You are on page 1of 460

LIGHT SCATTERING B Y ICE CRYSTA LS

Fundamentals and Applications

This research volume outlines the scientific foundations that are central to our current under-
standing of light scattering, absorption and polarization processes involving ice crystals.
It also demonstrates how data from satellite remote sensing of cirrus clouds (comprising
various ice crystal sizes and morphologies) can be combined with radiation parameteriza-
tions in climate models to estimate the role of these clouds in temperature and precipitation
responses to climate change. Providing a balanced treatment of both the fundamentals and
applications, this book synthesizes the authors’ own work, as well as that of other leading
researchers in this area, in a coherent and logical presentation. Numerous illustrations are
included, including three-dimensional schematics, in order to provide a concise discussion
of the subject and enable easy visualization of the key concepts.
This book is intended for active researchers and advanced graduate students in atmo-
spheric science, climatology and remote sensing, as well as scholars in related fields such
as ice microphysics, electromagnetic wave propagation, geometric optics, radiative transfer
and cloud–climate interactions.

kuo-nan liou is a Distinguished Professor of Atmospheric Sciences and founding


director of the Joint Institute for Regional Earth System Science and Engineering at the
University of California, Los Angeles. Dr. Liou has received numerous awards including
the 1998 Jule G. Charney Award from the American Meteorological Society, for his pio-
neering work in the theory and application of radiative transport and its interaction with
clouds, and the 2013 Roger Revelle Medal from the American Geophysical Union for out-
standing contributions in atmospheric sciences. He is a member of the National Academy
of Engineering and of Academia Sinica.
ping yang is Professor and Head of the Department of Atmospheric Sciences, and the
David Bullock Harris Chair in Geosciences, at Texas A&M University, where his research
interests cover the areas of remote sensing and radiative transfer. He received the 2013
Ascent Award from the Atmospheric Sciences Section of the American Geophysical Union,
of which he is an elected Fellow.
L I G H T S C AT T E R I N G B Y
I C E CRY S TA L S
Fundamentals and Applications

KUO-NA N L IOU
University of California, Los Angeles

and

P I N G YA N G
Texas A&M University

with contributions by

YO S H I H I D E TA K A N O
University Printing House, Cambridge CB2 8BS, United Kingdom

Cambridge University Press is part of the University of Cambridge.


It furthers the University’s mission by disseminating knowledge in the pursuit of
education, learning and research at the highest international levels of excellence.

www.cambridge.org
Information on this title: www.cambridge.org/9780521889162

C Kuo-Nan Liou and Ping Yang 2016

This publication is in copyright. Subject to statutory exception


and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2016
A catalog record for this publication is available from the British Library
ISBN 978-0-521-88916-2 Hardback
Cambridge University Press has no responsibility for the persistence or accuracy
of URLs for external or third-party internet websites referred to in this publication,
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
Epigraph

To all the happy ice crystals in planetary atmospheres

Let there be light.


Let there be beautiful ice crystals in the air and mountain ranges.
And here come the reindeers and Santa Claus carrying Maxwell’s equations, and
light rays are shining in the wonderlands.
Let the glory of Geometric Optics for ice crystals, Newton’s optics, and sun’s
light rays rise again from the horizon.
Let ice crystals’ old friends – black carbon and dust – be not forgot for Auld
Lang Syne.
And ice crystals are carried by the ceaseless winds; and
After traveling thousands of miles up and down, the sky looks very blue.
Let there be space missions to tender ubiquitous light rays in the sky,
And all things considered, let light scattering by ice crystals in remote sensing
and climate change be a delight.
Contents

Preface page xii


1 Ice in the Earth’s atmosphere 1
1.1 Introduction to clouds 1
1.2 Geographical and temporal distributions of ice in the Earth’s atmosphere 5
1.2.1 A global perspective on clouds 5
1.2.2 An example of global cloud climatology 9
1.2.3 View from ground-based instruments 11
1.2.4 View from specific satellite instruments 14
1.3 Formation of ice crystals 17
1.3.1 The structure of ice crystals 17
1.3.2 Homogeneous and heterogeneous nucleation 19
1.3.3 Atmospheric ice nuclei 20
1.3.4 Secondary ice particle production 22
1.4 Growth of ice crystals 22
1.4.1 Growth by diffusion 22
1.4.2 Growth by accretion 29
1.5 Illustration of ice crystal morphology 30
1.5.1 Some historical perspectives 30
1.5.2 Classification of ice crystals 32
1.5.3 Computer generation of three-dimensional ice crystals 33
1.6 Ice crystal size and shape distributions in cirrus clouds 41
1.6.1 Introductory remarks 41
1.6.2 Ice crystal size and shape spectra 41
1.6.3 Correlation between ice crystal maximum dimension and width 43
1.7 Correlation between ice water content and ice crystal size 45
1.7.1 Introductory note 45
1.7.2 Correlation analysis 46

vii
viii Contents

1.8 Numerical modeling of cirrus cloud formation 50


1.8.1 A review 50
1.8.2 A two-dimensional cirrus cloud model 52
1.8.2.1 Ice microphysics 53
1.8.2.2 Radiative transfer 55
2 Fundamentals of light scattering by ice crystals 58
2.1 Introductory remarks 58
2.2 Maxwell’s equations, wave equations, and boundary conditions 60
2.2.1 Maxwell’s equations 60
2.2.2 Boundary conditions 62
2.2.2.1 Normal components 62
2.2.2.2 Tangential components 63
2.2.3 Wave equations and some analytic solutions 64
2.2.3.1 Spherical coordinates 64
2.2.3.2 Cylindrical coordinates 68
2.2.3.3 Spheroidal coordinates 72
2.3 Optical properties of ice: index of refraction 75
2.3.1 Complex index of refraction and dispersion of light 75
2.3.2 Optical properties of ice 77
2.3.3 Indices of refraction for ice 79
2.4 Definition of the single-scattering and polarization properties
of ice crystals 83
2.4.1 Representation of a simple wave 83
2.4.2 Representation of electromagnetic waves by Stokes parameters 86
2.4.3 Scattering phase matrix for an ice crystal 93
2.4.4 Extinction cross-section for a non-spherical ice crystal 95
2.4.5 Scattering phase matrix for an ensemble of ice crystals 97
2.4.6 Some examples of the single-scattering properties of ice spheres,
cylinders, and spheroids 101
2.5 Introduction to radiative transfer in ice crystal clouds: link to single
scattering 106
2.5.1 Formulation of scalar and vector radiative transfer equations 106
2.5.2 Radiative transfer in horizontally oriented ice particles 111
3 Principles of geometric optics for application to light scattering by ice crystals 115
3.1 Application of geometric optics to light scattering by ice crystals 115
3.1.1 Snell’s and Fresnel’s laws for geometric optics 118
3.1.2 Absorption effects in the context of geometric optics 120
3.1.3 Geometric ray tracing and the Monte Carlo hit-and-miss approach
to ice crystals 124
3.1.4 Illustration of ice optics in the atmosphere 127
3.1.4.1 22° and 46° halos 128
3.1.4.2 Sundogs and their geometry 129
Contents ix

3.1.4.3 Horizontally oriented plate and column crystals 131


3.1.4.4 An observed halo and arc complex pattern 134
3.2 Fraunhofer diffraction for light scattering by ice crystals 134
3.2.1 Sphere 135
3.2.2 Rectangular parallelepiped 136
3.2.3 Hexagonal cylinder 139
3.2.4 Spheroid 140
3.2.5 Some numerical results 142
3.3 Generalized geometric optics for light scattering by ice crystals 146
3.3.1 Introduction 146
3.3.2 Conventional approach 147
3.3.3 Improved geometric-optics approach using the electromagnetic
equivalence theorem 158
3.3.4 Ray-by-ray approach to extinction and absorption calculations 163
3.3.5 Contributions of surface waves to spheres: the edge effect 165
3.3.6 The geometric-optics and surface-wave approach for spheres 168
3.3.7 Application of the geometric-optics and surface-wave approach to
ice crystals 169
3.3.7.1 Fundamentals 169
3.3.7.2 Geometric optics 170
3.3.7.3 Addition of surface waves 171
3.3.7.4 Comparison with available FDTD and DDA results 173
3.3.8 Geometric-optics and surface-wave approach coupled with the
Rayleigh–Gans–Debye approximation 173
3.4 A unified theory of light scattering by ice crystals based on generalized
geometric optics 178
3.4.1 The essence of a unified theory of light scattering by ice crystals 178
3.4.2 Laboratory scattering and ice microphysics experiments with ice
crystals 185
3.4.3 Applications to snow grains contaminated by black carbon (BC)
and dust 192
4 Other useful approaches to light scattering by ice particles 197
4.1 Finite-difference time domain method 197
4.1.1 A brief review 197
4.1.2 Numerical discretization of Maxwell’s curl equations 200
4.1.3 Absorbing boundary conditions and electromagnetic fields in the
frequency domain 204
4.1.4 Near-to-far-field mapping and optical property calculations 207
4.1.5 Validation of the method and some pertinent results 209
4.1.6 Pseudo-spectral time domain method 215
4.2 T-matrix numerical method 222
4.2.1 Fundamentals of the T-matrix approach 222
4.2.2 Invariant imbedding T-matrix method 226
x Contents

4.3 Discrete dipole approximation 238


4.3.1 Fundamentals of the DDA approach 238
4.3.2 Comparison among DDA, FDTD, and PSTD 244
5 Application of light scattering by ice crystals to remote sensing 247
5.1 Introduction to remote sensing of ice clouds in the Earth’s atmosphere 248
5.1.1 Atmospheric composition and structure 248
5.1.2 Atmospheric absorption spectrum 250
5.1.3 Relative position of the sun–satellite system and radiative transfer 253
5.1.4 Bidirectional reflectance spectra of ice 254
5.1.5 The A-Train satellite constellation 257
5.2 Determination of cirrus cloud optical depth and ice crystal size 260
5.2.1 Definition of optical depth and mean effective size 260
5.2.2 Two-channel inversion technique 262
5.2.3 The importance of ice crystal phase function in the two-channel
approach 267
5.2.4 The current MODIS operation program for ice cloud studies 269
5.2.5 Global mapping of cirrus clouds 277
5.2.5.1 MODIS 277
5.2.5.2 AIRS 279
5.2.5.3 CERES-2 283
5.2.6 Detection of thin cirrus 285
5.3 Remote sensing of cirrus cloud vertical size profile 288
5.3.1 The importance of cirrus cloud vertical profiles 288
5.3.2 Development of a retrieval algorithm for vertical sizing 291
5.3.3 Validation of the vertical sizing algorithm 295
5.4 Remote sensing of cirrus clouds using reflected polarization 301
5.4.1 An historical review of polarization measurements 301
5.4.2 Reflected polarization of sunlight from ice crystal clouds 303
5.5 Principle of backscattering depolarization for differentiation of ice
crystals and water drops 306
5.5.1 Theoretical basis 307
5.5.2 Issues of orientation and theoretical results 310
5.5.3 Experimental arrangements and results 313
5.5.4 Application to lidar (laser radar) 314
5.6 Remote sensing of cirrus using reflected line spectra in
1.38 and 0.76 µm bands 318
5.6.1 A brief review of the use of line spectra for Earth
remote sensing 318
5.6.2 Reflected line spectra in the 1.38 µm band 319
5.6.3 Reflected line spectra in the 0.76 µm band 323
5.6.4 Retrieval of cirrus cloud parameters using reflected line spectra 325
Contents xi

6 Application of light scattering by ice crystals to climate studies 327


6.1 Physical basis for the spectral single-scattering parameterization
of ice crystals 328
6.1.1 Physical foundations 329
6.1.2 A generalized single-scattering parameterization for ice crystals 332
6.1.3 Absorption line formation in the scattering of light by ice crystals 333
6.1.4 Single-scattering properties of combined ice crystals and
absorbing gases 337
6.2 Radiative transfer in cirrus cloudy atmospheres 339
6.2.1 The subject of radiative transfer for application to climate studies 339
6.2.2 Radiative transfer model results versus aircraft observations for
cirrus clouds 341
6.2.3 Radiative transfer model results compared with satellite
observations 345
6.3 Radiative forcing of cirrus clouds 350
6.3.1 A discussion of the radiative forcing of cirrus clouds 350
6.3.2 Theoretical analysis for cirrus radiative forcing 351
6.4 Climatic effects of cirrus clouds 357
6.4.1 Investigation using one-dimensional climate models 357
6.4.1.1 A one-dimensional cloud–precipitation–climate model 359
6.4.1.2 A one-dimensional climate–cirrus-cloud model 362
6.4.2 Ice microphysics sensitivity investigation using GCMs 365
6.4.2.1 The effect of interactive ice crystal size on precipitation
simulations 365
6.4.2.2 The effect of ice crystal shape on cloud radiative forcing
simulations 367
6.5 Climate issues associated with cirrus clouds 370
6.5.1 Contrails and contrail cirrus 370
6.5.1.1 Early studies 371
6.5.1.2 Contemporary programs and investigations 374
6.5.2 The role of cirrus in upper-troposphere and lower-stratosphere
exchange 380
6.5.2.1 Introductory notes 380
6.5.2.2 Cirrus clouds and the mechanism of stratosphere
dehydration 382
6.5.3 Thin cirrus clouds and climate 386
6.5.3.1 Thin cirrus detection and comparison with GCM output 386
6.5.3.2 Some notes on high thin clouds in global radiative and
climate forcing 387
References 392
Index 427
Preface

The preparation of Light Scattering by Ice Crystals: Fundamentals and Applications began
about seven years ago. We thought that sufficient material should be available to compose
a high-level text reflective of the complex and intricate domain of ice crystals in the Earth’s
atmosphere and their interaction with “light” from the sun and that emitted from the Earth
and the atmosphere, with applications to remote sensing and climate studies. This text was
supposed to be a three- to four-year project; however, after sifting through the literature
for about two years, gaps emerged on various subjects, including both fundamentals and
applications. For this reason, we conducted additional research in an attempt to bridge
various gaps that are essential, from our perspective, to the unification of all subjects in a
coherent and logical manner with reference to light scattering by ice crystals. Accordingly,
we are pleased to present this text for active researchers and advanced graduate students
who are interested in general areas of atmospheric physics, atmospheric radiative transfer,
atmospheric optics, computational modeling, cloud–climate interactions, and remote sens-
ing of the atmosphere and oceans within the purview of atmospheric and climate sciences.
It is intended to complement other researchers who work in the field of light scattering by
non-spherical particles, which includes ice crystals.
“Ice in the Earth’s atmosphere,” the title of Chapter 1, plays a key role in the hydrological
cycle and precipitation processes. Furthermore, ice clouds in the upper troposphere through
their solar albedo and infrared greenhouse effects are critical elements in determining
surface and atmospheric temperature patterns within the context of greenhouse warming
and climate change induced by man-made perturbations in greenhouse gases and regional air
pollution. In Chapter 1, we introduce cloud classification, a global view of clouds in general,
and cirrus clouds in particular, followed by discussion of the formation and growth of ice
crystals. We then illustrate the complex nature of ice crystals with reference to their size and
three-dimensional morphology based on findings obtained from laboratory experiments and
aircraft observations, to establish a correlation between ice water content and ice crystal
size. This correlation is important in developing radiative transfer parameterizations in
climate models and in understanding the role of ice in climate radiative forcing. Lastly,
we present a two-dimensional cirrus cloud model to illustrate interactions of winds, ice
microphysics, and radiative transfer.

xii
Preface xiii

In Chapter 2, “Fundamentals of light scattering by ice crystals,” a number of fundamental


subjects are presented in relation to light scattering by ice crystals. We discuss the scope
and boundaries of light scattering by ice crystals and present the fundamental Maxwell’s
equations, leading to vector wave equations whose solutions require the imposition of
boundary conditions. We show that exact analytic solutions of vector wave equations exist
only for spherical, circular cylindrical, and spheroidal coordinates. The optical properties
of ice are then introduced, followed by a discussion that defines the single-scattering and
polarization properties of non-spherical ice crystals, including the meaning of the scattering
phase matrix. We subsequently discuss the link between single-scattering properties of ice
crystals, deduced from the independent scattering concept, and the transfer of radiation,
including multiple scattering and emission within ice crystal clouds.
Chapter 3, entitled “Principles of geometric optics for application to light scattering
by ice crystals,” presents the geometric-optics approach to light scattering by ice crystals,
starting with an overview of the essence of geometric optics, including diffraction and
surface waves, from several historical perspectives. We then illustrate fascinating ice optics
produced by randomly and horizontally oriented ice particles by means of Monte Carlo
geometric ray tracing. Subsequently, we demonstrate that exact solutions for diffraction
involving a number of ice crystal shapes can be analytically derived, followed by discus-
sion of conventional and improved geometric-optics approaches, and, based on a number of
postulations, introduce surface-wave contributions – the edge effect – for spheres with mod-
ification to hexagonal ice crystals. Lastly, we present a unified theory of light scattering by
ice crystals on the basis of the geometric-optics surface-wave approach. In this discussion,
theoretical phase functions are compared with those determined from controlled laboratory
light scattering and spectroscopic experiments, as well as application to light absorption
and scattering by snow grains internally contaminated by black carbon and dust particles
wherein stochastic processes are further introduced.
In Chapter 4, “Other useful approaches to light scattering by ice particles,” we confine
our presentations to three contemporary numerical approaches to light scattering by non-
spherical particles within the purview of their applications to light scattering by ice crystals,
namely the finite-difference time domain method, the T-matrix numerical method, and the
discrete dipole approximation. Single-scattering and phase matrix results determined from
these methods for applicable ice crystal size and shape ranges have been used to cross check
and calibrate those computed from a number of geometric-optics approaches. Moreover, we
develop numerical techniques to improve the first two methodologies in terms of particle
size applicability, as well as application to intricate particle shapes.
The subject of “Application of light scattering by ice crystals to remote sensing” is
presented in Chapter 5, wherein we first discuss atmospheric composition and structure,
the atmospheric absorption spectrum, sun–satellite geometry, radiative transfer, and the
contemporary A-Train satellite constellation, which are important for cirrus cloud detection
and quantification. From that base, we then present the subject of retrieving the optical depth
and ice crystal size of cirrus clouds using reflected visible and near-infrared radiation and
illustrate the importance of the phase function of ice crystals. A discussion follows on
xiv Preface

detecting thin cirrus and vertical sizing of cirrus cloud layers. We subsequently cover the
subjects of remote sensing of ice clouds using reflected polarization and the principle of
backscattering depolarization to differentiate ice crystals and water droplets. Lastly, we
present reflected line spectra in the 1.38 µm band and the oxygen A-band for inferring the
composition and optical properties of high clouds.
Chapter 6, the last chapter of the text, comprises discussions of “Application of light
scattering by ice crystals to climate studies.” Herein, we present the physical foundations
for parameterization of the extinction and absorption coefficients and phase function, for
ice crystals imbedded in gaseous absorption line formation. We then discuss delta-two-
stream and delta-four-stream approximations for efficient radiative flux transfer in non-
homogeneous plane-parallel atmospheres and compare theoretical results with aircraft
and satellite observations. On this basis, we present radiative forcing of cirrus clouds
from the viewpoint of theoretical calculations and point out the prevalence of the infrared
greenhouse effect over its solar albedo counterpart. This is followed by a presentation of
climatic effects of cirrus clouds from the perspective of one-dimensional climate models,
where we point out that cloud cover, ice water path, and ice crystal size are influenced by
temperature increases in greenhouse warming scenarios. Also discussed are examples of the
impacts of microphysics on precipitation and radiative forcings using results analyzed from
global climate model simulations. We then present a number of climatic issues associated
with cirrus clouds, including contrails and induced contrail cirrus produced by high-flying
aircraft, a man-made perturbation, their role in upper troposphere and lower stratosphere
exchanges, and the usefulness of optically thin cirrus data determined from modern satellite
instrumentation for ice cloud parameterization development in climate models.
In view of the above, the subject of light scattering by ice crystals as presented in this
text has made definitive contributions to fundamental understanding of and insight into
light scattering, absorption, and polarization processes involving ice crystals. Moreover,
light scattering by ice crystals has provided a new dimension and valuable datasets to the
development of satellite remote sensing of ubiquitous cirrus clouds comprising various ice
crystal sizes and morphologies, as well as to radiation parameterizations for these clouds
in climate models to investigate uncertainties surrounding their role in temperature and
precipitation responses to global warming and climate change.
The senior author wishes to acknowledge the National Science Foundation for its sup-
port over the last 30 years or so of his basic research on light scattering by ice crystals
and related subjects in radiative transfer at the University of Utah and the University of
California, Los Angeles. In particular, he would like to thank the Atmospheric Sciences
Section for a Creativity Award based on contributions to “Light Scattering by Ice Crystals:
Theory and Experiment” (1996) through the recommendation of the late Ronald Taylor and
subsequent support from R. R. Rogers, Bradley Smull, and Chungu Lu, Program Directors
for Dynamical and Physical Meteorology. Additionally, he thanks Roberto Peccei, a par-
ticle physicist and UCLA Emeritus Vice Chancellor for Research, and Joseph Rudnick, a
condensed matter physicist and Senior Dean of the UCLA College of Letters and Science,
for unwavering support in his pursuit of academic and research excellence and the founding
of the Joint Institute for Regional Earth System Science and Engineering at UCLA. He is
Preface xv

also exceedingly grateful to Wilfried Brutsaert and Richard Goody for their encouragement
and advice.
The co-author is grateful to Kuo-Nan Liou, Warren Wiscombe, George Kattawar, and
Jerry North for long-term mentorship. He would also like to thank Hal Maring and Lucia
Tsaoussi, NASA Program Managers, and Chungu Lu, a National Science Foundation
Program Director, for their encouragement in his pursuit of excellence in light scattering
research over the years.
We are indebted to several individuals whose efforts have directly impacted the final
production of this volume: Tara Fickle, a former graduate student in the UCLA English
Department and a part-time editor from 2008 to 2014, for her assistance with typing,
editing, and computer graphics; Janice Amar, retired Chief Operating Officer at the UCLA
Joint Institute for Regional Earth System Science and Engineering and a recalled Senior
Editor in 2015, for her assistance with final editing, permission requests, and compliance
with Cambridge University Press format requirements; Christine Cho for general support
to the authors on various aspects of the book project; Steve Ou for gathering references
and other materials for a number of sections in Chapters 5 and 6; Yu Gu for assistance
with the discussion in Section 1.8; and Lei Bi and Zhibo Zhang for contributions to
the presentation of Chapter 4 and Subsection 5.2.4, respectively. It would not have been
possible to complete the book project without the dedicated support of Yoshihide Takano,
contributing author, who has also made significant contributions to the field, in carrying
out additional calculations, checking fundamental equations, gathering references from
the Internet, as well as making revisions to three-dimensional and color presentations for
numerous figures in the text. We thank Qiang Fu, Professor of Atmospheric Sciences at
the University of Washington, who has made pioneering contributions in areas presented in
Chapter 6, for his critical review and constructive comments. Lastly, we wish to thank Bryan
Baum, Qiming Cai, Andrew Heymsfield, Yongxiang Hu, Hironobu Iwabuchi, Brian Kahn,
George Kattawar, Michael King, Patrick Minnis, Michael Mishchenko, Steven Platnick,
and Manfred Wendisch for research collaboration during the course of our work on light
scattering by ice crystals.
Finally, we acknowledge the generosity of the American Geophysical Union and John
Wiley & Sons (Figures 1.7, 1.8, 1.23, 1.24, 3.32a, 3.32c, 3.33, 5.9, 5.10, 5.11, 5.22,
5.23, 5.24, 5.25, 5.32, 6.1, 6.9, 6.12, 6.15, 6.18, 6.19, 6.20), the American Meteorological
Society (Figures 1.4, 1.5, 1.16, 1.17, 1.22, 1.25, 3.5, 4.5, 5.14, 5.26, 5.30, 5.31, 5.33, 6.3,
6.5, 6.6, 6.7, 6.13), Elsevier (Figures 1.3b, 1.13, 2.7, 2.9, 3.4, 3.11, 3.12, 3.18, 3.19, 3.20,
3.21, 3.22, 4.10, 4.13, 4.17, 4.18, 4.19, 4.20, 4.21, 4.24, 6.16), the European Geophysical
Union (Figures 5.18, 5.27), Hokkaido University (Figures 1.14, 1.15), the Institute of
Electrical and Electronics Engineers (Figures 5.17, 5.19), Nature Publishing Group (6.17),
Taylor & Francis Ltd. (Figure 1.9), the Optical Society of America (Figures 2.4, 2.8, 2.20,
3.28, 3.29, 3.30, 3.31, 4.3, 4.4, 4.6, 4.7, 4.9, 4.11, 4.12, 4.28), Oxford University Press
(Figures 1.10, 5.34), Springer-Verlag (Figure 4.8), and colleagues who have furnished
unpublished material – Les Cowley (Figure 3.9b), Edwin W. Eloranta (Figure 1.6), Marko
Riikonen (Figure 3.9a), Ulrich Schumann (Figure 1.3a), and Stefan Bauer, Janek Zimmer,
and Uwe Reiss (Figure 1.1) – for permission to use this material in this text.
1
Ice in the Earth’s atmosphere

“Let there be light,” and there was light.∗

The fact that the Earth’s sky appears blue is a consequence of the scattering of “light” by
molecules according to the theory of Rayleigh scattering. The fascinating halos and arcs
we see mixed within blue sky result from light scattering by ice crystals.
Our presentation of light scattering by ice crystals begins with an overview of clouds.
This is followed by a global view of ice distribution in the Earth’s atmosphere; formation
and growth of ice crystals; ice crystal morphology, size, and distribution; and a discussion
of cirrus cloud modeling with a linkage to some of these topics.

1.1 Introduction to clouds


I BRING fresh showers for the thirsting flowers,
From the seas and the streams;
I bear light shade for the leaves when laid
In their noonday dreams.
From my wings are shaken the dews that waken
The sweet buds every one,
When rocked to rest on their mother’s breast,
As she dances about the sun.
I wield the flail of the lashing hail,
And whiten the green plains under,
And then again I dissolve it in rain,
And laugh as I pass in thunder.
...
I am the daughter of Earth and Water,
And the nursling of the Sky;
I pass through the pores of the ocean and shores;
I change, but I cannot die.
For after the rain, when with never a stain


A quotation from Genesis 1:3 to the extent to introduce the term, “Let there be light.”

1
2 Ice in the Earth’s atmosphere

The pavilion of Heaven is bare,


And the winds and sunbeams with their convex gleams
Build up the blue dome of air,
I silently laugh at my own cenotaph,
And out of the caverns of rain,
Like a child from the womb, like a ghost from the tomb,
I arise, and unbuild it again.
(Percy B. Shelley, “The Cloud” (1820))

Clouds are formed when water evaporates from oceans and other surfaces (lakes, ponds,
moist land surfaces) carried by convection, orographic, or frontal lifting and rises into the
upper, colder part of the atmosphere. Formation of clouds generally requires the interaction
of water vapor with a type of aerosol referred to as condensation nuclei or ice nuclei. A
cloud becomes visible once the water vapor has been cooled by the condition of water
or ice saturation. However, some very thin clouds cannot be seen by the human eye, and
are classified as subvisual clouds. Clouds are normally produced in a region referred to
as the troposphere (lower atmosphere), the lowest layer of the atmosphere, where weather
activities occur. Clouds are regulated by the hydrological cycle, which involves evaporation,
cloud formation, precipitation, runoff, and large-scale circulation.
In accordance with the World Meteorological Organization (WMO) definition, clouds
are conventionally classified in terms of their position and appearance (shape) in the
atmosphere. Clouds with a base height above 6 km are designated as high clouds, a category
that includes cirrus (Ci), cirrostratus (Cs), and cirrocumulus (Cc). On the basis of the U.S.
1976 Standard Atmosphere classification, 6 km corresponds to a temperature of about
249 K, which is 24 K below the freezing temperature (273 K). Thus, these clouds contain
exclusively ice particles. Cirrus clouds tend to be wispy and transparent (Figure 1.1, upper
left panel). In midlatitude, large numbers of this type of clouds are generally associated with
an approaching storm system. The upper right panel in Figure 1.1 illustrates the appearance
of Cc. Regional convective instability gives this cloud type a rolled or rippled appearance.
The upper middle panel in Figure 1.1 shows the appearance of Cs, which consists of mostly
continuous, wide sheets of clouds that cover a large area of sky and, when it is associated
with frontal systems, is a precursor to rain or snow.
The middle group of clouds, with heights between about 2 km (275 K) and 6 km,
consists of altocumulus (Ac) and altostratus (As), in which ice particles and water droplets
can coexist. The appearance of Ac, displayed in the middle left panel of Figure 1.1, is
a general indication of convective instability at the level of its formation. This cloud can
bring precipitation, usually in the form of virga, a type of precipitation trail that does not
reach the ground. Clouds of As (Figure 1.1, middle right panel) are formed when a stable
air mass is lifted to the level of condensation along a frontal system, which can produce
precipitation.
Low clouds, which are classified as having base heights below 2 km, include stratus
(St), stratocumulus (Sc), and fair-weather cumulus (Cu). These clouds contain exclusively
1.1 Introduction to clouds 3

Figure 1.1 The appearance of high (Ci, Cs, Cc), middle (Ac, As), and low (Sc, St, Cu) clouds in the
Earth’s atmosphere. These pictures are reprinted from the cloud atlas at http://www.clouds-online
.com. Copyright information: http://www.clouds-online.com/imprint.htm.

water droplets. Sc clouds have a lumpy appearance (lower left panel, Figure 1.1). They
commonly form in an unstable air mass following a cold front, and can produce light rain
or drizzle. St clouds (lower middle panel, Figure 1.1) form near the surface, usually over
coastal oceans, have a clearly defined base, and can produce drizzle. Cu clouds (lower right
panel, Figure 1.1) are the product of convective air mass instability and are often associated
with fair-weather conditions.
Other cloud types are associated with substantial vertical development. Cumulonimbus
(Cb) clouds are vertically developed cumulus produced by strong convectively unstable
conditions, principally occurring over tropical oceans and land in the summer. They appear
very dark gray with a cloud base height of about 1 km, and the cloud top can extend to
the top of the troposphere and occasionally into the lower stratosphere (15–18 km). They
generally produce thunderstorms, rain showers, and sometimes hail particles. The upper
portion of these clouds may contain pure ice or a mixture of ice and water. Nimbostratus
(Ns) is a type of cloud that tends to bring constant precipitation and low visibility. They
normally form from altostratus clouds and can thicken into lower levels during precipitation.
Figure 1.2 Cloud types defined in the vertical direction. High clouds have base heights above 6 km, middle clouds are located between about 2 km and
6 km, and low clouds have base heights below 2 km. The vertically developed cumulonimbus can grow up to 15–18 km in the tropics. Nimbostratus are
normally evolved from altostratus clouds.
1.2 Distributions of ice in the Earth’s atmosphere 5

Figure 1.2 depicts the approximate vertical positions in the troposphere of all cloud types
described above, including Cb and Ns, which produce precipitation that reaches the ground.
Cb clouds generate a significant number of anvils in the tropics. Drifting with the winds,
they can last for hours, and subsequently become high-level clouds before dissipating.
Upper-level ice crystal clouds produced by jet aircraft are known as contrails or conden-
sation trails (Appleman 1953). Contrails are visible line clouds resulting from water vapor
emissions that form behind aircraft flying in sufficiently cold air. Persistent contrails often
develop into more extensive contrail cirrus, particularly in ice-supersaturated air masses
in which ice supersaturation is generally too low to allow cirrus clouds to form naturally.
Contrails may enhance the extension of the natural cirrus cover in adjacent areas where
relative humidity is too low for the spontaneous nucleation of ice crystals. A comprehensive
analysis of jet aircraft contrails over the United States and Europe using satellite infrared
imagery has been carried out by Minnis et al. (1998a) and discussed in the IPCC report
(IPCC Report 1999) in conjunction with their climatic impact, but see Subsection 6.5.1
for further discussion. Figure 1.3a shows an unusual spiral contrail formed by a circling
aircraft observed in a NOAA-14 Advanced Very High Resolution Radiometer (AVHRR)
image west of Denmark at 1236 UT, May 22, 1998. Figure 1.3b displays a typical contrail
in the form of limited diffusion amidst a blue sky.
Finally, we would like to note that clouds have also been observed in the upper atmo-
sphere. A type of ice crystal cloud known as a polar stratospheric cloud (PSC) has often
been observed in the polar stratosphere between about 15 and 20 km. These clouds have fre-
quently been detected over the Arctic and Antarctic by limb-viewing satellite instruments
during the winter months when the ambient temperature falls below about 195 K. The
condensation of both water vapor and nitric acid (HNO3 ) results in the formation of HNO3
trihydrates, which serve as nuclei for ice crystal growth. In the stratosphere between about
20 and 30 km, a type of thin cloud usually resembling the cirrus form, and referred to as
mother-of-pearl (nacreous) clouds, has been detected. These clouds appear to be generated
by ice deposition on frozen particles of sulfuric acid. In the mesosphere between about 50
and 55 km, a cloud type known as noctilucent clouds has been observed and is well known.
These clouds are very tenuous and resemble cirrus.

1.2 Geographical and temporal distributions of ice in the Earth’s atmosphere


1.2.1 A global perspective on clouds
Clouds are global in nature, as evidenced from satellite cloud pictures. We shall confine
our discussion to the formation and dissipation of high and middle clouds based on Geo-
stationary Operational Environmental Satellite (GOES) thermal infrared (IR) images from
the perspective of large-scale weather activities, as presented in Liou (1986). Shown in
Figure 1.4a is a full disk IR picture taken at 2345 GMT, February 23, 1984. Warmer areas
are darker, while cooler areas are lighter. Temperature normally decreases with height in the
troposphere; thus, the whitest areas can safely be assumed to be high clouds. This picture
(a)

(b)

Figure 1.3 (a) An unusual spiral contrail, formed by a circling aircraft, surrounded by high clouds,
observed in a NOAA-14 Advanced Very High Resolution Radiometer (AVHRR) image west of
Denmark at 1236 UT, May 22, 1998 (Schumann 2002). (b) A typical contrail amidst a blue sky (Gao
et al. 2006 and courtesy of Randall Friedl of JPL/NASA).
1.2 Distributions of ice in the Earth’s atmosphere 7

(a)

(b)

Figure 1.4 (a) Full disk thermal infrared picture at 2345 GMT February 23, 1984, illustrating
globally distributed clouds in general and cirrus clouds in particular. (b) Same as (a), but for February
25, 1984 (after Liou 1986).
8 Ice in the Earth’s atmosphere

was taken over the Northern Hemisphere during daytime in the later part of winter, when
polar regions are generally outside the frame of the satellite camera, thus minimizing the
possibility that white areas may be associated with cold surface temperatures.
High-level cirrus clouds are globally distributed and present at all latitudes irrespective
of land, sea, or season. These clouds undergo continuous changes in area coverage, thick-
ness, texture, and position. The most striking cirriform cloud feature shown in Figure 1.4a
is the large spiral-comma-shaped pattern west of the Washington State coastline, associated
with a major surface cyclone located to the northeast of the cloud center. Lower clouds
and precipitation associated with the large-scale rising motion are present under much of
this high-cloud canopy. To the north of this cloud band and over the northernmost areas of
the Pacific, another bright cirriform area is associated with a complex of surface lows and
frontal systems. Further to the west, the leading edge of a large cirriform cloud mass is mov-
ing into the IR picture. This cloud mass represents a major storm developing off the coast of
Japan.
The most impressive area of cirriform cloudiness is seen over the Pacific Ocean between
Hawaii and Mexico, and is related to a huge, although rather weak, trough aloft linked to the
subtropical jet stream that curves southward near about (35°N, 160°W). The brightness of
these cirriform clouds coupled with light gray areas, which represent middle clouds below,
indicates an active zone of weather activity. Further east, broken, largely transverse bands of
cirriform clouds are spreading eastward into Mexico. The equatorial area is characterized
by strong, predominantly diurnal convection over western South America, with a large
production of anvil cirrus and a zone associated with the Intertropical Convergence Zone
(ITCZ) that extends across the Pacific at 10°S. In the central Pacific, a collection of
mesoscale and synoptic scale clusters of cumulonimbus, some imbedded in areas of middle
clouds, is producing the brightest (coldest) cirriform cloudiness in the picture.
The dominant cloudiness in the Southern Hemisphere is associated with cirriform clouds
produced by the strong cold front that extends north-northwestward from an occluded front
anchored in an intense low centered at (57°S, 140°W). In addition, some spiral-shaped
cirrus and middle clouds are shown in the vicinity of (31°S, 104°W); these are associated
with a low-pressure system that has been cut off from the westerlies. On the western edge
of the picture, a weak cold front with a thin cirrus band approaches New Zealand. To the
north of this front, a band of cirrus stretching north-northwestward from (30°S, 165°E) is
associated with a surface low at (19°S, 158°E) and a strong wind shear aloft.
In the ensuing 48 hours, the cirriform cloud pattern west of Washington State, as depicted
in Figure 1.4b, changes from a well-defined spiral coupled with the original surface low to a
disorganized, blotchy mass of less bright clouds in the midst of the dissipation–reformation
process, and finally to a redevelopment of brighter, more organized masses coming together
with a major storm development. A huge cirriform spiral, characteristic of extratropical
cyclones, is seen in the north Pacific. The thin line of cirrus on the poleward side of the cold
frontal band stretches from (47°N, 135°W) to (35°N, 154°W). This line occurs adjacent to
the polar jet stream that cuts across the frontal zone near the west coast of central British
Columbia. In conjunction with a huge cirriform frontal band and a spiral center at (43°N,
1.2 Distributions of ice in the Earth’s atmosphere 9

161°E), another major storm is also evident, which was just coming into view from the
west as shown in Figure 1.4b. In the Southern Hemisphere there has been relatively little
change in the 48-hour period between Figures 1.4a and 1.4b, due to a large-scale dynamic
blocking situation. The cirrus cloud band near (55°S, 175°W) in Figure 1.4b is associated
with a strong, new cold front beginning to sweep northeastward.
The preceding discussion makes it clear that the formation, maintenance, and dissi-
pation of high and middle clouds which contain ice particles are principally modulated
by large-scale weather features and disturbances. In the tropics, these phenomena are
related to deep-cumulus outflows associated with strong convection, characteristic of this
region. Thin and subvisual cirrus with optical depths  0.1 have not been identified in the
foregoing presentation due to the limitations of IR temperature techniques. Presented in
Subsection 1.2.2 are the clouds that were seen by specific satellite instruments and the
associated data gathered and analyzed for cloud climatology studies.

1.2.2 An example of global cloud climatology


In what follows, we present an example of global cloud climatology derived from the NOAA
High Resolution Infrared Spectrometer (HIRS) polar-orbiting satellite data from the period
1979–2001 during winter and summer (Wylie et al. 2005). The specific technique, called
the CO2 slicing method, which uses two channels in the 15 µm CO2 band and a 10 µm
window channel, has been developed to determine the statistics of cloud cover, height, and
emissivity. Many processing procedures were required to produce a global map of cloud
parameters (Wylie et al. 1994). Figure 1.5 illustrates the frequency of all clouds as well
as high clouds above 6 km. Clouds are most frequently found in the ITCZ and the midlat-
itude storm belts of the North Atlantic, North Pacific, and Antarctic Oceans, as noted in
Figures 1.4a and 1.4b. Clouds are less frequent between the subtropical high-pressure zones
over the oceans and the subtropical deserts over land.
High clouds are observed in about one-third of the HIRS observations. Their coverage
shows an annual cycle over land with the maximum occurrence during summer in each
hemisphere. The ITCZ is a region of more frequent high clouds, as are the midlatitude
storm belts in the Northern Hemisphere summer. The subtropical high-pressure systems
are evident in the region of less frequent high-cloud cover. Over the Indonesian region,
the ITCZ expands in coverage from winter to summer, whereas in the central Pacific
Ocean, it shows extension during the winter months. For the Southern Hemisphere, the
eastern Pacific Ocean off South America and the eastern Atlantic Ocean off Africa remain
relatively free of high clouds throughout the year. The North American high-cloud cover
shows little seasonal change. The high-cloud cover results derived from the CO2 slicing
method illustrate that high clouds are ubiquitous in the tropics, occurring with more than
70% frequency.
We wish to note that many past attempts have been made to classify the global dis-
tribution of clouds for climate study based on the emitted IR radiation in the 10 µm
window, plus the reflected visible radiation during daytime. The classification of cloudy
All Clouds
December, January, and February June, July, and August

High Clouds

Frequency of Clouds
0% 50% 100%

Figure 1.5 The frequency of all clouds and of high clouds above 440 hPa from 1979 to 2001, from HIRS data during boreal winter [Dec–Jan–Feb (DJF)]
and summer [Jun–Jul–Aug (JJA)] (after Wylie et al. 2005; see text for further discussion).
1.2 Distributions of ice in the Earth’s atmosphere 11

pixels by measured radiances is a complex decision-making process based on computer


algorithms. The best example is the International Satellite Cloud Climatology Program
(ISCCP), a WMO project developed in the 1980s (Schiffer and Rossow 1983, Rossow and
Schiffer 1999). ISCCP has also produced many useful cloud products, which have been
compared with similar results determined from the HIRS CO2 slicing approach discussed
above.
More recently, Minnis et al. (2011a) presented a cloud climatology based on NASA’s
Clouds and the Earth’s Radiant Energy System (CERES) project, analyzing data from
the Tropical Rainfall Measuring Mission (TRMM) Visible and Infrared Scanner and the
MODerate-resolution Imaging Spectroradiometer (MODIS) instruments on board the Terra
and Aqua satellites covering the period 1998 through 2007. King et al. (2013) developed a
cloud climatology from MODIS using over 12 years of continuous observations from Terra
and over 9 years from Aqua. Kahn et al. (2014) reported a cloud climatology determined
from cloud products of the Atmospheric Infrared Sounder (AIRS) and Advanced Microwave
Sounding Unit (AMSU) instrument suite. Cloud climatology products in these studies
include cloud-top temperature, ice cloud fraction, effective ice particle size, and ice cloud
optical depth. Additional discussions are presented in Subsection 5.2.5.

1.2.3 View from ground-based instruments


High-level clouds have been detected by ground-based instruments and their compositions
retrieved and quantified by lidar (laser radar) and millimeter (mm)-wave radar. In this
subsection we present two examples based on these two unique sensors.
During the 1960s, the advent of the laser as a source of energy opened up numerous pos-
sibilities for remote sensing of the atmosphere. Laser energy at optical frequencies is highly
monochromatic and coherent. With the innovation of Q-switching techniques, very short
pulses of high power could be generated, which prompted the development of backscat-
tering lidar techniques for detection of the composition and structure of clouds, aerosols,
and minor gases in the atmosphere. The lidar backscattering equation for probing the atmo-
sphere is similar to that of radar. Many innovative remote sensing techniques have been
developed, including the use of multiple-wavelength lidar to determine the composition
of minor gases and aerosol concentrations by means of differential absorption techniques,
Doppler approaches to infer the motion of particulates and molecules, and the Raman
scattering technique, which produces a weak scattering at a shifted wavelength for water
vapor measurements. In the context of cirrus cloud study, backscattering and depolarization
techniques have been used to retrieve cloud position and thickness, to differentiate ice and
water clouds, to detect ice particle orientation, and to infer particle size and optical depth
with the assistance of other sensors (see, for example, Sassen 1991). The development of a
lidar system on a space platform is discussed in the following paragraphs.
At visible wavelengths, lidar can penetrate through high clouds, which are normally opti-
cally thin with optical depths less than about 3. However, it cannot penetrate through lower
clouds, which have optical depths generally greater than 10–30. Illustrated in Figure 1.6
12 Ice in the Earth’s atmosphere

Top View

24 km

12 km
6 km

North

24 km

Side View
Figure 1.6 Three-dimensional observations of cirrus clouds in the vertical direction and in a hori-
zontal plane, from lidar backscattering (after E. Eloranta, personal communication 1995).

is a three-dimensional backscattering image of cirrus clouds by a scanning lidar system


across the prevailing winds. From this example, cirrus clouds appear to be highly inhomo-
geneous in the horizontal and vertical directions, and frequently have layer structure in the
vertical.
Conventional meteorological radars typically operate at centimeter (cm) wavelengths to
interpret backscattering signals from precipitating clouds that contain large raindrops and
snowflakes on the order of mm to cm sizes. Non-precipitating clouds, however, generally
consist of particles that are much smaller than mm sizes and do not produce significant
radar backscattering signals for their detection. The technical development of mm-wave
radar and its application to cloud studies is a subject of modern research. The selection of
wavelengths for a radar operated in the Earth’s atmosphere is restricted to those spectral
regions where absorption by atmospheric gases is small. The windows of H2 O and O2
absorption are located at 35 GHz (8.6 mm, Ka -band), 94 GHz (3.2 mm, W-band), 140 GHz
(2.14 mm, F-band), and 220 GHz (1.36 mm, G-band). The first mm-wave radar designed for
meteorological use was a 35 GHz system for cloud deck monitoring. It was subsequently
equipped with Doppler and polarization capabilities and employed for the observation of
1.2 Distributions of ice in the Earth’s atmosphere 13

(a) mm-wave radar reflectivity dBZe


-20
Height (km)

10
-30
8
-40

6 -50

18:00 Time (UTC) 17:00


(b) 1.38 μm reflectance %
2.6
36 .6ON 1.6
0.6
98oW 97.5oW 97oW
(c) 8.6 - 11 μm brightness temperature difference K
0

36.6 NO -1
-2
98oW 97.5oW 97oW
(d) P-parameter results

36.6ON Cirrus
Clear
98oW 97.5oW 97oW

Figure 1.7 A case study from the SGP-ARM site on February 11, 2001 (1730 UTC). Plotted in (a)
is the 1-hour mm-wave radar reflectivity time series from Lamont, Oklahoma. The MODIS/Terra
1.38 µm reflectance and 8.6–11 µm BTD are displayed in (b) and (c), respectively, while (d) shows
the cirrus detection P-parameter results (after Roskovensky and Liou 2003a).

clouds. A 94 GHz system with a shorter wavelength was further shown to provide effective
observations of the liquid/ice water content of cloud (in general in this text, “/” is to be
taken to mean “and”) and has been successfully deployed in a small satellite, referred to
as CloudSat, launched in April 2006. CloudSat is one of the constellation satellites in the
A-Train, discussed in Subsection 5.1.5.
Because water vapor concentration decreases rapidly with height, mm-wave radar gen-
erally operates at 35 GHz. Figure 1.7a shows an example of the 35 GHz mm-wave radar
reflectivity time series with a one-hour average observed from DOE’s Atmospheric Radia-
tion Measurement (ARM) site located at Lamont, Oklahoma, on February 11, 2001 (1730
UTC) presented in Roskovensky and Liou (2003a). The radar reflectivity depicts a distinct
two-layer distribution of cirrus clouds moving with the westerly winds. The radar time
series is given backwards, so that the earliest time matches the spatial location at the right
side of the other plots, while the latest time matches the left side. These plots are the 1.38 µm
reflectance (Figure 1.7b) and 8.6–11 µm brightness temperature difference (Figure 1.7c)
from MODIS data that can be used to detect the presence of cirrus clouds, and a thin cirrus
(optical depths between 0.1 and 0.9) mapping technique referred to as the P-parameter
method, developed in the paper (Figure 1.7d).
14 Ice in the Earth’s atmosphere

1.2.4 View from specific satellite instruments


A number of satellite instruments have been used to identify ice clouds and retrieve their
microphysical and optical properties in the atmosphere. However, the following exam-
ples present vertical images and profiles of ice clouds inferred from Cloud-Aerosol Lidar
and Infrared Pathfinder Satellite Observations (CALIPSO), a small satellite launched in
April 2006 specifically to study the impact of clouds and aerosols on the Earth’s radia-
tion budget and climate. It flies in formation with six other satellites in the international
A-Train constellation for coincident Earth observations (see Subsection 5.1.5). For inter-
pretation purposes, we also employ images and products determined from the Moderate-
Resolution Imaging Spectroradiometer (MODIS), referred to in Section 1.2.2, which was
designed to make 0.25–1 km resolution observations in 36 visible and infrared bands.
MODIS was launched in December 1999 on the Terra platform and in May 2002 on the
Aqua platform, which leads the A-Train.
In Figure 1.8 we show the vertical profiles of contrails and cirrus clouds derived from
the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) backscattering signal
operated at the 0.532 µm wavelength (Iwabuchi et al. 2012). The horizontal and vertical
resolutions of CALIOP data used for this presentation are 1 km and 60 m, respectively,
for altitudes between 8 and 20 km. With the help of location information determined from
MODIS data analysis, contrails were searched in the CALIOP profile. Figure 1.8b shows
CALIOP profiles corresponding to the MODIS image in Figure 1.8a.
Contrails were identified based on limited vertical extent with a remarkable backscatter-
ing, as compared with surrounding cirrus. The red vertical lines in the plots denote contrail
locations detected by MODIS image analysis. Young contrails A and B were isolated from
cirrus clouds, having confined widths within 1–2 km. The vertical extent is approximately
250 m for A and 450 m for B. A cirrus cloud with a width of about 30 km is present just
beneath contrail B, which is also shown in the MODIS image. Contrail C is not clearly
exhibited in this profile. Contrail D is imbedded in cirrus, while contrail E extends from
cirrus: both are mature, with moderate vertical and horizontal extents.
Figure 1.9a (Ou et al. 2012) illustrates another example of the vertical feature mask of ice
clouds from CALIOP polarized echoes. Between the two vertical dashed lines, which denote
the southern and northern bounds of the selected domain, clouds (light blue), extending to
5 km and mixed with dust layers (orange), have been identified using lidar depolarization
technique. These CALIOP results correspond to the MODIS/Aqua scene for March 31,
2007 at 0500 UTC over northeastern and eastern China and the East China Sea. Images for
the MODIS (RGB) overpass on March 31, 2007 at 0500 UTC over northeastern and eastern
China and the East China Sea are shown in Figure 1.9b. This case contains a significant
presence of dust layers mixed with clouds. Over northeastern China, brightness temperature
difference values between 11 and 12 µm (BTD11-12) are around −4 K, indicating a heavy
load of dust aerosols. This is a technique used to differentiate between cirrus clouds and dust
layers. Major cirrus clouds were present over northeastern China, Mongolia, and southern
Japan, where cloud temperatures are lower than 243 K. Over the remaining cloudy areas,
1.2 Distributions of ice in the Earth’s atmosphere 15

(a)

(b)

Figure 1.8 (a) An image of MODIS brightness temperature difference (BTD) between 11 µm and
12 µm channels over the Northeast Pacific on January 11, 2009. The cyan line shows the locations of
MODIS-CALIOP collocated pixels. The magenta line segments denote contrails detected by manual
inspection. (b) CALIOP profiles of attenuated total (lower) and perpendicular (upper) backscattering
coefficients (km−1 sr−1 ) at 532 nm wavelength corresponding to the scene in the MODIS image. The red
vertical lines show contrail locations detected by MODIS image analysis. Only cloudy cross-sectional
elements are plotted (after Iwabuchi et al. 2012).
16 Ice in the Earth’s atmosphere

(a) Vertical Feature Mask UTC: 2007-03-31 04:56:26.4 to 2007-03-31 05:09:55.0 Version: 3.01 Nominal
30
7
25 6
Altitude (km)

20 5
4
15 3
3 (L)
10
2
5 2 (L)
1
0

1 = clear air 2 = cloud 3 = aerosol 4 = stratospheric layer 5 = surface 6 = subsurface 7 = totally attenuated 8 = low/no confidence

(b) RGB (c) τ3 (Ocean)

Latitude (oN)

Longitude (oE)
(d) τ3 (Land, DB) (e) re
Latitude (oN)

Latitude (oN)

Longitude (oE) Longitude (oE)

Figure 1.9 (a) Result of the vertical feature mask obtained from CALIOP polarized echoes. Between
the two vertical dashed lines, which denote the southern and northern bounds of the selected domain,
clouds mixed with dust have been identified using the lidar depolarization technique. (b) Images for
the MODIS (RGB) overpass on March 31, 2007 at 0500 UTC over northeastern and eastern China
and the East China Sea. (c) Aerosol optical depths over ocean. (d) Aerosol optical depths over land.
(e) Cloud effective particle sizes. The rectangular domain contains dust over both land and water with
imbedded clouds (after Ou et al. 2012).

