

13.5.2 Direct and Diffuse Transmittance

Recall from Section 7.4.4 that the total transmittance t through a
cloud layer is actually the sum of two components: the direct trans-
mittance tdir and the diffuse transmittance tdiff :

t = tdiff + tdir . (13.67)

 403

1
ω=1
ω=0.999
ω=0.99
0.8 ω=0.9

Transmittance
0.6

0.4

0.2

0
0 10 20 30 40 50 60 70 80 90 100
τ*

Fig. 13.5: Cloud transmittance t as a function of cloud optical thickness τ ∗ for

g = 0.85 and varying values of the single scatter albedo ω̃.

0.8

0.6
Albedo

ω=1
ω=0.999
0.4 ω=0.99
ω=0.9

0.2

0
0 10 20 30 40 50 60 70 80 90 100
τ*

Fig. 13.6: Same as Fig. 13.5, but the plotted quantity is the cloud albedo r.
404

1
ω=0.999
ω=0.99
ω=0.9
0.8
Absorptance

0.6

0.4

0.2

0
0 10 20 30 40 50 60 70 80 90 100
τ*

Fig. 13.7: Same as Fig. 13.5, but the plotted quantity is the cloud absorptance a =
1 − r − t.
 General Case 405

In particular, tdir describes the fraction of radiation incident on the

top of the cloud that passes through to the bottom without either
being scattered or absorbed, while tdiff describes the fraction that
was scattered at least once before emerging from the cloud base. The
version of the two-stream equations we derived does not explicitly
distinguish between the two types of transmission. Nevertheless,
the information is there, for those who know how to get at it.
We start by considering the special case that the layer is nonscat-
tering, in which case ω̃ = 0, r∞ = 0, and Γ = 2. Equation (13.66)
then reduces to
∗
t = tdir = e−τ /µ , (13.68)

where once again µ = 0.5. As already mentioned earlier, we’re basi-

cally saying that the total (flux) transmission in this case behaves as
if we were illuminating our nonscattering layer with a single direct
beam from an effective zenith angle of 60◦ . Note also that we are
claiming that t in the above equation is the same as the direct trans-
mittance tdir . This is because in the absence of scattering, tdiff is by
definition zero.
But wait: if (13.68) gives tdir for nonscattering layer of optical
thickness τ ∗ , then it does so as well for a scattering layer! This is
because tdir , by definition, depends only on τ ∗ and not on ω̃ or g.
So we can subtract tdir from t to get the following expressions for
the diffuse component of the transmittance:



 0 ω̃ = 0 ,



 1 − r∞2
−τ ∗ /µ
∗ −e 0 < ω̃ < 1 ,
tdiff = e − r∞ e
Γτ ∗ 2 − Γτ (13.69)



 1


∗
− e−τ /µ ω̃ = 1 .
1 + (1 − g ) τ ∗

where we have used (13.64) and (13.66) for the total transmittance
in the conservative and nonconservative cases, respectively.
Figure 13.8 shows how the total transmission is partitioned be-
tween tdir and tdiff as functions of τ ∗ for a cloud with ω̃ = 1 (conser-
vative scattering). Note that when a cloud is optically thin (τ ∗  1),
most transmission is direct, and the diffuse component is small. As
406 13. Radiative Transfer with Multiple Scattering

total
diffuse
1 direct

τ* 2

4
0 0.2 0.4 0.6 0.8 1
Transmittance

Fig. 13.8: Diffuse, direct, and total transmittance as a function of τ ∗ in a nonab-

sorbing cloud with g = 0.85.

the layer becomes optically thicker, tdiff first rapidly increases, be-
coming the major contributor to the total transmittance, and then it
trails off slowly, long after tdir has gone to zero.
A good real-world example of this behavior can be observed in
the thickening layer of clouds that often precedes a warm front. In
its early stages, the layer consists of an optically thin veil of cirro-
stratus that does not significantly block the direct rays of the sun
(tdir ≈ 1) and also does not scatter enough sunlight to make itself
clearly visible against the blue sky (tdiff ≈ 0). In fact, you may have
trouble telling it’s even there at all, unless it reveals its presence with
a telltale halo.5 As the cirrostratus layer thickens, it makes the sky
whiter (tdiff initially increases) while the sun’s disk rapidly dims
(tdir decreases). Eventually, as the layer changes from cirrostratus
to altostratus and finally nimbostratus, the sun’s disk becomes in-
visible (tdir ≈ 0) and the sky becomes an ever-darker shade of gray
(t ≈ tdiff → 0).

5 The halo around the sun often seen in the presence of cirrostratus is due to a

local maximum, near Θ = 22◦ , in the scattering phase function p(Θ) for randomly
oriented hexagonal ice crystals — see Section 4.3.1.
 General Case 407

13.5.3 Semi-Infinite Cloud as Approximation

Our first application of the two-stream solution was to find the
albedo of a semi-infinite cloud. It turned out that that albedo is
given by r∞ , which is a function of ω̃ and g. Our more general solu-
tion (13.65) reveals that as you increase τ ∗ , the albedo first increases
rapidly and then more slowly. It eventually “saturates” at a value
approaching r∞ and does not increase significantly with further in-
creases in τ ∗ .
Although a real cloud is never truly semi-infinite, it is reason-
able to ask for what value of τ ∗ a cloud might viewed as effectively
semi-infinite, in the sense that further increases in τ ∗ don’t signifi-
cantly change the cloud’s overall radiative properties. Another way
of posing the same question is to ask how thick a cloud must be,
in order that the underlying atmosphere and/or surface has no sig-
nificant influence on the cloud-top albedo, as seen for example by a
satellite imager. If a cloud layer is effectively semi-infinite at a par-
ticular wavelength, then the cloud-top radiance at that wavelength
can be predicted from the cloud properties alone.
There is more than one way to measure the degree to which a
cloud’s radiative properties have approached the semi-infinite limit.
However, the most straightfoward is the cloud’s total transmittance:
if t ≈ 0, then we can be certain that almost none of the photons
emerging from cloud-top will have ever “seen” the lower boundary
and therefore can’t have been influenced by the details of where,
exactly, that boundary is found (i.e., what the precise value of τ ∗ is)
or its reflective properties.
Of course, t is never exactly zero for finite τ ∗ , so we have to
phrase the question as follows: For what minimum value of τ ∗ is
t < f , where we let f be largest value of the transmittance that we
still consider to be “negligible?” For example, if we set f = 0.01,
then we are saying that up to 1% transmittance will be tolerated
without invalidating the description “effectively semi-infinite.”
Setting t = f = 0.01 in (13.66) and solving for τ ∗ allows us
to express the minimum optical thickness as a function of ω̃ and
g. Representative results are plotted in Fig. 13.9. What we see is
that a strongly absorbing cloud may be effectively semi-infinite for
rather small τ ∗ , while a strongly scattering cloud layer continues to
transmit at least 1% of the radiation incident on the opposite side
408 13. Radiative Transfer with Multiple Scattering

Minimum Thickness of a ’Semi-Infinite’ Cloud

1000

g=0.85

100
g=0
τ*

10

1
0.0

0.9

0.99

0.999

0.9999

0.99999

0.999999

0.9999999
ω

Fig. 13.9: The minimum cloud optical thickness τ ∗ that qualifies as approximately
“semi-infinite”, based on the (arbitrary) requirement that the total transmittance
be less than 1%.

until τ ∗ reaches a value of several hundred.

One conclusion we can draw from this result is that even rel-
atively thin water clouds may be treated as opaque at thermal in-
frared wavelengths for which ω̃  1, while the same clouds may
be far from opaque at visible wavelengths, despite having roughly
the same total optical thickness τ ∗ in both the bands.

Problem 13.7: A certain cloud consists of drops of radius 10 µm.

The total column liquid water path L is 0.05 kg m−2 . Assume an
extinction efficiency Qe ≈ 2 for all wavelengths of interest, and g ≈
0.85.
(a) Compute the optical thickness τ ∗ [see (7.74)].
(b) Determine the range of ω̃ for which the cloud may be treated
as effectively ‘semi-infinite’, based on the information in Fig. 13.9.
 Similarity Transformations† 409

(c) Examine Fig. 12.10 and determine the approximate range

of wavelength over which the cloud cannot be considered ‘semi-
infinite.’ Identify the associated spectral band(s).