BTD11-12 values are around 1 K, indicating mostly water clouds. The distributions of cirrus
and water clouds are consistent with cloud-phase mask results from MODIS. In addition to
clouds, Figures 1.9c,d display aerosol optical depth maps over ocean and land, respectively,
where the results over land were obtained using the Deep Blue algorithm developed by
Hsu et al. (2004). A domain was selected within which dust aerosols, along with ice and
water clouds, are present over both ocean and land in the regions of northeastern China
and the East China Sea. Dust aerosols over ocean and land were located in the southern
1.3 Formation of ice crystals 17

and northern halves of the selected domain. Cloud particle radii are retrieved from MODIS
visible and near-IR channels and are displayed in Figure 1.9e.

1.3 Formation of ice crystals


Having presented a macroscopic view of ice clouds in the atmosphere, we shall now broach
the subject of ice cloud formation from the perspective of microphysics at the fundamental
level. The formation and growth of ice crystals are subjects that have developed and evolved
in the last few decades. A number of textbooks (Fletcher 1962, Byers 1965, Mason 1971,
Hobbs 1974, Rogers and Yau 1989, Pruppacher and Klett 1996, Wallace and Hobbs 2006,
Wang 2013) present detailed information on these subjects. However, we shall focus our
discussions on several topics to the extent that they are specifically relevant to our discussion
of light scattering and absorption by ice crystals and radiative transfer in ice clouds.

1.3.1 The structure of ice crystals


Before beginning our discussion of the initial formation of ice crystals, a few words are in
order regarding the structure of ice. Ice Ih is the hexagonal crystal form of ordinary ice,
or frozen water. Essentially all ice in the Earth’s biosphere is ice Ih. The water molecule
has three thermodynamic phases – vapor, liquid, and ice. The water vapor molecule has an
asymmetric top bending (triatomic) configuration with the oxygen atom in the middle and
one hydrogen atom on each side, with a bonding angle of 104.45°. The distance between
the oxygen and hydrogen atoms is 0.958 Å. The oxygen and hydrogen atoms have a number
of isotopic forms. Ordinary water is generally considered to be 1 H2 16 O or simply H2 O.
The water vapor molecule has three rotational constants coupled with three fundamental
vibrational modes, which produces countless pure rotational and vibrational–rotational
absorption lines covering the entire infrared and visible spectrum, critical to the discussion
of radiative transfer in planetary atmospheres.
The basic structure of ice Ih was originally proposed by Pauling (1935) and has since
been well established. The geometric arrangements of hydrogen and oxygen atoms in a
crystalline solid were later determined from analysis of the X-ray diffraction pattern. The
oxygen atoms, indicated by open circles in Figure 1.10a, are arranged in a hexagonal lattice
with each having four nearest neighbors at the corners of a rectangular tetrahedron. The
hydrogen atoms, shown as dark dots, are covalently bonded to the nearest oxygen atoms
to form H2 O molecules; at the same time, these molecules are connected to one another
by hydrogen bonds. Each molecule offers its hydrogen atoms to two other molecules and
accepts hydrogen bonds from another two. There is no long-range order in the orientation
of H2 O molecules or hydrogen bonds. In Figure 1.10a, the notation [0001] (one of the
Miller indices) is aligned with the c-axis and the picture is from a view perpendicular to
the c-axis. The unit cell of the average structure is marked by dashed lines connected by
upper-case letters in which the basic four oxygen atoms are shaded. The basic structure
18 Ice in the Earth’s atmosphere

Figure 1.10 (a) The crystal structure of ice Ih, with oxygen atoms denoted by white circles and
hydrogen atoms by black dots, viewed perpendicular to the c-axis. The unit cell of the average
structure is marked ABCDEFGH, while the four oxygen atoms contained within this cell are shaded.
(b) A two-dimensional view of the ice structure on the (101̄0) plane (after Petrenko and Whitworth
1999). The definition of the Miller indices is illustrated in Figure 1.13a.
1.3 Formation of ice crystals 19

then consists of a hexametric box where planes contain chair-form hexameters, similar to
sheets lying on top of one another.
The three-dimensional structure of ice Ih is complicated. Thus, a two-dimensional
layer perspective is shown in Figure 1.10b for easier visualization. There are two possible
hydrogen sites on each bond and four of these sites adjacent to each oxygen atom, such
that there are two hydrogen atoms adjacent to each oxygen atom and only one hydrogen
atom per bond. In this diagram, the Miller indices [1̄21̄0] and [101̄0] denote the directions
along one of the intersections and perpendicular to one of the prism faces, respectively (see
Figure 1.13a). By convention, a surface can be defined by four integers, which are placed in
order to complete the orientation definition; a bar over a number indicates an intersection
in a negative axial direction (e.g., see −a3 in Figure 1.13a). Crystal planes of the type
(112̄0), (101̄1), and (101̄2) would lead to rarely occurring dodecagonal, bullet pyramidal,
and pyramidal shapes, respectively (see the two lower diagrams in Figure 1.13a). For a
column or plate, the top and bottom are referred to as basal faces, which are perpendicular
to the c-axis, while the surrounding six sides are referred to as prism faces. The hexagonal
shape exhibited in the majority of ice crystals is evidently related to the basic hexagonal
symmetry within the crystal.
The crystal structure of Ih is stable down to −200°C and within atmospheric pressure.
It has a density of 0.917 g cm−3 , lower than liquid water, due to the low density of its
crystal lattice. Based on the structure of ice defined above, Subsection 1.3.2 presents the
homogeneous and heterogeneous nucleation of ice crystals.

1.3.2 Homogeneous and heterogeneous nucleation


Once a cloud extends to altitudes where the temperature is colder than 0°C, ice crystals
may form. There are two types of nucleation that can lead to the formation of ice, namely,
homogeneous and heterogeneous nucleation. Homogeneous phase transitions can lead to
the formation of ice. Homogeneous freezing of a pure liquid drop occurs when statisti-
cal fluctuations of water molecular arrangement produce a stable, ice-like structure that
can serve as an ice nucleus. Two possibilities determine the conditions for homogeneous
nucleation of freezing: the size of the stable nucleus and the probability of occurrence
of embryonic ice nuclei by random rearrangement of water molecules. These quantities
depend on the surface free energy of a crystal–liquid interface. Based on theory and exper-
iment, droplets smaller than 5 µm will freeze spontaneously at a temperature of about
−40°C. Larger droplets are predicted to freeze at slightly warmer temperatures. Some
liquid droplets exist in clouds as cold as −20°C.
Homogeneous deposition or sublimation occurs when vapor molecules form a stable ice
embryo via chance collisions. The surface free energy of a crystal–vapor interface is poorly
known; however, theory predicts that homogeneous nucleation of deposition should only
occur at extreme conditions of supersaturation. More than 20-fold supersaturation with
respect to ice is required at a temperature a few degrees below 0°C, and still higher super-
saturation at colder temperatures. Experimental confirmation of the theory of homogeneous
20 Ice in the Earth’s atmosphere

deposition appears to be impossible because liquid droplets will nucleate homogeneously


before the supersaturation reaches the high values required for ice. These droplets will
freeze spontaneously at temperatures colder than −40°C, making it impossible to recog-
nize ice crystals that might be formed by deposition. Homogeneous deposition cannot occur
in the atmosphere because the necessary extreme supersaturations never exist.
Ice crystals usually appear in a cloud in appreciable numbers when the temperature
decreases to below about −15°C, a range signifying heterogeneous nucleation. Water
vapor in contact with most materials freezes at these temperatures, while deposition occurs
on most surfaces at supersaturation and supercooling less than homogeneous nucleation
values. Thus, nucleation of ice in supercooled water or a supersaturated environment can
be enhanced by the presence of foreign surfaces or suspended particles.
The foreign material provides a surface onto which water molecules can impinge, stick,
bond together, and form aggregates with an ice-like structure. The larger the aggregate,
the more likely it is to be stable and to continue to exist. The probability of heterogeneous
nucleation by freezing or by deposition depends strongly on the properties of the substrate
material as well as on supercooling and supersaturation. The more tightly bound the water
molecules are to the substrate, the greater the probability of ice nucleation. In addition, if
the crystal structure of the substrate closely resembles that of an ice crystal plane, it will
increase the chances of ice nucleation. When the binding and the matching of the crystal
lattice are good, the supersaturation or supercooling required to nucleate ice on a substrate
may be much lower than that required for homogeneous ice nucleation.
Supercooled clouds in the atmosphere develop and exist in the presence of large numbers
of aerosol particles, a very small fraction of which can serve as ice nuclei at temperatures
considerably warmer than −40°C, an approximate threshold for homogeneous freezing.
There are a number of possible heterogeneous nucleation mechanisms (Rogers and Yau
1989). Ice may form directly from the vapor phase on suitable deposition nuclei, as shown
in Figure 1.11a. The lower panels show three modes of activation for freezing nuclei: some
serve first as centers for condensation, then as freezing nuclei (Figure 1.11b); some promote
freezing the instant they come into contact with a supercooled droplet (Figure 1.11c); others
cause freezing after becoming imbedded in a droplet (Figure 1.11d). A given particle (ice
nuclei, in general terms) may nucleate ice in different ways, depending on temperature and
saturation conditions, as well as being controlled by unknown probability in the course of
its history in the cloud.

1.3.3 Atmospheric ice nuclei


Observations have pointed to clay minerals, especially kaolinite, a common material found
in many soil types, as a major component of atmospheric ice nuclei. The nucleation threshold
is at a relatively warm temperature of −9°C. Snowflakes collected as they fall to the ground
are usually found to contain particles that appear to be centers for crystal growth. These
particles have been identified by electron microscopy to be kaolinite, with sizes ranging
from 0.1 µm to 4 µm. Another source of ice nuclei has been revealed by the discovery that
1.3 Formation of ice crystals 21

(a) Heterogeneous
Deposition

(b) Condensation
Followed by Freezing

(c) Contact Freezing

(d) Immersion Freezing

Figure 1.11 A schematic depiction of the processes for ice nucleation mechanisms involving atmo-
spheric ice nuclei and ice crystal formation: (a) direct heterogeneous deposition, (b) condensation in
the drop, followed by freezing, (c) contact with the drop, followed by freezing, and (d) immersion
into the drop, followed by freezing.

the bacteria in decaying plant leaf material can be effective nuclei at warm temperatures.
Research has shown that common soil particles can be active at temperatures warmer than
the threshold for kaolinite, which may explain the presence of submicron-sized nuclei of
some minor organic substance. The bacterium itself has also been found to serve as an ice
nucleus at a temperature as warm as −1.3°C, despite the fact that its nucleating ability is a
rare and changeable property. The overall significance of biogenic nuclei remains a subject
of research. Ice nuclei active at −4°C have also been observed in sea water rich in plankton.
In addition, meteoric material has been found to be an important source of atmospheric ice
nuclei. Finally, we note that the most effective material to have been used as an artificial
nucleating agent is silver iodide, discovered by Vonnegut (1947).
Measurements at coastal sites indicate more nuclei in air from sources over land than
over the ocean. The concentrations of nuclei over land also tend to decrease with altitude,
consistent with a source at the surface. Even at the South Pole, the particulates in snowflakes
are found to be clay minerals. It is clear that more than one kind of material can serve as
atmospheric ice nuclei, depending on temperature, humidity, and proximity to sources;
one may be more important than the others in a given cloud. Although the atmosphere
has an abundance of condensation nuclei, ice nuclei are scarce, regardless of their origin.
For this reason, the supercooling of cloud water to −15°C or colder appears to be rather
common.
Moreover, worldwide measurements of ice nucleus concentrations as a function of tem-
perature indicate that they tend to be higher in the Northern than the Southern Hemisphere.
On average, the number concentration N of ice nuclei per liter of air active at temperature
T tends to follow the empirical relationship ln N = a(T ∗ − T), where T ∗ is the threshold
temperature at which ice nuclei are active, typically at about −20°C (Fletcher 1962), and
22 Ice in the Earth’s atmosphere

a varies from 0.3 to 0.8 based on numerous studies. In urban air, the total concentration of
aerosol is on the order of 108 per liter, and only about one of them has been observed to act
as an effective nucleus at −20°C.

1.3.4 Secondary ice particle production


The occurrence of ice crystals in clouds is related to cloud type, temperature, and cloud
evolution. Observations confirm that the colder the cloud temperature, the greater the
likelihood that some ice crystals are present along with the supercooled water droplets.
Moreover, the probability of ice being present is 100% when the top temperature of the
cloud is below about −13°C. The first crystals to appear in a cloud must form on ice nuclei.
An exception to this generalization may be cirrus clouds, which form at low temperatures
(between about −20 and −60°C) in such a manner that homogeneous freezing may occur
as soon as the liquid phase appears. Additional crystals may then be produced by secondary
processes in which the primary crystals are “multiplied.”
Two mechanisms have been recognized as contributing to secondary ice particle pro-
duction: when ice crystals fracture and when freezing drops shatter or splinter. With respect
to the former, it is commonly observed that many crystals fragment in clouds. These are
produced when graupel particles overtake and collide with fragile, slower-falling dendritic
crystals. However, the process of crystal fracture appears to be difficult to quantify and
hence it is difficult to determine the rate of production of secondary particles as well as
the conditions favoring fragmentation. With respect to the latter, the mechanism of ice
multiplication is considered to be very effective under suitable conditions for production of
secondary particles when supercooled drops grow to an appropriate size and are captured
by graupel particles. At temperatures warmer than −3°C, drops tend to spread over the
ice surface instead of freezing as discrete drops. At temperatures colder than −8°C, the
freezing appears to proceed rapidly, starting with an outer shell of ice. Between these tem-
peratures, observations have revealed that small particles are ejected by drops larger than
25 µm in diameter. This mechanism of secondary particle production, sometimes called
rime-splintering, may account for high concentrations of ice crystals that are sometimes
observed in maritime cumulus clouds with temperatures warmer than −10°C.

1.4 Growth of ice crystals


1.4.1 Growth by diffusion
After ice crystals are formed in a cloud, they can undergo growth by diffusion processes,
particularly if supercooled drops coexist. On the basis of the Clausius–Clapeyron equation
derived from the conservation of the Gibbs function during phase transition, we have

des Les
= , (1.4.1a)
dT Rv T 2
1.4 Growth of ice crystals 23

(a)
12 a Evaporation (> 0oC)
b Sublimation Water
10 c Melting c
a* Evaporation (< 0oC) a
Pressure (hPa)

6 l $
Triple Point
4
Ice Water Vapor
2 a* b

0
-40 -30 -20 -10 0 10
(b)
30

25
100 x Δe sat (hPa)

20

15

10

5 Max
$
0
0 -10 -20 -30 -40 -50
Temperature (oC)
Figure 1.12 (a) Saturation vapor pressure as a function of temperature for three thermodynamic
phases of water molecules (vapor, water, ice) indicated by the regions in a two-dimensional diagram
based on the Clausius–Clapeyron equation. The triple point is denoted by a heavy dot and the solid
and dashed lines below 0°C represent ice and water, respectively. (b) Differences of saturation vapor
pressure between water and ice as a function of temperature, with the maximum at about −12°C.

where es denotes the saturation vapor pressure at a temperature T, L is the latent heat, and
Rv is the gas constant for water vapor. Integration leads to
es (T )
= exp[L(1/T0 − 1/T )/Rv ], (1.4.1b)
es0
where es0 (= 6.11 hPa) is the value at the triple point with a temperature of T0 (= 273.15 K)
at which vapor, liquid water, and ice coexist (Figure 1.12a). Because the latent heat for the
transition between vapor and ice is larger than that between vapor and water, the saturation
vapor pressure over ice is smaller than over its supercooled water counterpart. Consequently,
24 Ice in the Earth’s atmosphere

ice crystals will grow at the expense of supercooled droplets with a maximum growth rate
at a temperature of −12°C (Figure 1.12b). The growth of ice particles by this process is
referred to as the Wegener–Bergeron–Findeisen process (Wegener 1911, Bergeron 1935,
Findeisen 1938).
To quantify the mass increase in an ice particle, we must consider the diffusion process.
The vapor field in a cloud and ambient air can be described by the vapor density or absolute
humidity. At any point in the vapor field, the concentration of molecules, n, can be defined by
the conventional time-dependent second-order diffusion equation nt = D2 n, where
D is the molecular diffusion coefficient. Subject to the steady-state condition and imposing
boundary conditions at the positions of the ice particle and ambient air, the growth equation
for an ice crystal can be expressed in terms of vapor density in the form
dm
= 4πCD[ρv (∞) − ρv (s)], (1.4.2a)
dt
where ρ v () and ρ v (s) denote the vapor density of the ambient air and at the ice
crystal surface, respectively. To account for the non-spherical nature of ice crystals,
a parameter C has been introduced to denote the equivalent electrical capacitance (in
length units), which is a function of the size and shape of the particle. For a sphere,
C = r. For a circular disk of radius r, which can be used as an approximation for plate-type
ice crystals, C = 2rπ. Ice needles may be approximated by the formula for a prolate
spheroid of major and minor semi-axes a and b such that C = Aln [(a + A)b], where
A = (a2 − b2 )½ . For an oblate spheroid, we have C = aεarcsin ε, where the ellipticity
ε = (1 − b2 a2 )½ (Rogers and Yau 1989, Pruppacher and Klett 1996). Actual ice crystals
have more complex shapes such as commonly occurring plane dendrites and plates, which
can be reasonably approximated by a circular disk of equal area, while needles can be
approximated by long prolate spheroids.
In association with diffusion is the release of latent heat, which tends to raise the ice
crystal surface temperature above the ambient value. Analogous to mass diffusion, the
diffusion of heat away from the ice crystal is given by
dQ
= 4πCK(Ts − T ), (1.4.2b)
dt
where Ts is the temperature at the crystal surface, T is the ambient temperature, and K is the
thermal conductivity coefficient. Under stationary growth conditions, we have LI dmdt =
dQdt, where LI is the latent heat associated with the transformation between ice and water
vapor. Thus, the value of ρv (s) is determined according to this balance between the rates of
latent heating and heat transfer away from the surface, as follows:
[ρv (∞) − ρv (s)] K
= , (1.4.3)
(Ts − T ) L1 D
Using the Clausius–Clapeyron equation and the definition of saturation ratio with respect
to ice, Si = eei , where e is the vapor pressure of the ambient air and ei is the saturation
1.4 Growth of ice crystals 25

pressure over ice, an analytical expression for crystal growth can be derived and is given in
the following form:
dm 4πC(Si − 1)
= , (1.4.4)
dt (fK + fD )
where the terms in the denominator are defined by

fK = L2I /(KRv T 2 ), fD = Rv T /[Dei (T )], (1.4.5)

with ei (T) being the vapor pressure over ice at a temperature T. The growth rate varies
inversely with pressure and is a function of the temperature, with the maximum growth at
about −15°C over a wide range of pressure.
The preceding theoretical development for ice crystal growth based on diffusion of
water vapor cannot explain the different growth habits of ice or the transitions from one
habit to another. The more fundamental molecular-kinetic approach is required to explain
the formation of these habits or shapes. The surface of a growing crystal, as described in
Subsection 1.3.1, is made up of flat terraces of different heights, terminating at ledges and
separated by steps. Molecules of water that impinge onto the crystal surface are bound
more strongly at the ledges than on terraces. When unattached molecules move over the
crystal surface, they orient preferentially at ledges, leading to growth by the lateral motion
of surface steps. The growth rate is determined by the rates at which steps are generated
and advance, which in turn are controlled by temperature and supersaturation. The main
mechanism for the motion of steps is the diffusion movement of molecules across terrace
sites up to and into ledge sites. The kinetic effects determine the crystal habit, and tend to
slow the crystal growth rate, based entirely on the continuous vapor diffusion theory.
Temperature and saturation conditions determine not only growth rate but also ice crystal
habits, which are basically hexagonal structures with a range of different axis ratios. Ice
crystals have one common basic shape, a six-fold symmetric (hexagonal) prism with two
basal planes of type (0001) and six prisms of type (101̄0) in terms of Miller indices, as
shown in the two upper diagrams of Figure 1.13a. The orientation of the crystal lattice is
defined by the intersections of the planes with the hexagonal axes a1 , a2 , a3 , and c. Faces
such as (112̄0), (101̄1), and (101̄2) grow quickly to become the edges and corners of a
crystal, while faces of the type (0001) and (101̄0) grow slowly and become the bounding
faces of the crystal. Laboratory experiments have revealed that the rate of propagation of
the basal faces growing along the c-axis relative to that of the prism faces (growth along
the crystallographic direction of type [101̄0]) varies with temperature and supersaturation.
Primary ice crystal habits include column, plate, and dendrite, as shown in Figure 1.13b.
When an ice crystal grows and moves around in the cloud, its habit will be modified as a
consequence of changing temperature and saturation conditions. Sector stars can be formed
when plates develop peripheral dendritic structure, while capped columns are produced
when columns develop plates on their ends. The intricate stellar shapes have also been
frequently observed as a consequence of variations on the dendritic form (see Figures 1.14
and 1.15).
26 Ice in the Earth’s atmosphere

(a)
c

[0001]
a3

(0001) a2
(0001)
- -
[1210]
- - -1 0 ) -
(1010) (1010) (01 [ 0 11 0 ]
-a3
a1 -
-
[1010] [ 11 2 0 ]
--
[ 2 11 0 ]
(0001)
(1012)
-

- -
( 1 0 11 )
( 1 0 11 )

-
(1010)
- -2 0 )
(1010) ( 11

(b)

[0001]

--
[ 2 11 0 ]
--
[ 2 11 0 ]

Figure 1.13 Schematic representation of different habits of ice crystals. (a) The orientation of crystal
lattice planes in terms of the intersections of the planes with the hexagonal axes a1 , a2 , a3 , and c
is illustrated in the upper left picture. Shown at upper right is the representation of coordinates and
planes by Miller indices. The two basal planes are denoted by (0001), while (101̄0) denotes one of
the six prism planes depicted in the diagram. The numbers in square brackets denote directions; a bar
over a number stands for an intersection in a negative axial direction. Shown in the lower diagram
are representations of bullet and pyramid (after Wolff 1957). (b) Three primary ice crystal habits are
column, plate, and dendrite.
Figure 1.14 Temperature and humidity conditions for the growth of natural snow crystals. The sketches indicate the types of snow crystals (after Magono
and Lee 1966).
Figure 1.15 Meteorological classification of all types of ice and snow crystals as suggested by Magono and Lee (1966). Some of these
crystals are extremely complex.
1.4 Growth of ice crystals 29

1.4.2 Growth by accretion


After diffusion growth, ice crystals can further grow via collision and coalescence, referred
to as accretion. Ice crystal growth by collision and capture is controlled by its fall speed, u,
which has been generally determined from laboratory experiments in terms of maximum
dimension as follows (Langleben 1954):
u = kLn , (1.4.6)
where L is the maximum diameter in cm and u is in cm s−1 . For dendrites, k  160 and
n  0.3, while for columns and plates, k  234 and n  0.3. The mass (in grams) and size
of different ice crystals are usually related by empirical formulas of the form
m = aLb , (1.4.7)
where a and b are fitting coefficients for a number of ice crystal shapes that have been
determined from laboratory experiments (Mason 1971). For example, a = 1.9 × 10−2 and
b = 3 for thin hexagonal plates, and a = 9.4 × 10−4 and b = 2 for stellar crystals.
In the process of growth by accretion, there are issues concerning collection efficiency.
The first is related to the aerodynamic problem of collision efficiency, while the second is
in regards to the question of whether sticking occurs, given a collision. Collection has been
shown to be complex, with the crystal perimeter a preferred area for collisions. Because
freezing is likely to occur on contact with supercooled drops, the coalescence efficiency
is generally expected to be unity; that is, after a collision, coalescence will follow. In the
process of crystal aggregation, however, the collection efficiency is less well understood.
Open structures like dendrites are more likely to stick, given a collision, than crystals of
other shapes, and sticking has been found to be more likely at relatively warm temperatures.
Despite a number of uncertainty factors, the ice particle accretion growth may be derived
and is approximately governed by the following general form:
dm
= Ē CWC πR 2 u, (1.4.8)
dt
where CWC stands for the cloud water content (in g cm−3 ; can be ice or supercooled water),
R is an equivalent radius of the collector crystal, Ē denotes the mean collection efficiency,
and u is the difference in fall speed between the collector ice particle and small ice
crystals or supercooled water droplets.
In a mixed cloud, ice particles can increase in mass by capturing supercooled droplets,
which subsequently freeze onto them upon contact and form a coating of rime. This process,
referred to as growth by riming, leads to the formation of various rimed structures, including
rimed needle, rimed column, rimed plate, and rimed stellar. When riming proceeds beyond
a certain stage, the original shape is destroyed and the rimed particle is called graupel
(ice water). If the freezing is not immediate, denser layer structures can be formed –
centimeter-sized hail particles are extreme examples.
Aggregation is the clumping together of ice crystals to form snowflakes. When two ice
particles collide, the probability of adhesion is determined by temperature and the types
30 Ice in the Earth’s atmosphere

of ice particles. Complex crystals, such as dendrites, tend to adhere to one another due to
entanglement upon collision, whereas two solid plates will tend to rebound. The probability
of two colliding crystals adhering increases with increasing temperature; it is particularly
likely above temperatures of about −5°C, at which ice surfaces are said to become sticky.
Some graupel and aggregated particles will be illustrated in Subsection 1.4.1.

1.5 Illustration of ice crystal morphology


1.5.1 Some historical perspectives
In accordance with Needham and Lu (1961), the hexagonal shape (six sides) of snowflakes
was first noted by Chinese writer Han Ying in a book entitled “Han Shih Wai Chuan”
(Moral Discourses Illustrating the Han Text of the Book of Odes), written around 135 BC,
the time of the Western Han dynasty. Han Ying stated that:

Flowers of plants and trees are generally five-pointed, but those of snow, which are called “Ying,”
are always six-pointed.

In many classical Chinese writings, especially poems, six is a symbolic number, correlated
with the element water and the direction North, while five is associated with the Earth and
the Center. Hsiao Tung, a sixth-century Crown Prince of the Liang Empire (the third of the
Southern Dynasties during the Southern and Northern Dynasties period), gave an excellent
example in a poem:

The ruddy clouds flout in the four quarters of the cerulean sky
And the white snow-flakes show forth their six-petalled flowers.

There was no crystallographic explanation of the hexagonal nature of the crystals; however,
Chinese writers and poets attempted to define six-sided structure based on the principle
of the Yin-Yang and Wu-Shing (five elements) theory. It was believed that two opposing
forces operated in the entire universe, as well as among interactions between all of its
elements. Yin and Yang are said to be associated with the female (or negative aspects of
nature) and the male (or positive aspects of nature), respectively. It was noted that Yin and
Yang are also related to even and odd numbers, respectively. In principle, all elements in
the universe can be categorized as either Yin or Yang. Water is considered to possess the
spirit of Yin, a female attribute, and is associated with an even number. It happened without
explanation that the number six was assigned to the element water in ancient Chinese times.
For example, the great twelfth-century Chinese medieval philosopher of the Sung Dynasty,
Chu Hsi, wrote,

Six generated from Earth is the perfect number of water, so as snow is water condensed into crystal
flowers, these are always six-pointed.

In a text, Master Chu also attempted to further explain the formation of snow:
1.5 Illustration of ice crystal morphology 31

The reason why “flowers” or crystals of snow are six-pointed is because they are only sleet split open
by violent winds (and sleet being half-frozen rain, i.e. water) they must be six-pointed. Just so, if you
throw a lump of mud on the ground, it splashes into radiating angular petal-like form. Now six is a
Yin number, and thai-yin-hsuan-ching-shih (selenite crystal) is also six-pointed, with sharp prismatic
angular edges. Everything is due to the numbers inherent in Nature.

From the standpoint of work on ice nucleation in the twentieth century, it is interesting to
note that Master Chu connected snowflakes to the translucent hexagonal crystals of gypsum
or calcium sulfate, types of minerals that can serve as effective ice nuclei. The idea of snow
being six-pointed and related to Yin appears to have been accepted in the continuous
writings of ancient Chinese naturalists as a fact of nature or as a symbolic explanation, but
no further study on its hexagonal structure has been made. Although not within the context
of ice nucleation, Ho and Needham (1959) have discussed in some detail the observed halos
and parhelia (see Subsection 3.1.4) reported in ancient Chinese literature.
It appears that the subject of snow crystal shape was not formally documented by the
early Arabs or Classical Greeks. The first European observations are found in the writing
in 1260 of Albertus Magnus, who thought that snowflakes were star-shaped and fell only in
February and March. Almost 300 years later, Olaus Magnus, a Scandinavian bishop, wrote
in 1555 about snowflakes but appeared to overlook their essential hexagonal symmetry.
It was not until 1611 that the great German astronomer Johannes Kepler, in his book A
New Year’s Gift of Hexagonal Snow (Kepler 1611), described the hexagonal symmetry
of snowflakes and attempted to explain this structure on an atomistic basis, leading to a
modern discussion of cubical and hexagonal shapes in relation to mathematical theories of
close-packing of spheres.
After Kepler, Descartes (1637) presented sketches of snowflake crystals based on obser-
vations he had made. We also note that a light ray, which undergoes minimum deviation
from a spherical particle to explain, in part, the generation of rainbows, is referred to as a
Descartes ray. Further to this work, Erasmus Bartholinus published sketches of snowflakes
showing the branching of hexagonal stars in 1660.
With the invention of the microscope in the latter half of the seventeenth century,
advances in the study of snowflakes in Europe were rapid. Robert Hooke in 1665 published
a microscopic view of snowflakes, followed by Martens (1675), who went to the Arctic
as a ship’s barber on a whaler, and documented the correlations between the shapes of
snow crystals and meteorological conditions, and Donato Rossetti, a Canon of Livorno
in Italy, who in 1681 published a classification of snowflakes and a detailed drawing of
a hexagonal plate crystal. In 1761, Johann Wilcke of Sweden appears to have been the
first experimentalist to artificially produce snow crystals, using iodoform and camphor as
nucleation agents.
The first systematic classification of snowflake morphology was made by Scoresby
(1820), and described the shapes of the columnar and complex forms of ice crystals.
Scoresby also noted a relationship between snowflake shape and air temperature, an impor-
tant connection first suggested by Guettard (1762) and more definitively established by
32 Ice in the Earth’s atmosphere

the work of Fritsch (1853). In Japan, Toshitsura Doi (1832, 1839), a feudal lord of Koga,
presented numerous excellent sketches of snow crystals. The last great collection of snow
crystal pictures before the advent of modern science is attributed to Glaisher (1855). The
microphotography technique was established and used extensively in Europe to study snow
crystals and classify them into planar, columnar, and combinations of the two by G. Hell-
mann (1893). Bentley and Humphreys (1931) presented 6000 photomicrographs of snow
crystals taken by Bentley over a period of nearly 50 years in Vermont, U.S.A.

1.5.2 Classification of ice crystals


Several twentieth-century studies have presented the dependence of the crystal shape on
temperature. However, the most comprehensive investigations were undertaken by Ukichiro
Nakaya and his associates in Japan in the late 1930s, using controlled conditions to grow
ice crystals in the laboratory. The well-known Nakaya diagram illustrates that the habits of
ice crystals can be largely grouped in terms of temperature and the saturation condition of
water vapor in the cloud chamber. Since then, the field of ice microphysics has been well
developed at temperatures between about −5 and −35°C with low ice saturation values. In
the range of −10 to −20°C and above ice saturation, hollow column ice is seen to form,
and when water vapor supply increases, pristine plates, thick plates of skeleton forms, and
crystals with broad and sector-like branches are produced below water saturation, where
we also see hollow column, solid thick plate, solid bullet, and long solid needle. Above
water saturation, we see the formation of solid long needle, hollow column, sheath, and
scroll between −5 and −10°C. Between −10 and −20°C, we see a variety of plate types,
including thick plate of skeleton form, crystals with sector-like branches, fern-like, ordinary
dentritic, and stellar crystals, as well as hexagonal plate. For temperatures between −20
and −40°C, hollow column, hollow bullet, bullet rosette (combination of bullets), and
aggregate (combination of columns) are seen. When a sufficient number of small water
droplets are available, rimed needle, column, plate, and stellar crystal as well as graupel
can be observed.
All of the cloud physics texts cited previously (e.g., Pruppacher and Klett 1996) present
ice crystal morphology as a function of temperature and ice and water saturations with
a degree of similarity, as shown in Figure 1.14 (Magono and Lee 1966). Moreover, Ono
(1970) conducted an extensive investigation of ice crystal habits in atmospheric ice clouds,
which had temperatures between −2 and −32°C, based on the replicator results of more
than 10 000 ice crystals sampled from aircraft. These crystal habits were found to be in
agreement with those gathered in the laboratory (Nakaya 1954, Hallett and Mason 1958,
Kobayashi 1961) and similar to those displayed in Figure 1.14. Finally, the identification
of ice crystal morphology (or shape or habit) described above can be seen in Figure 1.15,
where a diverse variety of complex ice crystal shapes and snowflakes which occur in the
atmosphere and near the surface are displayed (Magono and Lee 1966), with nomenclature
defined in the figure.
More recently, Walden et al. (2003), using a conventional replicator technique, collected
ice crystals over the Antarctic in winter, many of which were identified as hollow and
1.5 Illustration of ice crystal morphology 33

solid columns (diamond dust), plates, pyramids, bullet clusters, and spheroids (blowing
snow). Bailey and Hallett (2004, 2009) presented a comprehensive morphology (or habit
or shape) diagram for atmospheric ice particles. The authors derived these morphology
results through both laboratory studies and a number of field observations, utilizing a new
ice particle probe called the Cloud Particle Imager (CPI), and stated that most ice crystals
are defective and irregular in shape to varying degrees and are mostly polycrystalline
at temperatures below −20°C. They proposed a new habit diagram which retained the
traditional descriptions of habits presented in Figure 1.14 for temperatures above −18°C.
The dominant habits range from plates (0 to −4°C) to columns (−4 to −8°C) to plates (−8
to −22°C). But for lower temperatures, their results revealed that the habit is dominated
by polycrystals of various forms, with two distinct habit regions, involving platelike from
−20 to −40°C and columnar from −40 to −70°C. The new diagram also emphasizes
that most individual crystals are complex, irregular, and imperfect in appearance to some
extent, including single crystals such as plates and columns that are most common at low
ice supersaturation. Also, it was pointed out that the majority of very small ice crystals
growing at low ice supersaturation are compact faceted polycrystals, not spheroids, as had
been suggested in previous publications.
For the purpose of illustration, we provide two examples of the observed crystal habit.
Lawson et al. (2006) collected numerous ice particles at the South Pole Station in summer
and sorted the CPI images of ice crystals into nine habit categories, including small plates
and faceted polycrystals (resembling spheroids), plates, rosettes, budding rosettes, long
columns, thick plates/short columns, small irregulars, blowing snow, and complex crystals
with side planes (Figure 1.16; see also Walden et al. 2003). Figure 1.17, taken from Bailey
and Hallett (2004), illustrates the common types of polycrystals of varying sizes, observed
between −20 and −40°C, collected from numerous laboratory and in situ observations
published by a number of investigators and data centers.
In Figure 1.18, we present a general depiction of ice crystal size and shape associated
with temperature in the atmosphere. Ice crystal size and shape captured by a replicator
balloon sounding system in Marshall, Colorado, on November 10, 1994 are shown as a
function of height and relative humidity. The relative humidity was measured by a cryogenic
hygrometer (dashed line) and Vaisala RS80 instruments (solid line and dots). Also shown
is temperature as a function of height. It is apparent that at the cloud top, pristine and
small columns and plates are predominant. At the cloud bottom, however, irregular bullet
rosettes and aggregates are produced, presumably due to accretion processes associated
with vertical mixing and gravitational pulling.

1.5.3 Computer generation of three-dimensional ice crystals


In order to carry out light scattering and absorption calculations and analysis, ice crys-
tal shape, size, and orientation must be defined mathematically in accordance with their
geometry. Takano and Liou (1989a, 1995) and Yang and Liou (1998a) have developed
computational techniques to generate ice crystal shapes used in light-scattering calcula-
tions involving various types of ice crystals. In the following discussion, we present a
34 Ice in the Earth’s atmosphere

Figure 1.16 Examples of CPI images of ice crystals sorted into habit categories during Antarctic
summer (after Lawson et al. 2006).
1.5 Illustration of ice crystal morphology 35

Figure 1.17 Laboratory and in situ examples of commonly observed polycrystals between −20 and
−40°C obtained from a number of experiments and field aircraft observations (after Bailey and Hallett
2004).

number of computer-generated 3-D ice crystal shapes introduced in previous sections,


including a solid column, a solid plate, a droxtal, two spheroids, a hollow column, a bullet,
a bullet rosette, an aggregate ice crystal, a dendrite, and a Koch snowflake. We divide the
ice crystal shapes into four groups, as shown in Figure 1.19.
36 Ice in the Earth’s atmosphere

14

13 -70
12
-60
11

Temperature (oC)
-50
Height (km)

10

9 -40

8 -30
7
-20
6
-10
5

4 0
0 10 20 30 40 50 60 70 80 90 100
Relative Humidity (%)

Figure 1.18 Ice crystal size and shape as a function of height and relative humidity as captured
by a replicator balloon sounding system in Marshall, Colorado on November 10, 1994. The relative
humidity was measured by a cryogenic hygrometer (dashed line) and Vaisala RS80 instruments (solid
line and dots). Also shown is temperature as a function of height (Liou 2002, courtesy of Andrew
Heymsfield).

Figure 1.19a shows three quasi-spherical ice particles: droxtal, prolate spheroid, and
oblate spheroid. For a droxtal, only the radius of a circumscribed sphere R and the two
angles θ 1 and θ 2 are required to define its geometric properties. For example, crystal lengths
a1 , a2 , L1 , L2 , and h are defined by

a1 = R sin θ1 , a2 = R sin θ2 , ⎪

L1 = R cos θ1 , L2 = R cos θ2 ,⎪ (1.5.1)

h = a1 (L1 − L2 )/(a1 − a2 ).

Other geometric properties of a droxtal, such as the position vectors of vertices and the
unit vectors normal to the individual faces, can be expressed in terms of R, θ 1 , and θ 2 .
For example, the unit vector n̂n = (nnx , nny , nnz ) associated with the six upper trapezoidal
faces of a droxtal is defined as follows:
 nπ  ⎫
nnx = sin α sin ,⎪⎪
3  ⎪
 nπ ⎬
nny = sin α cos ,⎪ (1.5.2)
3 ⎪


nnz = cos α,
1.5 Illustration of ice crystal morphology 37

Figure 1.19 Computer-generated ice crystal shapes: (a) quasi-spherical ice particles, (b) column
types, (c) plate types, and (d) bullet rosette types. Note that 2a = D, the ice crystal width. See text
for further discussion.

where n = 1, . . . , 6. The parameter α is an angle between a unit vector to an upper


trapezoidal face and the z-axis, as shown in Figure 1.20 (far left), and is given by

2(h + L1 − L2 )
α = sin−1 . (1.5.3)
[3a 2 + 4(h + L1 − L2 )]1/2
The geometry of a spheroid is defined by
x2 y2 z2
+ + =1 (1.5.4)
a2 a2 c2
38 Ice in the Earth’s atmosphere

z
unit vector z z z z
a of normal
g’ q’

x x x x
f’
O O O O O

y y y y
position vector
of vertex
(a2, 0, -L2) (a) (b) (c) (d)

Figure 1.20 The figure at far left presents the definitions of a unit vector, a position vector, and the
angle α involving a droxtal. The four pictures on the right show (a) a column in Cartesian coordinates
Oxyz, which can undertake three rotations to form an aggregated ice crystal; (b) rotation by an angle
γ ʹ with reference to its c-axis; (c) tilt of the c-axis from the z-axis by an angle θʹ in the x-z plane; and
(d) rotation of the c-axis about the z-axis by an angle φʹ.

where a and c are the lengths of the semi-axes, as shown in Figure 1.19a. The semi-axis
length a is the equatorial radius of a spheroid and c is the distance from the center to the
pole along the symmetry axis. When a < c, the spheroid is referred to as prolate (center);
when a > c, it is called oblate (far right).
Figure 1.19b consists of a solid column, a hollow column, and a solid column aggregate.
Let the radius of a hexagon be a and its half-height be c (= L2, where L is the length).
The equations which define the six hexagonal surfaces (1–6) can be expressed as
 nπ   nπ  √
3
sin x + cos y− a = 0, n = 1, 2, 3, 4, 5, 6, (1.5.5)
3 3 2
while the top and bottom surfaces (7 and 8) are defined by

cos[(n − 7)π]z − c = 0, n = 7, 8. (1.5.6)

For a hollow column with a depth of d, the basal planes expressed in Eq. (1.5.5) are replaced
by two sets of six slant hollow planes in the form

 
 
n n+1 n n+1
− sin π − sin π (±d)x + cos π − cos π (±d)y
3 3 3 3
 n 
n 
n+1
+ cos π sin π − sin π
3 3 3 (1.5.7)
n 
n 
n+1
− sin π cos π − cos π
3 3 3
× a(z ∓ c ± d) = 0, n = 1, 2, 3, 4, 5, 6.
In this equation, the term (z − c + d) corresponds to the top hollow face, while the term
(z + c − d) corresponds to the bottom hollow face.
To form an aggregated ice particle (far right in Figure 1.19b), we consider a hexagonal
ice crystal element whose c-axis is parallel to the z-axis of a Cartesian coordinate system
Oxyz, as shown in Figure 1.20a. An ice crystal can be rotated to any orientation by three
1.5 Illustration of ice crystal morphology 39

n=1 n=2 n=3 n=4

Figure 1.21 Computer generation of a Koch snowflake based on the procedure described in the text.

consecutive coordinate transforms: (1) rotation by an angle γ ʹ with reference to its c-axis
as in Figure 1.20b, (2) tilting of the c-axis from the z-axis by an angle θ ʹ in the x-z plane
as in Figure 1.20c, and (3) rotation of the c-axis about the z-axis by an angle φʹ as in
Figure 1.20d. This procedure is repeated for a number of hexagonal elements attached in
an arbitrary manner to produce an aggregated crystal. The spatial position of a specific
element can be determined by the coordinate values of the particle’s center: x0 , y0 , and z0 .
In Figure 1.19c, the equation defining the geometry of a solid plate (left) is the same
as that for a solid column; see Eqs. (1.5.5) and (1.5.6). For dendrites (center), we have
attached six branches to a hexagonal plate. Let the branch height bt and branch width bb
be defined as shown in Figure 1.19c; then each branch plane (12) can be expressed in the
form

π   π 

n+m
(a + bt) sin n − a sin n + bb sin π
3 3 3
  π 
× x − (a + bt) cos n

3

π  π  n+m
− (a + bt) cos n − a cos n + bb cos π
3 3 3
 π 
× y − (a + bt) sin n = 0, n = 1, 2, 3, 4, 5, 6, m = 2 or 4. (1.5.8)
3
The shape of a snowflake (right) can be constructed using the following procedure. We start
with an equilateral triangle with sides of unit length as an initiator, then add to the middle
third of each side a -shaped peninsula with three side lengths of 13, resulting in a Star of
David (a six-pointed star). Subsequently, on each of the 12 sides of the resulting six smaller
triangles, we carry out the same additions to yield 48 sides, and continue the process n
times, which will lead to the formation of a Koch snowflake (von Koch 1904). Usually,
n = 4 is sufficient for demonstration purposes. A Koch snowflake, which has a fractal
dimension of log4log3 = 1.262, develops only on the side planes; the two basal planes are
flat. Thus, a 3-D Koch snowflake contains two flat basal planes and 3 × 4n highly irregular
side planes (with 6-sided symmetry) associated with n fractal iterations (see Figure 1.21
for n = 4).
40 Ice in the Earth’s atmosphere

2a

(a) (b)
L a
χ l

O t
O
d
O O χ

Side View Top View Side View Top View

(c) (d) l3
l1
L

l2

bb a1 a2
O a3
O
O O

bt
a

Side View Top View Side View Top View

Figure 1.22 Side and top views of (a) a 3-D hollow column (O denotes the center); (b) a bullet
rosette (χ is the angle between bullets); and (c) the dendrite defined in Figure 1.19. Also shown is
(d) a capped column mapped onto side and top planes (after Takano and Liou 1995).

In order to conduct light scattering exercises, we must define the required surface area
and volume.
√ Let the length of an equilateral triangle be a. Then, its 2-D area is given by
A0 = ( 34)a2 . The area generated by n iterations is then given by

An = An−1 + (3 × 4n−1 /32n )A0 . (1.5.9)

The volume of a Koch snowflake is simply Vn = An L, where L is the length defining √ the two
basal planes. Also, the radius of a Koch snowflake can be defined by a∗ = ( 33)a. For
comparison purposes, we may also define an equivalent sphere with a radius aeq such that
Vn = (4π3)aeq 3
. The Koch snowflake is a concave shape which differs from the conventional
convex ice particles such as plates and columns.
In Figure 1.19d, consider a solid bullet rosette with six branches (far left). For a single
bullet, one of the hollow depths d is replaced by −t, where t is the tip length of the bullet.
Consider a single bullet whose tip is at the origin and c-axis is along the z-axis. The bullet is
rotated by an angle in the zenith direction and by another angle in the azimuthal direction.
Then four or six rotated bullets can be combined to produce a bullet rosette. To construct a
hollow bullet rosette with six branches, we may replace the solid bullet by a hollow bullet
as shown in the far right, where the hollow depth is defined as H.
1.6 Size and shape distributions in cirrus clouds 41

To further define the shapes of a number of irregular ice crystals displayed in Figure 1.19,
we illustrate in Figure 1.22 the top and side views (a) of a hollow column, (b) a four-bullet
rosette, and (c) a dendrite. Also shown is (d) a horizontal view of a capped column, which
frequently occurs in tropical cumulonimbus. All the relevant geometric parameters and
angles are defined in this figure and preceding discussions.

1.6 Ice crystal size and shape distributions in cirrus clouds


1.6.1 Introductory remarks
A substantial number of ice crystals in the atmosphere have been observed in cirrus clouds,
due to their high location and low temperature. The formation, maintenance, and dissipation
of these clouds are principally governed by large-scale synoptic features and disturbances.
In the tropics, they are related to deep-cumulus outflows. The ice crystal shapes depend on
temperature and relative humidity, as well as whether they undergo collision and coales-
cence processes in the clouds.
Weickmann (1948) observed that at humidity levels approaching water saturation, ice
crystals in cirrus and cirrocumulus have prismatic skeleton shapes that occur in hollow and
cluster crystals (bullet rosettes). Between ice and water saturation, ice crystals in cirrus
grow in the form of prisms. In cirrostratus, where relative humidity is normally close to
ice saturation, ice crystals are primarily individual with full crystals such as columns,
prisms, and plates. In midlatitude cirrus, where substantial aircraft observations have been
conducted, ice crystal shapes are conventionally classified as a function of temperature
(Heymsfield and Platt 1984, Heymsfield and Iaquinta 2000). For most cirrus clouds, hollow
columns and hexagonal plates are the most abundant types near the cloud top. Spatial ice
crystals such as bullet rosettes are the predominant forms above about −40°C, while hollow
or solid columns prevail below about −50°C. Between these temperatures, convective cirrus
clouds contain a majority of spatial crystal forms, whereas stable cirrus clouds are primarily
composed of hollow columns. As discussed in Subsection 1.6.2, numerous field campaigns
have been designed to understand the composition and structure of cirrus clouds.

1.6.2 Ice crystal size and shape spectra


Typical ice crystal size distributions for cirrus clouds have been developed on the basis of
a number of intensive field observations from aircraft platforms. Following is a summary
of these programs and the ice crystal size and shape data that were collected.
At midlatitude, the First ISCCP Regional Experiment (FIRE) for cirrus study was
conducted in 1986 (I) over Madison, Wisconsin, and in 1991 (II) over Coffeyville, Kansas.
During FIRE I, ice crystal size distributions and habit images were gathered from 2D-C and
2D-P image probes, which measured ice crystal sizes from 50 to 100 µm and from 1000 to
3000 µm, respectively. During FIRE II, balloon-borne replicators were used to capture ice
crystal size ranging from about 10 to about 1000 µm with a resolution of 2 µm (Miloshevich
42 Ice in the Earth’s atmosphere

and Heymsfield 1997). The European cirrus experiment carried out in 1989 also provided
useful ice crystal size and shape data. A fourth midlatitude cirrus field experiment was
carried out in the spring of 2000 during an intensive observation period in the vicinity
of Lamont, Oklahoma under the auspices of the Department of Energy’s Atmospheric
Radiation Measurement (ARM) program. During this experiment, the University of North
Dakota Citation aircraft equipped with 2D-C and 2D-P probes was used to measure ice
crystal size and shape and other microphysical data. During the Subsonic Aircraft: Contrail
and Cloud Effect Special Study (SUCCESS) conducted in 1996, useful ice crystal and
microphysical datasets were obtained for cirrus, contrails, and contrail cirrus. Some of the
contrail ice crystals were analyzed by Liou et al. (1998).
Observations of ice crystal size and shape distributions for cirrus over the tropics were
made during the Central Equatorial Pacific Experiment (CEPEX) conducted in 1993. Air-
craft microphysical measurements were carried out by a 2-D optical probe (30–300 µm)
and video ice particle sampler (VIPS, < 30 µm). Based on available measurements in the
tropics, ice crystal sizes ranged from about 10 to 2000 µm and presented in four dominant
shapes: bullet rosettes, aggregates, hollow columns, and plates, similar to those occurring
in midlatitude. Smaller ice crystal sizes generally occur in cloud-top (colder temperature)
conditions, whereas larger ice crystal sizes are associated with warmer temperatures and/or
the developing stage of clouds associated with convection (Rolland et al. 2000). Tropical
ice crystal data were also available from the Kwajalein Experiment carried out in 1999
under the auspices of the Tropical Rainfall Measuring Mission (TRMM). The cloud tem-
peratures ranged from 0 to −70°C with optical depths between 20 and 30 associated with
deep cumulus convection. The most recent tropical cirrus experiment was conducted in
a field campaign called the Cirrus Regional Study of Tropical Anvils and Cirrus Layers
(CRYSTAL) Florida Area Cirrus Experiment (FACE) in the summer of 2002 using NASA
WB57 high-flying aircraft.
Ice crystal size data were collected via CPI and video VIPS, as well as forward scattering
spectrometer probe (FSSP) for ice crystals measuring from a few micrometers to about
30 µm in the −58 to −76°C temperature range. To characterize ice crystal size distribution
and account for habit factor, a mean effective ice crystal size has been defined as 1.5 times
the ratio of ensemble particle volume and the particle projected area, as follows (Foot 1988,
Francis et al. 1994, Wyser and Yang 1998, Yue et al. 2007):
 
3
De (l, k) = [
i fi,l (L)Vi (L)]nk (L)dL [
i fi,l (L)Ai (L)]nk (L)dL, (1.6.1)
2 k k

where the i and k indices stand for habit and size distributions, respectively; Vi (L) and
Ai (L) are the volume and projected area, respectively, of an ice crystal with a maximum
dimension L for the ith habit; fi,l (L) is the fraction of the ith habit in the lth habit distribution;
nk (L) dL is the number density of ice crystals in the size interval (L, L + dL) for the kth size
distribution. The factor 32 takes into account the spherical counterpart in terms of volume
and area. In this manner, there is only one unique De value for a given set of the lth habit
distribution and kth size distribution (Yue et al. 2007).
1.6 Size and shape distributions in cirrus clouds 43

Figure 1.23 depicts the ice crystal size distributions for a variety of cirrus covering a
number of temperature and cloud development conditions. Twelve size models for cirrus
were developed on the basis of the preceding data, with a large spectrum of De ranging
from 24 to 124 µm; for example, 30 µm for cirrus with a temperature of −60°C, 42 µm for
cirrostratus, 58 µm for a cirrus with a temperature of −40°C, and 124 µm for a developing
cirrus uncinus. This type of classification of ice crystal size distribution is useful for the
development of look-up tables for the purpose of retrieving ice crystal size and optical depth
from satellite radiance measurements. In this presentation, we also note that ice crystal size
in cirrus clouds spans values from a few to thousands of micrometers. Also depicted in
Figure 1.23 are four size distributions for contrails, which range from a few to about 50 µm
(Liou et al. 1998).
Finally, ice crystal shape and size data in Arctic ice and stratiform clouds associated with
frontal systems were collected during the First ISCCP Regional Arctic Cloud Experiment.
Ice crystals larger than 40 µm were collected, and inspection of their shapes shows a
combination of pristine and irregular types, including solid and hollow columns, prisms,
plates, aggregates, and branched particles (Korolev et al. 1999). In the Arctic region, ice
crystal data were also collected in the fall of 2004 during the DOE’s ARM Mixed-Phase
Arctic Cloud Experiment (MPACE) at its North Slope Alaska site. Additionally, we note
that the extensive collection of ice particles at a surface station in the Antarctic illustrates
the prevalence of long needle ice crystal types (Grenfell and Warren 1999).

1.6.3 Correlation between ice crystal maximum dimension and width


Ice particles are three-dimensional in nature and intercept light from all directions. Due
to their irregularity, the size and shape of an ice crystal have generally been defined by
its maximum dimension on the basis of two-dimensional photographs and/or images, as
displayed in Section 1.5. However, the maximum dimension is not sufficient for the purposes
of light scattering and absorption analysis: an additional dimension is needed to define the
ice crystal volume and effective geometric cross-section (Yang et al. 2000a).
Laboratory and field observations have established that the maximum dimension, L,
and width, D, are related for certain types of ice crystals (Auer and Veal 1970, Heymsfield
1972). Reanalyzing the data gathered by Auer and Veal, Mitchell and Arnott (1994) derived
the following empirical relationship for columns:

D = 2aLb , (1.6.2)

where a = 0.35 and b = 1 for L < 100 µm, and a = 3.48 and b = 0.5 for L > 100 µm. As
illustrated in the morphology section (Section 1.5), many columns exhibit hollow structure,
which must be accounted for in light-scattering calculations. We may postulate that the
cavity depth in a column varies randomly, given by d = 2ξ <d>, where ξ denotes a
random number uniformly distributed in (0, 1) and < > signifies a mean value. A value of
<d>L = 0.25 has been used by Yang et al. (2000a) in light-scattering calculations.
Figure 1.23 Representative ice crystal size distributions for cirrus clouds as a function of ice crystal maximum dimension covering a range of
temperature and cloud type. Also shown are four size distributions for contrails (after Baum et al. 2000b, Rolland et al. 2000, and Liou et al. 1998
with modifications).
1.7 Correlation between water content and crystal size 45

For plates, the maximum dimension refers to D, while L is the length along the c-axis.
Their relationship has been developed and is given by Pruppacher and Klett (1996) and
Mitchell and Arnott (1994) as follows:

L = a(D/2)b , (1.6.3)

where a = 2.4883 and b = 0.474 for D2 ranging from 5 to 1500 µm.
For bullet rosettes, the L–D relationship can also be defined by the empirical equation
for columns, but with a = 1.1552 and b = 0.63. Moreover, the pyramidal tip, t, of a bullet
element may be specified via the following relationship:

3
t= tan α, (1.6.4)
4D
where α is the inclination angle with respect to the major axis of the bullet elements.
According to a series of photographs taken and analyzed by Greenler (1980), α  28°.
For aggregates, the constructing procedure has been developed by Yang and Liou (1998a)
using a relative unit to specify ice crystal dimension. Furthermore, to account for surface
inhomogeneity of the preceding ice particles, we may use a two-dimensional Gaussian
probability function (Cox and Munk 1954) to represent the roughness of the ice surface in
light scattering calculations.
Lastly, we wish to note that the spatial orientation of ice crystals in cirrus is a non-
negligible factor in addressing light scattering and polarization properties. Jayaweera and
Mason (1965) have studied the behavior of freely falling cylinders in a viscous fluid and
found that if the ratio of diameter to maximum dimension is less than 1, cylinders tend
to fall with their long axes horizontal. Observations by Ono (1969) of natural ice clouds
in the atmosphere revealed that columnar and plate crystals fall with their major axes
oriented horizontally. Platt et al. (1978) demonstrated that in cirrus clouds at approxi-
mately −15°C, ice crystals are predominately plate shape, oriented horizontally, based
on lidar backscattering measurements. In particular, it was noted that the return signals
from ice clouds of a vertically pointed laser beam retain the polarization of the inci-
dent energy and that, for this to occur, the plates must be perpendicular to the laser
beam.

1.7 Correlation between ice water content and ice crystal size
1.7.1 Introductory note
In recent years, developments in cloud modeling have included prognostic equations used
to predict ice water content (IWC) for high-level clouds formed in climate and general
circulation models (GCMs). This is a milestone accomplishment from the standpoint of
incorporating a physically based cloud microphysics scheme into these models, and equally
essential from the perspective of studying cloud–radiation interactions. However, cloud par-
ticle size is also an independent parameter that affects radiation transfer. For example, for
46 Ice in the Earth’s atmosphere

a given IWC in clouds, smaller particles will reflect more sunlight than larger counterparts,
an effect recognized by Twomey et al. (1984) and Liou and Ou (1989) that is associated
with aerosol–cloud indirect effects. As presented in preceding sections, ice crystal size and
shape in the Earth’s atmosphere are complex and intricate. After initial homogeneous and/or
heterogeneous nucleation involving suitable aerosol particles and atmospheric conditions,
ice crystal growth is governed by diffusion processes and subsequent actions by means
of collision and coalescence. These processes are complicated by the nature of the ice
crystal’s hexagonal and irregular shape. Incorporating a fully interactive ice microphysics
scheme based on first principles into a GCM appears to be an extremely difficult compu-
tational task. Innovative parameterization based on theory and observation to determine
ice crystal size from the model-generated IWC must be developed for GCM and climate
applications.
It has been common practice in GCMs to prescribe a mean effective ice crystal size
De (e.g., Gu et al. 2003). A number of GCMs have also used temperature to determine
De (Kristjansson et al. 2005, Gu and Liou 2006). This approach is rooted in earlier
ice microphysics observations from aircraft, and attests to the fact that small and large
ice crystals are related to cold and warm temperatures in cirrus cloud layers. Ou and Liou
(1995) developed a parameterization equation relating cirrus temperature to mean effective
ice crystal size based on a large number of midlatitude cirrus microphysics data presented
by Heymsfield and Platt (1984). Ou et al. (1995) reduced large standard deviations in the
size–temperature parameterization by incorporating a dimensional analysis between IWC
and De . Using CEPEX data, McFarquhar et al. (2003) developed a De parameterization as
a function of IWC for use in a single-column model.

1.7.2 Correlation analysis


In this subsection, ice crystal size and shape distributions obtained from in situ measure-
ments conducted in numerous field campaigns have been used to understand the correlation
between ice water content (IWC) and mean effective ice crystal size (De ). In the correlation
analysis, we have divided available datasets in accordance with three geographical areas
(tropics, midlatitude, and Arctic) because of their distinct ice cloud formation processes
(Liou et al. 2008). A significant fraction of tropical cirrus are generated from towering
cumulus convection, but the majority of midlatitude cirrus clouds are primarily related to
large-scale frontal and synoptic systems and mesoscale topographical forcings. In Arctic
regions, the formation of ice clouds appears to be directly related to cold temperature,
large-scale transport of sensible and latent heat, and boundary layer turbulence. We first
introduce the ice water content, defined as

IWC = V (L)ρi n(L)dL, (1.7.1)

where n(L) is the ice crystal size distribution, V(L) is the volume of an individual ice crystal
that accounts for the shape factor, ρ i is the density of ice, and L is the ice crystal maximum
1.7 Correlation between water content and crystal size 47

dimension. We may define a mean effective ice crystal size related to IWC in the following
general form:
 
De = V (L)n(L)dL A(L)n(L)dL


= IWC ρi A(L)n(L)dL = IWC/ρi Ac , (1.7.2)

where we have neglected the constant factor 32 in Eq. (1.6.1), A(L) is the cross-sectional
area for an individual ice crystal, and Ac represents the total projected area for a given
ice crystal size and habit distribution. Heymsfield and McFarquhar (1996) found that Ac 
aIWCb , where a and b are empirical coefficients. The Ac –IWC relation revealed that a
direct correlation between De and IWC exists. Larger (smaller) values of IWC imply larger
(smaller) values of De , which is in agreement with ice crystal growth by means of diffusion
and accretion. The definition of this mean effective size efficiently accounts for ice crystal
size and shape distributions in light-scattering calculations (Fu and Liou 1993, Yang et al.
2000a). However, their relationship is not unique, but is constrained by Ac . The procedures
developed by Yue et al. (2007) and Yang et al. (2000a, 2005) were followed for IWC, De ,
and Ac calculations required in correlation development.
Uncertainty in the measurement of small ice crystals (<100 µm) from aircraft platforms
has been an important issue in scientific discussion. Shattering of millimeter-sized ice
particles due to collision with the probe can artificially enhance the concentration of small
ice crystals (see Heymsfield et al. 2006). In view of the uncertainty in small ice crystal
measurements, we have conducted three independent IWC–De correlation experiments:
(1) maximizing small ice crystals; (2) reducing the concentration of these smaller ice
crystals (Nsm ) by one order of magnitude; and (3) reducing Nsm by two orders of magnitude.
Experiment 2 was used as the base run, while the other two give a possible range of
parameterized De due to uncertainties in small ice crystal measurements.
A total of 40 469 in situ measurements of ice crystal size distribution in the tropics were
available for analysis, including 5460 from CRYSTAL-FACE and 35 009 from CEPEX. The
former data were collected by instruments on board NASA’s WB57 on nine different dates
during July 2002. CEPEX data included measurements from a 2-DC probe for ice crystals
larger than about 100 µm and a parameterization of Nsm based on VIPS measurements
(McFarquhar and Heymsfield 1997) obtained in March and April 1993. These instru-
ments were on board Aeromet Learjet. Both the original CEPEX and CRYSTAL-FACE
datasets were comprised of ice crystal measurements averaged over 30-second intervals.
Due to the low sampling rate of CRYSTAL-FACE data, size distributions derived from
this experiment were averaged over 5 minutes. After selecting datasets having more than
five-channel size measurements to ensure proper size average, only 261 CRYSTAL-FACE
and 11 032 CEPEX cases were used in the correlation study. From the analysis of CEPEX
data, about 34% and 66% of ice crystals are solid columns and bullet rosettes/aggregates,
respectively.
48 Ice in the Earth’s atmosphere

3
10
(a) Polar Ice Clouds

102

1
10
Mean Effective Diameter (De, μm)

MPACE (10/17-10/18, 2004)


Best Fit
3
10
(b) Midlatitude Cirrus

2
10

10
1 -65oC < T < -40oC
ARM (3/9/2000)
-40oC < T < -20oC
ARM (4/18/1997)
Best Fit
3
10
(c) Tropical Cirrus

2
10

1
10 CRYSTAL-FACE (7/2002)
CEPEX (3-4/1993)
Best Fit
0
10
10-6 10-5 10-4 10-3 10-2 10-1 100 101

Ice Water Content (IWC, g m–3)


Figure 1.24 Mean correlation curves with standard deviations (vertical bars) for IWC and De for (a)
the Arctic region, (b) midlatitude, and (c) the tropics, using ice crystal data collected during a number
of field campaigns listed in this figure and discussed in the text (after Liou et al. 2008).

Figure 1.24c displays 11 293 data points in a 2-D logarithmic domain. The IWC values
cover a range from 10−6 to 1 g m−3 , while De ranges from 20 to 200 µm. A χ 2 best
fit was used for these observed data to obtain the best parameterization equation in a
polynomial form, as follows:

ln(De ) = a + b ln(IWC) + c[ln (IWC)]2 , (1.7.3)

where a = 5.4199, b = 0.352 11, and c = 0.012 680. Uncertainty in Nsm can result in
deviations from the base run ranging from 5% to 40%, as IWC decreases, much larger
1.7 Correlation between water content and crystal size 49

than the standard deviation from statistical uncertainty. We have also made attempts to
correlate De and in situ temperature measurements (−70 to −20°C); however, the 2-D
data points are extremely scattered and do not show a consistent pattern. In this case, it
appears that temperature is not a suitable variable for De determination, probably due to
the predominant convective nature of cloud systems in the tropics, where vertical velocity
is a more significant parameter in regulating De .
A total of 4033 in situ size distribution measurements were obtained for midlatitude cases
taken at the DOE/ARM Southern Great Plain site. Ice crystals larger than 100 µm were
measured by the 2D-C probe on board the UND Citation. Aircraft datasets were composed
of measurements averaged over a 5-second interval. We selected size distributions that
had measurements greater than 5-size channels, resulting in 3919 cases. IWC ranges from
10−4 to 10−1 g m−3 , while De has values from 30 to 140 µm. Although habit and shape
information from ARM data sources was not available, previous studies revealed that for
midlatitude cirrus clouds, ice crystal shape spans from bullet rosettes and aggregates (60%)
to hollow columns (20%) to plates (20%) for L > 70 µm. For L < 70 µm, shapes are 50%
bullet rosettes, 25% plates, and 25% hollow columns (Baum et al. 2000b).
Correlations between De and IWC were improved by dividing the temperature into
two groups: −40 to −20°C (warm cirrus) and −65 to −40°C (cold cirrus), as shown
in Figure 1.24b. For warm cirrus, the correlation coefficients for parameterization are:
a = 5.2375, b = 0.131 42, and c = 0. For cold cirrus, we have a = 4.3257, b =
0.265 35, and c = 0.021 864. De for warm cirrus is generally larger than that for cold
cirrus, and the range of De and IWC for midlatitude cirrus is narrower than its tropical
counterpart.
In the Arctic region, analysis was based on in situ data collected during the DOE’s
ARM MPACE experiment at the North Slope of Alaska site in the fall of 2004, as shown
in Figure 1.24a. Ice clouds were observed on two days, October 17 and 18, consisting of a
total of 1705 cases. After a data quality check, only 468 cases were selected, which were
largely from the UND Citation 2D-C measurements. The data points were averaged over
a 30-second period to ensure adequate statistical sampling. For ice particles <100 µm, a
gamma distribution was used to extrapolate SDs to 2 µm based on the empirical coefficients
derived by Boudala et al. (2002). In terms of habit, Korolev et al. (1999) gave the percentage
of pristine and irregular habits in different temperature bins for high-latitude ice clouds.
Because temperatures were much lower during MPACE observations, ice clouds were
assumed to contain relatively more pristine particles (20%), with a ratio of columns to
plates of 100/20, resulting in 3.3% plates and 16.7% columns. The remaining irregular
ice particles include 40% bullet rosettes and 40% aggregates. The correlation coefficients
for parameterization are as follows: a = 4.8510, b = 0.331 59, and c = 0.026 189. The
values of IWC and De in Arctic ice clouds range from 10−2 to 1 g m−3 and from 50 to
120 µm, respectively. Nsm sensitivity experiments show that small particles contribute 5 to
20% as IWC increases, which is smaller than in the tropical and midlatitude cases. Arctic
ice cloud IWC and De have narrower ranges than those in the other two regions. We were
unable to obtain an adequate correlation between De and temperature, which ranges from
50 Ice in the Earth’s atmosphere

−57 to −17°C for this dataset, possibly due to less stratification of the polar temperature
profile.

1.8 Numerical modeling of cirrus cloud formation


1.8.1 A review
Most of the ice particles in the Earth’s atmosphere reside in cirrus clouds. In Section 1.2,
we showed that cirrus cloud systems are normally associated with synoptic and mesoscale
disturbances and are partly related to deep cumulus outflows. Day-to-day experience reveals
that cirrus clouds have a fibrous appearance (see Figure 1.1). As discussed in Section 1.2,
satellite imagery suggests that large portions of the tropics are covered by extensive cirrus
systems which vary spatially and temporally. Cirrus formations associated with mesoscale
complexes may have large vertical extents, with bases near 5–6 km and tops between 12
and 16 km, which evolve during the life cycle of the complexes and are modulated by
large-scale disturbances. At various stages, the cirrus system may involve a mid-level or
a single elevated deck in the upper atmosphere (Houze et al. 1981). Tropical anvils are
relatively stable and long-lived. Unlike water clouds, these high anvils contain irregular ice
crystals at low concentrations and can be optically thin and non-black. The influence of
cirrus clouds on the atmospheric radiation field depends on their solar and thermal infrared
radiative properties, which in turn are modulated by cloud composition and position. Cloud–
radiation feedback is important in the formation, maintenance, and dissipation of tropical
anvils produced by strong diurnal convection in the equatorial region.
Outflow cirrus clouds from tropical towers appear to be maintained in a convectively
active state by radiative flux gradients within the clouds (Danielsen 1982). These clouds are
warmed and lifted into the stratosphere by net radiative heating. Danielsen has suggested that
extended anvils would become radiatively destabilized by cooling at tops and warming at
bases. The resulting increase in the lapse rate within the anvils would drive convective fluxes
by providing an upward flux of water vapor. The additional moisture would promote rapid
ice crystal growth and fallout, hence serving as a dehydration mechanism for the tropical
stratosphere. Ackerman et al. (1988) have computed heating rates in typical tropical anvils
and found that the difference between heating rates at the cloud bottom and top ranges
from 30 to 200 K d−1 , leading to convective instability in the anvils. Radiative heating
could have important consequences for upward mass transport in the tropics. Using the
radiative heating rates presented by Ackerman et al. (1988), Lilly (1988) has analyzed the
dynamic mechanism of the formation of cirrus anvils using a mixed layer model, and has
shown that destabilization of the layer could be produced by the strong radiative heating
gradient. Although these studies have shed some light on the dynamics of tropical anvils,
a comprehensive understanding of their evolution and the role of radiative heating during
their evolution requires an analysis of results produced by an appropriate cloud model.
Global cirrus cloud fields play an important role in the radiation field of the Earth–
atmosphere system and, hence, significantly affect the atmospheric thermal structure and
climate (Liou 1986, 1992). This importance has been highlighted by a number of intensive
1.8 Numerical modeling of cirrus cloud formation 51

composite field observations: the First International Satellite Cloud Climatology Project
(ISCCP) Regional Experiment (FIRE) I in October–November 1986; FIRE II in November–
December 1991; the European experiment on cirrus (ICE/EUCREX) in 1989; and Subsonic
Aircraft: Contrail and Cloud Effect Special Study (SUCCESS) in April 1996. However, it
appears that fundamental understanding of the mechanisms for the formation, maintenance,
and dissipation of cirrus in the atmosphere remains limited and requires in-depth studies
from both observational and modeling perspectives.
Several attempts have been made to develop numerical models to investigate the role
of various physical processes in the life cycle of cirrus clouds using a one-dimensional
(1-D) model, in which the vertical velocity is prescribed. The 1-D model permits the use
of a highly complex microphysics module to study the evolution of ice clouds (Heymsfield
and Sabin 1989). Inclusion of the length of semi-major axes and the aspect ratio to define
hydrometers was developed by Chen et al. (1997) in a 1-D model to study cirrus clouds.
Employing a 1-D model, Detwiler and Ramaswamy (1990) investigated the evolution of
microphysics and heating rate profiles of a quiescent cirrus deck in a motionless atmosphere
without including nucleation and collision. Jensen et al. (1994a, 1994b) developed a 1-D
model which incorporated the explicit simulation of ice crystal size distributions, a number
of microphysical processes, and three groups of species, including condensation nuclei,
activated solution drops, and ice crystals, the source of which was primarily homogeneous
nucleation of the (NH4 )2 SO4 droplet.
Cirrus clouds are a dynamic and thermodynamic system that involves the intricate
coupling of microphysics, radiation, and dynamic processes (Gultepe and Starr 1995). A
multidimensional setting is thus required for interaction and feedback studies. In their
pioneering work, Starr and Cox (1985a, 1985b) developed a two-dimensional (2-D) model
and showed that the effects of radiative processes and vertical transport are both significant
in cirrus cloud formation and maintenance. Gu and Liou (1997a, 1997b, 2000) constructed a
2-D cirrus model that includes a second-order turbulence closure and found that turbulence
can significantly modulate the supersaturation condition and hence affect the rate at which
water vapor is converted to ice. Lin (1997) and Khvorostyanov and Sassen (1998a, 1998b)
have incorporated explicit microphysics schemes in 2-D models for the simulation of ice
crystal size distributions.
The modeling of anvil cirrus clouds has normally been connected with 3-D cumulonim-
bus simulations. Tao and Moncrieff (2009) have undertaken a comprehensive review of the
development of the cloud resolving model (CRM) from 1-D, 2-D, and 3-D perspectives
over the last four decades. The 3-D CRM has been used to investigate the evolution of cir-
rus clouds with reference to understanding the interaction of dynamics, microphysics, and
radiative transfer. In Subsection 1.8.2, we present a 2-D cirrus cloud model developed by Gu
and Liou (2000) to demonstrate the linkage to cloud fields with reference to the formation
and growth of ice crystals (Sections 1.3 and 1.4), ice crystal morphology (Section 1.5), ice
crystal size and shape distributions (Section 1.6), and light scattering and absorption by
ice crystals (Chapters 3 and 6). Cloud fields are governed by dynamic and thermodynamic
processes in the atmosphere in addition to microphysics and radiation interactions.
52 Ice in the Earth’s atmosphere

1.8.2 A two-dimensional cirrus cloud model


When modeling stratiform cirrus clouds, we consider two-dimensional space and use the
Boussinesq approximation, which treats a fluid as incompressible; that is, air density is
considered to be constant except when it is coupled with gravity in the buoyancy term
of the vertical momentum equation. Satisfying this approximation requires that the depth
of convection be much smaller than the scale height of the basic state, which is about
8 km. The convective region of cirrus studied has a depth of around 1 km during the entire
simulation period. The Coriolis term is neglected in view of the fact that the motion in
cirrus is small-scale.
Consider a cloud-free initial state which is horizontally uniform and at rest except for
a constant vertical motion w0 . Since turbulence plays an important role in the formation
and dissipation of stratus and stratiform clouds (Moeng 1986, Liu and Krueger 1998),
a second-order turbulence closure was incorporated in the formation of the cirrus cloud
model. Also, the explicit evolution of ice crystal size distribution was accounted for in cloud
formation. Based on the preceding considerations, the mean field equations for dynamic
and thermodynamic variables in the x-z plane can be expressed in the following forms (Gu
and Liou 2000):
∂u ∂u ∂u 1 ∂p ∂ 2 ∂
= −u − (w0 + w) − − u − u w  , (1.8.1)
∂t ∂x ∂z ρ ∂x ∂x ∂z
∂w ∂w ∂w 1 ∂p ∂   ∂
= −u − (w0 + w) − + gB − u w − w 2 , (1.8.2)
∂t ∂x ∂z ρ ∂z ∂x ∂z
where u and w are velocity components in the horizontal and vertical directions, respec-
tively. The unsubscripted variables represent the resolved components with respect to
a prescribed background (initial) state, which is denoted by subscript 0. Variables with
primes represent unresolved turbulent fluctuations. B is the buoyancy term, which can be
represented by the potential temperature θ , the specific humidity of water vapor q, and the
ice water mixing ratio qc in the form
θ
B= + ε q − qc , (1.8.3)
θ0
where ε = (1 − ε)/ε, and ε is the ratio of the molecular weight of water to that of dry air.
The basic equations for θ , q, and the number densities of ice crystals for the ith size
group ni can be written as
∂θ ∂θ ∂(θ0 + θ ) ∂   ∂
= −u − (w0 + w) + π−1 (QC + QR ) − u θ − w θ  , (1.8.4)
∂t ∂x ∂z ∂x ∂z
∂q ∂q ∂(q0 + q) ∂   ∂
= −u − (w0 + w) −C− u q − w q  , (1.8.5)
∂t ∂x ∂z ∂x ∂z


∂ni ∂q dmi ∂ ∂
=− ni − [uni ] − [(w + w0 − Uf )ni ]
∂t ∂m dt ∂x ∂z


∂ni ∂   ∂
+ − u ni − w  ni  , (1.8.6)
∂t nuc ∂x ∂z
1.8 Numerical modeling of cirrus cloud formation 53

where π = (p0 /pr )R/Cp , with pr = 1000 mb; Cp is the specific heat at constant pressure;
mi is the mean mass of an individual particle in the ith size group; and Uf is the terminal
velocity of an ice crystal. The first term on the right-hand side of Eq. (1.8.6) represents
the net accumulation of ice crystal number density of a particular size due to diffusional
growth. The second term is the contribution from horizontal advection, while the third
term is the vertical convergence of ice crystal number density determined by vertical air
motion and ice crystal terminal velocity Uf . The fourth and fifth terms denote the nucleation
source and eddy diffusion contributions, respectively. To obtain solutions for the evolution
of ice crystal size distributions, the ice crystal length spectrum is discretized into a finite
number of intervals (bins) of 15 µm. A partial differential equation can then be expressed
for each bin and solved numerically. It is clear that several physical factors are involved
in the governing equations: radiation, saturation ratio, air temperature, and horizontal and
vertical motion.
Radiative heating, QR , can be computed from the radiative transfer scheme discussed in
Subsection 1.8.2.2. The ice water mixing ratio, qc , the heating rate due to phase change, QC ,
and the deposition or sublimation between water vapor and ice water, C, can be connected
to the ice crystal size spectrum in the forms

N
qc = ni mi , (1.8.7)
i=1

Ls C
QC = , (1.8.8)
Cp

N

dmi ∂n1
C = Cdif + Cnuc = ni + m1 , (1.8.9)
i=1
dt ∂t nuc

where the ice crystal size spectrum has been discretized into N bins; Ls is the latent heat of
sublimation; and Cdif and Cnuc represent sources and sinks of moisture due to diffusional
growth and nucleation of ice crystals, respectively. The ice crystals initially generated by
nucleation are denoted by the mass of the smallest size group, m1 . Thus, Cnuc is the product
of m1 and the rate of change of number density of ice crystals due to nucleation in this
size group. When the nucleation process is terminated, Cnuc = 0. In the following two
subsections, we discuss ice microphysics, interactions between radiation and an individual
ice crystal, and the radiative heating rate produced in cloud fields.

1.8.2.1 Ice microphysics


The subject of ice microphysics includes ice nucleation, diffusional growth, and terminal
velocity. Ice nucleation processes can be generally classified into four modes: homogeneous
freezing nucleation, heterogeneous freezing nucleation, heterogeneous deposition nucle-
ation, and contact freezing nucleation (see Figure 1.11). Homogeneous freezing nucleation
usually occurs at or below −40°C, while heterogeneous freezing is likely to be more
effective when sufficient water is involved (Chen and Lamb 1994). The nucleation process
requires the presence of ice nuclei (IN). Fletcher (1962) developed a relationship between
54 Ice in the Earth’s atmosphere

the effective IN concentration and temperature as follows: NIN = N0 exp[b (TSTD − T)],
where N0 and b are constants and TSTD = 273.15 K. This expression implies that NIN is a
well-defined function of temperature. However, Meyers et al. (1992) found that Fletcher’s
equation tends to under-predict ice crystal concentrations for temperatures warmer than
−20°C, while over-predicting them for colder clouds. Gagin (1972) and Huffman (1973)
found that the number concentration of effective IN increases with increasing supersatu-
ration over ice, Si , in the form NIN = cSik , where c and k are constants determined from
Huffman (1973) and Chen and Lamb (1994). Heterogeneous deposition nucleation can be
taken into account using this relationship. The effective INs nucleate immediately, implying
that their appearance is the same as the creation of new ice particles. The nucleation process
is assumed to be irreversible; that is, dNIN
0.
The time rate of change of mass for an individual ice crystal denoted in Eq. (1.8.9) can
be obtained from the classic mass growth theory, including the conservation of water mass
and total energy at the ice crystal surface (see Subsection 1.4.1). The diffusional growth for
an individual ice crystal is governed by the following two equations:
dm
+ 4πC  D  f1 ρ = 0, (1.8.10)
dt
dm
Ls + HR − 4πC  K  f2 T = 0, (1.8.11)
dt
where m is the mass of an individual crystal; C  is the equivalent radius for diffusive
processes of a non-spherical crystal; D  and K  are the diffusion coefficients for water
vapor and heat, respectively; f1 and f2 are the combined ventilation and surface curvature
factors, respectively; ρ and T are the density and temperature differences between the
ice crystal and the environment, respectively; and HR denotes the radiative effect on the
individual ice crystal. Analytic expressions for mass growth rate and ice crystal temperature
can be derived by solving Eqs. (1.8.10) and (1.8.11) simultaneously. Moreover, using the
Clausius–Clapeyron equation, the ideal gas law, and Kelvin’s expression for vapor pressure,
we can obtain the following equation for ice crystal growth:
dm
= AS − BH R , (1.8.12)
dt
where
2s 4πf2 K  C  Ls
S = (si − 1) − , A= , B= ,
ρi Rw T r B Rw T 2 B 2
L2s f2 K  Rw T
B = + (1.8.13)
Rw T 2 f1 D  es
and si is the ice saturation ratio, s is the surface free energy of ice, ρi is the bulk ice density,
T is the temperature in kelvin, Rw is the gas constant for water vapor, r is half of the ice
crystal width, and es is the saturation vapor pressure with respect to ice at temperature
T. The values for the unknown coefficients were taken from Laube and Holler (1988),
Ramaswamy and Detwiler (1986), Heymsfield (1972), and Auer and Veal (1970). The
1.8 Numerical modeling of cirrus cloud formation 55

mean mass of ice crystals of each size group can be calculated based on parameterization
relationships.
The terminal velocity of ice particles, Uf , defined in Eq. (1.8.6) can be expressed by the
following relationship (see Subsection 1.4.2):
Re η
Uf = , (1.8.14)
ρL
where η denotes the air viscosity; L is the ice crystal characteristic length (maximum
dimension), which can be computed from the empirical relationship given by Auer and
Veal (1970); and Re is the Reynolds number, which can be calculated from the unified
Davis number (Böhm 1989).

1.8.2.2 Radiative transfer


Following Fu and Liou (1993), the single-scattering properties of hexagonal ice crystals
can be computed from the parameterization using a mean effective size, De , to repre-
sent the ice crystal size distribution in radiative transfer calculations. Detailed ice crystal
shapes, which are a function of temperature and supersaturation, were not accounted for
in the study. Based on physical principles, the extinction coefficient normalized by IWC,
the single-scattering albedo, and the expansion coefficients for the phase function may be
expressed as third-degree polynomials in 1De . The coefficients in the polynomials that
are wavelength-dependent are determined by numerical fitting to the exact results obtained
from the light-scattering and absorption programs using 11 observed ice crystal size distri-
butions (Takano and Liou 1989a, Fu and Liou 1993). See Section 6.1 for further discussion.
The single-scattering parameterization was performed for 6 solar and 12 IR bands. The
radiation parameterization program is driven by the mean effective ice crystal size and ice
water path, which is the product of cloud thickness and IWC determined from

IWC = ρ n(mi )mi , (1.8.15)
i

where n(mi ) is computed from the model and ρ is the ice density.
For an individual ice crystal, the heating produced by the radiation effect must be
the product of the absorption cross-section and the net radiative flux density, and can be
expressed as follows:
 ∞
HR = σa,λ (mi )[Fλ+ + Fλ− − 2πBλ (T )]dλ ∼
= σa (mi )[F + + F − − 2σ T 4 ], (1.8.16)
0

where HR is the rate of radiative energy gained by an individual particle, which is determined
separately for each size bin, F+ and F− are the upward and downward net radiative fluxes,
and Bλ (T) is the Planck function at temperature T. To simplify the calculations, we may use
a mean absorption cross-section of an ice crystal, σa , which can be obtained from

I
σa = σa,i /I , (1.8.17)
i=1
56 Ice in the Earth’s atmosphere

9
10 min
20 min
40 min
8 60 min
120 min
Height (km)

5
0 2 4 6 8 10
Ice Water Content (mg m-3)

Figure 1.25 Vertical profile of the horizontally averaged IWC simulated from a 2-D cirrus cloud
formation model at five time steps (after Gu and Liou 2000).

where I is the total number of spectral bands in the parameterization and σ a,i is the
absorption cross-section of an ice crystal for a given wavelength, determined from radiation
parameterization. The radiative flux within a volume element of cloud can be obtained from
the radiation scheme. In this manner, the ice crystal size distribution is fully interactive with
radiation via the ice crystal mean effective size and IWC. The shapes of ice crystals were
assumed to be hexagonal solid columns and plates; other habits depicted in Figure 1.19
were not accounted for. Also, the 2-D cirrus cloud model did not account for collision and
coalescence events, which are critical for the formation of precipitation.
Lastly, we present in Figure 1.25 the vertical profile of IWC, the horizontally averaged
IWC, during a 120 min cirrus cloud simulation. Ice crystals initially form by deposition
nucleation associated with saturated air produced by dynamic and thermodynamic forcings.
At t = 10 min, IWC increases from 0 to about 5.7 mg m−3 . The nucleation process stops
when the supersaturation ratio begins to decrease due to deposition. Ice crystals are then
subject to diffusion growth, advection, sedimentation, and eddy diffusion. Ice crystals
become larger and fall to lower levels. At t = 20 min, IWC reaches its maximum value
of about 9 mg m−3 , and the location of its maximum shifts from 7.55 km at t = 10 min to
7.15 km at t = 20 min. At the same time, the cloud base is moving downward. After ice
crystals descend into the drier lower region, sublimation occurs and the simulated maximum
IWC decreases to about 4.2 mg m−3 at t = 40 min. After t = 60 min, a near steady state
is reached. Observations of cirrus IWC in weak synoptic forcing situations have been
1.8 Numerical modeling of cirrus cloud formation 57

reported to be between 3 and 8 mg m−3 over a horizontal path longer than 3 km and with
a temperature range of −30 to −40°C (Heymsfield 1975). The model results thus appear
to compare reasonably with the previous observations. The updrafts in the cloud layer are
larger than the downdrafts in the earlier stage of the simulation, with maximum values
of about 30 and 5 cm s−1 , respectively, at t = 20 min. Toward the end of the simulation,
updrafts and downdrafts are of the same magnitude, with a value of about 15 cm s−1 . The
simulation results are comparable to the observed vertical velocities in cirrostratus, with
typical values ranging from 5 to 20 cm s−1 and a maximum of about 60 cm s−1 (Heymsfield
1977).
In summary, in the development of a numerical model for the formation and evolution of
cirrus clouds, we have shown that ice microphysical properties in terms of ice nucleation,
diffusional growth, and terminal velocity, which are related to ice crystal shape and size,
become significant. Moreover, in order to evaluate cloud radiative heating and the heating
of individual ice particles, the basic scattering and absorption data for various ice crystal
sizes and shapes are required for accurate calculations of the transfer of solar and thermal
infrared radiation in association with model development.
2
Fundamentals of light scattering by ice crystals

In this chapter we present a number of fundamental subjects related to light scattering by


ice crystals. In Section 2.1, we discuss the scope of light scattering by ice crystals from
the perspectives of the independent scattering concept, the same light frequency before and
after scattering events, and the coupled absorption and induced polarization. Section 2.2
covers the fundamental Maxwell’s equations, leading to vector wave equations and required
boundary conditions for solutions. We show that exact analytic solutions of vector wave
equations exist only for spherical, circular cylindrical, and spheroidal coordinates. The
optical properties of ice are subsequently introduced in Section 2.3, followed by a discussion
in Section 2.4 of the definition of the single-scattering and polarization properties of non-
spherical ice crystals. Lastly, in Section 2.5 we discuss the link between the single-scattering
properties of ice crystals, deduced from the independent scattering concept, and the transfer
of radiation, which includes multiple scattering and emission in clouds containing an
ensemble of ice crystals.

2.1 Introductory remarks


In Chapter 1, we presented the formation of ice crystals in the atmosphere and the complexity
of their structure, shape, and size. We showed that the shape and size of an ice crystal are
primarily governed by temperature and supersaturation, but that they generally have a
basic hexagonal structure. In the atmosphere, if ice crystal growth involves collision and
coalescence, the crystal’s shape can be extremely complex. Observations based on aircraft
optical probes and replicator techniques for widespread midlatitude, tropical, Arctic, and
contrail cirrus show that these clouds are composed largely of ice crystals in the shape of
bullet rosettes, solid and hollow columns, plates, and aggregates, various forms of dendrites,
and ice crystals with irregular surfaces. The size of ice particles normally ranges from a few
micrometers to thousands of micrometers. In the presence of sunlight, some of the defined
ice crystals produce fascinating optical phenomena, including 22◦ and 46◦ halos, sundogs,
and numerous arcs and bright spots.
The scattering of light is a physical process by which a particle in the path of an
electromagnetic wave continuously abstracts energy from the incident wave and reradiates

58
2.1 Introductory remarks 59

that energy in all directions. Therefore, a particle may be thought of as a point source of
scattered energy. Scattering is a fundamental physical process associated with light and its
interaction with matter. In the presentation of light scattering by ice crystals, the following
three limitations apply.
First, in atmospheric scattering, it has normally been assumed that the light scattered
by molecules and particulates has the same frequency (or wavelength or wavenumber) as
the incident light. Shift frequencies produced by high-energy laser light such as Raman
scattering will not be within the purview of the present discussion.
Second, as illustrated in Chapter 1, the number density of ice crystals in clouds is
relatively small and ice crystals are separated by distances which are much larger than their
sizes, such that each would scatter light in exactly the same way if all other particles did not
exist. This is referred to as independent scattering. The independent scattering assumption
greatly simplifies the problem of light scattering involving a spectrum of particles within
the context of a “sufficiently small” volume so that the single-scattering properties of ice
crystals (as well as water droplets and aerosols) can be defined for application to radiative
transfer calculations. The concept of independent scattering allows for the use of energy
quantities instead of electric fields in the analysis of the propagation of electromagnetic
waves in planetary atmospheres.
Third, in a scattering volume which contains many particles, each particle is exposed to,
and also scatters, the light which has already been scattered by other particles. Scattering
more than once is referred to as multiple scattering. In principle, in the presence of particles,
light can undergo an infinite number of scattering events. In connection with the concept
of independent scattering discussed above, we will cover multiple scattering in ice crystal
clouds to the extent of introducing random and horizontal orientations of ice crystals. The
link between light scattering (single scattering events) and multiple scattering (radiative
transfer) will be discussed.
Scattering is frequently accompanied by absorption. Grass looks green because it scatters
green light while it absorbs red and blue light. The absorbed energy is converted into some
other form, and it is no longer present as red or blue light. Both scattering and absorption
remove energy from a beam of light traversing a medium. The beam of light is said to be
attenuated, and this attenuation is referred to as extinction. Thus, extinction is a result of
scattering plus absorption. In a non-absorbing medium, scattering is the sole process of
extinction. Absorption of sunlight by ice crystals is generally negligible at ultraviolet (UV)
and visible wavelengths. However, substantial absorption occurs at infrared and microwave
wavelengths.
All scattering processes lead to polarization in association with the propagation of
light beams. A light beam is characterized by certain polarization configurations that are
defined by the vibration of the electric vector and by the phase difference between the two
components of this vector. The polarization of a light beam cannot be visualized even in
the visible spectrum and requires a specific instrument for its detection. In Section 2.2, we
shall introduce Maxwell’s equations, wave equations, and boundary conditions, basic to the
discussion of the propagation of electromagnetic waves and their interaction with matter.
60 Fundamentals of light scattering by ice crystals

2.2 Maxwell’s equations, wave equations, and boundary conditions


2.2.1 Maxwell’s equations
The state of excitation established in space by the presence of electric charges constitutes
an electromagnetic field, which is governed by the fundamental Maxwell’s equations. The
field is represented by the electric vector E and the magnetic induction vector B. Addi-
tionally, a second set of vectors involving the electric current density j, the electric
displacement D, and the magnetic vector H must be introduced in order to describe
the effect of the electromagnetic field on material objects. Subject to the condition that
the physical properties of the medium are continuous at every point in its neighborhood, the
space and time derivatives of these five vectors are related by the following four equations
(Maxwell 1865):
1 ∂D 4π
∇×H= + j, (2.2.1a)
c ∂t c
1 ∂B
∇×E=− , (2.2.1b)
c ∂t
∇ · D = 4πρ, (2.2.1c)

∇ · B = 0, (2.2.1d)
where t denotes time and c is the velocity of light in vacuum, which is approximately
equal to 3 × 1010 cm s−1 . Equation (2.2.1c) can be considered as a defining equation for
the electric charge density ρ, while Eq. (2.2.1d) implies that no free magnetic poles exist.
Following Born and Wolf (1975), we use the Gaussian system of units in this presentation.
Since div curl (∇ · ∇×) of any vector is equal to 0, e.g., ∇ · ∇ × H = 0, a div (dot
product) operation on Eq. (2.2.1a) leads to
1 ∂D
∇·j=− ∇· . (2.2.2)
4π ∂t
Differentiating Eq. (2.2.1c) with respect to t leads to
∂ρ
+ ∇ · j = 0. (2.2.3)
∂t
This equation represents the equation of continuity in an electromagnetic field where charge
is conserved in the surroundings of any point.
In order to uniquely determine the field vectors from a given distribution of currents
and charges, the preceding equations must be augmented by relationships describing the
behavior of substances under the influence of the field. The electron conduction response
to the electric field and the two constitutive relations describing the aggregate response of
bond charges to the electric field and corresponding magnetic induction can be defined by
the following three equations:
j = σ E, D = εE, B = μH, (2.2.4)
2.2 Maxwell’s equations and boundary conditions 61

where σ is referred to as specific (or electrical) conductivity, ε is the permittivity, and


µ is the magnetic permeability. Substances for which σ
= 0 are called conductors. For
example, metals are excellent conductors in which conductivity decreases with increasing
temperature. However, in some classes of materials, known as semiconductors, conductivity
over a wide range increases with temperature. Substances for which σ is negligibly small are
referred to as dielectrics. In this case, their electric and magnetic properties are completely
determined by ε and µ. For most substances (air, for example), the magnetic permeability
µ is practically unity.
Maxwell’s equations connect the field vectors through simultaneous differential equa-
tions. To obtain individual differential equations for the vectors, we shall confine our
attention to a part of the field where there are no charges (ρ = 0) or currents (j = 0), and to
a medium which is homogeneous, so that ε and µ are constants. Subject to these conditions,
we obtain the so-called source-free Maxwell equations:
ε ∂E
∇×H= , (2.2.5a)
c ∂t
μ ∂H
∇×E=− , (2.2.5b)
c ∂t
∇ · E = 0, (2.2.5c)
∇ · H = 0. (2.2.5d)
In the following discussion, we will use Eqs. (2.2.5a)–(2.2.5d) to derive the electromagnetic
wave equation. Note that Eqs. (2.2.5c) and (2.2.5d) can be directly obtained by performing
a dot product operation on Eqs. (2.2.5a) and (2.2.5b).
Consider now a plane electromagnetic wave in a periodic field defined by a circular
frequency ω such that
E → Eeiωt , H → Heiωt , (2.2.6)

where i = −1 and we have used exp(iωt) in the derivation in order to have a negative
value for the imaginary part mi of the refractive index (see Subsection 2.3.1). Using these
transformations, Eqs. (2.2.5a) and (2.2.5b) can then be written as
∇ × H = ikm2 E, (2.2.7a)
∇ × E = −ikH, (2.2.7b)
where k = 2π/λ(= ω/c) is the wavenumber denoting the propagation constant in vacuum,

ω is the circular frequency, λ is the wavelength in vacuum, m = ε is the complex
refractive index of the medium at the frequency ω, and μ(∼ 1) is the permeability of air
noted previously.
By performing the curl operation on Eq. (2.2.7b), we obtain
∇ × ∇ × E = −ik∇ × H. (2.2.8)
62 Fundamentals of light scattering by ice crystals

(a) n2 n (b)
A t2

e2 n
Δh t
Δh ΔA
nA S
e1 t1
e2 S Δs

e1 n1

Figure 2.1 Illustration of the derivation of (a) normal boundary conditions for the magnetic induction
vector B and the electrical displacement vector D and (b) tangential boundary conditions for the
electric vector E and the magnetic vector H. All notations are defined in the text.

Since ∇ × ∇ × E = ∇(∇ · E) − ∇ 2 E and ∇ · E = 0, we can substitute ∇ × H from


Eq. (2.2.7a) to obtain the following equation:
∇ 2 E = −k 2 m2 E. (2.2.9)
Likewise, we can derive the following relationship from Eqs. (2.2.7a) and (2.2.5d) for the
magnetic vector H:
∇ 2 H = −k 2 m2 H. (2.2.10)
The preceding two equations illustrate that the electric vector and the magnetic induction
in a homogeneous medium satisfy the following vector wave equation:
∇ 2 A + k 2 m2 A = 0, (2.2.11)
where the vector A can be either E or H. In Subsection 2.2.2, we shall introduce the
boundary conditions that are required for solving the vector wave equation.

2.2.2 Boundary conditions


Across any surface separating two media, shape changes may occur in the parameters
characterizing the media so that the field vectors are expected to undergo corresponding
changes. Under the conditions of no charges (ρ = 0), no currents (σ = 0), and magnetic
permeability µ  1, the two media are then defined only by the permittivity, ε1 and ε2 , or
by the complex indices of refraction, m1 and m2 .