13.6 Similarity Transformations†

In the expressions (13.65) and (13.66) that we derived for albedo
and transmittance in the general case, there is an implicit depen-
dence on three radiative quantities: the optical depth τ, the single
scatter albedo ω̃, and the scattering asymmetry parameter g. Yet all
of this dependence is embodied in only two free variables: r∞ and
the product of Γ with τ ∗ . It follows that any two cloud layers hav-
ing the same values of both r∞ and Γτ ∗ are radiatively equivalent to
at least the accuracy of the two-stream approximation.
Because r∞ itself is a function of both ω̃ and g, there is an infin-
ity of combinations of these parameters that map to the same value of r∞ .
So if you were to measure the albedo at the top of a semi-infinite
cloud and find, for example, that r∞ = 0.80, you’d have no way of
knowing whether you were dealing with g = 0 and ω̃ = 0.988, or
with g = 0.8 and ω̃ = 0.998, or with any other combination that pro-
duces the same albedo. What you can uniquely determine, however,
is the similarity-transformed (or adjusted) single scatter albedo, which is
defined as
1−g
ω̃ ≡ ω̃ . (13.70)
1 − gω̃

It tells you what single scatter albedo in combination with isotropic

scattering (g = 0) would give you the same r∞ as your actual g and
ω̃. Note the effect of g > 0 (the usual case) is to make ω̃ < ω̃, except
of course when ω̃ = ω̃ = 1.
Similarly, the similarity-transformed (or adjusted) optical depth is
defined as

τ ≡ (1 − gω̃ )τ . (13.71)
410 13. Radiative Transfer with Multiple Scattering

It tells you what optical depth in combination with isotropic scattering

(g = 0) and adjusted single scatter albedo ω̃ will give you the same
value of Γτ as would the “real” τ in combination with your cloud’s
actual ω̃ and g. For g > 0 and ω̃ > 0, we find that τ < τ.

Problem 13.8: Verify the above interpretations of (13.70) and

(13.71) by showing that r∞ (ω̃ , 0) = r∞ (ω̃, g) and that Γ(ω̃ , 0)τ =
Γ(ω̃, g)τ.

The physical interpretation of (13.71) is straightforward. If g is

close to unity, then scattered radiation will tend to continue more
or less in the original direction of travel, almost as if it hadn’t been
scattered at all. Therefore, radiation will be able to traverse a greater
optical depth without being absorbed than would be the case if g
were smaller. The definition of τ gives an “effective” optical depth
that takes into account this phenomenon.
The interpretation of (13.70) is only slightly more subtle. The
idea here is that if g > 0, then photons incident at cloud top will
have a harder time “turning around” from their original downward
path than would be the case for g = 0. On average, a greater num-
ber of scattering events will have to occur in order for a photon to
have a good chance of re-emerging from the cloud top and con-
tributing to the albedo. Of course, the greater the number of scatter-
ings, the greater the fraction of photons that will be absorbed first,
if ω̃ < 1. The definition of ω̃ takes into account the role of g in the
overall absorptive properties of the cloud.

Problem 13.9: Given ω̃ = 0.99 and g = 0.85, compute ω̃ and r∞ .

13.7 Clouds Over Non-Black Surfaces

In order to obtain our solutions (13.39) and (13.40) to the two-stream
equations, we had to supply two boundary conditions. We chose to
 Clouds Over Non-Black Surfaces 411

take I ↓ (0) = I0 and I ↑ (τ ∗ ) = 0. The latter condition states that the

lower boundary is perfectly black; i.e., there is no radiation incident
on the cloud base from below.
If the lower boundary were in fact non-black, then radiation
transmitted by the cloud would reach the surface, partially reflect
upward and impinge on the cloud base from below. Some of that
radiation would be transmitted back through the cloud. Some of the
remainder would be reflect back downward, increasing the illumi-
nation of the surface, and so on, ad infinitum. The net result would
be (a) an increase in the total downward flux incident on the surface
and (b) an increase in the albedo at cloud top.
In principle, we could re-solve the two stream equations with a
new lower boundary condition to account for a non-black surface.
The new boundary condition would be I ↑ (τ ∗ ) = rsfc I ↓ (τ ∗ ), where
rsfc is albedo of the surface. But there is a simpler way, at least if we
are content with finding the modified fluxes at the upper and lower
boundaries.
Imagine that we have already used (13.65) and (13.66) to find
the reflectivity r and total transmittance t of a cloud layer with spec-
ified τ ∗ , ω̃, and g. The reflectivity and transmittance are intrinsic
to the cloud itself, because there is no contribution from the lower
boundary. The upward flux of radiation from cloud top is

F ↑ (0) = F0 r , (13.72)

where F0 is the incident solar flux. The downward flux below cloud
base is
F ↓ (τ ∗ ) = F0 t . (13.73)
Now let’s place the same cloud layer over a surface with albedo
rsfc . Of the downward flux given by (13.73), a fraction rsfc is reflected
back toward the cloud. A fraction t of that is transmitted through the
cloud, adding to the original F↑ (0) given by (13.72). An additional
fraction r is reflected back downward toward the surface. A fraction
rsfc of that is reflected back upward, where it again contributes to an
increase in F ↑ (0) as well to the flux reflected back downward from
cloud base.
Figure 13.10a depicts the first few terms in the infinite series of
reflections between a cloud layer and the surface. The total upward
flux at cloud top is now the sum of the cloud-reflected component
412

a) Combining cloud layer with reflecting surface

1 r t2 rsfc t2 r rsfc2 t2 r2 rsfc3

Cloud

t t rsfc t r rsfc t r rsfc2 t r2 rsfc2 t r2 rsfc3

Surface

b) Combining two cloud layers

1 r1 t12 r2 t 12r 1r22 t12 r12 r23

Cloud #1

t1 t 1r2 t1 r1 r2 t 1r 1r22 t1 r12 r22 t1 r12 r23

Cloud #2

t1 t 2 t1 t2r1 r2 t1 t2r12 r22

Fig. 13.10: Depiction of the first few transmittance and reflectance contributions
when multiple reflections occur between (a) a cloud layer and a non-black surface,
and (b) two cloud layers.
 Clouds Over Non-Black Surfaces 413

and each of the transmitted contributions due to successive reflec-

tions from the surface:

F ↑ (0) = F0 (r + rsfc t2 + rsfc

2
rt2 + rsfc r t + rsfc
3 2 2
r t +...)
4 3 2
. (13.74)

We can rewrite the above equation as

F ↑ (0)
r̃ ≡ = r + rsfc t2 [1 + rsfc r + (rsfc r )2 + (rsfc r )3 + . . . ] . (13.75)
F0
This does not look like a very convenient solution, unless you hap-
pen to remember the following power series expansion:

1
= 1 + x + x2 + x3 + . . . , (13.76)
1−x
in which case you’ll recognize that an equivalent closed-form ex-
pression for r̃ is

rsfc t2
r̃ = r + . (13.77)
1 − rsfc r

Using similar reasoning, we get

t
t̃ = , (13.78)
1 − rsfc r

where
F↓ (τ ∗ )
t̃ ≡ . (13.79)
F0

Problem 13.10: Derive (13.78).

Let’s consider first the modified cloud-top albedo r̃ given by

(13.77). We note that if either the cloud transmittance t or the surface
reflectivity rsfc are equal to zero, then the second term on the right
vanishes and we’re back to the original reflectivity r computed for
414 13. Radiative Transfer with Multiple Scattering

a black lower boundary. The second thing we notice is that if both

t and rsfc are not zero, then the second term on the right is greater
than zero, implying an enhancement of the albedo relative to the
original r. The third thing we notice is that if t = 1, then r must be
zero (because r + t + a = 1); therefore r̃ = rsfc . The last case is of
course equivalent to having no cloud at all. All of these inferrences
make physical sense.
Recall that t̃ is the ratio of the flux incident on the surface below
the cloud to the original flux F0 incident at cloud top. If t = 0, then t̃
is also zero. If either rsfc or r is zero, then t̃ = t. Again, these results
make sense.
But now consider the case that rsfc > 0 and the cloud absorp-
tance a = 0 so that r = 1 − t. We then get