2.2.2.1 Normal components


Solution of the vector wave equations requires suitable boundary conditions (Born and
Wolf 1975). The boundary surface separating the two media is denoted by the symbol S
and the associated normal vector n, as shown in Figure 2.1a. Consider an infinitesimally
thin transition layer such that the parameters change rapidly but continuously from their
values near S in (1) to their counterparts in (2). Within this transition layer as well as in (1)
and (2), the field vectors and their first derivatives are assumed to be continuous bounded
2.2 Maxwell’s equations and boundary conditions 63

functions of position and time. We may construct a small cylinder of height h and cross-
sectional area A, and let the directions of the surface normal unit vectors, n1 and n2 , at the
bottom and the top of the cylinder, be as indicated in Figure 2.1a. Integrating the divergence
equations (2.2.1c) and (2.2.1d) over the volume V = h A of the cylinder and using
Gauss’ divergence theorem, we have
 
∇ · B dV = B · dA = 0,
V
  (2.2.12)
∇ · D dV = D · dA = 0,
V

where the integrals over A are over the closed cylinder surface. To evaluate the surface
integrals, we assume that the cylinder height h → 0 so that the contribution of the walls
may be neglected. In this manner, we may write
  
B · dA = B1 · n1 dA + B2 · n2 dA = 0,
A A
   (2.2.13)
D · dA = D1 · n1 dA + D2 · n2 dA = 0.
A A

Since n2 = −n1 = n as shown in Figure 2.1a, it follows that


 
(B2 − B1 ) · ndA = 0, (D2 − D1 ) · ndA = 0. (2.2.14)
A A

Assuming that the field vectors are constant within an infinitesimal area element A,
we can then obtain the boundary conditions for the normal components of B and D as
follows:

(B2 − B1 ) · n = 0, (D2 − D1 ) · n = 0. (2.2.15)


These equations state that the normal components of B and D across any discontinuous
surface are continuous.

2.2.2.2 Tangential components


To express the behavior of the tangential components of B and D at the surface S, we shall
consider a closed surface A and the associated normal vector nA , as depicted in Figure 2.1b.
The tangential unit vectors along the lower and upper boundaries of A are denoted by t1
and t2 . We can perform the integration of Eqs. (2.2.1a) and (2.2.1b) (j = 0) and use the
Stokes integral theorem to obtain the following two equations:
  
1 ∂B
∇ × E · nA dA = E · ds = − · nA dA,
A A c ∂t
   (2.2.16)
1 ∂D
∇ × H · nA dA = H · ds = · nA dA,
A A c ∂t

where the first surface integrals have been replaced by closed line integrals.
64 Fundamentals of light scattering by ice crystals

The line integrals can be decomposed into two parts in each equation. In the limit
h → 0, contributions from the corresponding sides vanish. Thus, we have
  
1 ∂B
E1 · t1 ds + E2 · t2 ds = − · nA dA,
s s A c ∂t
   (2.2.17)
1 ∂D
H1 · t1 ds + H2 · t2 ds = · nA dA.
s s A c ∂t

For a small surface A, we have t2 = −t1 = nA × n, so that


 
1 ∂B
(E2 − E1 ) · nA × nds = − · nA dA,
s A c ∂t
  (2.2.18)
1 ∂D
(H2 − H1 ) · nA × nds = · nA dA.
s A c ∂t

For small values of A, all integrands can be considered to be constant. It follows that
1 ∂B
(E2 − E1 ) · nA × n = − · nA h,
c ∂t
(2.2.19)
1 ∂D
(H2 − H1 ) · nA × n = · nA h.
c ∂t
The right-hand sides of these equations vanish in the limit of h → 0, because the field
vectors and their derivatives with respect to time remain bounded. Finally, we obtain the
results

n × (E2 − E1 ) = 0, n × (H2 − H1 ) = 0, (2.2.20)

which state that the tangential components of the field vectors are continuous across a
discontinuous surface S. Equations (2.2.15) and (2.2.20) are necessary for the solution of
the differential vector wave equations denoted in Eqs. (2.2.9) and (2.2.10).

2.2.3 Wave equations and some analytic solutions


In light scattering problems, the solution of the vector wave equation denoted in Eq. (2.2.1)
subject to appropriate boundary conditions can be obtained only if a suitable coordinate
system can be defined to conform the particle geometry. In the following subsections, we
present solutions to the vector wave equation for the cases of a sphere, a cylinder, and
a spheroid, in which spherical, cylindrical, and spheroidal coordinates can be imposed,
respectively.

2.2.3.1 Spherical coordinates


We fully realize that the detailed solution for light scattering by a sphere has been presented
in historical papers by Lorenz (1890), Mie (1908), and Debye (1909), and is now referred
to as the Lorenz–Mie or simply the Mie theory. However, we wish to capture some new
2.2 Maxwell’s equations and boundary conditions 65

ar
z

af

r P(r, q, f)
z
θ
y
φ aq
x
y

x
Figure 2.2 Definitions of the parameters in spherical coordinates, where x = r sin θ cos φ, y =
r sin θ sin φ, z = r cos θ ; and ar , aθ , and aφ are orthogonal vectors with P a point on the sphere.

developments to introduce the manner in which the exact solution of the vector wave
equation can be derived from spherical coordinates. We shall largely follow the formulation
developed in Liou (1977, 2002). Referring to Eq. (2.2.11), we may write the scalar wave
equation in the form
∇ 2 ψ + k 2 m2 ψ = 0, (2.2.21)
where ψ is a scalar which satisfies the scalar wave equation. Consider the vectors Mψ and
Nψ in spherical coordinates (r, θ , φ) (Figure 2.2), defined by the following two equations:

∂ 1 ∂ 1 ∂
Mψ = ∇ × [ar (rψ)] = ar + aθ + aφ × [ar (rψ)]
∂r r ∂θ r sin θ ∂φ
(2.2.22)
1 ∂(rψ) 1 ∂(rψ)
= aφ − aφ ,
r sin θ ∂φ r ∂θ
mkNψ = ∇ × Mψ

2 (2.2.23)
∂ (rψ) 1 ∂ 2 (rψ) 1 ∂ 2 (rψ)
= 2
+ m 2 2
k (rψ) ar + aθ + aφ .
∂r r ∂r∂θ r sin θ ∂r∂φ
These equations satisfy the vector wave equation defined in Eq. (2.2.11) subject to
Eq. (2.2.21). The terms ar , aθ , and aφ are unit vectors in spherical coordinates.
Let u and v be two independent solutions of the scalar wave equation defined in
Eq. (2.2.21). The electric and magnetic field vectors can then be expressed as
E = Mv + iNu , (2.2.24)
H = m( −Mu + iNv ). (2.2.25)
66 Fundamentals of light scattering by ice crystals

These two equations will satisfy Eqs. (2.2.9) and (2.2.10). Using Eqs. (2.2.22) and (2.2.23),
E and H can then be written explicitly in spherical coordinates as follows:



i ∂ 2 (ru) 1 ∂(rv) i ∂ 2 (ru)
E= + m 2 2
k (ru) ar + + aθ
mk ∂r 2 r sin θ ∂φ mkr ∂r∂θ

(2.2.26)
1 ∂(rv) 1 ∂ 2 (ru)
+ − + aφ ,
r ∂θ mkr sin θ ∂r∂φ



i ∂ 2 (rv) m ∂(ru) i ∂ 2 (rv)
H= + m k (rv) ar + −
2 2
+ aθ
k ∂r 2 r sin θ ∂φ kr ∂r∂θ

(2.2.27)
m ∂(ru) i ∂ 2 (rv)
+ + aφ .
r ∂θ kr sin θ ∂r∂φ
We now return to the scalar wave equation defined in Eq. (2.2.21), which can be expressed
in spherical coordinates as follows:

1 ∂ 2 ∂ψ 1 ∂ ∂ψ 1 ∂ 2ψ
r + sin θ + + k 2 m2 ψ = 0. (2.2.28)
r 2 ∂r ∂r r 2 sin θ ∂θ ∂θ r 2 sin θ ∂φ 2
We now define the solution of this equation in terms of three independent variables in
the form
ψ(r, θ, φ) = R(r)(θ )(φ). (2.2.29)
Upon substituting Eq. (2.2.29) into Eq. (2.2.28) and dividing the entire equation by
ψ(r, θ , φ), we obtain

1 1 ∂ 2 ∂R 1 1 ∂ ∂ 1 1 ∂ 2
r + sin θ + + k 2 m2 = 0.
r 2 R ∂r ∂r r 2 sin θ  ∂θ ∂θ r 2 sin2 θ  ∂φ 2
(2.2.30)
Subsequently, we may multiply Eq. (2.2.30) by r2 sin2 θ to obtain


2 1 ∂ 2 ∂R 1 ∂ ∂ 1 ∂ 2
sin θ r + sin θ sin θ + k m r sin θ +
2 2 2 2
= 0.
R ∂r ∂r  ∂θ ∂θ  ∂φ 2
(2.2.31)
Since the first three terms in this equation consist of only the variables r and θ , but not φ,
Eq. (2.2.31) can be valid only under the condition that
1 d 2
= const = −2 , (2.2.32)
 dφ 2
where, for mathematical convenience, the constant has been chosen to be −2 , and  denotes
an integer. Combining Eqs. (2.2.31) and (2.2.32), we then have

2 1 ∂ 2 ∂R 1 ∂
sin θ r + sin θ sin θ + k 2 m2 r 2 sin2 θ − 2 = 0. (2.2.33)
R ∂r ∂r  ∂θ
2.2 Maxwell’s equations and boundary conditions 67

Upon dividing Eq. (2.2.33) by sin2 θ , we obtain the following equation:



1 ∂ ∂R 1 1 ∂ ∂ 2
r2 + k 2 m2 r 2 + sin θ − = 0. (2.2.34)
R ∂r ∂r sin θ  ∂θ ∂θ sin2 θ
This equation consists of two independent variables, namely R and . Thus, we must have

1 d dR
r2 + k 2 m2 r 2 = const = n(n + 1), (2.2.35)
R dr dr

1 1 d d 2
sin θ − = const = −n(n + 1) (2.2.36)
sin θ  dθ dθ sin2 θ
to satisfy Eq. (2.2.34), where n is an integer. Again, the selection of the constant
−n(n + 1) is for mathematical convenience. After rearranging Eqs. (2.2.32), (2.2.35),
and (2.2.36), we obtain the following three equations:


d 2 (rR) n(n + 1)
+ k m −
2 2
(rR) = 0, (2.2.37)
dr 2 r2


1 d d 2
sin θ + n(n + 1) −  = 0, (2.2.38)
sin θ dθ dθ sin2 θ

d 2
+ 2  = 0. (2.2.39)
dφ 2
The single-value solution of Eq. (2.2.39), a second-order differential equation, is given by

 = a cos φ + b sin φ, (2.2.40)

where a and b are arbitrary constants. Furthermore, Eq. (2.2.38) is the well-known
equation for spherical harmonics. For mathematical convenience, we may introduce a new
variable µ = cos θ so that Eq. (2.2.38) can be rewritten in the form



d 2 d 2
(1 − μ ) + n(n + 1) −  = 0. (2.2.41)
dμ dμ 1 − μ2
The solutions to Eq. (2.2.41) can be expressed using the associated Legendre polynomials,
which are the spherical harmonics of the first kind, in the form

 = pn (μ) = pn ( cos θ ). (2.2.42)

Finally, in order to solve the remaining Eq. (2.2.37), we set



kmr = ρ, R = (1/ ρ)Z(ρ), (2.2.43)

to obtain the following equation:




d 2Z 1 dZ (n + 1/2)2
+ + 1− Z = 0. (2.2.44)
dρ 2 ρ dρ ρ2
68 Fundamentals of light scattering by ice crystals

The solution of Eq. (2.2.44) can be expressed as the general cylindrical function of order
n + ½ in the form
Z = Zn+1/2 (ρ). (2.2.45)
It follows that the solution of Eq. (2.2.37) can be expressed as

R (kmr) = Zn+1/2 (kmr)/ kmr. (2.2.46)
Combining Eqs. (2.2.40), (2.2.42), and (2.2.46), the elementary wave functions at all points
on the surface of a sphere can now be expressed as

ψ(r, θ, φ) = Zn+1/2 (kmr)pn ( cos θ )(a cos φ + b sin φ)/ kmr. (2.2.47)
Each cylindrical function denoted in Eq. (2.2.46) can be expressed as a linear combina-
tion of two cylindrical functions of standard type in terms of the Bessel functions Jn+½ (ρ)
and Nn+½ (ρ). We may define two standard functions as follows:
 
ψn (ρ) = πρ/2Jn+1/2 (ρ), χn (ρ) = − πρ/2Nn+1/2 (ρ). (2.2.48)
The functions ψn are regular in every finite domain of the ρ plane, including the origin,
whereas the functions χn cannot represent waves inside the sphere. On the basis of the
definitions in Eq. (2.2.48), Eq. (2.2.46) can be rewritten in the form
rR (kmr) = cn ψn (kmr) + dn χn (kmr), (2.2.49)
where cn and dn are arbitrary constants. Equation (2.2.49) represents the general solution
of Eq. (2.2.37). It follows that the general solution of the scalar wave equation (2.2.21) can
be expressed as
∞ 
 n
rψ(r, θ, φ) = pn (cos θ )[cn ψn (kmr) + dn χn (kmr)](a cos φ + b sin φ).
n=0 =−n
(2.2.50)
Subsequently, the electric and magnetic field vectors of electromagnetic waves can be
derived from Eqs. (2.2.26) and (2.2.27). In the case when cn = 1 and dn = i, we have

ψn (ρ) + iχn (ρ) = πρ/2H (2)1 (ρ) = ξn (ρ), (2.2.51)
n+ /2
(2)
where Hn+ 1/2 is the half-integral-order Hankel function of the second kind. It has the
property of vanishing at infinity in the complex plane, suitable for the representation of
scattered waves.
The complete solution for light scattering of a plane wave by a homogeneous sphere
requires the coordinate transformation of electric vectors, matching the incident, scattered,
and transmitted waves at spherical boundaries, and mapping of the solution to the far field,
leading to the well-known Lorenz–Mie theory.
2.2.3.2 Cylindrical coordinates
The exact solution for the scattering by an infinite homogeneous dielectric circular cylinder
for normal incidence – that is, a light beam perpendicular to the axis of the cylinder – was
2.2 Maxwell’s equations and boundary conditions 69

az
z

af
P(r, f, z)

ar
z

y
x φ r
y

x
Figure 2.3 Definition of the parameters in cylindrical coordinates, where x = r cos φ, y = r sin φ,
z = z; and ar , aφ , and az are orthogonal vectors with P a point on the cylinder.

derived by Rayleigh (1918). The solution for arbitrary oblique incidence was solved by
Wait (1955). Subsequent numerical investigations concerning the scattering of light beams
by infinite circular cylinders have been carried out by van de Hulst (1957), Greenberg et al.
(1967), Kerker (1969), and Liou (1972a). Liou (1972b) used circular cylinders to model
light scattering by randomly oriented ice particles in the atmosphere.
In the cylindrical coordinate system shown in Figure 2.3, Mψ defined in Eq. (2.2.22)
can be written in the form
1 ∂ψ ∂ψ
Mψ = ∇ × (az ψ) = ar − aφ . (2.2.52)
r ∂φ ∂r
Moreover, using Eq. (2.2.23), we have

mkNψ = ∇ × Mψ

∂ ∂ψ ∂ 1 ∂ψ
=− − ar + aφ (2.2.53)
∂z ∂r ∂z r ∂φ


1 ∂ ∂ψ 1 ∂ 1 ∂ψ
+ r − − az .
r ∂r ∂r r ∂φ r ∂φ
It follows that


1 ∂ 2ψ 1 ∂ 2ψ 1 ∂ ∂ψ 1 ∂ 2ψ
Nψ = ar + aφ − r 2 + 2 az . (2.2.54)
mk ∂z∂r r ∂z∂φ r ∂r ∂ r r ∂φ 2
70 Fundamentals of light scattering by ice crystals

On the basis of Eqs. (2.2.24), (2.2.25), (2.2.52), and (2.2.54), the electric and magnetic
vectors can be expressed in the forms

1 ∂v i ∂ 2u
E= + ar
r ∂φ mk ∂z∂r


∂v i ∂ 2u i 1 ∂ ∂u 1 ∂ 2u (2.2.55)
+ − + aφ − r + 2 2 az ,
∂r mkr ∂z∂φ mk r ∂r ∂r r ∂φ

m ∂u i ∂ 2v
H= − + ar
r ∂φ k ∂z∂r


∂u i ∂ 2v i 1 ∂ ∂v 1 ∂ 2v
+ m + aφ − r + 2 2 az . (2.2.56)
∂r kr ∂z∂φ k r ∂r ∂r r ∂φ
The scalar wave equation in cylindrical coordinates is given by

1 ∂ ∂ψ 1 ∂ 2ψ ∂ 2ψ
r + 2 2
+ + m2 k 2 ψ = 0. (2.2.57)
r ∂r ∂r r ∂φ ∂z2
This equation is separable by letting

ψ(r, φ, z) = R(r)(φ)Z(z). (2.2.58)

On substituting Eq. (2.2.58) into Eq. (2.2.57) and dividing the entire equation by ψ(r, φ,
z), we obtain


1 d dR(r) 1 d 2 (φ) 1 d 2 Z(z)
r + 2 + + m2 k 2 = 0. (2.2.59)
rR(r) dr dr r (φ) dφ 2 Z(z) dz2
Since the first two terms in Eq. (2.2.59) contain variables r and φ, the only possible solution
of this equation is
1 d 2 Z(z)
+ m2 k 2 = c1 , (2.2.60a)
Z(z) dz2
where c1 is a certain constant. Rearranging this equation, we have
d 2 Z(z)
+ (m2 k 2 − c1 )Z(z) = 0, (2.2.60b)
dz2
where the term m2 k2 − c1 > 0. The solution of Eq. (2.2.60b) is an exponential function:

Z(z) = e−ihz , (2.2.61)

where we have set the arbitrary constant h = (m2 k2 − c1 )1/2 and used a negative sign in the
exponential function.
In view of Eqs. (2.2.59) and (2.2.60a), we must also have the following equation:


1 d dR(r) 1 d 2 (φ)
r + 2 + c1 = 0. (2.2.62)
rR(r) dr dr r (φ) dφ 2
2.2 Maxwell’s equations and boundary conditions 71

On multiplying by r2 throughout this equation, we get




r d dR(r) 1 d 2 (φ)
r + c1 r 2 + = 0. (2.2.63)
R(r) dr dr (φ) dφ 2
This equation contains two independent terms, so that we must have
1 d 2 (φ)
= −c2 , (2.2.64)
(φ) dφ 2
where c2 is a certain constant. The solution of this differential equation can be expressed
in terms of an exponential function in the form

(φ) = e±i c2 φ
. (2.2.65)

We may set c2 equal to an integer n such that the solution is a periodic function in φ with
a period of 2π in the form
(φ) = einφ , n = 0, ±1, ±2, . . . (2.2.66)
On the basis of Eqs. (2.2.63) and (2.2.64) and the definition of c2 = n2 , the first term in
Eq. (2.2.63) can be written as follows:


r d dR(r)
r + c1 r 2 − n2 = 0. (2.2.67a)
R(r) dr dr
After rearrangement, we have

d 2 R(r) 1 dR(r) n2
+ + c1 − 2 R(r) = 0. (2.2.67b)
dr 2 r dr r
Considering a positive c1 so that c1 = j2 and setting jr = x, Eq. (2.2.67b) can be written in
the form

d 2 R(x) 1 dR(x) n2
+ + 1 − R(x) = 0. (2.2.67c)
dx 2 x dx x2
The solution of Eq. (2.2.67c) is any Bessel function of order n, denoted by Zn (x), as
follows:
R(r) = Zn (j r), (2.2.68)
where j = (m2 k2 − h2 )12 . By combining Eqs. (2.2.61), (2.2.66), and (2.2.68), the general
solution of the scalar wave equation in cylindrical coordinates can be written as follows:
ψ(r, φ, z) = Zn (j r)einφ e−ihz . (2.2.69)
To find the complete solution for light scattering of a plane wave by an infinite circular
cylinder, we may follow the procedures outlined for the spherical case, including coordinate
transformation of the electric vectors; matching of the incident, scattered, and transmitted
waves at the cylindrical boundaries; and mapping of the solution to the far field. Separa-
tion of the TE and TM modes in the internal and scattered waves corresponding to each
72 Fundamentals of light scattering by ice crystals

η=
η=0

os π_
3
- co 3
ξ > ξ0

=c
s
π_

η
η (I)
= π_
-c ξ0 s 6
os _π co
6 η=
(II) ξ < ξ0
y z
z
F O F η =1
η = -1

1
se de
Ca Mo
TE
(i) H
(i) E
2
se ode
(i) H Ca M OF = OF = 
TM
(i) E

Figure 2.4 Coordinate system (η, ξ, φ) for the scattering by a prolate spheroid with a semifocal
distance . The z-axis is chosen as the axis of revolution. The incident plane contains the incident
direction and the z-axis. The x-axis is along the incident plane. For the transverse magnetic (TM)
mode, the E vector is at the incident plane, while for the transverse electric (TE) mode, the H vector is
along the incident plane. The incident angle ζ is the angle in the incident plane between the incident
direction and the z-axis (after Asano and Yamamoto 1975).

polarization mode of the incident wave is required to determine unknown coefficients in


the wave functions (see e.g., Liou 1972a).

2.2.3.3 Spheroidal coordinates


An exact scattering solution, similar to the Lorenz–Mie theory of light scattering by spheres
and the Rayleigh–Wait theory of light scattering by cylinders, can be derived for particles
with spheroidal shapes (Asano and Yamamoto 1975, Asano and Sato 1980). Prolate and
oblate spheroids may be used to approximate ice cylinders and plates, respectively. The
spheroidal coordinates can be obtained by the rotation of an ellipse about an axis of sym-
metry. Using this rotation, the spheroidal coordinate systems are defined by the following
parameters: η, the angular coordinate; ξ , the radial distance; and φ, the azimuthal angle.
Shown in Figure 2.4 are prolate spheroidal coordinates, which can be related to Cartesian
coordinates by the following transformation:

x = (1 − η2 )1/2 (ξ 2 − 1)1/2 cos φ,


y = (1 − η2 )1/2 (ξ 2 − 1)1/2 sin φ, (2.2.70)
z = ηξ,
2.2 Maxwell’s equations and boundary conditions 73

where  is the semifocal distance, −1 ≤ η ≤ 1, 1 ≤ ξ ≤ ∞, and 0 ≤ φ ≤ 2π. For the


oblate system, the ξ term in x and y should be replaced by (ξ 2 + 1)1/2 , with 0 ≤ ξ ≤ ∞.
The size and shape of an ellipse are defined by the semifocal distance and the eccentricity e.
1/2
For the prolate and oblate systems, e = 1/ξ0 and e = 1/(ξ02 + 1) , respectively, where ξ 0
is the value of ξ at the surface. If  = 0, the spheroidal coordinates reduce to the spherical
coordinates. In the far field, ξ → , η → cos θ , and ξ → r.
If ψ satisfies the scalar wave equation defined in Eq. (2.2.21), vectors associated with
electric and magnetic fields can be expressed in the spheroidal coordinate system such that
they satisfy the vector wave equations denoted by ψmn , and the solutions to the vector wave
equation can be obtained from the following two equations:
Mmn = ∇ × (a · ψmn ), (2.2.71)
mkNmn = ∇ × Mmn , (2.2.72)
where m (not the subscript index) is the complex refractive index and k = 2πλ, where
λ is the wavelength. The solution of the scalar wave equation can be expressed in terms
of scalar spheroidal wave functions. The vector a is an arbitrary constant unit vector
(a = e) or the position vector (a = r). The vectors defined in Eqs. (2.2.71) and (2.2.72)
are solenoidal, such that the field vectors E and H can be expressed in terms of these two
vectors in spheroidal coordinates.
The scalar wave equation (2.2.21) is separable in the spheroidal coordinate system in
terms of three second-order ordinary linear differential equations based on three variables:
η, ξ , and φ. The solutions of the equation in φ are eimφ , cos mφ, and sin mφ, where m =
0, 1, 2, . . .
The variable η satisfies the following equation:


d 2 dSmn (η) m2
(1 − η ) + λmn ∓ c η −
2 2
Smn (η) = 0, (2.2.73)
dη dη 1 − η2
where c =  · mk, m is the complex refractive index, k is the wavenumber, Smn is the
spheroidal angular function of order m and degree n of the first kind, which can be defined
in the range −1 ≤ η ≤ 1, and λmn is a separation constant, which is a function of c.
The eigenvalues λmn (c) and λmn (−ic), and the associated eigenfunctions Smn (c, η) and
Smn (−ic, η), correspond to the prolate and oblate coordinate systems, respectively. The
eigenfunctions can be expressed in terms of the associated Legendre functions of the first
kind as follows:


 mn
Smn (±c, η) = dr (±c)Pm+rm
(η), (2.2.74)
r=0,1

where we have defined c = +c and −c = −ic, and drmn (±c) are the expansion coefficients
related to the prolate (+c) and oblate (−c) systems. The prime on the summary symbol
signifies that the summation is over only even (odd) values when n − m is even (odd). It
follows that the spheroidal angle functions are dependent not only on the angular component
but also on the properties of the medium c. Note that, for spherical geometry, the angular
74 Fundamentals of light scattering by ice crystals

functions reduce to Pnm (η), independent of c [see Eq. (2.2.42)]. Based on the theory of
Sturm–Liouville differential equations, the spheroidal angle function has the following
orthogonal properties:
 1 
0 (n
= n ),
Smn (η)Smn (η)dη = (2.2.75)
−1 mn (n = n ),

where the normalization constant




 2(r + 2m)!  mn 2
mn = d . (2.2.76)
r=0,1
(2r + 2m + 1)r! r

The radical functions Rmn (ξ ) satisfy the following differential equation:





d dRmn (ξ ) m2
(ξ 2 ∓ 1) − λmn − c2 ξ 2 ± 2 Rmn (ξ ) = 0. (2.2.77)
dξ dξ (ξ ± 1)
(j )
This equation is normalized such that, for cξ → ∞, four asymptotic forms for Rmn (j = 1,
2, 3, 4) can be obtained.
The appropriate expansions in terms of spherical Bessel functions for the prolate (+c)
and oblate (−ic) systems can be expressed in the form
  
 (r + 2m)!

Rmn (±c, ξ ) = 1
(j ) mn
dr (±c)
r=0,1
r!
m/2 
ξ2 ∓ 1  (r+n−m) (r + 2m)! mn (j )
× i dr (±c) Zm+r (±cξ ) , (2.2.78)
ξ2 r=0,1
r!

where we have again replaced c by +c (prolate) and −ic (oblate) by −c, and ξ = iξ for
(j )
oblate, for simplicity of presentation. The notation Zn (cξ ) is the nth-order spherical Bessel
function (i = 1), Newmann (j = 2) and Hankel functions of the first kind (j = 3) and the
second kind (j = 4). Because of the properties of the spherical Bessel functions, Rmn (1)
is
used to represent the wave inside the spheroid and the incident wave. At large distances
(3)
from the spheroid, Rmn is suitable for representing the scattered wave.
On the basis of the preceding solutions for the three independent variables representing
spheroidal coordinates, complete solutions for the scalar wave can now be written as
follows. For the prolate system, we have

e cos mφ
(j )
ψmn (c, η, ξ, φ) = Smn (c, η)Rmn (c, ξ )
(j )
. (2.2.79)
o sin mφ

For the oblate system, the solutions can be obtained by replacing c by −ic and ξ by iξ .
The notations e and o refer to even and odd dependence on φ; i.e., cos mφ and sin mφ,
respectively. The superscript j will take the values 1 or 3 for the incident and internal waves
and for the scattered wave, respectively.
2.3 Optical properties of ice: index of refraction 75

Finally, the field vectors E and H can now be expressed in terms of the vector spheroidal
wave functions based on Eqs. (2.2.71) and (2.2.72). These functions can be written as

r e r e
Mmn and Nmn , (2.2.80)
o o
which can be obtained from the scalar functions in Eq. (2.2.79) for the two types of spheroids
(prolate and oblate) by setting a = r as the position vector. The components of spheroidal
vector wave functions can be found in Flammer (1957). The unknown coefficients in the
wave functions must be determined from boundary conditions in which it is necessary to
separate TE from TM modes in the internal and scattered waves corresponding to each
polarization mode of the incident wave (Asano and Yamamoto 1975).
Examples of the single-scattering properties of ice spheres, ice cylinders, and ice
spheroids will be presented after discussion of ice optical properties and numerous defini-
tions given in reference to the scattering phase matrix, extinction efficiency, single-scattering
albedo, and asymmetry factor.

2.3 Optical properties of ice: index of refraction


2.3.1 Complex index of refraction and dispersion of light
Electric dipoles are generated when positive and negative charges are impelled to move in
opposite directions by an applied electric field in a dielectric. The product of the charges
and the separation distance between the positive and negative charges is referred to as the
dipole moment, which when divided by unit volume is denoted as the polarization P. The
displacement vector D (charge per area) in Maxwell’s equations in a dielectric is defined
by
D = εE = E + 4πP. (2.3.1)
The permittivity of the medium is then given by
ε = 1 + 4πP · E/E 2 . (2.3.2)

In terms of ε and the permeability μ, the velocity of light can be written as c = 1/με.
The permeability μ of air or water is nearly equal to the permeability μ0 (= 1) of vacuum,
i.e., μ ≈ μ0 . Furthermore, the index of refraction is defined as the ratio of the velocity of
light c0 in vacuum to that in the medium:

m = c0 /c ≈ ε = (1 + 4πP · E/E 2 )1/2 . (2.3.3)
The polarization vector for N dipoles can be written in terms of the polarizability α as
follows:
P = N αE. (2.3.4)
Substituting Eq. (2.3.4) into Eq. (2.3.3) leads to
m2 = 1 + 4πN α. (2.3.5)
76 Fundamentals of light scattering by ice crystals

The polarization vector defined in Eq. (2.3.4) can also be defined in terms of the charge
e of the electron and a vector distance r in the form
P = N er. (2.3.6)
In view of Eqs. (2.3.4) and (2.3.6), we have the following relationship:
αE = er. (2.3.7)
Furthermore, the force generated by electric and magnetic fields is given by the following
Lorentz force equation:
F = e[E + (μ/c)v × H], (2.3.8)
where v denotes the velocity of an electron, which is very small compared to the velocity
of light. Hence, the force produced by the magnetic field may be neglected. The force on a
vibrating system in terms of displacement D = εE is produced by three actions: electron
acceleration; the damping force which carries away energy when vibrating electrons emit
electromagnetic waves proportional to electron velocity; and the restoring force of vibra-
tions, which is proportional to the distance r. From Newton’s second law, we can express
balance forces as follows:
F eE d 2r dr
= = +γ + ξ r, (2.3.9)
m∗e m∗e dt 2 dt
where m∗e is the mass of the electron and γ and ξ are the damping and restoring coefficients,
respectively. In scalar form, the balance equation can be expressed as
d 2r dr eE
2
+γ + ξr = ∗ . (2.3.10)
dt dt me
The homogeneous solution of this second-order differential equation is given by
r = r0 eiωt = r0 ei2πν̃t , (2.3.11)
where r0 denotes the value when r → 0, ω is the circular frequency, and ν̃ is the frequency.
Substituting Eq. (2.3.11) into Eq. (2.3.10), we obtain
[(ξ − 4π2 ν̃ 2 ) + i2πν̃γ ]r = eE/m∗e . (2.3.12)

The natural (or resonant) frequency is defined as ν̃0 = ξ /2π. Thus, we have
er e2 1
α= = ∗ 2 2 
E me 4π ν̃0 − ν̃ 2 + i2πγ ν̃
  (2.3.13)
e2 ν̃02 − ν̃ 2 i γ ν̃
= ∗   −   .
me 4π2 ν̃ 2 − ν̃ 2 2 + γ 2 ν̃ 2 2π 4π2 ν̃ 2 − ν̃ 2 2 + γ 2 ν̃ 2
0 0

Let the real and imaginary parts of the index of refraction (or refractive index) be mr and
mi , respectively, so that the complex index of refraction is defined in the form
m = mr − imi , (2.3.14)
2.3 Optical properties of ice: index of refraction 77

where mr and mi are positive real numbers. From Eqs. (2.3.5) and (2.3.13), we can then
obtain the following equations:
4πN e2 ν̃02 − ν̃ 2
m2r − m2i = 1 + ∗  2  , (2.3.15)
me 4π2 ν̃ − ν̃ 2 2 + γ 2 ν̃ 2
0

2N e2 γ ν̃
2mr mi =   . (2.3.16)
m∗e 4π2 ν̃ 2 − ν̃ 2 2 + γ 2 ν̃ 2
0

For air, the real part mr ≈ 1 and the imaginary part mi  (mr − 1). Additionally, in the
neighborhood of the resonant frequency, ν̃ ∼ = ν̃0 , so that (ν̃ 2 − ν̃02 ) = (ν̃ + ν̃0 ) · (ν̃ − ν̃0 ) ∼
=
2ν̃0 (ν̃ − ν̃0 ). The half-width of the natural broadening is dependent on the damping and is
given in the form αN = γ /4π, while the line strength S = πN e2 /(m∗e c). It follows that the
real part can be derived and is given in the form
N e2 ν̃ − ν̃0
mr − 1 = − ∗
. (2.3.17)
4πme ν̃0 (ν̃ − ν̃0 )2 + αN2
The absorption coefficient kν̃ is directly related to the imaginary part mi and can be written
as follows:
4πν̃0 mi S αN
kν̃ = = . (2.3.18)
c π (ν̃ − ν̃0 )2 + αN2
Figure 2.5 illustrates (mr − 1) and kν̃ as functions of frequency. The former increases
as the frequency increases when (ν̃0 − αn ) > ν̃. This mode is referred to as normal disper-
sion, under which light is dispersed by a prism into its component colors. For the region
(ν̃0 + αN ) > ν̃ > (ν̃0 − αN ), (mr − 1) decreases with increasing frequency, referred to as
anomalous dispersion. For the range ν̃ > (ν̃0 + αN ), normal dispersion takes place again;
however, (mr − 1) is smaller than unity. The absorption coefficient with the half-width αN
is referred to as the normal or Lorentz profile (Lorentz 1906).

2.3.2 Optical properties of ice


Theoretical computations of the radiative properties of ice clouds require knowledge of
laboratory measurements of the complex refractive index m(ν) = mr (ν) − imi (ν), where ν
is the wavenumber in a vacuum (ν = ν̃0 /c). In terms of wavelength, we have λ = 1/ν. The
real part of the refractive index determines the phase speed of the electromagnetic wave,
while the imaginary part of the refractive index is related to the absorption coefficient kν
defined in Eq. (2.3.18) in the wavenumber domain.
As shown in Eqs. (2.3.2) and (2.3.3), permittivity ε is related to polarization P. In
general, polarization refers to the variation of the electric vector. For example, we refer to
linear polarization to imply that the electric vector vibrates in a line defined in space (see
Subsection 2.4.1). In a related issue, we shall define the velocity of light in a medium. The
velocity of light in an isotropic material is the same in all directions. However, crystals are
78 Fundamentals of light scattering by ice crystals

mr -1
aN kν

-6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6
(ν - ν0)/aN

Figure 2.5 Real and imaginary parts of the complex index of refraction as a function of the nor-
malized frequency, where mr is the real refractive index, α N is the half-width, kν̃ is the absorption
coefficient, and (ν̃ − ν̃0 ) is the frequency deviation from the reference frequency ν̃0 .

generally optically anisotropic and a light beam is broken up into two waves which travel
with different velocities through the crystal. One of the two waves, called the ordinary
wave, travels with the same velocity in all directions in the crystal; therefore, its vector
surface is a sphere.
The velocity of the other wave, called the extraordinary wave, varies with the direction
of propagation through the crystal and its vector surface is an ellipsoid of revolution. For
optically uniaxial crystals, the spherical vector surface coincides with the ellipsoidal vector
surface at only two points, and these are at the ends of the axis of revolution of the ellipsoid,
as shown in Figure 2.6. In this case the velocity of the ordinary wave is equal to that of the
extraordinary wave when light travels along the axis of revolution of the ellipsoid, which
is the optical axis of the crystal. If the extraordinary wave is slow relative to the ordinary
wave, the crystal is said to be optically positive (Figure 2.6a). If the reverse is true, the
crystal is said to be optically negative (Figure 2.6b). The ordinary and extraordinary waves
are polarized at right angles to each other. The electric vector of the ordinary wave always
vibrates perpendicular to the optical axis, while that of the extraordinary wave is in the plane
defined by the propagation direction and the optical axis. This phenomenon is referred to
as birefringence.
Ice is a uniaxial, doubly refracting, and optically positive crystal. Its optical axis coincides
with the crystal’s c-axis. Two principal indices of refraction are defined for uniaxial crystals.
One is based on the velocity of the extraordinary wave in a direction normal to the wave
2.3 Optical properties of ice: index of refraction 79

(a) Optic axis (b) Optic axis

Ordinary
wave

Extraordinary
wave

Figure 2.6 Spherical ellipsoidal vector surfaces for an optically uniaxial crystal; the difference
between the surfaces is exaggerated in the two diagrams: (a) denotes the positive crystal, while (b)
denotes the negative crystal.

front when the wave travels perpendicular to the optical axis, and the other is the refractive
index for the ordinary wave. Thus, for a positive uniaxial crystal, the principal index of
refraction me for the extraordinary wave is defined as
me = c/v1 , (2.3.19)
where v1 is the minimum normal velocity of the extraordinary wave. The index of refraction
mo for the ordinary wave is given by
mo = c/v2 , (2.3.20)
where v2 is the velocity of the ordinary wave. The indices of refraction me and mo of ice
differ only in the third decimal place at a temperature of −3°C for visible wavelengths
(Merwin 1930, Hobbs 1974).
According to Born and Wolf (1975), the effective index of refraction meff for extraordi-
nary waves can be obtained from
1 cos2 κ sin2 κ
2
= 2
+ , (2.3.21)
meff mo m2e
where κ is the angle between the c-axis of the crystal and the propagation direction of
the waves. As shown in Takano and Liou (1989a), the effect of ice birefringence on the
single-scattering properties of hexagonal ice crystals is negligible. Thus, for all practical
purposes it suffices to use the average refractive index in atmospheric scattering, absorption,
and polarization calculations.

2.3.3 Indices of refraction for ice


The determination of mr and mi for ice begins with transmission measurements through
thin ice films in a laboratory setting. On the basis of Beer’s law, the transmission of light
through a medium for a given wavenumber ν, neglecting multiple scattering, can be written
80 Fundamentals of light scattering by ice crystals

as follows:
Iν (s)
Tν = = exp(−kν · ), (2.3.22)
Iν (0)
where Iν (0) is the incident intensity, Iν (s) is the emergent intensity at a position s,  is
the thickness of a thin ice film, and kν , the absorption coefficient in units of (length)−1 , is
equal to 4πνmi , as noted previously. Experiments to measure the absorption coefficient,
related to the imaginary refractive index, employed geometry with normal incidence, i.e.,
perpendicular to the thin film. In Eq. (2.3.22), we have neglected the reflection component,
which is small in normal cases. The absorption coefficient so defined includes the effects of
both the true absorption of energy and the scattering of energy. Complications caused by film
thickness, the required absorbing substrate, and other pertinent parameters have been noted
by Toon et al. (1994). Spectral measurements are conducted to cover a range of wavenumber.
An initial estimate of optical constraints is made to determine film thicknesses, and the best
value of imaginary index is obtained based on measured transmissions by minimizing the
mean square error between measured and calculated transmission for a number of films of
varying thickness.
The real refractive index can be obtained from the Kramers–Kronig relationship
(Kramers 1927, Kronig 1926) in the forms
 ∞ 
2 ν Im[ε(ν  )] 
Re[ε(ν)] = 1 + P dν , (2.3.23a)
π 0 ν2 − ν2
 ∞
2ν Re[ε(ν  )] − 1 
Im[ε(ν)] = − P dν , (2.3.23b)
π 0 ν2 − ν2
where P implies that the principal value of the integral is taken; Re and Im represent the
real and imaginary parts, respectively; and the integration covers all wavenumbers.
The real and imaginary parts of the refractive index for ice have been comprehensively
reviewed and tabulated by Irvine and Pollack (1968), based on the available laboratory mea-
surements of reflection and transmission. Bertie et al. (1969) have measured the absorptance
and reflectance of a film of ice in the range of 8000 to 30 cm−1 at a temperature of −173°C
and have derived real and imaginary indices of refraction. Schaaf and Williams (1973)
have performed similar measurements for ice at −7°C in the wavenumber range of 5000
to 300 cm−1 . Seki et al. (1981) have measured the reflectance of a single crystal of hexag-
onal ice at a temperature of 80 K in the 0.044 to 0.207 µm ultraviolet wavelength range.
A review of the ice refractive index has been given by Warren (1984). Toon et al. (1994)
determined the infrared optical constants of H2 O based on the transmission of infrared light
through thin films of varying thickness over a frequency range from 7000 to 500 cm−1 at
temperatures below 200 K. Additional measurements were reported by Kou et al. (1993),
Clapp et al. (1995), Gosse et al. (1995), Grundy and Schmitt (1998), Zhang et al. (2001),
Rajaram et al. (2001), Curtis et al. (2005), and Mätzler (2006). Several of these authors
performed measurements at temperatures lower than 210 K. Warren and Brandt (2008) have
updated the values reported in Warren (1984). A summary of these measurements in terms
of spectral and temperature ranges is depicted in Figure 2.7.
300

250
Temperature (K)

200

150

100
10-2 10-1 100 101 102 103 104 105 106 107

Wavelength (μm)

Figure 2.7 Available datasets for the imaginary part of the refractive index of ice as a function
of spectral coverage and temperature range, obtained by Warren and Brandt (2008), Gosse et al.
(1995), Toon et al. (1994), Grundy and Schmitt (1998), Zhang et al. (2001), Rajaram et al. (2001),
Clapp et al. (1995), Curtis et al. (2005), and Mätzler (2006) (after Iwabuchi and Yang 2011). Several
measurements were made at temperatures much lower than the melting temperature.

Warren & Brandt (2008)


Warren (1984)

10 0

10 −2

10 −4

10 −6

10 −8

0 1 2 3 4 5

Figure 2.8 Right panel: Real and imaginary indices of refraction for ice as a function of wavelength
from 0.2 to 100 µm. The insert is the imaginary refractive index from 0.2 to 5 µm covering the solar
spectrum. These datasets were compiled by Warren (1984) and updated by Warren and Brandt (2008).
Left panel: all data points from these two sources, presented in the domain of real and imaginary
indices of refraction.
82 Fundamentals of light scattering by ice crystals

Real Index of Refraction (a)

10-1 100 101 102 103 104 105 106


Imaginary Index of Refraction

(b)

10-1 100 101 102 103 104 105 106


Wavelength (μm)

105 104 103 102 101 100 10-1 10-2


Wavenumber (cm-1)
Figure 2.9 (a) Real and (b) imaginary parts of the refractive index of ice for a temperature range
from 160 to 270 K as a function of wavelength to 106 µm (after Iwabuchi and Yang 2011). Also shown
is the corresponding wavenumber.

Spectral ice indices of refraction are fundamental parameters that determine the relative
scattering and absorption properties of ice particles due to a single-scattering event for
a given wavelength and ice crystal size. If there is no absorption, real-part values are
responsible for the scattering processes. When absorption is involved, the amount of energy
scattered and absorbed depends on both real and imaginary parts, as well as on wavelength
2.4 Definition of scattering properties for ice crystals 83

and on particle size, shape, and composition. In order to compute solar and thermal infrared
radiative fluxes that are absorbed, reflected, or transmitted through ice clouds, accurate
values of the real and imaginary indices of refraction are required. They are equally
important in the development of remote sensing techniques to infer the size and shape
of water and ice particles and cloud optical depth.
Figure 2.8 illustrates ice refractive index datasets as a function of wavelength from 0.2
to 100 µm, as compiled by Warren (1984) and Warren and Brandt (2008). These datasets
have been widely used for calculations of the scattering and absorption properties of
ice crystals. A logarithmic scale is used for the lower right panel such that differences
between the two compilations are not clearly shown, except in the ultraviolet and visible
regions. The left panel displays the variation of the imaginary part versus the real part on a
uniform scale. It is clear that the two compilations show large differences in the microwave
region.
Figure 2.9 illustrates the real and imaginary parts of the refractive index of ice for
a range of temperature from 160 to 270 K in the wavelength domain from 0.05 µm to
100 cm. Also shown is the equivalent wavenumber domain. The dependence of the real
part on temperature appears to be small in all wavelength ranges. However, a significant
dependence of the imaginary part on temperature is shown for wavelengths longer than
about 20 µm, particularly in microwave regions.

2.4 Definition of the single-scattering and polarization properties of ice crystals


2.4.1 Representation of a simple wave
By means of scattering processes, ice crystals in the path of electromagnetic waves con-
tinuously abstract energy from the incident waves and reradiate that same energy in all
directions. Thus, we shall first describe the representation of electromagnetic waves. An
electromagnetic wave is characterized by electric and magnetic vectors E and H, respec-
tively, which form an orthogonal set with the direction of propagation of the wave. In any
medium, E and H are related, and it is customary to use E in scattering discussions. We
say light is polarized in a certain direction when the vibration of the electric vector E
concentrates in that direction. It follows that the direction of polarization is defined as the
direction of the electric vector.
The flow of energy and the direction of wave propagation are represented by the Poynting
vector S, as depicted in Figure 2.10. This vector is given by (Born and Wolf 1975, in
Gaussian units)
c
S= E × H, (2.4.1)

where |S| has units of flux density. The electric field vector E can be decomposed into
two components, El and Er , representing the electric vectors parallel (l) and perpendicular
(r) to a plane through the direction of propagation. The plane so defined is called the
84 Fundamentals of light scattering by ice crystals

E S

El
E

Er H
Figure 2.10 Propagation of an electromagnetic wave in terms of electric and magnetic vectors and
the definition of the Poynting vector S. An electric vector E in space can be arbitrarily decomposed
into two orthogonal components, referred to as parallel (l) and perpendicular (r) to a plane through
the direction of propagation.

plane of reference, and its selection is generally arbitrary. In practice, we select the plane
containing the incident and scattered beams as the common plane of reference for the two
beams.
Consider an electromagnetic wave that propagates in the z direction with a propagation
constant k (= 2πλ), where λ is the wavelength, and a circular frequency ω (= kc),
and with positive amplitudes and phases (al , ar ) and (δl , δr ) for the electric field of an
electromagnetic wave in the l(El ) and r(Er ) directions, respectively. Thus, we have

El = al e−iδl e−ikz+iωt , Er = ar e−iδr e−ikz+iωt , (2.4.2)

where El and Er are complex, oscillating functions. We may set ζ = kz − ωt and take the
cosine representation for the case when the plane wave is time harmonic, so that

El = al cos(ζ + δl ), Er = ar cos(ζ + δr ). (2.4.3)

It follows that

El /al = cos ζ cos δl − sin ζ sin δl ,


(2.4.4)
Er /ar = cos ζ cos δr − sin ζ sin δr .

Multiplying the first and second equations by (sin δr , cos δr ) and (sin δl , cos δl ), respec-
tively, and subtracting one from the other, we obtain

(El /al ) sin δr − (Er /ar ) sin δl = cos ζ sin(δr − δl ),


(2.4.5)
(El /al ) cos δr − (Er /ar ) cos δl = sin ζ sin(δr − δl ).
2.4 Definition of scattering properties for ice crystals 85

r
(al cosδ,ar)

(al,ar cosδ )

2ar l

(-al,-ar cosδ)

(-al cosδ ,-ar)


2al

Figure 2.11 Geometric representation of an elliptically polarized electromagnetic wave. The term δ
is the phase difference and al and ar are the amplitudes of the electric field in the l and r directions,
respectively. Intercepts with the rectangle are indicated by crosses.

By squaring and then adding the above two equations, we obtain


(El /al )2 + (Er /ar )2 − 2(El /al )(Er /ar ) cos δ = sin2 δ, (2.4.6)
where the phase difference δ = δr − δl .
Equation (2.4.6) represents the equation of a conic, the associated determinant of which
is given by
 
 1/al2 −cos δ/(al ar )  sin2 δ

  = 2 2 ≥ 0. (2.4.7)
 −cos δ/(al ar ) 1/ar2  al ar
Thus, the conic equation in (2.4.6) represents an ellipse. For this reason, an electromagnetic
wave is referred to as an elliptically polarized wave, as illustrated in Figure 2.11. The
ellipse is inscribed in a rectangle whose sides are parallel to the coordinate axes and have
lengths of 2al and 2ar . The ellipse touches the sides at the points (±al , ±ar cos δ) and
(±al cos δ, ±ar ).
Two special cases are of particular importance. If δ = mπ (m = 0, ±1, ± 2, . . . ), then
Eq. (2.4.6) becomes

El Er 2 El Er
± = 0, i.e., =∓ . (2.4.8)
al ar al ar
This equation describes two perpendicular lines. The wave for this case is said to be linearly
polarized. On the other hand, if δ = mπ/2 (m = ±1, ±3, . . .) and al = ar = a, we have
El2 + Er2 = a 2 . (2.4.9)
This equation describes a circle, and we call the wave in this case circularly polarized.
The polarization is called right-handed when sin δ > 0, and left-handed when sin δ < 0.
Right-handed and left-handed refer to the direction of rotation (direction of fingers) when
86 Fundamentals of light scattering by ice crystals

δ=0 sin δ < 0

δ=π sin δ > 0

Figure 2.12 Representation of linear (left) and circular (right) polarization on a plane, where δ is
the phase difference between El and Er , and left- and right-hand circular polarization are defined
according to the sign of sin δ.

the thumb is pointed in the direction of propagation. Geometrical representations of linear


and circular polarization are shown in Figure 2.12.
In Subsection 2.4.2, we introduce the Stokes vector to define the complete polarization
properties of an electromagnetic wave.

2.4.2 Representation of electromagnetic waves by Stokes parameters


To define the complete polarization properties of electromagnetic waves, four param-
eters are required: El , Er , δ l , and δ r . In reference to intensity, a set of four quanti-
ties referred to as the Stokes parameters, first introduced by Stokes (1852), has been
defined. Because intensity is proportional to the absolute square of the electric field, we
may define the following four parameters, neglecting a constant of proportionality, as
follows:
I = El El∗ + Er Er∗ , (2.4.10a)

Q = El El∗ − Er Er∗ , (2.4.10b)

U = El Er∗ + Er El∗ , (2.4.10c)

V = −i(El Er∗ − Er El∗ ), (2.4.10d)


where an asterisk denotes the complex conjugate value. The parameters I, Q, U, and V are
real quantities that satisfy I 2 = Q2 + U 2 + V 2 . Furthermore, based on the two electrical
2.4 Definition of scattering properties for ice crystals 87

y
x

χ
b
O
l

Figure 2.13 Geometric representation of elliptical polarization of a light beam in which the direction
of propagation is into the paper, a and b are the lengths of the semi-major and semi-minor axes,
respectively; χ is the orientation angle between the Ol and Ox axes; and β is the ellipticity angle
whose tangent is the ratio of the ellipse traced by the endpoint of the electric vector, i.e., tan β = ± ba,
where + and − stand for right- and left-handed polarization, respectively. See also Figure 2.11.

fields defined in Eq. (2.4.2), we have


I = al2 + ar2 , (2.4.11a)
Q = al2 − ar2 , (2.4.11b)
U = 2al ar cos δ, (2.4.11c)
V = 2al ar sin δ. (2.4.11d)
The Stokes parameters can be expressed in terms of the geometry defining an ellipse.
Let β denote an angle whose tangent is the ratio of the axes of the ellipse traced by the
endpoint of the electric vector, as displayed in Figure 2.13. If the semi-major and semi-
minor axes of the ellipse are given by a and b, respectively, then tan β = ±b/a. Also, let χ
be the orientation angle between the major axis of the ellipse and the l direction. When the
plane waves are time harmonics, we may express the electric field vectors along the l and
r directions in terms of amplitude and phase using the cosine representation, as shown in
Eq. (2.4.3).
Let x and y denote the directions along the major and minor axes, respectively, Then the
electric fields in the x-y plane may be written




Ex cos χ sin χ El
= , (2.4.12a)
Ey −sin χ cos χ Er
88 Fundamentals of light scattering by ice crystals

where Ex and Ey may also be expressed in terms of amplitudes (a, b) and an arbitrary
phase δ0 using cosine and sine representations such that they satisfy the elliptical equation
in the forms
Ex = a cos(ξ + δ0 ), (2.4.12b)
Ey = ±b sin(ξ + δ0 ). (2.4.12c)
We may define an angle β, referred to as the ellipticity angle, as follows:
a = a0 cos β, ∓b = a0 sin β. (2.4.13a)
Equations (2.4.12b) and (2.4.12c) can then be written in the forms
Ex = a0 cos β cos(ξ + δ0 ), (2.4.13b)
Ey = −a0 sin β sin(ξ + δ0 ), (2.4.13c)
where a02 is equivalent to I. Using Eqs. (2.4.12a), (2.4.13b), and (2.4.13c), we obtain




El cos χ −sin χ a0 cos β cos(ξ + δ0 )
=
Er sin χ cos χ −a0 sin β sin(ξ + δ0 )

(2.4.14)
cos β cos χ cos(ξ + δ0 ) + sin β sin χ sin(ξ + δ0 )
= a0 .
cos β sin χ cos(ξ + δ0 ) − sin β cos χ sin(ξ + δ0 )
The two components in Eq. (2.4.14) can be expressed as
El = a0 [cos β cos χ (cos ξ cos δ0 − sin ξ sin δ0 )
+ sin β sin χ (sin ξ cos δ0 + cos ξ sin δ0 )]
(2.4.15a)
= a0 [(cos β cos χ cos δ0 + sin β sin χ sin δ0 ) cos ξ
+ (−cos β cos χ sin δ0 + sin β sin χ cos δ0 ) sin ξ ],
Er = a0 [cos β sin χ (cos ξ cos δ0 − sin ξ sin δ0 )
− sin β cos χ (sin ξ cos δ0 + cos ξ sin δ0 )]
(2.4.15b)
= a0 [(cos β sin χ cos δ0 − sin β cos χ sin δ0 ) cos ξ
− (cos β sin χ sin δ0 + sin β cos χ cos δ0 ) sin ξ ].
From Eq. (2.4.3), we have
El = al (cos ξ cos δl − sin ξ sin δl ), (2.4.16a)
Er = ar (cos ξ cos δr − sin ξ sin δr ). (2.4.16b)
Comparing Eqs. (2.4.15a) and (2.4.15b) with Eqs. (2.4.16a) and (2.4.16b), the following
relationships emerge:
a0 (cos β cos χ cos δ0 + sin β sin χ sin δ0 ) = al cos δl , (2.4.17a)
a0 (−cos β cos χ sin δ0 + sin β sin χ cos δ0 ) = −al sin δl , (2.4.17b)
a0 (cos β sin χ cos δ0 − sin β cos χ sin δ0 ) = ar cos δr , (2.4.17c)
−a0 (cos β sin χ sin δ0 + sin β cos χ cos δ0 ) = −ar sin δr . (2.4.17d)
2.4 Definition of scattering properties for ice crystals 89

Eliminating δl by performing square operations leads to

al2 = a02 (cos2 β cos2 χ + sin2 β sin2 χ ). (2.4.18a)

Similarly, eliminating δr yields

ar2 = a02 (cos2 β sin2 χ + sin2 β cos2 χ ). (2.4.18b)

It follows from Eqs. (2.4.18a) and (2.4.18b) that

I = a02 = al2 + ar2 = Il + Ir , (2.4.19a)

Q = Il − Ir = a02 [cos2 β(cos2 χ − sin2 χ ) + sin2 β(sin2 χ − cos2 χ )] = I cos 2β cos 2χ .


(2.4.19b)

Moreover, substituting Eqs. (2.4.17a)–(2.4.17d) into the expression for U = 2al ar cos δ,
we derive the following equations:

U = 2al ar (cos δl cos δr + sin δl sin δr )


= 2a02 [(cos β cos χ cos δ0 + sin β sin χ sin δ0 )(cos β sin χ cos δ0
− sin β cos χ sin δ0 ) + (cos β cos χ sin δ0 − sin β sin χ cos δ0 )
× (cos β sin χ sin δ0 + sin β cos χ cos δ0 )].
= 2a02 [cos2 β sin χ cos χ cos2 δ0 − sin β cos β cos2 χ sin δ0 cos δ0
+ sin β cos β sin2 χ sin δ0 cos δ0 − sin2 β sin χ cos χ sin2 δ0
+ cos2 β sin χ cos χ sin2 δ0 + sin β cos β cos2 χ sin δ0 cos δ0 (2.4.20)
− sin β cos β sin χ sin δ0 cos δ0 − sin β sin χ cos χ cos δ0 ]
2 2 2

= 2a02 [cos2 β sin χ cos χ − sin2 β sin χ cos χ


+ sin β cos β(−cos2 χ sin δ0 cos δ0 + sin2 χ sin δ0 cos δ0
+ cos2 χ sin δ0 cos δ0 − sin2 χ sin δ0 sin δ0 )]
= 2a02 sin χ cos χ (cos2 β − sin2 β) = I sin 2χ cos 2β.

In a similar manner, substituting Eqs. (2.4.17a)–(2.4.17d) into the expression for the term
V = 2al ar sin δ yields
V = I sin 2β. (2.4.21)
In summary, the Stokes parameters can be expressed in terms of total intensity, ellipticity,
and orientation angles in the forms
I = Il + Ir , (2.4.22a)

Q = Il − Ir = I cos 2β cos 2χ , (2.4.22b)

U = I cos 2β sin 2χ , (2.4.22c)

V = I sin 2β. (2.4.22d)


Note that I and V are independent of the orientation angle χ .
90 Fundamentals of light scattering by ice crystals

V
I

2β U
Y
Q 2χ

Figure 2.14 Representation of Stokes parameters (I, Q, U, V), denoted by heavier lines, on a Poincaré
sphere. The angles β and χ are defined in Figure 2.13. When β is positive (negative), the polarization
is said to be right- (left-)handed.

Equations (2.4.22a)–(2.4.22d) may be represented in Cartesian coordinates on a sphere


referred to as the Poincaré sphere (Poincaré 1892), as illustrated in Figure 2.14. The radius
of the sphere is given by I, while the zenithal and azimuthal angles are defined by π/2 − 2β
and 2χ , respectively. Thus, Q, U, and V denote the lengths in the x, y, and z directions,
respectively. On this sphere, the northern and southern hemispheres represent right-handed
and left-handed elliptical polarization, respectively. The north and south poles denote right-
handed and left-handed circular polarization, respectively, and points in the equatorial plane
represent linear polarization. For a simple wave, we have I 2 = Q2 + U 2 + V 2 . Moreover,
from Eqs. (2.4.22a)–(2.4.22d), the plane of polarization and the ellipticity angle can be
expressed in terms of Stokes parameters as follows:

tan 2χ = U/Q, (2.4.23a)


1/2
sin 2β = V /(Q2 + U 2 + V 2 ) . (2.4.23b)

In the preceding discussion, we have assumed a constant amplitude and phase in repre-
senting the wave vibration using Eqs. (2.4.22a)–(2.4.22d). However, the actual light beam
consists of many simple waves in very rapid succession. Within a very short duration (e.g.,
seconds), millions of simple waves are collected by a detector. Consequently, measurable
intensities are associated with the superposition of many millions of simple waves with
independent phases. If the operator   denotes the time average over a time interval (t1 , t2 ),
2.4 Definition of scattering properties for ice crystals 91

then the Stokes parameters of the entire beam of light for this time interval may be expressed
as
   
I = al2 + ar2 = Il + Ir ,
   
Q = al2 − ar2 = Il − Ir ,
(2.4.24)
U = 2al ar cos δ,
V = 2al ar sin δ.

In this case, we can prove that I 2 ≥ Q2 + U 2 + V 2 . A light beam is generally charac-


terized by partial elliptical polarization. If the phase differences between the two electric
components are 0° or an integer multiple of 180° (i.e., β = 0), the light beam is linearly
polarized (V = 0). If, on the other hand, the amplitudes of the two electric components are
the same and their phase differences are an odd-integer multiple of 90°, then the light beam
is circularly polarized (Q = U = 0). When the ellipticity angle β is positive or negative,
the circular polarization is said to be right- or left-handed. These cases have been discussed
previously using electric vectors.
Additionally, we may represent the Stokes parameters in terms of detectable variables.
Consider a retardation ε in the r direction with respect to the l direction and let the component
of the electric field vector be in a direction making an angle  with the positive l direction.
For a simple wave at time t, the electric field can then be represented in the form

E(t; ; ε) = El cos  + Er e−iε sin 


(2.4.25)
= al cos e−iξ + ar e−i(δ+ε)−iξ sin .

The expressions for El and Er have been defined in Eq. (2.4.2). For a time interval (t1 , t2 ),
the time average intensity is then given by

I (, ε) = E(t; ; ε)E ∗ (t; ; ε)


   
= al2 cos2  + ar2 sin2  + 12 2al ar cos δ sin 2 cos ε (2.4.26a)
− 1
2
2al ar sin δ sin 2 sin ε,

where ∗ denotes the complex conjugate value; see Eq. (2.4.10a). Using Eq. (2.4.24) and
noting that Il cos2  + Ir sin2  = (I + Q cos 2)/2, we obtain

I (, ε) = 12 [I + Q cos 2 + (U cos ε − V sin ε) sin 2]. (2.4.26b)

On the basis of Eq. (2.4.26b), the Stokes parameter can now be expressed in terms of
retardation and polarization angle in the form (Liou 1975a)

I = I (0, 0) + I (π/2, 0),


Q = I (0, 0) − I (π/2, 0),
(2.4.27a)
U = I (π/4, 0) − I (3π/4, 0)
V = − [I (π/4, π/2) − I (3π/4, π/2)] .
92 Fundamentals of light scattering by ice crystals

Detector

Q
Light
Source
Θ
I0 P M S

Figure 2.15 A schematic diagram of the optical setup in a laboratory setting for the measurement
of F() in which the symbols P, M, Q, and A denote polarizer, modulator, quarter-wave plate, and
analyzer, respectively; and S denotes the center of scattering.

Consequently, the Stokes parameters of a light beam can be represented by a combination


of several polarizers oriented at the positions 0, π4, π2, and 3π/4 and a compensator
(e.g., a quarter-wave (π/2) plate), which can detect circular polarization defined in the
element V.
In conjunction with the experimental setup, several optical components can be used,
including linear polarizers, quarter-wave plates, and an electro-optic modulator, which all
affect the state of polarization of a light beam as it crosses the component. In mathematical
terms, this can be described as obtaining the Stokes vector of the transmitted beam by
multiplying the Stokes vector of the incident beam by a 4 × 4 transformation matrix, referred
to as the Mueller matrix (Shurcliff 1962). The Mueller matrix of an optical component is
defined in terms of the component’s orientation relative to a reference plane which, in turn,
is defined by an angle between the scattering plane (the horizontal plane in laboratory
experiments) and the principal axis of the component, and can be written as follows:
⎡ ⎤ ⎡ ⎤⎡ ⎤
I M11 M12 M13 M14 I0
⎢ Q ⎥ ⎢ M21 M22 M23 M24 ⎥ ⎢ Q0 ⎥
⎢ ⎥=⎢ ⎥⎢ ⎥ (2.4.27b)
⎣ U ⎦ ⎣ M31 M32 M33 M34 ⎦ ⎣ U0 ⎦ ,
V M41 M42 M43 M44 V0
where the terms Mij (i, j = 1–4) denote the 4 × 4 Mueller matrix elements. Referring to
Figure 2.15, the scattered Stokes vector can be expressed in terms of the incident Stokes
vector I0 and a number of optical component arrangements as follows (Coulson 1988, Kuik
et al. 1991):
I() = A Q F()M P I0 , (2.4.27c)
where A, Q, M, and P represent the Mueller matrices of the analyzer, quarter-wave plate,
modulator, and polarizer, respectively, which have orientation angles with respect to the
incoming light beam, F() is the transformation matrix defined in Eq. (2.4.32) below, and
 is the scattering angle. The analyzer is a polarizer inserted in front of the detector, while
2.4 Definition of scattering properties for ice crystals 93

the modulator uses the linear electro-optic effect such that the voltage over a cell can be
varied sinusoidally in time.
To quantify the polarization of a light beam, it is customary to define the following
parameter, referred to as the degree of polarization:
1/2
PO = (Q2 + U 2 + V 2 ) /I . (2.4.28a)

If the ellipticity is neglected, we may define the degree of linear polarization in the form

LP = −Q/I = −(Il − Ir )/(Il + Ir ). (2.4.28b)

Let there be light: Light occurring in nature is called natural light. It is light whose
intensity remains unchanged and is unaffected by the retardation of one orthogonal com-
ponent relative to another when resolved in any direction in the transverse plane. This
means that natural light is defined from Eq. (2.4.26b) as, I (, ε) = I /2. Its intensity is
hence independent of  and ε. Thus, the necessary and sufficient condition that light be
natural is Q = U = V = 0. Under this condition, the degree of polarization defined in Eq.
(2.4.28a) for natural light is zero. As a consequence, natural light is also referred to as
unpolarized light. Light emitted from the sun is unpolarized. However, after interacting
with molecules and particles through scattering events, the unpolarized sunlight generally
becomes partially polarized. Natural light characterized by Q = U = V = 0 can be shown
to be equivalent to a mixture of any two independent oppositely polarized streams of half
the intensity.
In the atmosphere, light is generally partially polarized and its Stokes parameters (I, Q,
U, V) can be decomposed into two independent groups characterized as natural light and
elliptically polarized light, as follows:
⎡ ⎤ ⎡ ⎤ ⎡ 2 ⎤
I I − (Q2 + U 2 + V 2 )1/2 (Q + U 2 + V 2 )1/2
⎢Q⎥ ⎢ 0 ⎥ ⎢ Q ⎥
⎢ ⎥=⎢ ⎥+⎢ ⎥ . (2.4.29)
⎣U ⎦ ⎣ 0 ⎦ ⎣ U ⎦
V 0 V
Having presented the essence of the physical and mathematical definitions of a light
beam, we shall now introduce the single-scattering (light scattered just once) properties of
ice crystals.

2.4.3 Scattering phase matrix for an ice crystal


Consider an ice particle of arbitrary shape and size. The scattered electric field at a distance
r from the particle must be related to the two components of the incident electric field
(Eli , Eri ). In the far field, the 2 × 2 amplitude matrix that transforms the incident electric
vector into the scattered electric vector may be written as

s

i

i
El exp(−ikr + ikz) S2 S3 El A2 A3 El
= = , (2.4.30)
Ers ikr S4 S1 Eri A4 A1 Eri
94 Fundamentals of light scattering by ice crystals

where z is the vertical direction in Cartesian coordinates, Sj (j = 1, 2, 3, 4) are the amplitude


functions, and the matrix comprising Aj (j = 1, 2, 3, 4) is referred to as the transformation
matrix. Because of the symmetry of spherical particles with respect to the incident beam,
S3 = S4 = 0.
Using the Stokes parameters defined above, the incident and scattered electric vectors
can be expressed in terms of their intensity components as follows:
⎡ ⎤ ⎡ ⎤
I I0
⎢Q⎥ ⎢ ⎥
⎢ ⎥ = F ⎢ Q0 ⎥ , (2.4.31)
⎣ U ⎦ k 2 r 2 ⎣ U0 ⎦
V V0

where the subscript 0 denotes the incident beam and the transformation matrix F is given
by
⎡ ⎤
F11 F12 F13 F14
⎢ F21 F22 F23 F24 ⎥
F=⎢ ⎣ F31 F32 F33 F34 ⎦ ,
⎥ (2.4.32)
F41 F42 F43 F44

where

F11 = 12 (M2 + M3 + M4 + M1 ); F12 = 12 (M2 − M3 + M4 − M1 ), (2.4.32a)


F13 = S23 + S41 ; F14 = −D23 − D41 , (2.4.32b)
F21 = 1
2
(M2 + M3 − M4 − M1 ); F22 = 1
2
(M2 − M3 − M4 + M1 ), (2.4.32c)
F23 = S23 − S41 ; F24 = −D23 + D41 , (2.4.32d)
F31 = S24 + S31 ; F32 = S24 − S31 , (2.4.32e)
F33 = S21 + S34 ; F34 = −D21 + D34 , (2.4.32f)
F41 = D24 + D31 ; F42 = D24 − D31 , (2.4.32g)
F43 = D21 + D34 ; F44 = S21 − S34 . (2.4.32h)

Expressions for the matrix elements in terms of the electric fields can be derived from
the definition of the Stokes parameters and are given by

Mk = |Sk |2 , (2.4.33a)

Skj = Sj k = (Sj Sk∗ + Sk Sj∗ )/2, (2.4.33b)

−Dkj = Dj k = (Sj Sk∗ − Sk Sj∗ )i/2, j, k = 1, 2, 3, 4. (2.4.33c)

The preceding matrix elements are real numbers. In light scattering and radiative transfer,
it is conventional to define the scattering phase matrix, P, such that its first element is
2.4 Definition of scattering properties for ice crystals 95

normalized to unity as follows:


 2π  π
P11 ()
sin ddφ = 1, (2.4.34)
0 0 4π
where  and φ denote the scattering and azimuthal angles, respectively.
To complete the definition of the phase matrix, we need to introduce the scattering
cross-section, which represents the amount of incident flux that is removed from the
original direction as a result of a single-scattering event. This flux is distributed isotropically
throughout the area of a sphere whose radius is R and whose center is the scatterer. The
scattering cross-section is related to the first element of the transformation matrix defined
in Eq. (2.4.32) in the form
  % 4 &
1 2π π 1 
σs = 2 Mk sin  d dφ. (2.4.35)
k 0 0 2 k=1

Using Eq. (2.4.35), the scattering phase matrix can be defined in terms of the transfor-
mation matrix as follows:
P 1
= F. (2.4.36)
4π σs k 2
Subsequently, the scattered Stokes parameters can be expressed in the form
⎡ ⎤ ⎡ ⎤
I I0
⎢Q⎥ ⎢ ⎥
⎢ ⎥ = eff P ⎢ Q0 ⎥ , (2.4.37)
⎣U ⎦ ⎣
4π U0 ⎦
V V0

where the effective solid angle associated with scattering is given by eff = σs /r 2 . If no
assumption is made about the shape and position of the scatterer, the scattering phase matrix
consists of 16 non-zero elements as follows:
⎡ ⎤
P11 P12 P13 P14
⎢ P21 P22 P23 P24 ⎥
P(general) = ⎢
⎣ P31 P32 P33 P34 ⎦ .
⎥ (2.4.38)
P41 P42 P43 P44

2.4.4 Extinction cross-section for a non-spherical ice crystal


The extinction cross-section for a spherical particle can be derived from the solution for the
Lorenz–Mie theory discussed in Subsection 2.2.3.1. In this subsection, we shall formulate
the extinction cross-section for a non-spherical ice particle with arbitrary orientation.
Consider the conceptual diagram shown in Figure 2.16. A polarized incident wave can
be resolved into two components, the TE and TM modes for which the electric vector
vibrates perpendicular and parallel to the incident plane, respectively. For the TE mode
96 Fundamentals of light scattering by ice crystals

Far field
Θ = 0O
i
E i s
E +E

2 2
| E i |2 - | E i + E s | = se | E i |

Figure 2.16 A conceptual diagram for the evaluation of extinction cross-section in an arbitrarily
oriented non-spherical ice particle. All notations are defined in the text.

(case 1) the scattered electric field in the forward direction from Eq. (2.4.30) is given by

s

El e−ikr+ikz S3 (0) e−ikr+ikz
E ( = 0) =
s
s = Eri = S(0)Eri , (2.4.39)
Er ikr S1 (0) ikr
where S (0) = [S3 (0), S1 (0)]T and the symbol T denotes the transpose of a matrix.
Further, consider a point (x, y, z) in the forward direction, i.e.,  = 0. In the far field,
since x z and y z, we have
1 x2 + y2
r = (x 2 + y 2 + z2 ) /2 ≈ z + . (2.4.40)
2z
Superimposing incident and scattered electric fields in the forward direction yields


S(0) −k(x 2 +y 2 )/2z
Er + E = Er I +
i s i
e , (2.4.41)
ikz
where Eir = [Eri , Eri ]T and the unit matrix I = [1, 1]T . The far-field combined flux density
in the forward direction is then proportional to


 2  2 2 S (0) −k(x 2 +y 2 )/2z
IEir + Es  ≈ Eri  I + Re e , (2.4.42)
kz i
where Re[ ] represents the real part of the argument. We may integrate the combined flux
density over the cross-sectional area A of a scattering particle to obtain the total power of
the combined image as follows:

I
 2 |Eir + Es |dxdy = A − σ e , (2.4.43)
E i 
r A

where A = [A, A] and σ e = [σ e,11 , σ e,12 ]T . We may set the integration limits to infinity
T

for mathematical convenience to obtain


 ∞ ∞
2πz
e−k(x +y )/2z dxdy =
2 2
. (2.4.44)
−∞ −∞ ik
2.4 Definition of scattering properties for ice crystals 97

It follows that the extinction cross-section for the TE mode (case 1) is given by
σe,1 = σe,11 + σe,12 = (4π/k 2 )Re[S1 (0)] + (4π/k 2 )Re[S3 (0)]
= (4π/k 2 )Re[S1 (0) + S3 (0)]. (2.4.45a)

Following the same procedure, the extinction cross-section for the TM mode (case 2) can
be derived as follows:
σe,2 = (4π/k 2 )Re[S2 (0) + S4 (0)]. (2.4.45b)
The preceding two equations are the generalized extinction cross-sections for a non-
spherical particle. For incident natural light, the extinction cross-section is the average of
the two linear polarization components:
σe = (σe,1 + σe,2 )/2. (2.4.45c)
For a group of randomly oriented non-spherical ice particles, to be discussed in the next
subsection, because of the sign cancelation of S3 (0) and S4 (0), Eq. (2.4.45c) becomes
approximately (Yang and Liou 1997, Liou et al. 2014)

σe ∼
= 2 Re[S1 (0) + S2 (0)]. (2.4.45d)
k

2.4.5 Scattering phase matrix for an ensemble of ice crystals


The preceding discussion was concerned with the scattering of light by an ice crystal.
Scattering by a group of ice crystals is determined by the orientation and size of individual
crystals with respect to the incident light beam. For continuity in the present discussion, we
shall consider a sample of non-spherical ice crystals of the same size, randomly oriented in
space. The scattering phase matrix can then be expressed as
 2π  π/2
1
P() = P (α  , γ  )σs  (α  , γ  ) sin α  dα  dγ  , (2.4.46)
2πσs 0 0
where α  and γ  are the orientation angles of a non-spherical particle with respect to the
incident light beam, and P denotes the scattering phase matrix for a single particle. In
this case, the scattering phase matrix is solely a function of the scattering angle. Further,
the scattering cross-section defined in Eq. (2.4.35) for randomly oriented non-spherical ice
crystals can be written in the form
 
1
σs = σs (α  , γ  ) sin α  dα  dγ  , (2.4.47)

where σs denotes the scattering cross-section for a single crystal. For a symmetric hexagonal
crystal, we may define two orientation angles with reference to the incident beam. Let α be
the complement of the zenith angle, and β be an angle that is mapped on √ the2x-y plane with
reference to the incident beam. The surface area of the basal plane is 3 3D /8, while that
of the prism plane
√ is DL. The cross-sectional area for the basal plane mapped along the
light beam is 3 3 (D 2 /8) sin α. The cross-sectional area for the prism plane also depends
on the position of the light beam with reference to six prime planes. For example, if the
light beam is at an angle of β = π/6, the mapped cross-section along the beam would be
98 Fundamentals of light scattering by ice crystals

at maximum. Thus, we should have DL cos α cos(π/6 − β). The geometric cross-section
for the entire hexagonal crystal is the sum of the preceding mapped cross-sections:

3 3 2
G(α, β) = D sin α + DL cos α cos(π/6 − β). (2.4.48)
8
Defining the average cross-section for randomly oriented ice crystals and carrying out the
integrations over α and β, we obtain
 
6 π/6 π/2
3 √ 2
Ḡ = G(α, β) cos αdαdβ =( 3D /4 + DL). (2.4.49)
π 0 0 4

Note that the surface area of a hexagonal crystal is given by S = 3( 3D 2 /4 + DL). It
follows that in the context of the optical extinction theorem, the extinction cross-section,
which is twice the geometric cross-section, is related to the surface area by σe = 2Ḡ = S/2.
Thus, the average geometric cross-section for randomly oriented plate/column crystals Ḡ
is one-fourth of their surface areas (Vouk 1948).
If ice crystals are randomly oriented in space, the law of reciprocity can be applied
(Perrin 1942, van de Hulst 1957). In this case, the directions of the incident and scattered
polarized beams can be reversed to achieve the same result. It follows that the amplitude
functions (S3 , S4 ) in Eq. (2.4.30) must be equivalent to (−S4 , −S3 ), so that


S2 −S4
S∗ = . (2.4.50)
−S3 S1
The transformation matrix corresponding to this 2 × 2 amplitude matrix can then be written
in the form [see Eq. (2.4.32)]
⎡ ∗ ∗ ∗ ∗ ⎤
F11 F12 F13 F14
⎢ ∗ ∗ ∗ ∗ ⎥
⎢ F21 F22 F23 F24 ⎥
∗ ⎢
F =⎢ ∗ ⎥, (2.4.51)
∗ ∗ ∗ ⎥
⎣ F31 F32 F33 F34 ⎦
∗ ∗ ∗ ∗
F41 F42 F43 F44
where the matrix elements are defined by
∗ ∗
F11 = 12 (M2 + M4 + M3 + M1 ); F12 = 12 (M2 − M4 + M3 − M1 ), (2.4.51a)
∗ ∗
F13 = −S24 − S31 ; F14 = D24 + D31 , (2.4.51b)
∗ ∗
F21 = 1
2
(M2 + M4 − M3 − M1 ); F22 = 12 (M2 − M4 − M3 + M1 ), (2.4.51c)
∗ ∗
F23 = −S24 + S31 ; F24 = D24 − D31 , (2.4.51d)
∗ ∗
F31 = −S23 − S41 ; F32 = −S23 + S41 , (2.4.51e)
∗ ∗
F33 = S21 + S34 ; F34 = −D21 − D34 , (2.4.51f)
∗ ∗
F41 = −D23 − D41 ; F42 = −D23 + D41 , (2.4.51g)
∗ ∗
F43 = D21 − D34 ; F44 = S21 − S34 . (2.4.51h)
2.4 Definition of scattering properties for ice crystals 99

The scattered and incident Stokes parameters in this case are related via the scattering phase
matrix in the form [see Eq. (2.4.37)]
⎡ ⎤ ⎡ ⎤⎡ ⎤
I P11 P12 P13 P14 I0
⎢ Q ⎥ eff ⎢ P21 P22 P23 P24 ⎥ ⎢ Q0 ⎥
⎢ ⎥= ⎢ ⎥⎢ ⎥. (2.4.52)
⎣U ⎦ 4π ⎣ P31 P32 P33 P34 ⎦ ⎣ U0 ⎦
V P41 P42 P43 P44 V0

Consider an arbitrarily small volume containing a sample of non-spherical particles


whose scattering phase matrix is P, and let there be the same number of reciprocal particles
such that its scattering phase matrix is P∗ [corresponding to F∗ ; see Eq. (2.4.36)]. Then,
the average scattering phase matrix P corresponding to this volume can be expressed as
(P + P∗ )/2 and is proportional to the following transformation matrix:
⎡ ⎤
P11  P12  P13  P14 
⎢ P21  P22  P23  P24  ⎥
P = ⎢ ⎥
⎣ P31  P32  P33  P34  ⎦ , (2.4.53)
P41  P42  P43  P44 

where

P11  = 12 (M2 + M4 + M3 + M1 ); P12  = 12 (M2 − M1 ), (2.4.53a)


P13  = 1
(S
2 23
+ S41 − S24 − S31 ); P14  = 1
2
(−D23 − D41 + D24 + D31 ), (2.4.53b)
P21  = 12 (M2 − M1 ); P22  = 12 (M2 − M3 − M4 + M1 ), (2.4.53c)
P23  = 12 (S23 − S41 − S24 + S31 ); P24  = 12 (−D23 + D41 + D24 − D31 ), (2.4.53d)
P31  = 12 (S24 + S31 − S23 − S41 ); P32  = 12 (S24 − S31 − S23 + S41 ), (2.4.53e)
P33  = S21 + S34 ; P34  = −D21 , (2.4.53f)
P41  = 12 (D24 + D31 − D23 − D41 ); P42  = 12 (D24 − D31 − D23 + D41 ), (2.4.53g)
P43  = D21 ; P44  = S21 − S34 . (2.4.53h)

In terms of the elements expressed as Pij , we have

P12  = P21 , P13  = −P31 , P14  = P41 , P23  = −P32,


P24  = P42 , P34  = −P43 . (2.4.54)

Using these six relationships, the scattering phase matrix can then be written as
⎡ ⎤
P11  P12  P13  P14 
⎢ P  P22  P23  P24  ⎥
⎢ 12 ⎥
P = ⎢ ⎥. (2.4.55)
⎣ −P13  −P23  P33  P34  ⎦
P14  P24  −P34  P44 
100 Fundamentals of light scattering by ice crystals

In this case, the scattering phase matrix contains ten independent elements. We shall
now consider a sample of randomly oriented ice particles, which have a plane of symmetry
and let the incident Stokes parameters be changed to (I0 , Q0 , −U0 , −V0 ). The scattered
Stokes parameters must also be changed to (I, Q, −U, −V ). Thus, we may replace U0 and
V0 by −U0 and −V0 to obtain
I = P11  I0 + P12  Q0 + P13  U0 + P14  V0
= P11  I0 + P12  Q0 − P13  U0 − P14  V0 . (2.4.56a)
For this equation to be valid, we must have P13  = P14 = 0. Likewise, we have, for the
Q, U, and V components,
Q = P21  I0 + P22  Q0 + P23  U0 + P24  V0
= P21  I0 + P22  Q0 − P23  U0 − P24  V0 , (2.4.56b)
U = P31  I0 + P32  Q0 + P33  U0 + P34  V0
= P31  I0 + P32  Q0 − P33  U0 − P34  V0 , (2.4.56c)
V = P41  I0 + P42  Q0 + P43  U0 + P44  V0
= P41  I0 + P42  Q0 − P43  U0 − P44  V0 . (2.4.56d)

Consequently, we must have P23  = P24  = 0, P31  = P32  = 0, and P41  =


P42  = 0 to satisfy Eqs. (2.4.56b)–(2.4.56d). It follows that the scattering phase matrix in
this case is given by
⎡ ⎤
P11  P12  0 0
⎢ P21  P22  0 0 ⎥
⎢ ⎥. (2.4.57)
⎣ 0 0 P33  P34  ⎦
0 0 P43  P44 
Moreover, from Eq. (2.4.54), we note that P21  = P12  and P43  = −P34 . Thus, the
scattering phase matrix for randomly oriented non-spherical ice particles, which obey the
law of reciprocity, can be expressed as follows (the symbol   is neglected for presentation
convenience):
⎡ ⎤
P11 P12 0 0
⎢ P12 P22 0 0 ⎥
P(ice crystals) = ⎢
⎣ 0
⎥. (2.4.58a)
0 P33 P34 ⎦
0 0 −P34 P44
It follows that the scattering phase matrix reduces to six independent elements. In the case
of spherical particles, S3 = S4 = 0, so that P22 = P11 and P44 = P33 . Therefore, there are
only four independent scattering phase matrix elements, as follows:
⎡ ⎤
P11 P12 0 0
⎢ P12 P11 0 0 ⎥
P(spheres) = ⎢
⎣ 0
⎥. (2.4.58b)
0 P33 P34 ⎦
0 0 −P34 P33
2.4 Definition of scattering properties for ice crystals 101

Consider now a spectrum of ice particles randomly oriented in space with a size dis-
tribution given by n(L), where L is the major axis of an ice particle. A parameter referred
to as the aspect ratio LD, where D denotes the width, is used to define the particle size.
We may employ a mean effective size in association with light-scattering calculations. The
scattering phase matrix associated with a suitable volume, which contains a spectrum of
ice particles defined by a size distribution n(L), can be obtained from
 L2  L2
P() = P(, L)σs (L)n(L)dL σs (L)n(L)dL, (2.4.59a)
L1 L1

where L1 and L2 are the lower and upper limits of ice particle length. Let the total number
of ice crystals in the cloud be N. Then the scattering (and extinction) cross-section for a
spectrum of ice crystals of different sizes and shapes can be obtained from the following:
 L2
σs,e = σs,e (L)n(L)dL/N . (2.4.59b)
L1

As stated previously, the condition under which the scattering phase matrix is composed
of six independent elements is that non-spherical particles are randomly oriented in space
in such a manner that every one of them has a plane of symmetry based on which the
law of reciprocity can be applied: that the incident and scattered beams are reversible. In
Subsection 1.5.2, we gave examples of typical ice crystal sizes and shapes. Assuming that
they are randomly oriented in space, it would be almost impossible to prove that every
one of these ice crystals has a plane of symmetry that would obey the law of reciprocity.
Nevertheless, numerical calculations for light scattering by ice crystals carried out by
Takano and Liou (1989a, 1995) show that the scattering phase matrix elements P13 , P14 ,
P23 , P24 , P31 , P32 , P41 , and P42 are practically zero and that P12 = P21 and P34 = −P43 .

2.4.6 Some examples of the single-scattering properties of ice spheres, cylinders,


and spheroids
In this subsection, we present a number of pertinent single-scattering results for ice spheres,
ice circular cylinders, and ice spheroids. The extinction efficiency Qext is defined as
σ e G, where G is the geometric cross-section with respect to the incident direction. For a
spherical particle with a radius of r, it is simply πr2 . Moreover, the single-scattering albedo,
representing the fraction of light that is scattered out of the incident light beam, is defined
as follows:

 = σs (scattering cross section)/σe (extinction cross section). (2.4.60a)

Thus, (1 – ϖ) represents the fraction of light that is absorbed. From the total scattered
intensity, we define a parameter referred to as the asymmetry factor in the form

1 1
g = < cos  > = P11 (cos ) cos  d cos , (2.4.60b)
2 −1
102 Fundamentals of light scattering by ice crystals

Extinction Efficiency Single-Scattering Albedo Asymmetry Factor


4 1.0 1.0

0.8 0.8
3
r = 10 μm

0.6 0.6
2
0.4 0.4

1
0.2 0.2

0 0.0 0.0

4 1.0 1.0

0.8 0.8
3
r = 100 μm

0.6 0.6

2
0.4 0.4

1
0.2 0.2

0.0 0.0
0
10−1 100 101 102 10−1 100 101 102 10−1 100 101 102

Wavelength (μm)

Figure 2.17 Extinction efficiency, single-scattering albedo, and asymmetry factor as functions of
wavelength from 0.2 to 100 µm for an ice sphere with a radius of 10 µm (top panels) and 100 µm
(bottom panels). These results are computed from the exact Lorenz–Mie solution.

where P11 () denotes the normalized phase function, and is directly related to the total
scattered intensity, as a function of the scattering angle . The g-factor is the first moment
of the phase function and represents the strength of forward scattering. The extinction
efficiency, single-scattering albedo, and asymmetry factor are the three basic parameters
for input to a radiative transfer program for radiative flux calculations. For remote sensing
purposes, however, we need the phase function for bidirectional reflectance analysis; and
if polarization is involved, information from the full scattering phase matrix is required.
Figure 2.17 depicts the extinction efficiency Qext , the single-scattering albedo ϖ, and
the asymmetry factor g for an ice sphere with radii of 10 and 100 µm as functions of 0.2
to 100 µm wavelength. The results were calculated from Lorenz–Mie theory, employing
the complex (real and imaginary) indices of refraction depicted in Figure 2.8. Qext for r =
10 µm can be explained in terms of the size parameter 2πrλ. For r = 100 µm, Qext is close
to 2 as a result of the limit of large size parameters when the wavelength becomes smaller.
Minima for ϖ at λ = 2.85, 10.75, and 42.76 µm when r = 10 µm result from the Christiansen
effect (Christiansen, 1884, 1885), which occurs when the real part of the refractive index
approaches 1, while the corresponding imaginary counterpart is substantially larger, leading
to the domination of absorption. For r = 100 µm at λ 3 µm, ϖ has a value of 0.5 due
to strong absorption. The maxima of g-factor for r = 10 µm occur in concurrence with
the minima of ϖ, and vice versa, at λ  23 µm, contributed from diffracted and externally
reflected rays, which are more dominant when ϖ is smaller. At λ  3 µm, the g-factor for
2.4 Definition of scattering properties for ice crystals 103

Extinction Efficiency Single-Scattering Albedo Asymmetry Factor


4 1.0 1.0

0.8 0.8
3
r = 10 μm

0.6 0.6
2
0.4 0.4

1
0.2 0.2

0 0.0 0.0

4 1.0 1.0

0.8 0.8
3
r = 100 μm

0.6 0.6

2
0.4 0.4

1
0.2 0.2

0.0 0.0
0
10−1 100 101 102 10−1 100 101 102
10−1 100 101 102
Wavelength (μm)

Figure 2.18 Extinction efficiency, single-scattering albedo, and asymmetry factor for an ice circular
cylinder with a radius of 10 µm (top panels) and 100 µm (bottom panels) at normal incidence.

r = 100 µm can be interpreted in terms of geometric optics as in the case of r = 10 µm. At


λ 3 µm, the g-factor becomes relatively flat due to strong absorption.
Figure 2.18 shows Qext , ϖ, and g for a circular cylinder at normal incidence computed
from a Lorenz–Mie-like program (Liou 1972a). Qext is similar to the values presented in
Figure 2.17 due to geometric similarity, except for the enhanced amplitude of sinusoidal
oscillations around 2 at λ  3 µm. ϖ for r = 10 µm is also similar to that of the sphere at
λ  30 µm, while ϖ for r = 100 µm is similar to that of the sphere over all wavelengths.
The g-factor for r = 10 µm is larger than that for the sphere, but for r = 100 µm it is larger
than the spherical case at λ  3 µm, although the results are comparable to the g-factor for
a sphere at λ 3 µm.
Figure 2.19a illustrates the phase function for polydisperse ice spheres having a gamma
size distribution with an effective radius re of 20 µm, corresponding to cirrostratus clouds.
We have selected four remote sensing wavelengths for presentation. At visible wavelengths,
noticeable features are a diffraction peak at the scattering angle  = 0°, the primary rainbow
feature at  = 136°, and a glory feature at  = 180°. At λ = 1.38 µm, the rainbow peak
reduces and spreads due to a small size parameter. At λ = 3.7 µm, the rainbow shifts to
 = 154°. Rainbow and glory intensities are reduced as a result of absorption. At
λ = 10 µm, the phase function consists of diffracted and externally reflected rays (see
Subsection 3.1.2) caused by strong absorption, rendering it featureless.
Figure 2.19b shows the corresponding degree of linear polarization (LP). At λ =
0.63 µm, a negative polarization at   70° results from transmitted rays without internal
104 Fundamentals of light scattering by ice crystals

0.8
(a) (b)
1x106
Cs (re = 20 μm) 0.6

Degree of Linear Polarization


1x104
0.4
Phase Function

1x10 2 1 =0.63 m 0.2

1
1.38 m
0
1x100
3.7 m
=0.63 µm
−0.2 1.38 µm
1x10−2 10 m 3.7 µm
10 µm
−0.4
0 60 120 180 0 60 120 180
Scattering Angle (deg.)

Figure 2.19 (a) Phase function and (b) degree of linear polarization for a group of spheres governed
by a gamma size distribution with an effective radius of 20 µm, corresponding to cirrostratus. In the
phase function diagram, the vertical scale applies to the lowest curve, while the upper curves are
shifted upward by a factor of 10.

reflection, while maxima at  = 96°, 136°, and 144° are produced, respectively, by exter-
nally reflected rays, rainbows, and supernumerary bows. Intense variation of LP around
 = 180° is produced by the glory feature. At λ = 1.38 and 3.7 µm, maxima produced
by external reflection at  = 100° and 112°, respectively, are reduced due to smaller size
parameters. At λ = 3.7 µm, the maximum produced by rainbows shifts to  = 154°. At λ =
10 µm, a single maximum occurs, produced by external reflection in conjunction with strong
absorption.
In Figure 2.20, some characteristics of the normalized phase function P11 for randomly
oriented spheroids are observed in backscattering regions; that is, a rather flat angular
distribution with a small increase at the backscattering. Features such as rainbows and
glories produced by scattering by large spheres are greatly reduced for spheroids of large
ab. The element −P12 P11 (= LP) gives the degree of linear polarization for the single
scattering of unpolarized incident light. It has positive values at scattering angles  120°
in contrast to negative values in the case of spheres.
The element P22 P11 represents the ratio of the intensity component depolarized or
cross-polarized to the total scattered intensity. The depolarization ratio for total intensity is
defined as  = 1 − P22 P11 , which is a measure of non-sphericity because P22 = P11 and
 = 0 for homogeneous spheres. For small angles  25°,  is very small. However, from
102 1.0 102 1.0

Prolate Oblate
m = 1.33 0.5 m = 1.33 0.5
1
101 a = 15 10 a = 15
a/b = 5 a/b = 5
0 0
0 0
10 10
-0.5 P43/P11 -0.5 P43/P11

10-1 (c) 10-1 (c)


P11 -1.0 P11 -1.0
1.0 1.0
(a) (a)
10-2 10-2
1.0 0.5 1.0 0.5
P22 /P11 P22 /P11
P44/P11 P44/P11
0.5 0 0.5 0

0 -0.5 P33/P11 0 -0.5 P33/P11


-P12/P11 -P12/P11
(b) (d) (b) (d)
-0.5 -1.0 -0.5 -1.0
0 60 120 180 0 60 120 180 0 60 120 180 0 60 120 180
Scattering Angle (deg.)
Figure 2.20 Angular distribution of the elements of normalized scattering matrix for randomly oriented prolate (left panels) and oblate (right panels)
spheroids with m = 1.33, α = 15, and ab = 5: (a) the normalized phase function P11 ; (b) the element P22 P11 and the degree of linear polarization
−P12 P11 ; (c) the element P43 P11 ; and (d) the elements P33 P11 and P44 P11 . The scattering matrix elements for area equivalent spheres are shown by
dotted lines (after Asano and Sato 1980).
106 Fundamentals of light scattering by ice crystals

side to backscattering regions,  values are relatively large in the case of spheroids. The
element P43 P11 for randomly oriented spheroids is close to that for equivalent spheres.
Also, good agreement extends to larger scattering angles than in the case of LP. For large
scattering angles, however, P43 P11 for spheroids is larger than that for equivalent spheres.
The element P44 P11 for randomly oriented spheroids is generally larger than P33 P11 . The
case of thin oblate spheroids is an exception, however. We find that differences between
P44 P11 and P33 P11 vary as a function of angle in a manner similar to P22 P11 , and that
 ≥ |P44 − P33 |/P11 with the equality valid at  = 0 and 180°. At these angles, P33 =
±P22 , as predicted from the symmetry relations discussed by van de Hulst (1957).

2.5 Introduction to radiative transfer in ice crystal clouds: link to single scattering
The preceding discussion applies to scattering by an ice crystal just once, including cou-
pled absorption and polarization processes and scattering by an ensemble of ice crystals,
assuming that the independent scattering concept is applicable such that each crystal scat-
ters the light beam independently (single scattering). In the latter case, we may consider
a sufficiently small volume in an ice cloud so that its single-scattering properties can be
defined on the basis of Maxwell’s fundamental electric and magnetic vector equations and
Stokes’ vector intensity (I, Q, U, V). We shall present the link between these two funda-
mentals and the transfer of radiation in planetary atmospheres, in which multiple scattering
(and absorption or emission) takes place. The term “light scattering” thus includes single
scattering and multiple scattering, the latter also referred to as “radiative transfer.”
The subject of radiative transfer covers a variety of fields, including astrophysics, applied
physics and optics, planetary science, atmospheric science, and meteorology, as well as
various engineering disciplines. Astrophysicists pioneered the development of this field
at the beginning of the twentieth century (Schuster 1905, Schwarzschild 1906, Eddington
1916, Menzel 1966). Prior to 1950, the subject of radiative transfer was also an important
research area in applied physics and nuclear engineering, associated with neutron transport.
In his landmark book, Chandrasekhar (1950) presented the subject of radiative transfer in
plane-parallel atmospheres as a branch of mathematical physics and developed numerous
solution methods and techniques, including the consideration of polarization. The principle
of radiative transfer has also been extensively employed by planetary scientists, particularly
in association with the remote sensing of planetary atmospheres by means of spectroscopy
and polarimetry (Chamberlain and Hunten 1987), as well as by atmospheric scientists
to study the transfer of solar and thermal infrared radiation in the Earth’s atmosphere
containing clouds and aerosols, for climate, weather, and remote sensing applications
(Goody and Yung 1989; Liou 1992, 2002; Mishchenko et al. 2000a, 2000b).