1−r
t̃ = > t = 1−r . (13.80)
1 − rsfc r

Multiple reflections between the ground and the cloud thus enhance
the downward flux at ground level, relative to what the flux would
have been with a black surface. In other words, making the ground
more reflective below you makes the sky brighter above you! If you
live in a part of the country that gets snowfall, you will undoubtedly
have noticed that an overcast sky is substantially brightened by the
presence of snow on the ground.
More surprising, perhaps, is that if rsfc = 1 and a = 0, then t̃ = 1,
implying that the downward flux below cloud base is then exactly as large
as it is above cloud top, even when a large fraction of the incident
radiation is reflected back to space before even passing through the
cloud! In other words, in the absence of absorption, the presence
of the cloud makes no difference whatsoever to the downward flux
measured by an observer at the surface!
Although the above conclusion might seem counterintuitive,
there is a simple physical explanation based on energy conservation:
if there is no loss of radiative energy due to absorption in or below
the cloud layer, then the upward flux of radiation below cloud base
must increase through multiple reflection until just as much energy
is lost by transmission upward through the cloud as is gained by
downward transmission of the incident flux above the cloud. Since
the transmittance of the cloud is the same in both directions, steady
 Multiple Cloud Layers 415

state is achieved when the two fluxes are equal. And since the up-
ward flux below cloud equals the downward flux when the surface
is perfectly reflective, the downward flux below cloud equals the
downward flux above cloud.

Problem 13.11: The above argument can be equally well applied

to the downward and upward flux (and thus intensity in the two-
stream approximation) of radiation at an arbitrary level τ within a
semi-infinite, nonabsorbing cloud. Specifically, we expect I ↓ (τ ) =
I ↑ (τ ) = I0 .
(a) Outline the physical argument, drawing an analogy to the
case of a finite cloud layer overlying a perfectly reflecting surface.
(b) Demonstrate the stated relationship, using the equations
given earlier for the intensity in a semi-infinite layer.
(c) Demonstrate the stated relationship, using the equations
given for the intensity in a nonabsorbing layer.

Problem 13.12: Assume that the incident flux of visible radiation

on the top of a stratiform cloud layer is F0 = 400 W m−2 . The cloud
itself has a transmittance t = 0.2 and does not absorb. The surface
albedo is initially rsfc = 0.05, but the cloud produces snowfall which
blankets the surface, eventually raising rsfc to 0.95.
(a) Create a table with two rows and six columns. The rows cor-
respond to “before snowfall” and “after snowfall”. The first three
columns will contain, respectively, the downward flux, upward flux,
and net flux at the surface. The last three columns will contain the
same quantities, but at cloud top.
(b) By what percentage did the downward flux at ground level
increase after the snowfall?
(c) For either case, compare the net flux at ground level with that
at cloud top. Is one greater than the other? Why or why not?

13.8 Multiple Cloud Layers

We previously considered what happens when a cloud layer that
partially reflects and partially transmits radiation is combined with
416 13. Radiative Transfer with Multiple Scattering

a non-black (i.e., partially or totally reflective) surface. We can un-

dertake a similar analysis to find the combined radiative properties
of two cloud layers, the first with reflectivity r1 and transmittance t1 ;
the second with r2 and t2 . Figure 13.10b depicts the first few terms
in the infinite series of reflections between the cloud layers. It can
be shown that the total reflectance of the two-layer combination is
given by

t21 r2
r̃ = r1 + , (13.81)
1 − r1 r2

and the total transmittance is

t1 t2
t̃ = . (13.82)
1 − r1 r2

Problem 13.13: Write out the derivation for the above two equa-
tions.

Equations (13.81) and (13.82) apply to a combination of just two

layers, but they can be used to compute the combined reflectance
and transmittance of any number of layers. First, find the reflectance
and transmittance of one adjacent pair of layers. Then treat this pair
as a single layer to be combined with a third layer, and so on, ad
infinitum.

Problem 13.14: Three nonabsorbing layers have transmittances

t1 = 0.2, t2 = 0.3, and t3 = 0.4.
(a) Compute their combined reflectivity and transmittance.
(b) Compare the computed transmittance with that predicted by
Beer’s Law for combinations of nonscattering layers, and explain the
difference.
(c) Under what condition(s) is (13.82) consistent with Beer’s Law,
and why?
 Accurate solution methods† 417

Although we implicitly assumed we were dealing with two dis-

tinct cloud layers separated by clear space, the space between the
layers plays no role (as long as it is nonscattering and nonabsorb-
ing, of course). Therefore, we can apply (13.81) and (13.82) equally
well to two contiguous layers of the same cloud, as illustrated by
the following problem:

Problem 13.15: A certain cloud layer has total optical thickness

τ ∗ = 12, single scattering albedo ω̃ = 0.99, and asymmetry parame-
ter g = 0.85.
(a) Use (13.66) and (13.65) to compute the total reflectivity and
transmittance of the cloud layer.
(b) For the same values of ω̃ and g, compute the reflectivities and
transmittances of layers having τ ∗ = 3 and τ ∗ = 9, respectively.
(c) Use (13.81) and (13.82) to combine your results from (b) in
order to find the total reflectivity and transmittance of the layer with
τ ∗ = 12. Do your results agree with your answers to (a)?