2.5.1 Formulation of scalar and vector radiative transfer equations


In accordance with the independent scattering concept introduced in Section 2.1, we may
define the intensity, Iλ , in units of energy per second per square meter per wavelength
per solid angle (e.g., W m−2 λ−1 sr−1 ), where W (watt) = joule per second, m = meter,
λ = wavelength in units of µm, and sr = steradian. See also Eq. (2.4.10a). For simplicity
2.5 Radiative transfer in ice crystal clouds 107

(a) Z (b)
Z
(m, f)
z

s i2
i1 (m ', f ')
P2
Θ
q q P1
q'
O Y O

f
f'
f−

Figure 2.21 (a) Definitions of the zenith angle θ and the azimuthal angle φ with reference to the
Cartesian coordinate system (x, y, z), where s represents a position vector in space. (b) Scattering
plane OP1 P2 with respect to the meridian planes OP1 Z and OP2 Z. All angles in the figure are defined
in the text.

of presentation, we set I = Iλ . It is understood that I represents monochromatic intensity


confined within a solid angle. Thus, we say that intensity acts like a pencil of radiation,
which is a function of position, defined by (x, y, z) in Cartesian coordinates, and direction,
defined by a solid angle in terms of zenith and azimuthal angles (Figure 2.21a). We shall
first consider I as a scalar and follow the formulation developed by Chandrasekhar (1950)
based on the energy conservation principle.
A pencil of radiation traversing an ice cloud will be weakened by its interaction with ice
crystals. If the intensity I becomes I + dI after traversing a thickness ds in the direction of
its propagation, then we have

dI = −kρI ds, (2.5.1)

where ρ is the density of the cloud and k denotes the mass extinction (scattering +
absorption) cross-section (in units of area per mass) for radiation of a given wavelength.
At the same time, the intensity can be strengthened by multiple scattering from all other
directions into the pencil plus self-emission from ice crystals under the Kirchhoff–Planck
thermal equilibrium condition, all at the same wavelength. We define the source function
coefficient j such that the enhancement in intensity due to multiple scattering and emission
is given by

dI = jρds, (2.5.2)

where the source function coefficient j has the same physical meaning as the mass extinction
cross-section. Combining the two radiation sources defined in Eqs. (2.5.1) and (2.5.2), we
obtain

dI = −kρI ds + jρds. (2.5.3)


108 Fundamentals of light scattering by ice crystals

For presentation convenience, we may define the source function J = jk to obtain
dI (s)
− = −I + J. (2.5.4)
kρds
This represents the general radiative transfer equation without imposition of a coordinate
system.
For atmospheric applications, we may consider that the atmosphere in localized portions
is plane-parallel such that intensity and atmospheric parameters (e.g., temperature and
gaseous profiles) are horizontally homogeneous, while variation only occurs in the vertical
direction. In this case, it is convenient to measure linear distances normal to the plane of
stratification (Figure 2.21a). If z denotes this component, then the general equation for
radiative transfer defined in Eq. (2.5.4) becomes
dI (z; θ, φ)
cos θ = −I (z; θ, φ) + J (z; θ, φ), (2.5.5)
kρdz
where θ denotes the zenith angle in reference to the upward normal and φ the azimuthal
angle in reference to the x-axis.
For convenience of analysis, we introduce the non-dimensional normal optical depth (or
simply optical depth) τ defined by the following integration:
 ∞
τ= k(z )ρ(z )dz , (2.5.6)
z

so that dτ = −kρdz. In terms of the optical depth and by measuring intensity downward
from the outer boundary (top of the atmosphere), we have
dI (τ ; μ, φ)
μ = I (τ ; μ, φ) − J (τ ; μ, φ), (2.5.7)

where µ = cos θ . This is the basic equation for radiative transfer in plane-parallel atmo-
spheres. The source function representing intensity enhancement during the transfer of
radiation includes three terms: (1) multiple scattering, (2) direct radiation from the sun
attenuated to the level τ , and (3) emission from ice crystals subject to the Kirchhoff–Planck
condition. All three of these components with the same wavelength are traversing in the
direction of intensity defined by (µ, φ).
Thus, the basic scalar equation for the transfer of diffuse intensity (i.e., scattering more
than once) as a function of optical depth τ and direction denoted by (µ, φ) in plane-parallel
atmospheres can be written in the form

dI (τ ; μ, φ) 
μ = I (τ ; μ, φ) − I (τ, μ , φ  )P (μ, φ; μ , φ  )dμ dφ 
dτ 4π 4π (2.5.8)

− F P (μ, φ; −μ0 , φ0 )e−τ /μ0 − (1 − ' ω)B[T (τ )],

where the second term on the right-hand side represents multiple scattering contribu-
tions from all directions (4π solid angle), where the single-scattering albedo  has been
defined in Eq. (2.4.60a); the phase function P (μ, φ; μ , φ  ) represents the normalized 3-D
2.5 Radiative transfer in ice crystal clouds 109

distribution of scattered intensity, which redirects the incoming intensity defined by (μ , φ  )
to the outgoing intensity defined by (μ, φ); the third term defines the solar flux F at the
top of the atmosphere, attenuated to the level τ and undergoing a single scattering event in
the direction (−μ0 , φ0 ) via the phase function P (μ, φ; −μ0 , φ0 ); and the last term is the
emission contribution under the local thermodynamic equilibrium condition such that emis-
sion is equal to absorption, where B[T(τ )] is the Planck function defined by the temperature
at the position τ .
In order to describe the radiation field completely at each point in space, the scalar
intensity I must be replaced by the vector intensity I = (I, Q, U, V ), in terms of the
four Stokes parameters defined by the intensity, the degree of polarization, the plane of
polarization, and the ellipticity of light beams as a function of incoming and outgoing
directions, as shown in Eqs. (2.4.22a)–(2.4.22d). Thus, the radiative transfer equation can
be written in vector form as follows:
dI(τ ; μ, φ)
μ = I(τ ; μ, φ) − J(τ ; μ, φ), (2.5.9)

where the source function J(τ ; μ, φ) is a vector consisting of four elements, which can
be obtained as follows. Consider a differential increment dJ(τ ; μ, φ; μ , φ  ), produced by
multiple scattering involving a pencil of radiation of solid angle d in the direction (μ , φ  ).
The diffuse intensity vector I(τ ; μ , φ  ), which generates the source term, is in reference
to the meridian plane OP1 Z, defined in Figure 2.21b. However, the scattering phase matrix
P() derived from the scattering theory is in reference to the plane of scattering OP1 P2 that
contains the incident and scattered beams. We must then transform I(τ ; μ , φ  ) to the plane
of scattering in order to obtain the proper source function by first redirecting I(τ ; μ , φ  )
to the plane of scattering by applying the transformation matrix L(−i1 ), such that the
angle −i1 is defined between the meridian plane OP1 Z and the plane of scattering OP1 P2 ,
and the minus sign signifies that the rotation of the plane is counterclockwise. Thus, the
contribution to the source function with reference to the plane of scattering at the point
P2 is given by the term denoted by the symbol (∗ ) =  P()L(−i1 )I(τ ; μ , φ  )d /4π,
where the approximation of a constant single-scattering albedo for a cloud is made. The
transformation matrix for the Stokes vector (I, Q, U, V ) defined on the x-y plane relative
to that on the l-r plane can be obtained on the basis of Eq. (2.4.12a), and it is given by
⎡ ⎤
1 0 0 0
⎢ 0 cos 2χ sin 2χ 0 ⎥
L(χ ) = ⎢
⎣ 0 −sin 2χ cos 2χ 0 ⎦ .
⎥ (2.5.10)
0 0 0 1

The Stokes parameters I and V are invariant in the transformation process. Based on matrix
manipulations, we also have L(χ1 )L(χ2 ) = L(χ1 + χ2 ) and the inverse matrix L−1 (χ ) =
L(−χ ).
To transform this vector, denoted by (∗ ) above, to the scattering direction (μ, φ) repre-
senting the meridian plane OP2 Z, we must again apply the transformation matrix L(π − i2 )
110 Fundamentals of light scattering by ice crystals

through the angle (π − i2 ) clockwise, where i2 denotes the angle between the meridian plane
OP2 Z and the plane of scattering OP1 P2 . Thus, the desired differential source function due
to the diffuse component can be expressed as

dJ(τ ; μ, φ; μ , φ  ) =  Z(μ, φ; μ , φ  )I(τ ; μ , φ  )d /4π, (2.5.11)

where the phase matrix is defined by

Z(μ, φ; μ , φ  ) = L(π − i2 )P()L(−i1 ). (2.5.12)

The source function vector for multiple scattering can be obtained by performing the
integration over all incoming directions signified by (μ , φ  ) as follows:
 2π  1
  1
J(τ ; μ, φ; μ , φ ) =  Z(μ, φ; μ , φ  )I(τ ; μ , φ  )dμ dφ  . (2.5.13)
4π 0 −1

We have used Z and P to represent the phase matrix and the scattering phase matrix,
respectively, to differentiate between the two. From spherical trigonometry, the angles i1
and i2 can be derived and are given by
 
sin(φ − φ  ) 1 − μ2 sin(φ − φ  ) 1 − μ 2
sin i1 = , sin i2 = . (2.5.14)
sin  sin 
Following the same procedure, the direct component of the source function vector associated
with the point source I (−μ, φ) = δ(μ − μ0 )δ(φ − φ0 )F , where δ is a delta function and
F is the downward solar flux vector, is given by
1
J(τ ; μ, φ) =  Z(μ, φ; −μ0 , φ0 )F e−τ /μ0 . (2.5.15)

Combining Eqs. (2.5.15) and (2.5.13), the transfer of sunlight, including full polarization,
can be written as follows:
 2π  1
dI(τ ; μ, φ) 1
μ = I(τ ; μ, φ) −  Z(μ, φ; μ , φ  )I(τ ; μ , φ  )dμ dφ 
dτ 4π 0 −1 (2.5.16)
1
−  Z(μ, φ; −μ0 , φ0 )F e−τ /μ0 + (1 −  )B(T )Ie ,

where B(T) is the Planck intensity at temperature T; and Ie = (I, Qe , 0, 0), with −Qe the
linear polarization component associated with emission. In the preceding formulation, we
have assumed that an optical depth can be defined for the medium. This applies to spherical
particles that are randomly located with the scattering phase matrix P given by Eq. (2.4.58b),
and to non-spherical particles that are randomly oriented with the scattering phase matrix
P given by Eq. (2.4.58a). In these cases, we may replace the phase function depicted in Eq.
(2.5.8) with the 4 × 4 phase matrix Z to account for the full polarization effect. The phase
matrix Z has a number of unique properties associated with the symmetry principle of light
beams. For the preceding cases, a light beam can reverse its direction with the final results
being the same, so that Z(−μ, −φ; −μ , −φ  ) = Z(μ, φ; μ , φ  ).
2.5 Radiative transfer in ice crystal clouds 111

θ
φ - φ' Out
θ'
Θ
In π-

O
α
α'
γ - γ'
Y

φ' Z' φ

Figure 2.22 Geometry of single scattering by a non-spherical ice particle. The coordinate system
(x , y , z ) refers to the crystal axis, while (x, y, z) is fixed in space. The angles (θ, θ  ) are zenith
angles associated with the incident and scattered beams with respect to the (x, y, z) coordinates, while
φ = (φ − φ  ) are corresponding azimuthal angles. The angles (α, α  ) are zenith angles with respect
to the orientation of the particle, while (γ , γ  ) correspond to azimuthal angles. These angles can be
transformed to (x , y , z ) coordinates in terms of a set of angles (, φ) and (α  , γ  ), for analysis of
the light scattering by horizontally oriented ice crystals.

2.5.2 Radiative transfer in horizontally oriented ice particles


The scattering of a light beam by a non-spherical ice particle is dependent on the directions
of the incoming and outgoing radiation, and on the particle’s orientation with respect to the
incoming beam. In order to formulate the transfer of solar radiation in a medium comprised
of horizontally oriented ice particles, we may begin by assuming that the intensity varies
only in the z direction, referred to as the plane-parallel condition.
Referring to Figure 2.22, we may select a fixed coordinate system (x, y, z) such that the
z-axis is along the zenith direction. Additionally, we let (x  , y  , z ) be a coordinate system
referred to the crystal axis, with z along the c-axis. Angles (φ  , φ, γ  , γ ) are azimuthal
angles corresponding to (θ  , θ, α  , α) defined in the figure, and  is the scattering angle.
The scattering parameters for a non-spherical ice particle, including the phase function
and the extinction and scattering cross-sections, may be expressed with respect to either
of these two coordinate systems. Thus, we may express them symbolically as follows:

P (α, γ ; μ , φ  ; μ, φ) = P (α  , γ  ; cos , φ); σe,s (α, γ ; μ , φ  ) = σe,s (α  , γ  ), (2.5.17)

where φ = φ − φ  . It is clear that the phase function is dependent on the directions of


the incident and scattered beams as well as the orientation of the non-spherical ice particle.
112 Fundamentals of light scattering by ice crystals

The extinction and scattering cross-sections, however, depend only on the direction of the
incident beam and the ice particle orientation.
For a sample of non-spherical ice particles randomly oriented in space, the average
single-scattering properties may be expressed in the forms
 2π  π/2
1
P (cos , φ) = P (α  , γ  ; cos , φ)σs (α  , γ  ) sin α  dα  dγ  ,
2πσs 0 0
 2π  π/2 (2.5.18)
1     
σe,s = σe,s (α , γ ) sin α dα dγ .
2π 0 0

The extinction and scattering cross-sections for randomly oriented non-spherical particles
are independent of direction. If all non-spherical ice particles have rotational symmetry
(e.g., circular cylinders), then the phase function is independent of the azimuthal angle φ.
Consequently, multiple scattering of the diffuse intensity in randomly oriented, symmetrical
non-spherical ice particles can be formulated using six independent scattering phase matrix
elements.
The spatial orientation of hexagonal and irregular ice crystals in cirrus clouds is a
significant factor in the transfer of radiation in the atmosphere. The fact that numerous
halos and arcs have been observed demonstrates that specific orientation of ice particles
must exist in some cirrus. Based on laboratory experiments, cylinders with a diameter-to-
length ratio of less than 1 tend to fall with their long axes horizontally oriented. Observations
of columnar and plate crystals in cirrus clouds have shown that these particles fall with
their major axes parallel to the ground. The orientation of ice particles in cirrus clouds
has been observed by numerous lidar measurements based on the depolarization technique
in the backscattering direction. The depolarization ratio of the backscattered return from
horizontally oriented plates is close to zero (see Section 5.5), but this ratio increases
significantly as the lidar scans a few degrees from the vertical. Specific orientation occurs
when the ice particles have relatively large sizes and defined shapes, such as columns and
plates do. However, if the ice crystals are irregular, as aggregates are, a preferred orientation
is unlikely to occur. Furthermore, smaller ice crystals in cirrus clouds where substantial
turbulence occurs tend to orient in three-dimensional space. Finally, it has been noted that
ice particle orientation and alignment are closely modulated by the electric field in clouds.
In the case of horizontally oriented ice crystals, their single-scattering parameters are
dependent on the direction of the incident light beam. Thus, the conventional formulation
for the multiple-scattering problem requires modification. Takano and Liou (1989b) used
realistic scattering parameters and the Stokes vector for horizontally oriented ice crystals
in association with the adding method for radiative transfer. Takano and Liou (1993)
further presented the theoretical formulation and numerical calculations for the transfer of
polarized thermal infrared radiation in optically anisotropic media with specific application
to horizontally oriented ice particles. In the following paragraphs, we present a unified
theoretical formulation that is applicable to both solar and thermal infrared radiative transfer,
including polarization, for horizontally oriented ice crystals.
2.5 Radiative transfer in ice crystal clouds 113

When ice particles are randomly oriented in a horizontal plane, we have α = π/2 from
Eq. (2.5.17). Thus, the phase function and cross-sections are dependent only on the incident
angle and may be symbolically written in the forms

  2 π/2
P (μ , φ ; μ, φ) = P (π/2, γ ; μ , φ  ; μ, φ)dγ , (2.5.19a)
π 0
 2π  π/2
1
σe,s (μ) = 2 σe,s (π/2, γ ; μ, φ)dγ dφ. (2.5.19b)
π 0 0

With the preceding understanding of the incident direction of a light beam with respect to ice
particle geometry, we may define a differential normal optical depth such that d τ̃ = −β̃e dz,
where the vertical extinction coefficient β̃e = βe (μ = 1) and z is the distance. The general
equation governing the transfer of the Stokes vector may be expressed in the form
dI(τ̃ ; μ, φ)
μ = k(μ)I(τ̃ ; μ, φ) − J(τ̃ ; μ, φ), (2.5.20)
d τ̃
where the actual extinction coefficient normalized by its vertical counterpart is defined by
k(μ) = βe (μ)/β̃e ,  in Eq. (2.5.16) is replaced by the single-scattering albedo defined by
 (μ) = βs (μ)/βe (μ), and βs is the scattering coefficient matrix, which has a form similar
to the 4 × 4 extinction coefficient matrix.
For horizontally oriented particles, the extinction coefficient is dependent on both the
energy characteristics of the incident beam and its state of polarization, referred to as the
dichroism of the scattering medium. This generally occurs when the light beam passes
through a cloud of aligned non-spherical particles associated with electric and/or magnetic
fields, as noted previously. Because of dichroism, the extinction coefficients corresponding
to the Stokes vector are represented by a 4 × 4 extinction matrix (Martin 1974, Mishchenko
1991, Liou and Takano 2002). If ice particles are randomly oriented in space such that
each one of them has a plane of symmetry and the law of reciprocity may be applied,
the scattering phase matrix P consists of only six independent elements, as shown in
Eq. (2.4.58a). In this case, k(μ) = 1 and βs , βe , and  are independent of µ. For all
practical purposes, we may use the extinction coefficient βe and the single-scattering albedo
ω̃ in scalar forms in the calculations for horizontally oriented column and plate crystals.
We may approach the radiative transfer problem for ice particles randomly oriented in
a horizontal plane using the single-scattering and optically thin approximations, and omit
the emission term. The phase function and single-scattering parameters are now dependent
on the direction of the incident beam. In the case of the transfer of solar radiation, we must
use the normal optical depth τ̃ and the cosine of the solar-zenith angle-dependent single-
scattering albedo,  (μ0 ). We must also distinguish between the reflection and transmission
of radiation from above and below, since the phase functions for horizontally oriented ice
particles differ in these two configurations (Takano and Liou 1989b, Liou and Takano
2002).
Figure 2.23 depicts an example of a non-dimensional reflected light beam that has been
scattered for horizontally oriented (2-D) and randomly oriented (3-D) columns using a
114 Fundamentals of light scattering by ice crystals

c d
e

3D 2D

0 0 0
0

0
0

le

5
5

le

10
10

g Zen 20 40

15
15

Zen 20 40 60 60 g

20
20

An An

25
25

30
30

35
al
35

ith A ith A al

40
40

45
45

120
uthΔf

50
120
50

uth

55
60
55

ngle 60 ngle

60
60

im Δf

65
65

70
70

,q i m ,q 80 180

75
80 180
75

80
z
80

Az

85
85

Figure 2.23 Non-dimensional reflected light at a wavelength of 0.63 µm for ice columns randomly
oriented in space (3-D) and oriented parallel to the ground (2-D columns) in the plane of zenith and
azimuthal angles. The cosine of the solar zenith angle and the optical depth used in the calculations
are 0.5 and 1, respectively. In the 3-D case, two features are seen: (a) the antisolar peak and (b) limb
brightening. For the 2-D column case, three peak features are: (c) the broad peak around the antisolar
point, (d) the broad peak around the subsun, and (e) the lower tangent arc. See also Subsection 3.1.4.

representative cirrostratus ice crystal size distribution with an optical depth of 1 in the
plane defined by the zenith (θ ) and relative azimuthal (φ − φ0 ) angles. The wavelength and
the cosine of the solar zenith angle used are 0.63 µm and 0.5, respectively. Note that the
domain for the reflectances is from 60° to 180° scattering angles, which does not cover
the commonly observed halos located at 22° and 46°. In the 3-D case, the maximum, at
θ ≈ 80◦ and close to the principal plane φ − φ0 = 0◦ , is related to limb brightening. A
small maximum associated with the antisolar peak is seen at θ ≈ 40◦ and φ − φ0 = 120◦ .
Otherwise, the reflectance variations are relatively small in the linear scale. In the case
of 2-D columns, we see numerous reflection maxima. The chief ones are: (1) the subsun
located at θ ≈ 60◦ in the principal plane produced by external reflections; (2) the lower
tangent arc located at θ ≈ 80◦ in the principal plane generated by two refractions; and (3)
the antisolar peak located at θ ≈ 60◦ and φ − φ0 = 180◦ , caused by internal reflections.
Much larger anisotropy occurs in this case than in the 3-D case. In realistic cirrus clouds,
we would anticipate that some of the large and defined ice particles (plates and columns)
are horizontally oriented (see also discussions in Subsections 1.6.2 and 3.1.4).
3
Principles of geometric optics for application to light
scattering by ice crystals

In Chapter 1, we showed that ice crystals are complex and intricate in terms of their
morphology and size – essentially no single ice crystal that has been observed in the Earth’s
atmosphere and laboratory setting exhibits the exact same size, shape, and composition
(internally mixed with ice nuclei). In Chapter 2, we followed by presenting the available
solutions for light scattering by sphere, circular cylinder, and spheroid upon which the
coordinate systems can be imposed to define these particle shapes. Thus, analytic solutions
can be derived from conventional wave equations and suitable boundary conditions to match
the incident, internal, and scattered electric and magnetic vectors in three-dimensional
space defined by coordinate systems. Computations can then be performed using solution
equations referred to as the “exact” method, to obtain the single-scattering properties of
ice spheres, cylinders, and spheroids. For instance, the Lorenz–Mie theory is the exact
“solution” for light scattering by spherical particles. Light scattering by non-spherical ice
crystals, however, cannot be resolved by the conventional approach to solutions in view
of their complex structure. This chapter presents the geometric-optics approach to light
scattering by ice crystals. We first review the essence of geometric optics from several
historical perspectives, including diffraction and surface waves.

3.1 Application of geometric optics to light scattering by ice crystals


In his pioneering optics experiments, Newton (1704) demonstrated that the refraction
of light through a prism leads to its dispersion into component colors, as depicted in
Figure 3.1a. Light seen by the retina of the human eye, referred to as white light, covers fre-
quencies between 4.3 × 1014 and 7.5 × 1014 s−1 (wavelength 0.4–0.7 µm). When incident
white light enters a prism and undergoes two consecutive refractions, the outgoing light dis-
plays seven color spectra – red bends the least, violet bends the most, and indigo, blue, green,
yellow, and orange colors bend in between in a sequential manner. In Subsection 2.3.1,
we discussed the physical basis for light dispersion, leading to the definition of complex
refractive index. Ice absorption is negligibly small at visible wavelengths (see Figure 2.8),
so the optical properties of light can be defined by the real refractive index. In all optical
phenomena discussed in the following paragraphs, water drops, ice crystals, and aerosols in

115
116 Geometric optics for light scattering by ice crystals

(a) Refraction (b) Refraction/Reflection


Sunlight

White light

Prism dispersion of light


(Newton 1704) Rainbow (Airy 1838)

(c) Surface wave (Edge effect) (d) Diffraction


Sunlight
Sunlight

Glory (van de Hulst 1947) Corona (Babinet 1837)

(e) Refraction (f) Refraction

Sunlight
Sunlight






22o halo (Huygens 1662) 46o halo (Mariotte 1681)

Figure 3.1 (a) The dispersion of white light into component colors by two refractions through a
prism. (b) The production of the primary rainbow from light rays undergoing two refractions and one
internal reflection in a raindrop. (c) The generation of glories based on the concept of surface waves,
referred to as the edge effect. (d) The formation of coronas by means of diffraction. The production of
(e) a 22° halo and (f) a 46° halo by two refractions through prism angles of 60° and 90°, respectively.

the Earth’s atmosphere act like a prism and produce visible color sequences. Indeed, geo-
metric optics in terms of tracing light rays is essential to understanding physical processes
associated with light scattering by ice crystals.
The best known atmospheric optical phenomenon is the rainbow, an inspiration for
art and mythology in human history. Rainbows in the sky are usually formed on summer
3.1 Application of geometric optics to light scattering 117

afternoons after a rain shower. They are produced by two refractions and one internal
reflection of the sun’s light rays within raindrops, as illustrated in Figure 3.1b. The most
frequently observed rainbow (first rainbow) is due to the sun’s light rays undergoing
minimum deviation, referred to as the Descartes ray, resulting in maximum intensity at an
observed angle of 137° between the sun (see the thick white line with arrows) and raindrops,
causing the white light to disperse into component colors (red outside and blue inside).
Thus, in order for a rainbow to be observed, the sun must be behind the observer. When the
raindrops are sufficiently large, it is possible to observe a secondary rainbow at an angle of
130° (blue outside and red inside). A supernumerary bow, which often appears inside the first
rainbow, is produced by wave interference that cannot be explained by geometric reflection
and refraction. The physical explanation for the formation of rainbows and supernumerary
bows is attributed to Airy (1838), and known as the Airy theory of rainbows.
When water clouds are present, an optical feature known as a glory may be observed
on mountain tops as well as from aircraft. This optical feature requires 180° backscattering
geometry. Geometric reflection and refraction cannot explain the occurrence of a glory
because water’s refractive index of 1.33 cannot produce minimum-deviation rays. van
de Hulst (1947) suggested that glories are caused by grazing-incident rays that undergo
one internal reflection, as shown in Figure 3.1c. He postulated that the 14.4° gap between
the scattering angles of 165.6° and 180°, associated with the 1.33 refractive index, could
be bridged by surface waves, which are produced by interaction of the incident waves at
grazing angles near the edges of a sphere and continuation of the wave motion along its
surface into the shadow region, referred to as the edge effect. Nussenzveig (2002) added
further insight into the edge effect. The color sequence (red outside and blue inside) is
produced by wave interference.
The formation of the optical feature known as a corona, frequently seen around the sun
or moon, can be explained by light diffraction associated with water or aerosol particles, as
illustrated in Figure 3.1d. According to Babinet’s principle (Babinet 1837), diffraction by
a cloud particle is equivalent to that produced by an aperture of the same size and shape.
Colorful rings appear around the sun when particle size is uniform; however, the color
sequence differs from the 22° halo discussed below. After a huge volcanic eruption in 1883,
color rings (red outside and blue inside) called Bishop’s rings (Bishop 1884) appeared
around the sun. These rings were produced by the diffraction of sunlight associated with
volcanic aerosols having about the same size through the optical interference of diffracted
waves.
Light scattering by ice crystals produces numerous fascinating optical features associated
with their hexagonal shapes and orientation properties, as illustrated in Subsection 3.1.4.
We have identified the two most noticeable features: 22° and 46° halos (Figure 3.1e,f),
which have color rings around the sun with red inside and blue outside (Greenler 1980).
It appears that Mariotte (1681) was the first to attribute the formation of 46° halos to
equilateral triangular prismatic ice crystals. Huygens (1662) appears to have been the first
to present a quantitative explanation for the formation of 22° halos, the most commonly
observed optical feature generated by ice crystals in cirrus clouds.
118 Geometric optics for light scattering by ice crystals

It is quite clear that the principles of geometric optics, including diffraction and surface
waves, have been followed to identify essentially all optical features that have been observed
in the Earth’s atmosphere and to provide physical and mathematical explanations as to how
these phenomena are produced. In Chapter 1 we used geometric-optics fundamentals to
develop numerous approximations for light scattering by ice crystals covering all possible
sizes from 0.1 to 1000 µm (see Figure 1.23) and shapes (see Figure 1.19). Results
derived from these approximations have been verified with respect to those computed
from the “exact” Lorenz–Mie theory and other numerical and computational methods
presented in Chapter 4. Additionally, we have assessed geometric-optics results vis-à-
vis the values determined from light-scattering and ice-microphysics experiments in a
controlled laboratory setting. This chapter presents a systematic and logical development
and application of the theories of geometric optics, diffraction, and surface waves to light
scattering by ice crystals. Snell’s and Fresnel’s laws are first introduced.

3.1.1 Snell’s and Fresnel’s laws for geometric optics


We shall consider a plane wave arriving at a boundary which separates two homogeneous
media with distinct optical properties. This wave splits into two waves: a transmitted wave
entering the second medium and a reflected wave propagating back into the first medium.
From the part of the wave that touches the surface plane, we may isolate a narrow light
beam referred to as a ray that is much smaller than the surface. Let v1 and v2 be the
velocities of propagation in the two media such that v1 > v2 , and let θi and θt be the angles
corresponding to incident and refracted waves. On the basis of the geometry depicted in
Figure 3.2, we have

sin θi /sin θt = v1 /v2 = m, (3.1.1)

where m is the refractive index for the second medium relative to the first medium. This
equation represents the essence of Snell’s law, which relates incident and refracted angles
to the indices of refraction. As background, this law was named after the Dutch astronomer
Willebrord Snellius (1580–1626); however, according to Born and Wolf (1975), Snel-
lius presented this law in an unpublished manuscript. Apparently, it was first accurately
described in 984 by Ibn Sahl, a scientist at the Baghdad court, in the manuscript “On the
Burning Instruments” (Rashed 1993).
In order to derive the Fresnel equations, we consider the electric vector of the incident
field Ei , with incidence denoted by the superscript i. The components of the incident
electric field vector perpendicular (r) and parallel (l) to the plane comprised of incident and
refracted rays mapped in rectangular coordinates, as shown in Figure 3.2, are given by

Exi = −Eli cos θi , Eyi = Eri , Ezi = Eli sin θi . (3.1.2)

From the Maxwell equations, the relationship between electric and magnetic vectors
√ √
can be shown to be H = εa × E, or E = − 1/εa × H, where a is a unit vector in
3.1 Application of geometric optics to light scattering 119

Incident Reflected
Err
Eri r
i
El El

t θi
v1
θr 1
X
2
v 2t
θt

t
El
Ert

Transmitted (or Refracted)

Figure 3.2 Graphical representation of the reflection and refraction of a plane wave traversing from
medium 1 to medium 2. The choice of positive directions for the parallel components (l) of the electric
vectors is depicted in the figure. The perpendicular components are at right angles into the plane of
reference. The terms v1 t and v2 t denote the distance in media 1 and 2, respectively. The incident
angle is denoted by θi , while the refracted angle is denoted θt . The reflected angle θr = π − θi .

the direction of propagation. Thus, the components of the magnetic vector are as follows

(μ ≈ 1, m = ε):
Hxi = −Eri cos θi m1 , Hyi = −Eli m1 , Hzi = Eri sin θi m1 , (3.1.3)
where m1 is the refractive index of the first medium with respect to vacuum.
In a similar manner, if Et and Er denote the transmitted (t) and reflected (r) electric
vectors, respectively, then the following relationships can be derived:
Ext = −Elt cos θt , Eyt = Ert , Ezt = Elt sin θt ,
(3.1.4)
Hxt = −Ert cos θt m2 , Hyt = −Elt m2 , Hzt = Ert sin θt m2 ,
Exr = −Elr cos θr , Eyr = Err , Ezr = Elr sin θr ,
(3.1.5)
Hxr = −Err cos θr m1 , Hyr = −Elr m1 , Hzr = Err sin θr m1 ,
where m2 is the refractive index of the second medium with respect to vacuum and θr =
180◦ − θi , as displayed in Figure 3.2.
The boundary conditions require that the tangential components of E and H be contin-
uous [see Eq. (2.2.20)]. Hence, we must have
Exi + Exr = Ext , Eyi + Eyr = Eyt ,
(3.1.6)
Hxi + Hxr = Hxt , Hyi + Hyr = Hyt .
120 Geometric optics for light scattering by ice crystals

Upon substituting all the electric and magnetic components into Eq. (3.1.6), we obtain the
following four relationships:
 
cos θi Eli − Elr = cos θt Elt , Eri + Err = Ert , (3.1.7a)
 i   
m1 cos θi Er − Err = m2 cos θt Ert , m1 Eli + Elr = m2 Elt . (3.1.7b)

Using these four equations, solutions for the electric components of the reflected and
transmitted waves in terms of the incident wave are given by

Elr = Rl Eli , Elt = Tl Eli , Err = Rr Eri , Ert = Tr Eri , (3.1.8)

where the amplitude coefficients can be obtained as follows:


cos θi − m cos θt m cos θi − cos θt
Rr = , Rl = , (3.1.9a)
cos θi + m cos θt m cos θi + cos θt
2 cos θi 2 cos θi
Tr = , Tl = , (3.1.9b)
cos θi + m cos θt m cos θi + cos θt
with m = m2 /m1 , the refractive index of the second medium relative to the first medium.
Equations (3.1.9a) and (3.1.9b) represent the well-known Fresnel formulas (Fresnel 1819).
The Poynting vector for electromagnetic waves [Eq. (2.4.1)] is expressed as S =
(c/4π)E × H. Because H can be expressed in terms of E, the flux density |S| =

(c/4π) ε|E|2 (μ = 1). Thus, the amounts of energy incident, reflected, and transmitted
per unit area of the boundary per unit time (e.g., W m−2 ) are given by

F i = |Si | cos θi = (c/4π)m1 |Ei |2 cos θi , (3.1.10a)


F = |S | cos θr = (c/4π)m1 |E | cos θr ,
r r r 2
(3.1.10b)
F = |S | cos θt = (c/4π)m2 |E | cos θt ,
t t t 2
(3.1.10c)

where θr = 180◦ − θi . The reflected and transmitted portions of energy in two polar-
2
ization components with respect to the incident energy are proportional to Rr,l and
Tr,l m cos θt / cos θi , respectively. We can prove that Rr,l + Tr,l m cos θt / cos θi = 1, based
2 2 2

on the principle of energy conservation. The transmitted (or refracted) parts of the energy
can then be expressed as (1 − Rr,l 2
).

3.1.2 Absorption effects in the context of geometric optics


The geometric-optics approach postulates that within a particle the effect of absorption on
the propagation direction of a ray can be neglected, so the refracted angle and ray path length
can be computed from Snell’s law and the particle’s geometry. This approach is correct if
absorption is weak, as in the case of ice and water at most solar wavelengths. For cases of
strong absorption, rays refracted into the particle are almost totally absorbed, so that the
geometric-optics method can also be used to compute external reflection. The preceding
argument is physically correct in the limits of weak and strong absorption; however, we
3.1 Application of geometric optics to light scattering 121

normal of interface
incident wave reflected wave
er
ei

qi qr
Air
Ice
qt

planes of constant planes of constant


amplitude phase

et
ea refracted wave

Figure 3.3 Illustration of a light beam traveling from air to ice, a medium with absorption. The plane
of constant amplitude of the refracted wave is parallel to the interface, which is referred to as the
equi-amplitude plane (Born and Wolf 1975). At the same time, the direction of phase propagation for
the inhomogeneous wave inside the medium is determined by Snell’s law. The notations ei , et , and
eα represent the directions of the incident wave, the refracted wave, and the normal to the interface,
respectively. The angles θ i , θ r , and θ t are defined in the figure.

shall consider the general absorption effect in the context of geometric optics based on
fundamental electromagnetic wave theory. The effect of the complex refractive index on
geometric optics has been formulated only for the Fresnel coefficients, as shown by Stratton
(1941) and Born and Wolf (1975). Following Yang and Liou (1995) and Liou (2002), we
present Snell’s and Fresnel’s laws under absorption conditions in terms of modified real
and imaginary indices of refraction. See also Dupertuis et al. (1994) and Chang et al.
(2005) for discussions of inhomogeneous wave properties in the context of geometric
ray-tracing.
Referring to Figure 3.3, consider the propagation of an incident wave from air into
ice. The wave vectors associated with the incident and reflected waves are real quantities
because these waves, which are outside the ice medium, must have the same properties.
However, the wave vector of the wave refracted into the ice is complex, and the result is
referred to as the inhomogeneity effect. The three wave vectors can be defined by

ki = kei , kr = ker , kt = kt et + ikα eα , (3.1.11)

where ei , er , et , and eα are unit vectors defined in Figure 3.3; k = 2π/λ, where λ is the
wavelength in air; and kt and kα are two real parameters which determine the complex
wave vector of the refracted wave. For non-absorptive cases, kα is zero. The corresponding
122 Geometric optics for light scattering by ice crystals

electric vectors can be expressed as

Ei (r) = Ai exp(ikr · ei ), (3.1.12a)


Er (r) = Ar exp(ikr · e ), r
(3.1.12b)
Et (r) = At exp(ikr · e + ikα r · e ),
t α
(3.1.12c)

where Ai , Ar , and At denote amplitude vectors. Furthermore, we may define the following
normalized parameters:

Nr = kt /k, Ñi = kα /k. (3.1.13)

At the interface of the two media, where the position vector is denoted rs , the phases
must be the same for incident, reflected, and refracted waves. Thus, from Eqs. (3.1.11) and
(3.1.13), we obtain

ei · rs = er · rs = Nr (et · rs ) + i Ñi (eα · rs ). (3.1.14)

Because the wave vectors for the incident and reflected waves are real, we must have

ei · rs = er · rs = Nr (et · rs ), eα · rs = 0. (3.1.15)

On the basis of the geometry defined by Eq. (3.1.15), a generalized form of Snell’s law
can be derived and is given by
sin θi
sin θi = sin θr , sin θt = , (3.1.16)
Nr
where θi , θr , and θt denote the incident, reflected, and refracted angles, respectively
(Figure 3.3). The vector eα in Eq. (3.1.15) is normal to the interface of the two media.
It follows that the planes of constant amplitude of the refracted wave are parallel to the
interface. To determine Nr and Ñi , we use the electric field of the refracted wave, which
must satisfy the wave equation in the form [see Eq. (2.2.9)]

∇ 2 Et (r) + k 2 (mr − imi )2 Et (r) = 0, (3.1.17)

where mr and mi are real numbers representing the real and imaginary parts of the refractive
index, respectively. Substituting Eq. (3.1.12c) into Eq. (3.1.17) and using Eq. (3.1.13) leads
to the following relationships:

Nr2 − Ñi2 = m2r − m2i , Nr Ñi cos θt = mr mi . (3.1.18)

If we let Ni = Ñi cos θt , then we obtain from Eqs. (3.1.16) and (3.1.18) the following
expressions:

2( 2 ) 2 *1/2 +1/2 mr mi
Nr = mr − m2i + sin2 θi + m2r − m2i − sin2 θi + 4m2r m2i , Ni = .
2 Nr
(3.1.19)

These two parameters are referred to as the adjusted real and imaginary indices of refraction.
3.1 Application of geometric optics to light scattering 123

(a) (b)

110o
130o

Figure 3.4 Solid lines are reflection/refraction determined from effective indices of refraction, while
dotted lines are based on Snell’s law, given by sin θt,1 = sin θi,1 /mr and sin θt,j = mr sin θi,j , j =
2, 3, 4 . . . The ice crystal’s complex index of refraction is (a) 1.3857 − i 0.422; (b)1.0925 − i 0.248
(after Yang and Liou 2009).

The refracted wave given in Eq. (3.1.12c) can now be rewritten in the form

Et (r) = At exp(−kNi la + ikNr et · r), (3.1.20)

where la = (eα · r)/ cos θt is the distance of propagation of the refracted wave along the
direction et . The direction of the phase propagation for the inhomogeneous wave inside the
medium is determined by Nr via Snell’s law, whereas the attenuation of wave amplitude
during propagation is determined by Ni . Consequently, the refracted wave can be precisely
traced. The Fresnel reflection and refraction coefficients defined in Eqs. (3.1.9a, b) in terms
of the adjusted real and imaginary indices of refraction can then be written as follows:
Nr cos θi − cos θt 2 cos θi
Rl = , Tl = , (3.1.21a)
Nr cos θi + cos θt Nr cos θi + cos θt
cos θi − Nr cos θt 2 cos θi
Rr = , Tr = . (3.1.21b)
cos θi + Nr cos θt cos θi + Nr cos θt
In these equations, the subscripts l and r denote the parallel and perpendicular polarized
components, respectively.
Yang and Liou (2009) further developed recurrence formulas for the effective indices of
refraction associated with high-order reflection/refraction events when the ray transmission
is from an ice crystal to air. Figure 3.4a depicts a 2-D case for propagation of an incident ray
into an ice hexagon using a complex index of refraction of 1.3857 − i 0.422, corresponding
to a 12 µm wavelength, while Figure 3.4b is for a complex index of refraction of 1.0925 −
i 0.248, corresponding to an 11 µm wavelength. The incident angles for the first-order
reflection/refraction event in the computation were set at 65° and 55° (i.e., θi,1 + θr,1 =
130◦ and 110°), as shown in Figure 3.4a and Figure 3.4b, respectively. The solid lines
show the results computed using the effective indices of refraction for corresponding
orders of reflection/refraction events. The dotted lines depict the results computed from
124 Geometric optics for light scattering by ice crystals

Table 3.1 The real and imaginary parts of the effective index of refraction for
Figure 3.4a (case 1) and Figure 3.4b (case 2).

Case 1 Case 2
mr = 1.3857 and mi = 0.422 mr = 1.0925 and mi = 0.248
θi,1 = 65◦ θi,1 = 55◦
Order of reflection/
refraction event, j Nr,j Ni,j Nr,j Ni,j

1 1.423 4 .410 83 1.121 3 .241 63


2 1.394 2 .419 44 1.094 0 .247 66
3 1.394 2 .419 44 1.094 0 .247 66
4 1.420 2 .411 75 1.094 0 .247 66
5 1.395 3 .419 10 1.118 3 .242 27

the simplified Snell’s law given by θi,j = θr,j , j = 1, 2, 3 . . .; sin θt,1 = sin θi,1 /mr ; and
sin θt,j = mr sin θi,j , j = 2, 3, 4 . . . under the condition that total reflection does not occur.
The effect of wave inhomogeneity on the propagation of a localized ray is substantial for
higher-order reflection/refraction events. However, its effect on the scattering properties of
the particle is relatively small because it is significant only when an ice particle’s absorption
is substantial. For a strongly absorbing particle, the energy carried by a higher-order ray
is much lower than that carried by a lower-order ray, in view of the fact that a particle’s
absorption coupled with external reflection dominate the scattered field.
In Table 3.1 the real and imaginary parts of the effective index of refraction are listed for
cases 1 and 2, corresponding to Figures 3.4a and 3.4b, respectively. The terms Nr,j and Ni,j
are larger and smaller than mr and mi , respectively. The dependence of the effective index
of refraction on ray history is also evident from the results listed in the table. There is no
systematic change in Nr,j and Ni,j as a function of the order of reflection/refraction events.
By matching these events for the results in this table with the geometries displayed in
Figure 3.4, the effective indices of refraction for two sequential internal reflections are the
same if two hexagonal faces for these reflections are parallel. In summary, the effective
indices of refraction are dependent on both ray history and particle shape.

3.1.3 Geometric ray tracing and the Monte Carlo hit-and-miss approach
to ice crystals
Use of the Monte Carlo method in connection with geometric ray-tracing was first developed
by Wendling et al. (1979) for hexagonal ice columns and plates. See also Marchuk and
Mikhailov (1967). Takano and Liou (1995) and Yang and Liou (1998a) further applied
the hit-and-miss Monte Carlo method to trace photons in complex ice crystals, including
contributions from absorption and polarization.
3.1 Application of geometric optics to light scattering 125

(a) (b)

(c) (d)

Figure 3.5 Positions of incident photons on each crystal, determined using random numbers:
(a) hollow column, (b) bullet rosette, (c) dendrite, and (d) capped column (after Takano and Liou
1995). See Subsection 1.5.3 for the computer-generated ice crystal sizes and shapes.

We may consider a bundle of parallel rays, representing a flow of photons, which


strike an ice crystal from a direction defined by a set of two angles (α, β), where α is the
complementary angle between the incident direction and the crystal’s c-axis and β is the
azimuthal angle with respect to this axis. Consider also a plane normal to this bundle of
incident rays and the geometric shadow of an ice crystal projected onto this plane. We may
let a rectangle (defined by dimensions X and Y) enclose this geometric shadow, such that
the center of this rectangle coincides with the center of the crystal. One of the sides, X,
is parallel to the geometric shadow of the crystal’s principal axis. A point (xi , yi ) can be
subsequently selected inside this rectangle using random numbers, RN (0–1), such that
   
xi = X RN − 12 , yi = Y RN − 12 . (3.1.22)

In this manner, xi varies from −X2 to X2, while yi varies from −Y2 to Y2. If the point
is inside the geometric shadow of the crystal, it is regarded as an incident point on the
crystal; otherwise, it is disregarded. If there are more than two crystal planes for a photon,
the point closer to the light source is regarded as the incident point.
Figures 3.5a–3.5d show the positions of incident photons for hollow column, bullet
rosette, dendrite, and capped column. The coordinates of an incident point (xi , yi ) in these
figures can be transformed to the coordinates (x, y, z) with respect to the body-framed
126 Geometric optics for light scattering by ice crystals

coordinate system (Takano and Asano 1983) and can be expressed in terms of xi , yi , α, β,
and one unknown, zi , in the form
⎡ ⎤ ⎡ ⎤⎡ ⎤
x sin α cos β −sin β cos α cos β xi
⎣ y ⎦ = ⎣ sin α sin β cos β cos α sin β ⎦ ⎣ yi ⎦ . (3.1.23)
z −cos α 0 sin α zi
A crystal surface can generally be defined by the following equation:

a0 x + b0 y + c0 z − d0 = 0, (3.1.24a)

where the coefficients a0 , b0 , c0 , and d0 are determined by the coordinates of three


points on the crystal surface. In particular, a plane containing the points P1 (x1 , y1 , z1 ),
P2 (x2 , y2 , z2 ), and P3 (x3 , y3 , z3 ) can be expressed as
 
 x − x1 y − y1 z − z1 
 
 x1 − x2 y1 − y2 z1 − z2  = 0. (3.1.24b)
 
x − x y − y z − z 
2 3 2 3 2 3

For example, one of the side planes of a dendritic crystal branch √ can be defined by an
equation obtained
√ from three of the four points: [−(a + bb)/2, 3(a − bb)/2, ±L/2] and
[−(a + bt)/2, 3(a + bt)/2, ±L/2], where +L/2 denotes the upper point and −L/2
denotes the lower point. See Figure 1.19 for the graphic definitions of L, a, bt, bb, t, d,
H, and other notation defining the four ice crystal shapes depicted in Figure 1.22. On
substituting (x, y, z) in Eq. (3.1.23) for (x, y, z) in Eq. (3.1.24a), the unknown, zi , can be
evaluated. Using this zi and (xi , yi ), the coordinates of the incident point (x, y, z) can be
calculated using Eq. (3.1.23).
Once the incident coordinates have been determined, the photons are traced using the
hit-and-miss Monte Carlo method. The Fresnel reflection coefficients, Rl and Rr , are first
calculated and compared with a random number, RN. If (|Rl |2 + |Rr |2 )/2 is greater than
RN, the photon is reflected; otherwise, it is transmitted. When a photon traverses a crystal, it
can be absorbed. We can account for absorption by means of a stochastic procedure. When
a photon enters a crystal, an absorption length, la , is generated from a random number such
that
ln(RN)
RN = exp(−2kNi la ), i.e., la = − . (3.1.25)
2kNi
The random number represents the probability of transmission of the photon. The absorption
path length, la , represents the distance traversed by the photon in the crystal before it is
absorbed. An actual path length, l, between an incident point and the next internal incident
point can then be calculated on the basis of Snell’s law and the specific ice-crystal geometry.
The transmission is given by T = exp(−2kNi l). If T ≤ RN ≤ 1, then the photon associated
with this RN is absorbed. Similarly, if l is greater than la , then the photon is absorbed;
otherwise, it is transmitted without absorption. This procedure is repeated as long as photons
travel inside the crystal.
3.1 Application of geometric optics to light scattering 127

After a photon is transmitted out of the crystal or reflected externally, it can re-enter
the crystal depending on the crystal’s shape. In this case, a new incident direction can be
calculated using the direction cosine of the scattered beam. The new incident coordinates
can also be determined from the new incident direction and the coordinates of an emergent
point of the photon on the crystal surface. The foregoing procedure is repeated until the
photon escapes from the crystal. When a photon re-enters the crystal, the scattering angle
and the scattering matrix are computed with respect to the original incident direction. In the
conventional method, the number of scattered photons per unit solid angle, 2π sin θ θ, is
counted as the phase function. The single-scattering albedo is obtained from the ratio of the
number of scattered photons to the number of incident photons. The Monte Carlo method
allows us to treat complicated ice crystals effectively and can be employed in connection
with the improved geometric ray-tracing approach.
The surfaces of ice crystals may not be exactly smooth, particularly if they undergo
collision and riming processes. Also, a careful electron-microscopic examination of some
polycrystalline ice crystals reveals rough structures on the surfaces. Halo and arc patterns
that are absent from some cirrus clouds could be caused by deviations of the ice crystal
surfaces from defined hexagonal structures. Some aspects of ice crystal surface roughness
can be incorporated into geometric ray-tracing by following an idea developed by Cox and
Munk (1954) for wavy sea surfaces. A rough surface may be thought of as comprised of
a number of small facets that are locally planar and randomly tilted flat surfaces. We may
use a two-dimensional Gaussian probability function to define the surface tilt as follows:
% &
1 zx2 + zy2
p(zx , zy ) = exp − , (3.1.26a)
πσ 2 σ2

where σ is a parameter determining the degree of roughness and the parameters zx and zy
are the slopes defined for a facet of rough surface along two orthogonal directions, given
by
∂z ∂z
zx = = tan θ cos φ, zy = = tan θ sin φ, (3.1.26b)
∂x ∂y
where θ and φ are the local polar angles defining the position of the tilt of the surface facet.
In general, surface roughness of ice particles has the effect of smoothing out the scattering
maxima that occur in the phase function (see Subsection 3.4.2).

3.1.4 Illustration of ice optics in the atmosphere


The geometric-optics approach to ray tracing provides a means of identifying all the fasci-
nating cloud optics we see in the atmosphere on a daily basis, as introduced in Section 3.1.
Rainbow and glory features produced by scattering by water drops can be explained using
geometric ray tracing. Ice crystals are much more complex in view of their non-spherical
and irregular shapes, as displayed in Section 1.5. Ice crystal non-sphericity also leads to
orientation issues, which cannot be directly observed using in situ sampling techniques.
128 Geometric optics for light scattering by ice crystals

(a) (b) A Incident ray


θ'
 
θi θ'i
 22o Halo θt θ't

Scattered ray

46o Halo

Figure 3.6 (a) The production of 22° and 46° halos from two refractions through a plate crystal
with prism angles A of 60° and 90°, respectively. (b) Definition of the minimum deviation, which
generates the maximum intensity; see text for the definition of the angles in this diagram.

Moreover, it should be pointed out that geometric optics is the only approach in light scatter-
ing that will provide physical explanations of the features that occur in the phase function
computed from “exact” solutions (e.g., the Lorenz–Mie theory) or numerical methods
(e.g., finite-difference time domain method). The fascinating optical phenomena produced
by defined ice crystals are numerous, and are presented in the following subsections.

3.1.4.1 22° and 46° halos


First, we discuss the production of 22° and 46° halos, which occur when ice crystal clouds
are present in the atmosphere. Referring to Figure 3.6, if a light beam from the sun (or
moon) passes through a prism of angle A in a plane at right angles to the refracting edge,
the deviation angle is

θ  = (θi − θt ) + (θi − θt ) = 2θi − A. (3.1.27a)

The minimum deviation (maximum intensity) occurs when


dθ  dθ  i
=0=1+ . (3.1.27b)
dθi dθi
Since A = θt + θt , we also have 1 + dθt  /dθt = 0. An obvious solution of the last two
equations is when θi = θi and θt = θt . Thus, from Eq. (3.1.27a), the incident angle at which
minimum deviation occurs is θi = (θ  + A)/2. Also, the angle of refraction at minimum
deviation is given by θt = A/2. Applying Snell’s law as denoted in Eq. (3.1.1) yields

sin[(θ  + A)/2] = m sin A/2. (3.1.28)

This equation represents the basic relationship for finding the minimum deviation from
a known index of refraction and prism angle. As illustrated in Figure 3.6, the possible prism
angles are 60°, 90°, and 120°. However, owing to the property of the sine function and the
fact that the refractive index of ice is about 1.31 in visible wavelengths, a prism angle of
120° cannot produce minimum deviation. The common halo has an angular radius of 22°,
3.1 Application of geometric optics to light scattering 129

produced by two refractions in a hexagonal prism. The halo of 46° is generated by two
refractions in a rectangular prism. Since the index of refraction varies with wavelength,
white light is dispersed into its component colors, with red refracted least and blue refracted
most. Halos are seen in the form of circles because of the random orientations of hexagonal
prisms in the atmosphere.

3.1.4.2 Sundogs and their geometry


Second, microphysics observations from laboratory ice clouds have revealed that hexagonal
plates and long columns fall through the air with their major axes parallel to the ground and
would most likely have a random orientation in the horizontal plane. A light beam which
reaches a sheet of ice columns or plates randomly oriented in the horizontal plane will then
produce bright spots along conical circles at the angle of minimum deviation around the
sun or the moon. When the sun is close to the horizon and thin cirrus clouds are present,
colored streaks may sometimes be observed at greater than 22°, at the same elevation as the
sun. This optical phenomenon is referred to as the parhelia of 22° or, commonly, sundogs
or mock suns. Sundogs consist of only two bright spots, since other light from the sun
deviated by clouds higher above the horizon cannot be observed. To evaluate their position,
we shall now refer to Figure 3.7, which illustrates the relevant geometry of a plate crystal
with respect to the sun’s position.
Let the lines AO and A O in Figure 3.7a be the incident (skew ray) and refracted
directions, respectively, and consider the enclosure AOCB. We can determine the following
relationships:
sin εi sin εt
sin θi = , sin θt = . (3.1.29a)
sin  sin 
The elevation angle εi denotes the position of the sun, and other angles are defined in the
figure. Substituting Eq. (3.1.29a) into mr sin θt = sin θi [Eq. (3.1.1), Snell’s law], we obtain

mr sin εt = sin εi . (3.1.29b)

On the basis of the geometry of the incident (skew ray) and refracted rays defined in
Figure 3.7a, we can further show that

cos θi = cos εi cos φi , cos θt = cos εt cos φt . (3.1.30)

Eliminating θ i and θ t from Eq. (3.1.30) and the Snell equation, we obtain

mr (cos εt /cos εi ) sin φt = sin φi . (3.1.31)

Set mr (cos εt /cos εi ) = mr on the left-hand side of Eq. (3.1.31). Then, from Eq. (3.1.29b),
this term can be rewritten as follows:

 m2r − sin2 εi
mr = . (3.1.32a)
cos εi
130 Geometric optics for light scattering by ice crystals

(a) A


Skew ray

εi A'
θi Δ
Q θt
 
Z εt  B
Sundog   B'
O φi φt C
(> 22o halo) 
OQBC XYZ
Y
X A'B', AB OQBC
(b) A

Emergent ray
Incident ray leaving prism


θ'/2 B
εi D
θ'h /2
O 

Figure 3.7 (a) The geometry of a skew ray that produces a sundog in a 3-D plate crystal, and
the definition of the sun’s elevation angle εi . (b) The definition of incident and emergent rays with
reference to the elevation angle and the minimum deviation, and their mapping onto a horizontal
plane.

When mapping on a horizontal plane, Eq. (3.1.31) can also be expressed in the form

mr sin φt = sin φi . (3.1.32b)

Referring to Eq. (3.1.27b), the 22° parhelion occurs if the following conditions are met:
(φi = φi ; φt = φt ). If A is the angle of the prism, we have the following relationship:

φt + φt = A. (3.1.33)

Using Eq. (3.1.33), the minimum deviation angle projected onto the plate’s basal plane
(QOCB) is given by

θh = φi − φt + φi − φt = 2φi − A. (3.1.34)

Thus, we obtain

φi = (θh + A)/2. (3.1.35)


3.1 Application of geometric optics to light scattering 131

From Eq. (3.1.33), we have φt = A/2. On substituting this and Eq. (3.1.35) into
Eqs. (3.1.31) and (3.1.32b), we obtain


1  A
sin (θh + A) = mr (cos εt /cos εi ) sin , (3.1.36a)
2 2


1  A
sin (θh + A) = mr sin . (3.1.36b)
2 2
Moreover, based on the geometry depicted in Figure 3.7b, we can show that

cos θ  = sin2 εi + cos2 εi cos θh . (3.1.37a)

This equation can be rewritten as follows:


θ θ
sin = cos εi sin h . (3.1.37b)
2 2
It follows that the angles θh and θ  can be computed from Eqs. (3.1.29b), (3.1.36a), and
(3.1.37b). For the prism angle A = 60° and the sun’s elevation angle εi = 30°, the angular
and azimuthal distances θh and θ  for a sundog are 24.54° and 28.41°, respectively, for
the 0.656 µm red wavelength and 25.40° and 29.41°, respectively, for the 0.405 µm violet
wavelength.

3.1.4.3 Horizontally oriented plate and column crystals


Third, in the case of horizontally oriented (2-D) plates, the scattered sunlight is confined
to four latitude belts due to specific geometry. Based on geometric ray-tracing, the zenith
angle θ ∗ can be computed from the incoming solar zenith angle θi in the forms (Takano
and Liou 1989b)
  1/2  1/2

π/2 − sin−1 m2r − sin2 θi , for θi > sin−1 m2r − 1  58◦
θ =  1/2  1/2 (3.1.38)
sin−1 m2r − cos2 θi , for θi < cos−1 m2r − 1  32◦ .
For example, if θi = 75◦ , the latitude belts correspond to zenith angles of
±75◦ and ± 27◦ with negative values representing mirror images. If the incident angle
is 27°, the four latitude belts are ±27◦ and ±75◦ . Due to the symmetrical property of 2-D
plates (randomly oriented with their c-axes vertical) with respect to the incoming light
beam, all scattered light is also confined to the four latitude belts. At the 75° emergent
zenith angle, the subsun, subsundog, 120° subparhelion, and antisolar peak optical features
can be observed. In addition to these optical phenomena, the anthelion (AN) located at the
180° azimuthal angle can also be observed, due to the scattering and coupling of the subsun
and antisolar peaks. At the 27° zenith angle, the Kern’s arc (KA) can be seen. The subsun
feature and sun pillars are produced by external reflections from the flat horizontal faces of
ice plates, and appear as streaks of white light above and/or below the sun.
The subpeak at θi = 82◦ in the transmitted light is caused by the lower sunvex Parry
arc of the subsun. For Parry columns (the c-axis and a pair of prisms facing horizontally),
132 Geometric optics for light scattering by ice crystals

Halos and arcs related to 22o


22o Halo, Upper & Lower 22o Halo, Sundog Subsundog
Tangent Arcs, Parry Arcs

 
  


Halos and arcs related to 46o


46o Halo 46o Halo 46o Halo 46o Halo
Circumzenithal Arc Circumhorizontal Arc Supralateral Arc Infralateral Arc

 

 
 


Other halos and arcs


Subsun Parhelic Circle Subparhelic Circle


 
  
   
 



120o Parhelion Helic Arc Subhelic Arc Wegener’s & Hastings’


Anthelic Arc
  

 
  
  


 

Figure 3.8 A summary of the origins of frequently observed halos and complex arcs produced by
horizontally oriented columns and plates, in terms of reflection and refraction. Dots denote points of
incoming, refracted, and outgoing light rays, while dashed lines denote ray paths inside a plate or
column crystal.

several optical features are observable. These include the subsun (SS) and lower sunvex
Parry arc (LSVP) in the reflected light, and the circumzenith arc (CZA) and upper suncave
(USCP) and sunvex (USVP) Parry arcs in the transmitted light. For 2-D columns (rotational
orientation about the c-axis in a horizontal plane), the lower tangent arc (LTA) is noticeable
in the reflected light. Aside from these features, the reflected and transmitted light of 2-D
columns is similar to that for randomly oriented columns.
The physical causes of the preceding optical features generated by “smooth” 2-D plate
and column crystals and other crystal types are displayed in Figure 3.8 in terms of external
reflections, two refractions, and internal reflections, based on the principles of geometric
optics. These features are summarized in three major groups: 22° halos and associated arcs,
46° halos and associated arcs, and other complex halos and arcs.
3.1 Application of geometric optics to light scattering 133

(a) Observation

Simulation
(b) q
p
m
n
o
l
k
f
g
e
i
b
d
h
j
c
r
a

Figure 3.9 (a) Halo and arc complex observed in Antarctica by Riikonen (1999). (b) A geomet-
ric ray-tracing halo simulation program developed by Les Cowley and Michael Schroeder (2009,
HaloSim computer ray tracing, Les Cowley – http://www.atoptics.co.uk) was employed for simula-
tion purposes. We have identified (a) 22° halo, (b) 46° halo, (c) 22° parhelion, (d) upper tangent arc,
(e) Parry arc, (f) circumzenithal arc, (g) supralateral arc, (h) infralateral arc, (i) Parry supralateral arc,
(j) Parry infralateral arc, (k) helic arc, (l) subhelic arc, (m) parhelic circle, (n) Wegener arc, (o) 120°
parhelion, (p) anthelion, (q) diffuse arc, and (r) a diamond shape, which is a Fraunhofer diffracted
pattern overlapping a weak sun pillar (courtesy of Yoshi Takano). See Figure 3.8 for the causes of
these fascinating halos and arcs.
134 Geometric optics for light scattering by ice crystals

3.1.4.4 An observed halo and arc complex pattern


Fourth, we present halo and arc complex patterns observed in Antarctica by Riikonen
(1999), as depicted in Figure 3.9a. Most of these patterns are rare and do not appear in
daily observations; however, they cover essentially all possible optical features produced by
ice crystals in the Earth’s atmosphere. Geometric ray-tracing computer simulation results
are displayed in Figure 3.9b, where the names of these fascinating optical phenomena
are: (a) 22° halo, (b) 46° halo, (c) 22° parhelion, (d) upper tangent arc, (e) Parry arc,
(f) circumzenithal arc, (g) supralateral arc, (h) infralateral arc, (i) Parry supralateral arc,
(j) Parry infralateral arc, (k) helic arc, (l) subhelic arc, (m) parhelic circle, (n) Wegener arc,
(o) 120° parhelion, (p) anthelion, (q) diffuse arc, and (r) a diamond shape, which is a
diffracted pattern overlapping a weak sun pillar. Physical causes of these features in terms
of reflection and/or refraction are illustrated in Figure 3.8.

3.2 Fraunhofer diffraction for light scattering by ice crystals


The theoretical development of diffraction begins with Babinet’s principle, which states
that the diffraction pattern in the far field, i.e., Fraunhofer diffraction (Fraunhofer 1821),
from a circular aperture is the same as that from an opaque disk or a sphere of the same
radius. Babinet (1837) stated the diffraction principle by citing the corona produced by
water droplets in the atmosphere. Babinet’s principle can take scalar or vector form.
The vector form accounts for the sign of electric and magnetic fields. However, when
the product of wavenumber and diffracting particle size is larger than about 1, the scalar
Babinet principle is sufficient for analysis (Jackson 1975). Moreover, the conditions for,
and limitations on, Fraunhofer diffraction have been discussed by Born and Wolf (1975).
The diffracted images for circular and rectangular apertures were presented in Born and
Wolf (1975). Diffracted images for other polygonal apertures were studied by Komrska
(1972, 1973) and Smith and Marsh (1974). Cai and Liou (1982) performed the computation
of diffracted intensity for randomly oriented hexagonal cylinders, while Takano and Asano
(1983) studied diffraction phenomena produced by randomly or horizontally oriented ice
crystals.
Liou et al. (1983b) extended their numerical computations on diffraction by hexagonal
cylinders to account for cubes and parallelepipeds. Takano and Liou (1989a) illustrated the
manner in which diffracted intensity can be added to reflected and/or refracted intensity
when the phase function for ice crystals produces a delta-function transmission at the 0°
forward scattering direction. Parviainen et al. (1994) discussed the connection of vertically
elliptical coronas with horizontally oriented pollens. While carrying out computation of the
Fraunhofer integral, Tränkle and Mielke (1994) simulated corona patterns using elliptical
shapes and produced strong brightening for birch and pine pollens. Additionally, Bi et al.
(2011a) showed good agreement between results determined from Fraunhofer diffraction
and a surface-integral method for an absorbing cube. Also, Hesse et al. (2012) presented
polar scattering patterns of diffracted and externally reflected intensities for a strongly
absorbing cube.
3.2 Fraunhofer diffraction by ice crystals 135

x
y

Plane waves
O' r
P
φ
r0
θ
z
O

r0 - r

A
u0 e-ikz

Figure 3.10 Diffraction of plane waves, denoted by u0 e−ikz , by a circular aperture with geometrical
area A, evaluated at the position P. The shaded area is perpendicular to the aperture defined at the x-y
plane. See text for notation definition.

In Subsection 3.2.1, we shall first review Fraunhofer diffraction from a circular


aperture.

3.2.1 Sphere
Let the z-axis be in the direction of propagation of the incident light, and let the wave
disturbance occur at a point P to the right of a geometric aperture of area A (Figure 3.10).
The distances from point P to point O (x, y) on the aperture area and to the origin O are
denoted r and r0 , respectively. In the far field, the light-wave disturbance at P can be derived
from Fraunhofer diffraction theory following Born and Wolf (1975) as follows:
 
up = −uc e−ikr dxdy, (3.2.1)
A
where uc = iu0 /rλ, u0 represents the disturbance in the original wave at point O on the
plane wave front, k = 2π/λ, and λ is the wavelength. The phase difference at P between
waves passing through points O and O can be expressed in the form
δ = k(r − r0 ) = k(x cos φ + y sin φ) sin θ, (3.2.2)
where θ is the angle between lines OP and the z-axis, and φ is the azimuthal angle on the
aperture. Substituting Eq. (3.2.2) into Eq. (3.2.1), we obtain
 
−ikr0
up (sphere) = uc e exp[−ik(x cos φ + y sin φ) sin θ ]dxdy. (3.2.3)
A
For a circular aperture, we can transfer rectangular coordinates (x, y) to polar coordinates
(ρ, ψ) such that x = ρ cos ψ and y = ρ sin ψ to obtain
 a  2π
up = −uc e−ikr0 exp[−ikρ cos(ψ − φ) sin θ ]ρdρdψ. (3.2.4)
0 0
136 Geometric optics for light scattering by ice crystals

Using the zero-order Bessel function defined by


 2π
1
J0 (y) = eiy cos α dα, (3.2.5)
2π 0
Eq. (3.2.4) can be expressed in the form
 a
up = −uc e−ikr0 2π J0 (kρ sin θ )ρdρ. (3.2.6)
0

Moreover, Bessel functions have the following well-known recurrence relationship:


d
[yJ1 (y)] = yJ0 (y), (3.2.7)
dy
where J1 (y) is the first-order Bessel function. Upon integration, we have
 y
y  J0 (y  )dy  = yJ1 (y). (3.2.8)
0

Thus, from Eqs. (3.2.6) and (3.2.8), we obtain


up (sphere) = −2uc e−ik0 AJ1 (y)/y, (3.2.9)
where the geometric shadow area A = πa 2 , y = x sin θ, and the size parameter x = ka.
The scattered intensity in terms of the incident intensity I0 = |u0 |2 is then given by
ip
Id = |up |2 = I0 , (3.2.10)
k2r 2
where the diffraction angular intensity function for a single sphere is defined by
x4
ip = [2J1 (y)/y]2 . (3.2.11)
4
Diffraction depends only on the particle size parameter and is independent of the index
of refraction. If we display [2J1 (y)/y]2 = D 2 versus y, a principal maximum of 1 at y = 0
(i.e., θ = 0) will show and oscillates with gradually diminishing amplitude as y increases.
When J1 (y) = 0 and D 2 = 0, a number of minima diffraction patterns are produced. The
positions of maxima are defined by the values of y that satisfy
d 2 d
D = [2J1 (y)/y]2 = 0. (3.2.12)
dy dy
The maxima in the diffraction patterns have been used to explain the position of the corona
produced by the diffraction of uniform sized spherical aerosols.
The following analytical solutions and analyses are based on the presentation given in
Takano et al. (2012).

3.2.2 Rectangular parallelepiped


Consider a plane wave incident on a rectangular parallelepiped from a direction with zenith
angle (π2 − α) and azimuthal angle β measured with respect to the coordinate system
3.2 Fraunhofer diffraction by ice crystals 137

(a) (b) y’
y’
5 y2 6
x’=a1y’-b1 x’=a2y’+b2
4 y2 8

x’ = a1y’ - b1 x’ = a2y’ + b2 y1
1
4
2a
1 y1 x’=a3y’+b 3
1’  5
3 2
2c z’
z’
O
O x’ ψ x’
2b
7
7’
3
7
10
2 2c 2a
6

9 8

y’
(c) x
(d)

v

a z’ y
ζ

b
x’ p
v cosζ q

c z O η c x’
ζ O z
ξa
y’ a
x

Figure 3.11 These four figures depict the geometric shadow of (a) a rectangular parallelepiped
normal to the incident direction; (b) a hexagonal cylinder normal to the incident direction (after
Takano and Asano 1983); (c) a spheroid along the semi-major axis with the semi-major and semi-
minor axes denoted by c and a, respectively, where c is along the rotational axis defined by z and the
angle ζ is defined in the figure; and (d) a three-axis ellipsoid projected onto a plane normal to the
incident direction (after Takano et al. 2012).

fixed to the parallelepiped, as shown in Figure 3.11a, where the x-, y-, and z-axes are
directed, respectively, through the centers of three sets of rectangular planes. The origin
O of the coordinate system is taken to be the center of the parallelepiped. The geometric
shadow of the parallelepiped as projected onto a plane normal to the incident direction can
be expressed in terms of the (x , y ) coordinates of its marginal vertices in a new coordinate
system where the z -axis is taken along the incident direction.
The transformation of the body-framed coordinate system X(x, y, z) into the new coor-
dinate system X (x , y , z ) can be written in the form

X = CDX, (3.2.13)
138 Geometric optics for light scattering by ice crystals

where the transformation matrices are given by


⎡ ⎤
sin α 0 −cos α
C=⎣ 0 1 0 ⎦, (3.2.14)
cos α 0 sin α
⎡ ⎤
cos β sin β 0
D = ⎣ −sin β cos β 0 ⎦ . (3.2.15)
0 0 1
Thus, if the size and shape of the parallelepiped and the propagation direction (α, β)
of the incident wave are known, the coefficients a1 , a2 , b1 , b2 , y1 , and y2 defined in
Figure 3.11a can be determined by Eq. (3.2.13). The shape of a parallelepiped is defined by
lengths a, b, and c, which are half-lengths of a rectangular parallelepiped along the three
axes. The size parameter is given by ka. Having defined a number of geometric variables,
we may rewrite the diffraction integral equation of Eq. (3.2.3) in the form
 
−ikr0
up (parallelepiped) = uc e exp[−i(Xx + Y y)] dxdy, (3.2.16)
Ap
where
X = k sin θ cos φ, Y = k sin θ sin φ. (3.2.17)
The integration of Eq. (3.2.16) over the geometric shadow area bounded by (2-3-4-
8-5-6-2), defined in Figure 3.11a, can be carried out by dividing the entire integration
domain into three sub-domains, including two trapezoids [(1 -4-8-5-1 ) and (3-7 -6-2-3)]
and one parallelogram (1 -5-7 -3-1 ). In this manner, we can follow the integration method
presented in Smith and Marsh (1974) to carry out these integrations, after which the
diffraction amplitude up /(uc e−ikr0 ) of a parallelepiped defined by the orientation angles
(α, β) at an arbitrary point P(θ, φ) can be obtained in the form
up (θ, φ; α, β; ka, b/a, c/a)/(uc e−ikr0 )
4 ) * ( +
= sin 12 (y2 − y1 )(a1 X + Y ) sin 12 [−a1 (y1 − y2 ) + 2b1 ] X − 12 (y1 + y2 )
X(a1 X + Y )
4 ) * ( +
+ sin 12 (y2 − y1 )(a2 X + Y ) sin 12 [a2 (y1 + y2 ) + 2b2 ] X + 12 (y1 + y2 )
X(a2 X + Y )
4
+ sin [y1 (a1 X + Y )] sin(b1 X) (3.2.18)
X(a1 X + Y )
where a1 , a2 , b1 , b2 , y1 , and y2 are explicit functions of α, β, a, b, and c as defined in the
following equations:
a1 = sin α tan β, a2 = − sin α/tan β, (3.2.19a)
b1 = sin α tan β(a sin β − b cos β) + a sin α cos β + b sin α sin β + c cos α, (3.2.19b)
b2 = −(sin α/tan β)(a sin β − b cos β) + a sin α cos β + b sin α sin β + c cos α,
(3.2.19c)
y1 = −a sin β + b cos β, y2 = a sin β + b cos β. (3.2.19d)
3.2 Fraunhofer diffraction by ice crystals 139

Using these relationships, the intensity function for a given orientation of a rectangular
parallelepiped, defined by angles α and β, can then be expressed as
2
k2
ip (θ, φ; α, β; ka, b/a, c/a) = |up /uc e−ikr0 |2 . (3.2.20)

In the case of a rectangular aperture, we may define the geometric shadow area as 2aL,
where 2a and L are the lengths of the rectangle. Referring to Eq. (3.2.3), we may set the
integrating boundaries x and y to be from −a to a and from −L2 to L2, respectively. A
separation of the two integrations leads to
 a  L/2
up (rectangle) = uc e−kr0 exp(−ikx cos φ sin θ )dx exp(−iky sin φ sin θ )dy
−a −L/2

sin(ka cos φ sin θ ) sin [k(L/2) sin φ sin θ ]


= uc e−kr0 4a(L/2) .
ka cos φ sin θ k(L/2) sin φ sin θ
(3.2.21)

The solution of diffraction due to the light-wave disturbance associated with a rectangle
in the far field is given by the product of sine functions. It follows that the intensity function
for diffraction can be expressed as

4k 4 a 2 (L/2)2 sin2 (ka cos φ sin θ ) sin2 [k(L/2) sin φ sin θ ]


ip [θ, φ; ka, k(L/2)] = .
π2 (ka cos φ sin θ )2 [k(L/2) sin φ sin θ ]2
(3.2.22)

Alternatively, if we let α = 0, β = 0, a = b, and c = L2 in Eq. (3.2.18), we may also


derive Eq. (3.2.22) directly.

3.2.3 Hexagonal cylinder


Consider a plane wave incident on a hexagonal cylinder from a direction with the zenith
angle (π/2 − α) and the azimuthal angle β measured in reference to the coordinate system
fixed to the cylinder, where the x- and z-axes are directed, respectively, along the a- and
c-axes of the hexagonal cylinder. The origin point O of the coordinate system is fixed at
the center of the cylinder. The geometric shadow of the cylinder projected onto a plane
normal to the incident direction is expressed in terms of the (x  , y  ) coordinates of its
marginal vertices in a new coordinate system such that the z -axis is taken to be along the
incident direction (see Figure 3.11b). The transformation of the body-framed coordinate
system X(x, y, z) into the new coordinate system X (x  , y  , z ) can then be written in the
form

X = BCDX, (3.2.23)

where the transformation matrixes C and D have been defined in Eqs. (3.2.14) and (3.2.15),
140 Geometric optics for light scattering by ice crystals

and the B matrix is given by


⎡ ⎤
cos ψ sin ψ 0
B = ⎣ −sin ψ cos ψ 0⎦. (3.2.24)
0 0 1
The relationship among the three angles α, β, and ψ is defined by the following equation:
 π ,
tan ψ = − tan β + sin α, (0 ≤ ψ ≤ π). (3.2.25)
3
Thus, if the size and shape of a hexagonal cylinder and the propagation direction (α, β)
of the incident wave are known, the coefficients a1 , a2 , a3 , b1 , b2 , b3 , y1 , and y2 defined
in Figure 3.11b can be determined by Eq. (3.2.23). The shape of a hexagonal cylinder is
defined by the length-to-radius ratio ca, where a and c are the semi-minor and semi-major
axes, respectively. The size parameter is given by ka.
Having defined a number of geometric variables, we may rewrite the integral equation
of Eq. (3.2.3) in the form
 
up (hexagon) = uc e−ikr0 exp[−ik(Xx  + Y y  )] dx  dy  . (3.2.26)
A

The geometric shadow area bounded by (1-6-5-4-10-9-8-7-1), shown in Figure 3.11b,


can be divided into three sub-domains: two trapezoids (1-6-5-4-1) and (7-10-9-8-7) and
one parallelogram (1-4-10-7-1). The operation of the matrix B in Eq. (3.2.23) can then be
applied so that the bases of the trapezoids are parallel to the x  -axis. In this manner, we can
follow the method of Smith and Marsh (1974) to carry out these integrations. It follows
that the diffraction amplitude up of a hexagonal cylinder defined by the orientation position
(α, β) at an arbitrary point P(θ, φ) can be obtained in the form:
up (θ, φ; α, β; ka, c/a)/uc e−ikr0
4 ) * ( +
= sin 12 (y2 − y1 )(a1 X + Y ) sin 12 [−a1 (y1 − y2 ) + 2b1 ] X − 12 (y1 + y2 )
X(a1 X + Y )
4 ) * ( +
+ sin 12 (y2 − y1 )(a2 X + Y ) sin 12 [a2 (y1 + y2 ) + 2b2 ] X + 12 (y1 + y2 )
X(a2 X + Y )
4
+ sin [y1 (a3 X + Y )] sin(b3 X). (3.2.27)
X(a3 X + Y )
Subsequently, the diffracted intensity for a given orientation of a hexagonal cylinder defined
by angles α and β is given by an expression similar to Eq. (3.2.20).

3.2.4 Spheroid
The geometric shadow of a spheroid projected onto a plane perpendicular to the incident
direction is generally an ellipse. We may specify the orientation of the spheroid by an angle
ζ between the incident direction and its rotation axis. An ellipse is a circle of radius a
multiplied by a factor ξ in the direction Ox (see Figure 3.11c). Thus, the elliptical aperture
3.2 Fraunhofer diffraction by ice crystals 141

Ae is equal to a circular aperture Ac multiplied by ξ in the direction along Ox. Referring to


Eq. (3.2.3) and changing variables such that x  = x/ξ and y  = y, we obtain
 
−ikr0
up (spheroid) = ξ uc e exp[−ik(x  ξ cos φ + y  sin φ) sin θ]dx  dy  . (3.2.28)
Ac

We may transform the rectangular coordinates (x , y ) to polar coordinates (ρ, ψ) such
that x  = ρ cos ψ and y  = ρ sin ψ. In this manner, Eq. (3.2.28) can be rewritten as follows:
 a  2π
up = ξ uc e−ikr0 exp[−kρl cos(ψ − σ ) sin θ]ρdρdψ, (3.2.29)
0 0

where σ = tan−1 (tan φ/ξ ) and l = (ξ 2 cos2 φ + sin2 φ)1/2 . In terms of the zero-order Bessel
function defined in Eq. (3.2.5), we have
 a
up = ξ uc e−ikr0 2π J0 (kρl sin θ )ρdρ. (3.2.30)
0

Referring to Eqs. (3.2.7) and (3.2.8) regarding the relationship between the zero- and
first-order Bessel functions and to Eq. (3.2.30), we can show that
2J1 (xl sin θ )
up = uc e−ikr0 Ae , (3.2.31)
xl sin θ
where the geometric shadow area Ae = ξ πa2 and the size parameter x = ka. Hence, the
scattered intensity in terms of the incident intensity I0 = |u0 |2 can be expressed as


ξ 2 x 4 2J1 (lx sin θ ) 2
ip (θ, φ; ζ ; ka, c/a) = . (3.2.32)
4 lx sin θ
To determine ξ , we note that the geometric shadow of a spheroid onto a plane perpendic-
ular to an incident direction is an ellipse. We may specify the orientation of the spheroid by
an angle ζ between the incident direction and its rotation axis, defined by the semi-minor
axis a and the semi-major axis along the rotation axis c, as shown in Figure 3.11c. Thus,
the spheroid and the incident light beam tangent to it are defined by the following two
equations:
x2 z2
2
+ 2 = 1, x = tan ζ z + v. (3.2.33a)
a c
Upon combining these two equations, we obtain

1 tan2 ζ 2v tan ζ v2
+ z 2
+ z + − 1 = 0. (3.2.33b)
c2 a2 a2 a2
Moreover, in order to have the incident beam tangent to the spheroid, the determinant D
must be equal to 0 such that
2
 v tan ζ 2 1 tan2 ζ v
D = − 2+ − 1 = 0, (3.2.33c)
a2 c a2 a2
142 Geometric optics for light scattering by ice crystals

which leads to the solution in the form v = ±(c2 tan2 ζ + a 2 )1/2 . Additionally, we find the
following relationship:
ξ a = |v| cos ζ = a[(c/a)2 sin2 ζ + cos2 ζ ]1/2 . (3.2.34a)
Thus, we obtain
ξ = [(c/a)2 sin2 ζ + cos2 ζ ]1/2 . (3.2.34b)
Once ξ is known, the scattered intensity due to diffraction by a spheroid can be evaluated
from Eq. (3.2.32).
The general case of a spheroid shape is a three-axis ellipsoid, which is governed by the
following geometric relationship:
x2 y2 z2
2
+ 2 + 2 = 1. (3.2.35)
a b c
When a three-axis ellipsoid is viewed from an arbitrary direction expressed by angles
α and β, its shadow is an ellipse, as shown in Figure 3.11d. The ellipse can be obtained
as follows. On solving X = (CD)T X , the coordinate system (x, y, z) can be expressed as
(x , y , z ). Substituting (x, y, z) into Eq. (3.2.35), we obtain a quadratic equation for the
coordinate system (x , y , z ). If the determinant D for the quadratic equation with respect
to z is 0, another quadratic equation in x and y can be obtained to represent a shadow
ellipse. Specifically, we obtain
Ax 2 + Bx  y  + Cy 2 = D. (3.2.36)
Upon rotation of the x and y coordinate axes by an angle η (= tan−1 [B/(A − C)]),
Eq. (3.2.36) can be simplified in the form
x  /p2 + y  /q 2 = 1. (3.2.37)
It follows that the semi-axis lengths p and q, and the angle η between the p-axis and x -axis
as well as its diffracted intensity can then be computed.
Finally, for randomly oriented non-spherical particles, the intensity function ip can be
obtained by performing integrations over the angles α, β, and φ in the form
 2π  2π  π
1
īp (θ ; ka, [b/a] , c/a) = ip (θ, φ; ka, [b/a] , c/a; α, β) cos αdαdβdφ.
4π2 0 0 0
(3.2.38)
The intensity function is now a function of the scattering angle and the particle’s geometric
parameters in reference to the incident light beam characterized by a certain wavelength.

3.2.5 Some numerical results


In Figures 3.12a–i, the intensity function is shown in logarithmic contours for a diffracting
body with the shape and orientation indicated in the inset. The wavelength used for the
calculation is 0.55 µm. Any diffracted intensity would have a maximum at the scattering
(a) 90
(b) 90
(c) 90
120 60
120 60 φ 120 60 φ φ
4 3
3 4

150 30 150 4 30 150 30


5 4
4 5
6
6 4 3 3
180 3 4 4 45 6 0 0 180 0 0 180 4 4 5 0 0

2 2 2
θ θ
4 4 4
210 210

330 330 210 330
6 6

8 8 8
240 300 240 300 240 300
270 270 270

(d) 90
(e) 90
(f) 90
120 60 φ 120 60 φ 120 60 φ

150 30 150 30 150 30


5
4
4
5
4 5 6 4
180 55 67 0 0 180 0 0 180 4 4 5 6 0 0
5
4
3 2 2 2
4 θ 3 3
4 4 4θ
θ 4
210 330 210 330 210 330
6 6 3 6

8 8 4 8
240 300 240 300 240 300
270 270 270

(g) (h) 90
(i) 90
90
60 120 60 φ 120 60 φ
120 φ
4
3
150 30 150 5 30 150 30
4

4 4 5 6
5 5 6
180 0 0 180 0 0 180 4 0 0
4 5
4 6 3
2 4 5 2 2
4
3
4 4θ 4 4
210 3 210 210

330

330 330
6
4
8 8 8
240 300 240 300 240 300
270 270 270

Figure 3.12 Diffracted intensity function contours on a logarithmic scale for an object with the shape and orientation
indicated by the inset, using an incident wavelength of 0.55 µm. The radial direction covers the scattering angle θ from
0° to 10°. The objects are (a) a circular aperture (a = 4.24 µm), (b) an elliptical aperture (a = 3 µm and c = 6 µm), (c) a
rectangular aperture (a = 3 µm and c = 6 µm), (d) a hexagonal aperture (a = 10 µm), (e) a cube (α = 45°, β = 45°, and
a = 3.8 µm), (f) a hexagonal cylinder (α = 60°, β = 30°, a = 4 µm, and c = 8 µm), (g) a hexagonal cylinder (α = 60°,
β = 20°, a = 4 µm, and c = 8 µm), (h) a rectangular parallelepiped (α = 30°, β = 30°, a = 2 µm, b = 7 µm, and c =
4 µm), and (i) a three-axis ellipsoid (α = 60°, β = 45°, a = 3 µm, b = 6 µm, c = 9 µm) (after Takano et al. 2012).
144 Geometric optics for light scattering by ice crystals

angle θ 1st (max) of 0°. The blackened parts in the figures associated with the dense contour
lines depict weak diffracted intensities. The diffraction pattern of a circular aperture is
concentric (Figure 3.12a) with the first and second minima located at θ 1st (min) = 4.5° and
θ 2nd (min) = 8.3°, respectively. These dark bands can be explained by the fact that the
first-order Bessel function J1 (y) has zeros at y = 3.832 and 7.016.
Figure 3.12b shows the diffraction pattern of an elliptical aperture whose area is equal
to that of the circle. The long axis of the elliptical contour is along the short axis of its
aperture, such that the aspect ratio is the same as that of the aperture. Figure 3.12c is for a
rectangular aperture circumscribing the ellipse in Figure 3.12b. The first maximum pattern
around θ 1st (max) = 0° has a rectangular shape with an aspect ratio equal to that of the
aperture but with a 90° rotation. Next to the first maximum pattern, four surrounding fans
of rectangular shapes are produced.
The regular hexagonal aperture produces a hexagonal pattern around θ 1st (max) = 0°,
surrounded by six fans (Figure 3.12d). However, there is a dark circle inside the hexagonal
pattern, which was also shown in the work of Smith and Marsh (1974). The scattering angle
θ 1st (min) of 2.11° at which the diffracted intensity becomes zero is identified by inspecting
the position at which the diffracted amplitude changes its sign from positive to negative.
This dark circle coincides with the first minimum of circular aperture of the same area
with a radius of 9.09 µm, corresponding to y = 3.832. The same dark circle also occurs
for a regular octagonal aperture. Figure 3.12e shows the diffracted pattern of a cube for an
oblique incident light beam. In view of the shape defined in the inset, its pattern resembles
that of a hexagon but its apex has a 30° rotation. For example, there is a dark circle-like
shape inside the hexagonal pattern containing the first maximum. Its approximate scattering
angle θ is 3.4°, close to the first minimum of a circular aperture of the same area, which
occurs at θ 1st (min) = 3.43°.
Figure 3.12f illustrates the diffraction pattern of a hexagonal cylinder for an oblique
incident light beam. In view of the projected shadow in the inset, its diffraction pat-
tern has the characteristics of both rectangular and hexagonal apertures. Figure 3.12g
shows another example of a hexagonal cylinder in which the incident direction in
Figure 3.12f is shifted by changing the rotational angle β in the horizontal direction from
30° to 20°. The diffraction pattern is also rotated by 10° and is not symmetric in reference
to the line defined by φ = 90° and φ = 270°, which differs from the patterns depicted in
Figures 3.12a–f. Figure 3.12h is for a rectangular parallelepiped such that the shape of
the projected shadow resembles a parallelogram. As a result, a parallelogram-like pattern
occurs around the incident direction and is surrounded by four fans. Figure 3.12i depicts
the diffraction pattern generated by a three-axis ellipsoid whose projected shadow is an
ellipse with a semi p-axis length of 6.6 µm, semi q-axis length of 4.0 µm, and an angle η
between the p-axis and xʹ-axis of 29.0°. In Figure 3.12b, we show the diffraction pattern
for an ellipse with its long axis along the q-axis.
Figure 3.13 depicts the normalized diffracted intensity as a function of the scattering
angle from 0 to 10° for randomly oriented spheroids, hexagonal cylinders, and rectangular
parallelepipeds with aspect ratios of 4 and 14 such that their cross-sectional areas are
3.2 Fraunhofer diffraction by ice crystals 145

(a) (b)
108 108

c/a = 4 c/a = 1/4

c/a = 4 c/a = 1/4


107 107
c/a = 4, a = b c/a = 1/4, a = b
Diffracted Intensity

106 106

105 105

104 104

103 103
0 2 4 6 8 10 0 2 4 6 8 10
Scattering Angle (deg.)

Figure 3.13 Diffracted intensity functions for randomly oriented spheroids, hexagonal cylinders,
and regular parallelepipeds: (a) aspect ratio = 4; (b) aspect ratio = 14. The cross-sectional area for
both cases is that of a sphere with a size parameter 2πaλ of 100, where λ is the wavelength and a is
the radius.

equal to that of a sphere with a size parameter of ka = 100. It is evident that the two
groups (4 and 14) display similar patterns but with numerous differentiable fluctuations.
In Figure 3.12, we demonstrate that for a non-spherical particle with a specific orientation
relative to the incident light beam, the scattering light produced by diffraction displays
significant fluctuations in 3-D space. However, when an ensemble of non-spherical particles
are randomly oriented in 3-D space, the diffracted light is symmetrical with respect to the
azimuthal angle and is a function of the scattering angle only. Intensity fluctuations in this
case are much smaller than for a particular orientation. Results for the diffraction of a
group of spheres having the same sizes show substantial max–min fluctuations (see also
Figure 3.13a). These fluctuations can be largely damped out by applying a size distribution
for spherical particles.
For application to ice crystals, we may use a size distribution to smooth out small
fluctuations. The semi-exact solution for the diffraction of irregular ice particles (bullet
rosette, aggregate, snowflake, and rough-surfaced crystal) may be obtained by performing
tedious numerical calculations once the shape is defined. As a first approximation, however,
we may use diffraction patterns for columns/plates with equivalent cross-sectional areas
(Cai and Liou 1982, Takano and Liou 1989a) for application to light scattering by ice
crystals.
146 Geometric optics for light scattering by ice crystals

3.3 Generalized geometric optics for light scattering by ice crystals


3.3.1 Introduction
The scattering of light by spheres, cylinders, and spheroids can be solved by the exact theo-
ries presented in Subsection 2.2.3, and computations can be performed for size parameters
that are practical for atmospheric applications. However, an exact solution for the scattering
of light by non-spherical ice crystals covering all sizes and shapes that occur in planetary
atmospheres does not exist in practical terms. It is unlikely that one specific method can
be employed to resolve all of the scattering problems associated with non-spherical and
inhomogeneous ice crystals.
The geometric-optics approach, as discussed in Section 3.1, is the asymptotic approx-
imation of fundamental electromagnetic theory, valid for light scattering computations
involving a target whose dimension is much larger than the incident wavelength. As also
noted in that section, the geometric ray-tracing method provides a means of identifying
numerous fascinating ice optical phenomena observed in the Earth’s atmosphere. The part
of reflection/refraction within the context of geometric optics is based on the principle that
light rays can be identified at a particle surface; however, the part of diffraction is not
necessarily governed by this localization principle.
The geometric-optics method, including geometric reflection/refraction and diffraction
contributions, has been extensively used to investigate the single-scattering properties of
non-spherical ice crystals (Wendling et al. 1979, Liou and Coleman 1980, Coleman and
Liou 1981, Cai and Liou 1982, Takano and Jayaweera 1985, Takano and Liou 1989a
and 1989b, Macke 1993, Hess and Wiegner 1994, Macke et al. 1996). Application of the
ray-tracing technique to light scattering by spheres can be traced back to studies by Liou
and Hansen (1971). The authors applied this technique to light scattering by spheres and
compared the phase function and polarization patterns computed from the geometric ray-
tracing method with the “exact” Lorenz–Mie theory for a size distribution. The patterns
computed from the ray-tracing technique are reasonably accurate when the modal size
parameter is larger than 100 for a polydisperse system of spheres.
The application of geometric ray-tracing to light scattering by hexagonal ice prisms was
first carried out by Jacobowitz (1971). In his study, a sufficiently large number of equally
spaced, parallel rays were traced through a hexagonal ice crystal. The external reflection and
two refractions coupled with various orders of internal reflection were summed to determine
the scattering pattern in the far field. Diffraction contributions due to the scattering of
incident light in the forward direction were computed by Kirchhoff’s formula. Hexagonal
prisms were assumed to be infinitely long along the major axis. To circumvent the unrealistic
infinitely-long assumption, Wendling et al. (1979) combined the Monte Carlo method and
the geometric ray-tracing technique to compute the phase function of finite hexagonal
columns. In these early studies, the polarization effect and phase interferences associated
with the incident, internal, and scattered rays were not accounted for in the calculation.
The paper by Cai and Liou (1982) appears to be the first study to include a 4 × 4 scattering
phase matrix and polarization configuration as well as phase interferences in the context
3.3 Generalized geometric optics for light scattering 147

of geometric ray tracing to compute the single-scattering properties of hexagonal columns


and plates.
The theoretical foundation for “conventional” geometric ray tracing and the associated
computational algorithm developed by Cai and Liou (1982) were later improved and refined
by a number of researchers. Takano and Liou (1989a, 1989b) considered the effects of the
ice crystal’s birefringence property, horizontal orientation, and size spectrum in associa-
tion with light-scattering calculations. The single-scattering properties of ice crystals with
horizontal orientation have also been investigated by Rockwitz (1989) and Noel et al.
(2001).
Furthermore, the ray-tracing method has been applied to various complex ice crystal
shapes by Takano and Liou (1995), Macke (1993), Macke et al. (1996), Iaquinta et al.
(1995), Muinonen et al. (1997), Peltoniemi et al. (1989), and Yang and Liou (1998a), as
well as to ice crystals with internal inclusions (e.g., air bubbles and soot) by Macke et al.
(1996) and Macke (2000). Application of the ray-tracing technique with implementation of
the Monte Carlo method to complex geometries has also been reported by Nousiainen et al.
(2003) and Grynko and Shkuratov (2003). Additionally, Borovoi et al. (2002) investigated
backscattering characteristics of arbitrarily oriented hexagonal ice crystals using the ray-
tracing method. Borovoi and Grishin (2003) discussed an effective ray-tracing algorithm for
computing phase matrix elements for ice particles. Alternate approaches to the conventional
ray-tracing method have been developed by Muinonen (1989) and Yang and Liou (1995,
1996a, 1997) in which the principles of geometric optics are applied to the computation
of the near field either on the surface of, or inside, the scattering particle. The near field
obtained from the ray-tracing technique is then mapped to the far field on the basis of either
a surface-integral- or volume-integral-based electromagnetic relationship.
In Subsections 3.3.2–3.3.4, we present conventional and improved approaches to light
scattering by ice crystals, the methods dealing with absorption in the context of geometric
optics, and the numerical implementation of the Monte Carlo method in geometric ray
tracing.

3.3.2 Conventional approach


When the size of an ice crystal is much larger than the incident wavelength, we may consider
a light beam as consisting of a bundle of separate parallel rays that undergo reflection and
refraction outside and inside the ice crystal with propagation directions determined by
Snell’s law at the surface. The total field is assumed to consist of diffracted and reflected
and refracted rays. Diffracted rays pass around the ice crystal while rays impinging on it
undergo local reflection and refraction, referred to as Fresnelian interaction. The energy
carried by the diffracted and Fresnelian rays is assumed to be the same as the energy
intercepted by the ice crystal cross-section projected along the incident direction.
The intensity of the scattered light within the small scattering-angle interval θ in the
scattering direction θ in the far field can be computed from the summation of the intensity
contributed by each individual ray emerging at between θ + θ /2 and θ − θ /2. It is
148 Geometric optics for light scattering by ice crystals

(conventional)

diffracted ray
Incident wave front reflected ray
e1r
incident ray
Q1 equivalent surface E&M theory
l l
Q0
far field
e1i =ei0 l
E&M currents
(improved)
Q2
Q3
l

e2t

transmitted ray
diffracted ray
transmitted ray
e3t

Figure 3.14 A graphical depiction of the principle of the geometric ray-tracing technique, including
diffraction, for computing the single-scattering properties of an ice crystal (conventional). The sym-
bols Q1 , Q2 , and Q3 denote the entrance points of light rays; and er1 , et2 , and et3 are the corresponding
directional vectors. Also shown is a conceptual diagram of the geometric-optics integral-equation
method, including geometric ray tracing within an ice crystal and the mapping of resulting currents
to the far field by means of fundamental electromagnetic theory (improved; see Subsection 3.3.3).

usually assumed that the interference is smoothed out when ice crystals are randomly
oriented. In this case, the extinction efficiency (the ratio of the extinction cross-section
to the average projected area of the particle) of the ice crystal is 2. Based on Babinet’s
principle, diffraction by an ice crystal may be regarded as the same as diffraction by an
opening, on an opaque screen perpendicular to the incident light, which has the same
geometric shape as the projected cross-section of an ice crystal. The Fraunhofer diffraction
approach can then be employed to compute the diffraction component for column and plate
crystals (Cai and Liou 1982, Takano and Liou 1989a).
To trace the reflected and refracted rays, let np (p = 1, 2, 3 . . . ) be the unit vectors locally
normal to the particle surfaces at the incident points Qp (p = 1, 2, 3 . . . ) facing the incoming
rays, as shown in Figure 3.14. For external reflection at point Q1 , the incident direction ei1
and the incident angle θ1i are given, respectively, by the following two expressions:

ei1 = ei0 , (3.3.1a)


 
θ1i = cos−1 − n1 · ei1 , (3.3.1b)

where ei0 denotes the initial incident direction (Figure 3.14). Following Snell’s law,
the directions of the externally reflected ray and the corresponding refracted ray are
3.3 Generalized geometric optics for light scattering 149

defined by

er1 = ei1 + 2 cos θ1i n1 , (3.3.2a)


 - 
et1 = ei1 /m + cos θ1i m − cos θ1t n1 , (3.3.2b)

where m is the refractive index of the scattering particle, and θ1t is the refractive angle given
by Snell’s law as follows:
 - 
θ1t = sin−1 sin θ1i m . (3.3.2c)

When the refractive index m is a complex number, simultaneous absorption and scattering
occur and the refracted wave within the particle is an inhomogeneous wave. In this case,
an adjusted refractive index must be used to trace the refracted rays (see Subsection 3.1.2).
For practical computations at visible and near-infrared wavelengths, the real part of the
refractive index may be used as an excellent approximation for the adjusted refractive
index to trace ray directions based on Eq. (3.1.19).
For internal reflection with orders of p = 2, 3, 4 . . . , the incident directions can be
defined in a similar manner and are given by the directions of either the first-order refracted
rays or internally reflected rays, as follows:

ei2 = et1 , (3.3.3a)

eip = erp−1 , p = 3, 4, 5 . . . (3.3.3b)

With some vector algebraic manipulations on the basis of Snell’s law, it can be shown that
the propagating directions of pth-order reflected and refracted rays are given by

erp = eip + 2 cos θpi np , (3.3.4a)


 
etp = meip + m cos θpi − cos θpt np , (3.3.4b)

where the incident and refraction angles, θpi and θpt , are defined by
 
θpi = cos−1 − np · eip , (3.3.5a)
 
θpt = sin−1 m sin θpi . (3.3.5b)

Total reflection occurs if the term m sin θpi in Eq. (3.3.5b) is larger than 1. In this case, a
refracted ray should not be expected so that the ray-tracing computation should be continued
only for the ray associated with total reflection. Equations (3.3.1a)–(3.3.5b) constitute a
closed set of equations for tracing the directions of all the reflected and refracted rays
associated with a given incident ray.
A localized plane electromagnetic wave is a transverse vector wave. Thus, the vector
property or the polarization configuration of electric fields associated with localized rays in
the ray-tracing computation must be accounted for, which requires the definition of various
auxiliary unit vectors. For the incident direction of an initial ray specified by a unit vector
ei0 , we define two unit vectors u0 and v0 , as shown in Figure 3.15, which are normal to the
150 Geometric optics for light scattering by ice crystals

first order ray (p=1)

u1i t
f1 fp ept up
n1 v1
e1r e1i vp
u1r θ1i
v1 θ pt
Air Air
upr
Ice Ice
epr
θ pi vp

np
θ1t v1
epi upi
vp
u1t
e1t p-th order rays (p>1)

Figure 3.15 Schematic diagrams for the direction of the incident, reflected, and refracted rays for
the first-order ray (p = 1) and pth-order rays (p > 1). The unit vectors e, u, and v are defined
with reference to polarization configuration: θ p and φ p (p = 1, 2, 3, . . . ) are zenith and azimuthal
angles, respectively: and the superscripts i and t denote incident and refracted (transmitted) directions,
respectively.

incident direction and satisfy the following relations:


u0 · v0 = 0 and v0 × u0 = ei0 . (3.3.6)
The unit vectors v0 , u0 , and ei0 defined in this equation constitute a right-handed coor-
dinate system. To define the initial rays in practice, we may specify the unit vectors v0 , u0 ,
and ei0 in the directions along the x-, y-, and z-axes of the incident coordinate system.
Because the unit vectors ei0 , u0 , and v0 are orthogonal to each other, the incident polariza-
tion configuration can be specified with respect to u0 and v0 such that the incident electric
field Ei0 can be written as follows:
Ei0 = E0u
i
u0 + E0v
i
v0 . (3.3.7)
Similarly, we may define three pairs of unit vectors (uip , vp ), (urp , vp ), and (utp , vp ) for
the ray directions along eip , erp , and etp (p = 1, 2, 3 . . . ), respectively. The unit vectors uip ,
urp , and utp are on the incident plane, as shown in Figure 3.15, while the unit vectors vp
(p = 1, 2, 3 . . . ) are defined as pointing out of the incident plane. These vectors can be
specified by the following expressions:
 
vp = eip × np /sin θpi , p = 1, 2, 3 . . . , (3.3.8a)
ui,r,t
p = ei,r,t
p × vp , p = 1, 2, 3 . . . (3.3.8b)
The unit vector vp is normal to the incident plane, a plane containing the incident
direction and the direction locally normal to the particle surface at the incident point, for
the pth-order reflection and refraction. The unit vectors uip , urp , and utp are parallel to
3.3 Generalized geometric optics for light scattering 151

the pth-order incident plane. The vector vp in Eq. (3.3.8a) cannot be uniquely specified if
sin θpi = 0. In this case, we select vp = vp−1 . With the aforementioned unit vectors defined,
the electric fields associated with the pth-order incident, reflected, and refracted rays can
be expressed as follows:

Ei,r,t
p = Epu up + Epv
i,r,t i,r,t i,r,t
vp . (3.3.9)

Consider now the external reflection and the first-order refraction. In order to apply
the Fresnel formulas, the electric field associated with the incident ray impinging on the
point Q1 must be specified with respect to ui1 and v1 . Also, the electric field associated
with the incident ray specified in Eq. (3.3.7) can be expressed in an alternative form as
follows:

Ei0 = E1u
i
ui1 + E1v
i
v1 . (3.3.10)