13.9 Accurate solution methods†

The two-stream method and its close relatives (e.g., the so-called Ed-
dington approximation) are often adequate for computing shortwave
fluxes, including albedo and total transmittance, in plane-parallel
clouds, given τ ∗ , ω̃, and g. However, you should not forget that it
is an approximate solution based on several fairly drastic assump-
tions, such as uniform intensity (I (µ, φ) = constant) in each hemi-
sphere. Ninety-nine percent of the time when you’re outside during
the daytime, your own eyes tell you that shortwave radiant inten-
sity I ↓ is a strong function of direction.6
You should therefore should not rely on the two-stream method
in either of the following cases:

• You require the ability to compute fluxes with high accuracy,

taking into account the details of the scattering phase func-
tion p(Θ) and the surface bidirectional reflectance function
ρ(θi , φi ; θr , φr ); or
6 The other one percent of the time, your eyes are probably closed.
418 13. Radiative Transfer with Multiple Scattering

• You require the ability to calculate radiant intensities in one

or more specific directions, such as those viewed by a remote
sensing instrument in space or on the ground.

For such applications, you need a suitable numerical method

for solving the full radiative transfer equation (11.13) for a plane-
parallel atmosphere. In addition, there are times when you can’t
safely ignore three-dimensional structure, in which case even the
plane-parallel assumption goes out the window and you have to
fall back on (11.8) as your starting point.
Fortunately, there are quite a few off-the-shelf radiative trans-
fer codes available that can be adapted to almost any problem that
you’re likely to encounter. It is therefore not usually necessary even
for radiation or remote sensing specialists to derive and program
their own radiative transfer codes, unless they have very special
needs. And most nonspecialists (i.e., the presumed majority of read-
ers of this book) will not have to deal with these codes at all, let
alone know much about what goes on “under the hood.”
One of the most popular and flexible plane parallel radiative
transfer codes in use today, called DISORT, is based on the discrete
ordinates method. The discrete ordinates method can be thought of
as a generalization of the two-stream method to an arbitrarily large
number of discrete “streams” of radiation in each hemisphere, each
one representing a different direction.
Other common radiative transfer methods suitable for multiple
scattering problems include the adding-doubling method, successive
orders of scattering, and the Monte Carlo method. The last of these
was briefly outlined in section 13.1.
The adding-doubling method may be thought of as a general-
ization of equations (13.77), (13.78), (13.81), and (13.82). In this
method, the scalar reflectance r and transmittance t for each layer
are replaced with N × N matrices R and T, which represent the cou-
pling of intensities between each of N discrete directions for the in-
cident radiation into each of the corresponding outgoing directions
for reflected and transmitted radiation, respectively. Division in the
above equations is replaced with the matrix inverse operator. If you
then start with a layer which is very thin, then the single scatter-
ing approximation applies, allowing R and T to be computed from
the scattering phase function alone. Thereafter, it is only necessary
 Accurate solution methods† 419

to combine layers successively until you achieve the desired total

optical thickness τ ∗ .
The method of successive orders of scattering is most suitable
when dealing with a cloud layer that is either not too thick or else
is moderately absorbing. In this method, you first determine the
radiation field that results from single scattering of the incident ra-
diation. This field is then used to determine the contribution from
radiation that is scattered a second time, and so on. Summing the
contributions from all orders of scattering gives the complete field
of multiply scattered radiation. Of course, the sum converges after
a reasonable number of iterations only if a significant fraction of the
radiation is lost from the cloud layer with each successive scatter-
ing, either due to absorption or because the radiation escapes from
the upper or lower boundary of the cloud.
All of the above numerical methods are covered in far greater
depth by L02 (Ch. 6), TS02 (Ch. 8), and GY89 (Ch. 8).