Equations (3.3.7) and (3.3.10) for the incident electric vector can be expressed in terms
of a rotational matrix in the form
   
i i
E1u E0u
i
= 1 i
, (3.3.11a)
E1v E0v

where the rotational matrix  1 (p = 1) is given by



i

u1 · u0 ui1 · v0 cos φ1 sin φ1
1 = = . (3.3.11b)
v1 · u0 v1 · v0 −sin φ1 cos φ1

Because the field components on the left-hand side of Eq. (3.3.11a) are specified with
respect to the incident plane, the Fresnel formulas can now be applied. The externally
reflected field, Er1 , is given by

Er1 = E1u
r r
u1 + E1v
r
v1 , (3.3.12a)

where
     
r i i
E1u E1u E0u
r
= R1 i
= R1  1 i
. (3.3.12b)
E1v E1v E0v

In Eq. (3.3.12b), R1 is the reflection matrix for external reflection given by




R1u 0
R1 = . (3.3.12c)
0 R1v

The elements of the reflection matrix in Eq. (3.3.12c) are given by the Fresnel coefficients
defined in Subsection 3.1.1 as follows:
m cos θ1i − cos θ1t cos θ1i − m cos θ1t
R1u = , R1v = . (3.3.13)
m cos θ1i + cos θ1t cos θ1i + m cos θ1t
152 Geometric optics for light scattering by ice crystals

Likewise, the electric field associated with the first-order refracted ray is given by
Et1 = E1u
t
ut1 + E1v
t
v1 , (3.3.14a)
     
t i i
E1u E1u E0u
t
= T1 i
= T1  1 i
, (3.3.14b)
E1v E1v E0v
where the refraction matrix T1 is defined in the form
    
2 1/2
T1u 0 1 − R1u 0
T1 = =   . (3.3.14c)
2 1/2
0 T1v 0 1 − R1v
In Eq. (3.3.14c), conservation of the energy for the ray due to changes in the refractive
index and ray cross-section in two media is accounted for in the refraction matrix (Cai and
Liou 1982).
For external reflection, the direction along the reflected ray is the scattering direction.
Thus, the scattering angle (note that  has been used throughout most of the text) can be
expressed as
 
θ1s = cos−1 ei0 · er1 . (3.3.15a)
The direction perpendicular to the scattering plane can be subsequently specified by
-
vs1 = ei0 × er1 sin θ1s . (3.3.15b)
If sin θ1s = 0 in Eq. (3.3.15b), implying forward (θ1s = 0◦ ) and backward scattering (θ1s =
180◦ ), then the vector vs1 cannot be defined. In this case, we select vs1 = v0 . Having the unit
vector vs1 defined, the direction parallel to the scattering plane is given by
us1 = er1 × vs1 . (3.3.15c)
We may express the electric field associated with the externally reflected ray with respect
to two directions parallel and perpendicular to the scattering plane in the form
Er1 = E1u
s s
u1 + E1v
s s
v1 . (3.3.16)
From Eqs. (3.3.12a, b) and (3.3.16), we can show that
   
s i
E1u E0u
s
=  1 R1  1
s
i
, (3.3.17a)
E1v E0v
where  s1 is a rational matrix given by
 
us1 · ur1 us1 · v1
 s1 = . (3.3.17b)
vs1 · ur1 vs1 · v1
To obtain the scattering matrix, the incident field must be defined with respect to the
directions parallel and perpendicular to the scattering plane in the form
 i 
Ei0 = E1su
i
e0 × vs1 + E1sv
i
vs1 . (3.3.18)
3.3 Generalized geometric optics for light scattering 153

Note that the unit vector ei0 × vs1 in Eq. (3.3.18) is parallel to the scattering plane. The
expression in Eq. (3.3.18) for the incident field is related to that in Eq. (3.3.7) as follows:
   
i i
E0u E1su
i
= 1 i
i
, (3.3.19a)
E0v E1sv

where  i1 is a 2-D rotational matrix given by


   
u0 · ei0 × vs1 u0 · vs1
1 =
i
  . (3.3.19b)
v0 · ei0 × vs1 v0 · vs1

Thus, we can express the scattered field in Eq. (3.3.17a) as follows:


   
s i
E1u E1su
s
=  1 R1  1  1
s i
i
. (3.3.20a)
E1v E1sv

Likewise, for the refracted rays with p = 2, we have


   
s i
E2u E2su
s
=  2 T2  2 T1  1  2
s i
i
. (3.3.20b)
E2v E2sv

Furthermore, for the orders p = 3, 4, 5 . . . , we have


   
s i
Epu Epsu
s
=  p Tp  p . . . R2  2 T1  1  p
s i
i
. (3.3.20c)
Epv Epsv

The definitions of  sp , Tp ,  p , Rp , and  ip are similar to the case for p = 1. In this


manner, both the incident and scattered electric field vectors are expressed with respect to
the scattering plane as shown in Eqs. (3.3.20a–3.3.20c). Thus, contributions of the scattered
rays to the transformation matrix defined in Eq. (2.4.30) can be expressed as follows. First,
for externally reflected rays (p = 1), we have
 (1) 
A2 A(1)
3
A =
(1)
=  s1 R1  1  i1 . (3.3.21a)
A(1)
4 A (1)
1

For transmitted rays (p = 2), the transformation matrix can be expressed as


 (2) 
A2 A(2)
3
A =
(2)
=  s2 T2  2 T1  1  i2 . (3.3.21b)
A(2)
4 A(2)
1

For p > 2, we can define the following transformation matrix as follows:


 (p)  ⎛ ⎞
A A
(p) 02
=  sp Tp  p ⎝ Rk  k ⎠ T1  1  ip ,
2 3
A(p) = (p) (p)
(3.3.21c)
A4 A1 k=p−1
154 Geometric optics for light scattering by ice crystals

where  1 has been defined in Eq. (3.3.11b) and  p has the same expression except that
index 1 is replaced by p. We may perform the transpose of Eq. (3.3.21c), resulting in the
following equation:
  %p−1 &
) (p) *T (p)
A2 A4
(p)
) i *T 0 ) *T ) *T
A = (p) (p)
=  p [ 1 ] T1
T
[ k ] Rk  p Tp  sp , (3.3.21d)
T
A3 A1 k=2

where the transpose of  p (p = 1, 2, 3, . . . ) is given by


 
cos φp −sin φp
[ p ] =
T
. (3.3.21e)
sin φp cos φp

If the sign of the rotation angle φp is changed, then the signs of the non-diagonal elements
in Eq. (3.3.21d) are also changed, since Tp and Rp are diagonal matrices. As a result, we
have the following transformation matrix:
  %p−1 &
A2
(p)
−A4
(p)
0
(p)∗
A = (p) (p)
=  ip  1 T1  k Rk  p Tp  sp . (3.3.21f)
−A3 A1 k=2

The right-hand side of Eq. (3.3.21f) represents a reverse order of terms in reference to
Eq. (3.3.21c) such that every ice particle position has a reciprocal position with respect
to the initial position. Consequently, incident and scattered light beams can be reversed.
This is referred to as the law of reciprocity, as discussed in Subsection 2.4.5. For external
reflected and diffracted rays, non-diagonal elements of the transformation matrix are zero
and so the addition of these two components to geometric refraction and reflection would
not alter the consideration of reciprocity.
In the preceding discussion, the phase change associated with ray optical paths has not
been accounted for. However, light rays that impinge on an ice crystal at different locations
must experience phase changes due to their different paths. For this reason, Cai and Liou
(1982) considered the phase interference of emerging rays in ray-tracing computations.
The follow-on studies reported in the literature essentially ignored the phase shifts asso-
ciated with ray paths. Takano and Jayaweera (1985) showed that phase interference can
be smoothed out when ice crystals are randomly oriented. For practical applications, we
generally assume that ice crystals are randomly oriented in radiative transfer computations.
If ice crystals are horizontally oriented, however, the single-scattering properties depend
not only on the scattering angle but also on the azimuth of the scattering plane. In this
case, the radiative transfer calculation can be quite involved (see Subsection 2.5.2). In addi-
tion to random orientation, integration over the size spectrum will smooth out fluctuations
produced by phase interferences in the scattering pattern for one ice crystal size.
For randomly oriented ice particles, the corresponding phase matrix has only six inde-
pendent elements, as shown in Subsection 2.4.3. Thus, for the pth-order emerging ray, its
3.3 Generalized geometric optics for light scattering 155

contribution to the phase matrix is given by the following expression (Takano and Jayaweera
1985):
⎡ (p) (p) ⎤
f11 f12 0 0
⎢ (p) ⎥
⎢ f12 f22 (p)
0 0 ⎥
F =⎢
(p) ⎢ ⎥, (3.3.22)
(p) (p) ⎥
⎣ 0 0 f33 f34 ⎦
(p) (p)
0 0 −f34 f44
where the matrix F(p) transforms the incident Stokes parameters to the scattered Stokes
parameters associated with the pth-order outgoing localized wave, and its elements are
defined as follows:
(p)  (p) (p) (p) (p) -
f11 = M1 + M2 + M3 + M4 2, (3.3.22a)
(p) (p)  (p) (p) -
f12 = f21 = − M1 + M2 2, (3.3.22b)
(p)  (p) (p) (p) (p) -
f22 = M1 + M2 − M3 − M4 2, (3.3.22c)
(p) (p) (p)
f33 = S12 + S34 , (3.3.22d)
(p) (p) (p)
f34 = −f43 = −D12 , (3.3.22e)
(p) (p) (p)
f44 = S12 − S34 , (3.3.22f)
and the terms in these equations are given by
 (p) 2
Mi = Ai  ,
(p)
(3.3.23a)
(p) (p)  (p) (p)∗ (p)∗ (p) -
Sij = Sj i = Ai Aj + Ai Aj 2, (3.3.23b)
(p) (p) √  (p) (p)∗ (p)∗ (p) -
Dij = −Dj i = −1 Ai Aj − Ai Aj 2, (3.3.23c)
where the subscripts i and j range from 1 to 4, and the asterisk indicates the complex con-
jugate. Thus, the scattering matrix associated with the various orders of external reflections
and transmissions for all the incident rays can be expressed in the following form:
 N ∞  N
(p)
Fray = σj Fj σj , (3.3.24)
j =1 p=1 j =1

where j denotes the external reflection, while the various orders of transmission are asso-
ciated with the jth initial ray, N is the total number of incident rays, and σj is the
cross-section of the jth initial ray. To speed up the computation, the preceding ray-tracing
algorithm can be augmented with the Monte Carlo hit-and-miss approach described in
Subsection 3.1.3.
In addition to contributions from reflected and refracted rays, diffraction also contributes
to the scattering of the incident wave. According to Babinet’s principle (see Section 3.2),
the diffraction pattern associated with an object is the same as that for an aperture with a
shape identical to the projection of the object on a plane normal to the incident direction.
The diffraction matrix obtained by the scalar Fraunhofer diffraction theory for a scattering
156 Geometric optics for light scattering by ice crystals

particle (see e.g., Jackson 1975) has been used extensively in geometric ray-tracing studies.
It can shown that the transformation matrix associated with diffraction can be written in
the form

dif 

A2 0 k2 (1 + cos θ ) 1 0
A =
dif
= exp(−ikr · ξ)d ξ ·
2
,
0 Adif
1 2π 2 0 1
projected area
(3.3.25)

in line with the A matrix defined for geometric reflection and refraction [see Eq. (3.3.21c)].
From Eq. (3.3.25), the contribution of diffraction to the scattering phase matrix, denoted
by Fdif , can be evaluated. To sum the contributions due to diffraction and Fresnel rays,
proper weighting factors must be accounted for, particularly in cases where an ice particle
is absorptive.
If an ice crystal is absorptive with respect to its incident wavelength, i.e., the imagi-
nary refractive index is non-zero, total absorption can be accounted for by considering the
absorption of individual rays. In general, the absorption cross-section of a particle depends
on the polarization configuration of the incident light. However, for randomly oriented
particles, the absorption cross-section is the average of absorption cross-sections corre-
sponding to two orthogonal polarization cases. Consider a case where polarization of the
incident light is specified as follows: (Ei0u , Ei0v ) = (1, 0). It follows that the intensity of the
first-order refracted field can be obtained from Eq. (3.3.14b) in the form
 t  ∗,T 
E0u , Et0v Et0u , Et0v  i i = (1, 0)(T1  1 )T (T1  1 )∗ (1, 0)T , (3.3.26)
(E ,E )=(1,0)
0u 0v

where the superscript T denotes the transpose of a matrix, and ∗ indicates the complex
conjugate. The intensity given by Eq. (3.3.26) is the amplitude of the Poynting vector in
which the refractive index and change in the ray cross-section due to refraction have been
implicitly accounted for in the refractive matrix given by Eq. (3.3.14c). A similar expression
can be derived for the case when the polarization of a field is given by (Ei0u , Ei0v ) = (0, 1).
Therefore, the contribution of the first-order refracted rays to the absorption cross-section
is given by

N
σabs,1 = 2−1 σj [1 − exp(−4πmi dj 1 /λ)]
j =1
)  ∗,T    ∗,T  *
× Etpu , Etpv Etpu , Etpv  i i + Etpu , Etpv Etpu , Etpv  i i
(E 0u ,E0v )=(1,0) (E
0u ,E0v )=(0,1)


N
= 2−1 σj [1 − exp(−4πmi dj 1 /λ)]
j =1
) *
× (1, 0)(T1  1 )T (T1  1 )∗ (1, 0)T + (0, 1)(T1  1 )T (T1  1 )∗ (0, 1)T (3.3.27a)

where the subscript j denotes the jth initial ray, dj 1 is the distance between the first incident
point (i.e., Q1 in Figure 3.14) and the second incident point (i.e., Q2 in Figure 3.14), σj
is the cross-section of the jth initial ray, mi [denoted as Ni in Eq. (3.1.19)], is the adjusted
imaginary refractive index, and λ is the incident wavelength in a vacuum. Likewise, the
3.3 Generalized geometric optics for light scattering 157

contribution of the pth-order reflected rays is


% &

N 
p−1
−1 −1
σabs,p = 2 σj [1 − exp(−4πmi djp /λ)] exp −4πmi λ dj L
j =1 L=1

× [(1, 0)(Tp  p . . . R2  2 T1  1 )T (Tp  p . . . R2  2 T1  1 )∗ (1, 0)T


+ (0, 1)(Tp  p . . . R2  2 T1  1 )T (Tp  p . . . R2  2 T1  1 )∗ (0, 1)T ]. (3.3.27b)

Thus, the absorption cross-section of an ice crystal can be expressed as follows:




σabs = σabs,p . (3.3.28)
p=1

In practice, the summation in Eq. (3.3.28) can be truncated for the terms with p > 10,
because the amount of energy carried by higher-order rays is relatively insignificant.
Equations (3.3.27a) and (3.3.28) provide explicit formulations for absorption cross-section
within the framework of the ray-tracing technique in which the polarization configuration
is fully accounted for. In the conventional ray-tracing method under the condition of equal
energy partition of diffraction and geometric reflection/refraction, the extinction cross-
section is twice the projected area of the scattering particle; that is, σext = 2σp , where σp is
the particle’s projected area on a plane normal to the incident direction.
One of the shortcomings of the conventional ray-tracing method is the production of
delta-transmission associated with the refraction of rays through two parallel faces of
pristine ice crystals. The delta-transmission phenomenon has been identified and discussed
in detail by Takano and Liou (1989a) and Mishchenko and Macke (1998). Let the portion of
the scattering cross-section associated with the delta-transmission be σδ . Then the scattering
cross-section can be separated into three terms as follows:

σsca = (σp − σabs − σδ ) + σδ + σp . (3.3.29a)

The first term corresponds to contributions from externally reflected rays and various
transmitted rays excluding delta-transmitted rays, the second term denotes contributions
from delta-transmitted rays, and the third term is associated with diffraction.
Let fδ be the ratio of delta-transmitted energy to the total scattered energy, defined by

fδ = σδ /σsca = σδ /(2σp − σabs ) = σδ /(σext − σabs ). (3.3.29b)

The scattered Stokes vector can be expressed as follows [see Eq. (2.4.37)]:
⎡ ⎤ ⎡ ⎤
I I0
⎢Q⎥ ⎢ ⎥
⎢ ⎥ = σs P ⎢ Q0 ⎥ , (3.3.30a)
⎣ U ⎦ k 2 r 2 ⎣ U0 ⎦
V V0

where (I0 , Q0 , U0 , V0 ) and (I, Q, U, V ) represent incident and scattered Stokes parameters,
respectively, and P is the normalized scattering phase matrix. Based on Fray in Eq. (3.3.24),
158 Geometric optics for light scattering by ice crystals

Fdif defined on the basis of Adif in Eq. (3.3.25), the expressions in Eqs. (3.3.29a, b), and
the associated physical meanings of these quantities and expressions, the normalized phase
function is given by (Yang and Liou 2006)

[(2σp − σabs )(1 − fδ ) − σp ] a Fray + 2(2σp − σabs )fδ δ(cos  − 1)Ī¯ + σp b Fdif
P() =
2σp − σabs


¯ 1 1
= 2fδ δ(cos  − 1)Ī + (1 − fδ ) − a Fray + b Fdif , (3.3.30b)
2 2
where  = σsca /σext is the single-scattering albedo. In Eq. (3.3.30b),  is the scattering
angle, Ī¯ is a unit 4 × 4 matrix, and the two parameters a and b are normalization factors
given, respectively, by the following two equations:
2 2
a= π , b=  π . (3.3.31)
Fray () sin  d Fdif () sin  d
0 0

With the normalization factors given in Eqs. (3.3.31), it can be shown that the phase matrix
in Eq. (3.3.30b) is normalized such that the first phase matrix element P11 (i.e., the phase
function) satisfies the normalization condition defined in Eq. (2.4.34).
The conventional ray-tracing technique follows the postulation that the energy attenuated
by a scattering particle is equally partitioned into two parts: extinction associated with
diffraction and extinction due to Fresnel reflection and refraction. In this case, the extinction
efficiency, the ratio of extinction cross-section to the projected area of an ice particle, is
2 regardless of the particle’s size and shape. In addition, computation of the far field by
directly applying the ray-tracing technique leads to delta-transmission (Takano and Liou
1989a) in the forward direction, as is evident from the presence of a delta function. To
overcome these shortcomings, Yang and Liou (1995, 1996b, 1997) have developed two
improved geometric-optics approaches, described in the following Subsections 3.3.3 and
3.3.4.

3.3.3 Improved geometric-optics approach using the electromagnetic


equivalence theorem
In principle, the laws of geometric optics are applicable to light scattering by a particle whose
size is much larger than the incident wavelength so that rays can be localized. In addition
to the requirement of the localization principle, the conventional geometric ray-tracing
technique assumes that the energy attenuated by the scatterer may be decomposed into
equal extinction from diffraction and from Fresnel rays. Also, the Fraunhofer diffraction
formulation used in geometric ray-tracing does not account for the vector property of
electromagnetic fields. Finally, direct calculations of the far field by ray-tracing will produce
a discontinuous distribution of the scattered energy, such as the delta transmission noted by
Takano and Liou (1989a).
3.3 Generalized geometric optics for light scattering 159

To circumvent a number of shortcomings in the conventional geometric-optics approach,


an improved method has been developed by Yang and Liou (1995, 1996a). The concept is
simple in that the energies determined from geometric ray-tracing at the particle surface are
collected and mapped to the far field. This differs from the conventional approach, which
collects energies produced by geometric reflections and refractions directly at the far field
through a prescribed solid angle (see Figure 3.14).
Tangential components of the electric and magnetic fields (see Section 2.2.2) on a surface
S that encloses the scatterer can be used to determine equivalent electric and magnetic
currents for the computation of a scattered far field on the basis of the electromagnetic
equivalence theorem (Schelkunoff 1943). In this theorem, the electromagnetic field detected
by an observer outside the surface would be the same as if the scatterer was removed and
replaced by equivalent electric and magnetic currents given by

J = ns × H, (3.3.32a)
M = E × ns , (3.3.32b)

where ns is the outward unit vector normal to the surface. For the far-field region, the
scattered electric field can be represented by the following integral equation:
 3
eikr k 2  r  r 4 r
Es (r) = M(r ) + × J(r ) exp −ikr · d 2 r , (3.3.33)
ikr 4π r S r r

where rr denotes the scattering direction, r is the reference position vector, r is the
position vector of the source point, and k is the wavenumber. The far-field solution can also
be determined by a volume integral involving the internal field.
By means of geometric ray-tracing, the electric field on the surface of a particle can
be evaluated after successive application of Fresnel reflection and refraction coefficients
parallel and perpendicular to a defined reference plane at the point of interaction, taking
into account the path length in 3-D geometry. If an ice crystal shape is of great complexity,
as with an aggregate, its surface can be defined as a cubic box so that the computation
of the electric field can be conducted on a regularly shaped surface. The electric field of
illuminated and shadowed sides can be defined as follows:

Ea (r) + Eb (r), illuminated side,
E(r) = (3.3.34a)
Eb (r), shadowed side,

where


Ea (r) = Ei (r) + Er1 (r), Eb (r) = Etp (r). (3.3.34b)
p=2

In these equations, Ei is the incident electric field, Er1 is the electric field for external
reflection, and Etp are the electric fields produced by two refractions and internal reflections
(p ≥ 2). Because the transverse electromagnetic wave condition is implied in ray-tracing,
160 Geometric optics for light scattering by ice crystals

the magnetic field for each reflection and refraction for r outside the ice particle can be
obtained from the following equation:

p (r) = ep × Ep (r).
Hr,t r,t r,t
(3.3.35)
In practice, mapping the near-field solution to the far field can be accomplished in its
entirety for Ea in Eq. (3.3.34b). However, for Eb in Eq. (3.3.34b), the mapping must be
performed ray-by-ray so as to take into account full phase interferences in the determination
of the phase function.
We shall begin with Maxwell’s equations to derive integral equations for absorption and
extinction cross-sections. For a dielectric medium with an incident harmonic wave whose
time dependence is given by exp(iωt), Maxwell’s curl equations defined in Eqs. (2.2.7a)
and (2.2.7b) can be written as
∇ × H = ik(εr − iεi )E, (3.3.36a)
∇ × E = −ikH, (3.3.36b)
where k = ω/c. By defining S as the complex Poynting vector and using the preceding two
equations along with vector algebra, we obtain the following two equations:
c
S= E × H∗ , (3.3.37a)

iω ωεi
∇ ·S= (εr E · E∗ − H · H∗ ) − E · E∗ , (3.3.37b)
4π 4π
where the asterisk denotes the complex conjugate. Taking the real part of Eq. (3.3.37b) and
integrating it over the region inside the particle leads to



 3   2 
Re ∇ · S(r )d r = Re nS · S(r )d r
V S
 (3.3.38)
ω   ∗  3 
=− εi (r )E(r ) · E (r )d r ,
4π V

where nS is the outward-pointing unit vector normal to the particle surface. Based on
the physical meaning of the Poynting vector and the principle of conservation of energy
(Jackson 1975), the surface integration term in Eq. (3.3.38) is the net rate at which elec-
tromagnetic energy intersects with the particle surface; that is, the energy absorbed by the
particle.
Furthermore, the incident electromagnetic flux is given by
c c
F0 = E0 · E∗0 = |E0 |2 . (3.3.39a)
4π 4π
Thus, the absorption cross-section of the particle can be expressed as

 
σa = −Re nS · S(r )d 2 r F0
S
 (3.3.39b)
k   ∗  3 
= εi (r )E(r ) · E (r )d r .
|E0 |2 V
3.3 Generalized geometric optics for light scattering 161

In what follows, we shall derive an expression for the extinction cross-section similar
to Eq. (3.3.39b). In this connection, we consider the Poynting vector, which is defined in
Eq. (2.4.1), and note that it can be decomposed into incident S0 , scattered Ss , and extinction
Se components as follows:
S = S0 + Ss + Se , (3.3.40a)
where the complex extinction component of the Poynting vector is given by
c
Se = (E0 × H∗ + E∗ × H0 ). (3.3.40b)

We note the following vector calculus identity:
∇ · (A × B) = B · (∇ × A) − A · (∇ × B) . (3.3.40c)
Hence, we obtain from Eqs. (3.3.40b) and (3.3.40c) the following relationship:
c
∇ · Se = ∇ · (E0 × H∗ + E∗ × H0 )

c
= [H∗ · (∇ × E0 ) − E0 · (∇ × H∗ ) + H0 · (∇ × E∗ ) − E∗ · (∇ × H0 )].

(3.3.40d)
Moreover, using Eqs. (3.3.36a) and (3.3.36b), we get
c
∇ · Se = [H∗ · (−ikH0 ) − E0 · (−ikε∗ E∗ ) + H0 · (ikH∗ ) − E∗ · (ikE0 )]


∗ (3.3.40e)
iω ∗ iω
= (ε − 1)E∗ · E0 = − (ε − 1) E · E∗0 .
4π 4π
Further, by using Gauss’ divergence theorem denoted in Eq. (2.2.12), which transfers area
integral to volume integral, we have



Re ns · Se (r  )d 2 r = Re ∇ · Se (r  )d 3 r
S V

  ∗
iω   ∗  3 
= −Re [ε(r ) − 1]E(r ) · E0 (r ) d r ,
V 4π
(3.3.40f)
and, noting that Re[(iA)∗ ] = −Im [A], we then obtain



 2  ω   ∗  3 
Re nS · Se (r )d r = Im [ε(r ) − 1]E(r ) · E0 (r )d r . (3.3.41a)
S 4π V

Consequently, the extinction cross-section is given by


 
ω   ∗  3 
σe = − Im [ε(r ) − 1]E(r ) · E0 (r )d r F0
4π V
  (3.3.41b)
k
= −Im [ε(r ) − 1]E(r ) · E∗0 (r )d 3 r .
|E0 |2 V
162 Geometric optics for light scattering by ice crystals

4 λ = 0.63 µm
L /a = 6


3

IGOM CGOM

2

FDTD
1



0 

100 101 102 103


Size Parameter kL

Figure 3.16 Comparison of the extinction efficiency computed from conventional (CGOM) and
improved (IGOM) geometric-optics approaches and from the finite-difference time domain (FDTD)
method (see Section 4.1) as a function of size parameter. On the basis of the conventional approach,
the extinction efficiency is 2 based on equal energy partition between geometric and diffracted light
rays. The presentations use a wavelength of 0.63 µm and randomly oriented ice columns of a uniform
size identified in the figure (after Liou et al. 2000, with modifications).

Finally, when the ray-tracing technique is applied to obtain the surface field, one must
properly account for area elements from which externally reflected and transmitted localized
waves make a contribution to the surface field. If the cross-section of the incident localized
wave is σi , the area of the particle surface for external reflection is
σ1r = −σi (ni · ei )−1 . (3.3.42a)
For the transmitted rays, the area is given by
 )  *−1
σpt = −σi n1 · et1 (n1 · ei ) np · etp , p = 2, 3, 4, . . . , (3.3.42b)
where all unit vectors have been defined in Eqs. (3.3.1a) and (3.3.2b). The radius of a ray’s
cross-section is on the order of k −1 so that the phase change over this cross-section is not
significant and we may properly account for the phase interference of localized waves by
using phase information at the centers of the rays. Because phase variation over a ray’s
cross-section can be neglected, numerical results are not sensitive to the ray’s cross-sectional
shape. Hence, we may use a circular shape in the calculations.
In Figure 3.16, we first compare the extinction efficiency obtained from the conventional
geometric-optics method (CGOM) and that computed from the improved geometric-optics
3.3 Generalized geometric optics for light scattering 163

method (IGOM). We then compare the results obtained from these two methods with those
determined from the finite-difference time domain (FDTD) method (see Section 4.1). On
the basis of the conventional approach, the extinction efficiency is 2, based on the equal
energy partition of geometric and diffracted light rays (see Subsection 3.3.1). The results
computed from IGOM and FDTD used randomly oriented solid columns of uniform sizes
with an aspect ratio of 6 and 0.63 µm wavelength in the computations. The size parameter
is defined by kL, with k = 2π/λ. The extinction ripples shown for IGOM are produced by
interference between geometric and diffracted rays, which converge to 2 for size parameters
approaching 1000. Because of computer time restrictions, FDTD computations were limited
to sizes smaller than 30, as denoted by crosses in the figure. The extinction efficiencies
obtained from IGOM and FDTD show excellent agreement in the size parameter range
20–30, but IGOM results deviate from FDTD for size parameters smaller than about 20.
Figure 3.17 compares the phase functions computed from IGOM and FDTD methods as
a function of the scattering angle for kL = 20 with aspect ratios of La = 6 (column) and 2
(plate). For the former, the ice crystal geometric cross-section is small, so the computation
of surface fields may not be sufficiently accurate. We see differences at scattering angles
larger than 70°. For the latter, because the ice crystal cross-section is increased by a factor
of 3, the accuracy of IGOM is greatly improved and the computed major features mimic
those produced by the FDTD method. Also shown in Figure 3.17 are the phase functions
computed by IGOM and FDTD for hollow columns and bullet rosettes. The depth of
the hollow pyramid is shown as d; the cross angle for the bullets is 90°; the tip length
of a bullet is t; and other parameters defining the geometry of the ice crystals are given
in the diagrams. The phase function patterns for solid- and hollow-column crystals are
similar, except in backscattering directions – a pronounced scattering maximum is shown
at a scattering angle of 155° in both cases. This feature does not appear in the case
of bullet rosettes. It appears that agreement of the phase function results computed from
two completely different approaches, IGOM and FDTD, is quite reasonable. The FDTD
method has inherent numerical limitations and becomes computationally expensive for size
parameters larger than 20 (see Section 4.1).

3.3.4 Ray-by-ray approach to extinction and absorption calculations


To further enhance the applicability of the geometric-optics method to small size parameters,
Yang and Liou (1997) developed a ray-by-ray approach, specifically for the calculation of
extinction and absorption cross-sections. The amplitude scattering matrix S(r), required
for single-scattering calculations, is the sum of the contributions from all localized rays as
follows:

S(r) = Sp (r), (3.3.43a)
γ p

where the vector r denotes the scattering direction. The first summation covers all the
incident rays impinging onto the sphere, denoted by γ , while the second summation is over
164 Geometric optics for light scattering by ice crystals

102 102
l=0.55μm l=0.55μm
kL=20 kL=20
Phase Function P11 101 L/a=6 101 L/a=2

Phase Function P11


100 100

10-1 10-1

FDTD FDTD

10-2 10-2
IGOM IGOM

10-3 10-3
0 60 120 180 0 60 120 180
Scattering Angle (deg.) Scattering Angle (deg.)

102 102
l=0.55μm l=0.55μm
kL=20 kL=5
101 d/L=0.25 101 d/L=0.5
L/a=6 a/L=0.25
Phase Function P11
Phase Function P11

100 100

10-1 10-1

FDTD
FDTD
10-2 10-2
IGOM
IGOM

10-3 10-3
0 60 120 180 0 60 120 180
Scattering Angle (deg.) Scattering Angle (deg.)

Figure 3.17 Comparison of phase functions computed by IGOM and FDTD methods at a 0.55 µm
wavelength for solid and hollow columns, plates, and bullet rosettes as a function of the scattering
angle (after Yang and Liou 1996a, with modifications).

the internal localized rays, denoted by subscript p ( = 1, 2, 3 . . . ), based on the localization


principle. The amplitude scattering matrix for an individual ray can be written in the
form


k2  (1 − ε)
Sp (r) = Sp,q (r) = Kp Up {exp[iζp+1 (r)] − exp[iζp (r)]} ,
q
4π q m − r · ep q

(3.3.43b)
3.3 Generalized geometric optics for light scattering 165

where the phase of a ray is defined by


⎛ ⎞

p−1
ζp (r) = k ⎝e0 · rQ1 + mj dj − r · rQp ⎠ , (3.3.43c)
j =1

and m denotes an average complex refractive index in a general inhomogeneous case and
can be evaluated by the Maxwell Garnett mixing rule (Chýlek et al. 1984). The term mj
represents the complex refractive index for the inhomogeneous layer in a sphere; ε is the
permittivity; k is the wavenumber; Q1 , Qp , and ep are defined in Figure 3.14; and dj is the
distance between the points Qj and Qj+1 , defined by |rQj +1 − rQj |. The summation over q
signifies the travel of a ray through homogeneous segments of an inhomogeneous particle.
The main objective is to effectively compute extinction and absorption cross-sections.
For this reason, we may simplify the preceding matrix formulation by replacing the factor
Kp Up  by ±1, so that the coordinate transformation for polarization is neglected. The
diagonal elements of the scattering matrix Sp (r) can then be written in the form

k2  1−ε
Sjj,p (r) = ± {exp[iζp+1 (r)] − exp[iζp (r)]} , (3.3.44)
4π q m − r · ep q

where the ± corresponds to the sign of the cumulative product of Fresnel coefficients,
p−1
tj2 rj ; rj can take a + or − sign; and j = 1 or 2, denoting parallel or perpendicular
components, respectively.
The extinction cross-section is defined by the sum of the two diagonal elements in the
forward direction as follows [see also Eq. (2.4.45d)]:

σe = Re[S1 (e0 ) + S2 (e0 )], (3.3.45a)
k2
where e0 denotes the incident direction and the absorption cross-section is given by
⎛ ⎞
 ∞ 
p−1
 p−1 p−1 -
σa = exp ⎝−2k mi,j dj ⎠ [1 − exp(−2kmi,p dp )] t12 r1 + t22 r2 2,
γ p=1 j =1

(3.3.45b)
where mi,j (or p) represents the imaginary part of the refractive index for an inhomogeneous
particle. For a spherical particle, Eq. (3.3.44) is evaluated as if the path of twice-refracted
rays (p = 1) were not deflected at all, so that the phase of a ray in this equation is given by
ζ2 = 2ka cos τi (m − 1), which is the dominant exponential term for extinction, where τi is
the incident angle and a is the radius of a sphere. Note that ζl = 1, and ζp (p > 2) terms
vanish due to phase cancelation.

3.3.5 Contributions of surface waves to spheres: the edge effect


For a spherical ice particle, the extinction and absorption efficiencies and the radiation pres-
sure evaluated from the geometric-optics approach generally deviate from results computed
166 Geometric optics for light scattering by ice crystals

Incident wave front


surface wave

reflected wave

Incident localized wave


 Q1
e0 refracted wave

e1
e3 
Q2
reflected
wave
e2
diffracted wave


Q3

Figure 3.18 A conceptual ray-by-ray geometric-optics approach which includes surface-wave con-
tributions in addition to diffracted waves and localized geometric reflection and refraction (referred
to as GOS) for light scattering and absorption by a sphere (after Liou et al. 2010, with modifications).

from the exact Lorenz–Mie theory, due principally to the neglect of surface waves along
the edge of the spherical particle. These waves are produced by the interaction of incident
waves at grazing angles near the edges of a sphere and continuation of the wave motion
along its surface into the shadow region. If the sphere is relatively small, the waves may
move around and encompass the entire spherical surface (Figure 3.18). Also shown are
diffracted waves and waves undergoing geometric reflection and refraction. It is important
to state that surface and diffracted waves are not subject to the principle of localization for
light rays that undergo reflection and refraction, as noted previously.
Nussenzveig and Wiscombe (1980) and Nussenzveig (1992) have presented physical
equations for the calculation of surface waves based on complex angular momentum theory,
which makes use of the transformation of the Debye expansion of two scattering functions
in a complex domain. This procedure allows the mapping of localized incident rays into
a complex domain such that the Airy integral can be incorporated in the analysis. In the
remainder of this subsection, we try to capture key elements of the contribution of surface
waves to light scattering processes.
In order of size parameter x, we may express the surface-wave term for extinction in the
form
−1/2
Qext = c1 x −2/3 + 2Im[(m2 + 1)(m2 − 1) ]x −1 − c2 x −4/3
(3.3.46)
−3/2
− c3 Im[eiπ/3 (m2 − 1) (m2 + 1)(2m4 − 6m2 + 3)]x −5/3 ,

where the coefficients c1 = 1.992 39, c2 = 0.715 35, c3 = 0.664 13, and Im denotes the
imaginary part of the term. We see that the size parameter appears in the denominator of
3.3 Generalized geometric optics for light scattering 167

Eq. (3.3.46), and for this reason Qext diverges when x approaches small values. Under
this condition, specific corrections must be made to ensure its physical continuity for
computational purposes.
The edge effect contribution to absorption is governed by the following two integrations:
 2  ya  yb
−1/3 −2/3 + − −
Qabs = 2 x ϕ(rj l )dy + [ϕ(rj l ) − ϕ(r̃j l )]dy , (3.3.47)
l=1 0 0

where l denotes two polarization components and the function ϕ is given by


ϕ(rj l ) = (1 − e−b )(1 − r2l )/(1 − r1l e−b ), (3.3.48)
where r2l and r1l are, respectively, the external and internal reflectivities for the state of
polarization l given by
rj l = |Rj l |2 , j, l = 1, 2; Rj,l = (−1)j (zj − uel )/(z + uel ), (3.3.49a)
 
z = cos θ, u = m cos θ , sin θ = m sin θ , el = 1,
(3.3.49b)
e2 = m−2 , z1 = z, z2 = z∗ ,
where z∗ denotes the complex conjugate of z and b = 4x Im(m cos θ  + θ  sin θ ). The term
rj±l in Eq. (3.3.47) can obtained from rjl such that z is replaced by
z± = −(2/x)1/3 eiπ/6 Ai (±ye2iπ/3 )/Ai(±ye2iπ/3 ), (3.3.50)
where Ai is the Airy function (Nussenzveig and Wiscombe 1980) and a prime on Ai signifies
a derivative, and θ is related to y by the following equation:
sin θ = 1 ± 2−1/3 x −2/3 y, (3.3.51)
where the + and − signs apply to the first and second integrals in Eq. (3.3.47), respectively.
The integration limits are given by
ya = 21/3 (m − 1)x 2/3 , yb = (x/2)2/3 . (3.3.52)
The term r̃j−l is obtained from rj−l such that z− is replaced by (2/x)1/3 y 1/2 .
The radiation pressure term required for calculation of the asymmetry factor for a
spherical particle can also be defined by the sum of two integrations, as follows:
2  ya  yb
Qpr = −2−1/3 x −2/3 Re (ρl+ − τl+ + 1)dy + [(ρl− − ρ̂l− ) − (τl− − τ̂l− )]dy ,
l=1 0 0
(3.3.53)

where Re denotes the real part of the term, which is generally negative, and
∗ 
ρl = f1 (z)R2l R2l ,
τl = f1 (z)f2 e−b (1 + R1l
∗ 
)(1 + R1l ∗
)(1 + R2l 
)(1 + R2l ∗ 
)(1 + R1l R1l f2 e−b )−1 ,
Rj 1 = (fj )−1 [m2 zj − u + (−1)j iM 2 ](m2 z + u + iM 2 )−1 ,
Rj 2 = (fj )−1 [(m2 + M 2 )zj − u + (−1)j iM 2 (1 − uzj )]
× [(m2 + M 2 )z + u + iM 2 (1 + uz)]−1 ,

f1 (z) = (1 + iz∗ )/(1 − iz), f2 = e−2iθ , (3.3.54)
168 Geometric optics for light scattering by ice crystals

with M 2 = m2 − 1. In all the parameters in Eq. (3.3.53) with ± upper indices, substitutions
denoted by Eqs. (3.3.50) and (3.3.51) are required. The terms ρ̂l− and τ̂l− are obtained from

ρl− and τl− such that z− is replaced by (2/x)1/3 ( y + i/4y).

3.3.6 The geometric-optics and surface-wave approach for spheres


A combination of the geometric-optics term and the surface-wave adjustment (referred to
as GOS) should, in principle, constitute a solution close to the “exact” solution derived
from the Lorenz–Mie (LM) theory such that

Qw (GOS) = Qw (GO) + Qw ≈ Qw (LM), (3.3.55)

where w denotes ext (extinction), abs (absorption), or pr (radiation pressure). In this equa-
tion, linearity between the geometric-optics (GO) and surface-wave () terms is implied.
The conventional geometric approach can be employed to calculate the asymmetry fac-
tor for homogeneous and layer spheres, which in combination with known extinction and
absorption coefficients can be used to evaluate a term referred to as radiation pressure
(van de Hulst 1957). Light rays can carry momentum as well as energy. The part of the
forward momentum that is removed from incident rays, which is not represented by the
forward momentum associated with scattered rays, is related to the hemispheric average of
the phase function weighted by cos θ , where θ is the zenith angle in Cartesian coordinates.
This term is referred to as the asymmetry factor. We can express the radiation pressure term
as follows:

Qpr (GOS) = Qext (GOS) − g(GOS)[Qext (GOS) − Qabs (GOS)], (3.3.56)

where g represents the asymmetry factor, Qext (= σe /πa 2 ) is the extinction efficiency,
Qabs (= σa /πa 2 ) is the absorption efficiency, and the index GOS denotes the terms evaluated
by the geometric-optics and surface-wave approach.
Based on Eq. (3.3.56), the asymmetry factor can be written in the form
Qext (GOS) − Qpr (GOS)
g(GOS) = . (3.3.57)
Qext (GOS) − Qabs (GOS)
The term Qext breaks down when x approaches zero, since the size parameter x appears
in the denominator in the right-hand side of Eq. (3.3.46). To investigate applicability of the
surface-wave term derived from complex angular momentum theory, we have conducted
an analysis of Qext in terms of the phase shift parameter, a function of both the refractive
index and the size parameter, defined as ρ = 2 |m − 1| x. The surface-wave adjustment term
Qext , which is normally on the order of 0.1, approaches an unrealistically large number by
a factor of more than 10 when ρ  2.86. The extinction efficiency results computed from the
ray-by-ray geometric-optics approach outlined above, without including the surface-wave
effect, match closely with those determined from the exact LM theory when ρ < 2.86, so
surface-wave adjustments are not required in this region.
3.3 Generalized geometric optics for light scattering 169

Extinction Efficiency Q ext Single-Scattering Albedo v Asymmetry Factor g


1.0 1.0
4
0.9
LM 0.9
m = 1.3 − i 0.0

GOS
3
0.8 0.8

2 0.7
0.7

1 0.6 0.6

0 0.5 0.5
1.0 1.0
4
0.9 0.9
m = 1.3 − i 0.01

3
0.8 0.8
2
0.7 0.7

1 0.6 0.6

0 0.5 0.5
0.8 1.0
4
0.7 0.9
m = 1.3 − i 0.1

3
0.6 0.8

2
0.5 0.7

1 0.4 0.6

0 0.3 0.5
0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100
Size Parameter x Size Parameter x Size Parameter x

Figure 3.19 Comparison of extinction efficiency, single-scattering albedo, and asymmetry factor
as a function of size parameter between the results computed from Lorenz–Mie (LM) theory and
the geometric-optics and surface-wave (GOS) approach for a real refractive index of 1.3 and three
imaginary indices of 0.0, 0.01, and 0.1 (after Liou et al. 2010).

We have carried out a series of comparisons between the extinction efficiency Qext , the
single-scattering albedo , and the asymmetry factor g determined from GOS, and those
computed from LM equations for homogeneous spheres with size parameters ranging from
1 to 100. Figure 3.19 displays comparison results for mr = 1.3 and mi = 0.0, 0.01, and 0.1.
Both Qext and  values computed from the GOS approach compare remarkably well with
the exact LM results. Maximum and minimum oscillations in the single-scattering patterns
are seen to be progressively damped when mi or absorption increases.

3.3.7 Application of the geometric-optics and surface-wave approach to ice crystals


3.3.7.1 Fundamentals
For application of the GOS approach to non-spherical ice crystals, we shall consider a gen-
eral aggregate particle comprised of a collection of core–shell spheres built by a stochastic
170 Geometric optics for light scattering by ice crystals

Surface wave

Reflected
photons

Incident photons

Refracted Diffracted wave


photons

Figure 3.20 A conceptual depiction of the geometric-optics and surface-wave approach to light
scattering by an ensemble of core–shell spheres (aggregates), representing general non-spherical
particles, including ice crystals. The geometric-optics components include the hit-and-miss Monte
Carlo photon tracing associated with external and internal reflections and refractions, and diffraction
based on Babinet’s optical principle for randomly oriented non-spherical particles. The surface waves
travel along the edge of the particles and propagate into shadow regions (after Liou et al. 2011 with
modifications).

process (Friedlander 2000), as illustrated in Figure 3.20. This diagram demonstrates the
concept of the traversing of surface waves, or the edge effects, which can touch the spheres
when they propagate in forward directions and can move into shadow regions of the aggre-
gate. Also shown are diffracted waves acting upon the projected cross-sectional area in
accordance with Babinet’s principle and passing around this area. In geometric reflec-
tion/refraction, we have employed the hit-and-miss Monte Carlo technique to treat the
interaction of light beams in terms of rays (see Subsection 3.1.3) in order to account for
the intricacy and complexity of an aggregate comprised of spheres with layered structures,
following the ray-by-ray approach described in Subsection 3.3.4.

3.3.7.2 Geometric optics


In conjunction with the conceptual depiction in Figure 3.20, consider a group of randomly
oriented ice crystals in space to replace the aggregate. Let A be the area of a square
sufficiently large to cover the geometric cross-section of an individual ice crystal and set
the square equal to L2 , where L denotes the crystal’s maximum dimension. Further, let Na
be the number of photons incident on the ice crystals, which is dependent on the particle’s
orientation, and Nt be the number of total photons used in the calculation. Thus, the effective
geometric shadow of an ice crystal on a plane perpendicular to the incident light beam,
whose propagation direction is along the z-axis, can be defined as follows:

As (α, β) = A [Na (α, β)/Nt ] (3.3.58)


3.3 Generalized geometric optics for light scattering 171

where α and β are angles that denote the orientation of an aggregate in a two-dimensional
plane with reference to the incident light beam.
It follows that for an ensemble of randomly oriented ice crystals, the extinction and
absorption efficiencies averaged over all orientations can be written in the form
   
Qext,abs = σe,a (α, β) cos αdαdβ As (α, β) cos αdαdβ. (3.3.59)
α β α β

The original theoretical development of diffraction begins with Babinet’s principle,


which states that the diffraction pattern in the far field (Fraunhofer diffraction) from a
circular aperture is the same as that from an opaque disk or sphere of the same radius,
as discussed in Section 3.2. Following this principle and in order to apply it to randomly
oriented non-spherical particles, we may use the effective geometric cross-sectional area
defined in Eq. (3.3.58). It follows that the diffracted intensity Id for an ensemble of randomly
oriented ice particles can be expressed as
   2
 
Id ∝  exp[−ikr(x, y, α, β) dxdy] cos αdαdβ, (3.3.60)

α β As

where r is the distance between a point in the shadow As and a point distant from the shadow.
Analytical solutions of Eq. (3.3.60) exist only for a number of simple non-spherical shapes,
such as column/plate, rectangular parallelepiped, and spheroid/ellipsoid (see Section 3.2).
However, “exact” numerical calculations can be carried out for irregular ice particles to
obtain the diffracted energy.
On the basis of the scattered intensities determined from geometric reflection/refraction
and diffraction for randomly oriented ice particles, the asymmetry factor g can be deter-
mined. We can then evaluate the term referred to as radiation pressure efficiency, the concept
of which can be applied to non-spherical particles as follows:

Qpr (GO) = Qext (GO) − g(GO)[Qext (GO) − Qabs (GO)] (3.3.61)

where GO stands for geometric optics and the extinction and absorption efficiencies Qext
and Qabs have been defined in Eq. (3.3.59).

3.3.7.3 Addition of surface waves


In the context of light scattering and absorption by a sphere, surface waves are generated by
interactions of incident waves at grazing angles near the sphere’s edge, and they propagate
along its surface into the shadow region, as shown in Figure 3.18. On the basis of theoretical
postulations and numerical calculations, Liou et al. (2011) demonstrated that a linear
combination (reflection/refraction and diffraction, denoted by GO) of the geometric-optics
term and coupled with the surface-wave adjustment (denoted by GOS) constitutes a solution
close to the exact Lorenz–Mie (LM) theory, so that

Qw (GOS) = Qw (GO) + f Qw ∼ Qw (LM), w = ext, abs, pr (3.3.62)


172 Geometric optics for light scattering by ice crystals

where Qw is the efficiency factor for extinction (ext), absorption (abs), or radiation pressure
(pr), and Qw is the surface-wave adjustment. Additionally, a parameter f has been added
into this equation for the case of non-spherical particles. In the previous analysis for spheres,
f was set to be 1. However, for large size parameters, greater than about 50, f is close to 0.
Once Qpr has been determined, the asymmetry factor from Eq. (3.3.57) applied to the case
of GOS can be expressed in the form

g(GOS) = [1 − Qpr (GOS)/Qext (GOS)]/ (GOS), (3.3.63)

where the single-scattering albedo  is defined by Qsca /Qext and Qsca = Qext − Qabs .
Surface waves differ from diffraction, which originates from an incomplete wave front
due to blocking of the incident wave by a group of randomly oriented ice particles and
is governed by the geometric cross-sectional area facing the incident light beam. Surface
waves propagate along and touch ice crystal surfaces and must therefore be governed by
absorption in addition to scattering. The theoretical foundation for surface waves exists
only for spheres, as discussed in Subsection 3.3.5. For non-spherical particles, a physically
based adjustment must be developed in order to effectively account for the shape factor.
Based on physical arguments, it is postulated that irregular surfaces will suppress or dampen
the propagation of these waves and reduce the effectiveness of their interaction with ice
particles – slow down the waves, so to speak. By virtue of its smooth surface, a perfect
sphere is the most efficient geometry for surface-wave propagation. Therefore, the parameter
f introduced in Eq. (3.3.62) must be less than 1 for non-spherical ice particles; it must be
close to zero for an elongated particle since the probabilities for the forward and backward
propagation of surface waves would be limited in this case. Volume (three-dimensional)
must also play a key role in the effectiveness of surface-wave contributions.
In the following analysis, consider two types of volume. First, we may define a volume
for an ice crystal as follows:
-
Vv = n × 4πrv3 3, (3.3.64a)

where rv is the equivalent radius of a sphere with the same volume as the individual ice
particle, and n is the total number of ice crystals. We may also define a volume corresponding
to the geometric cross-sectional area As , defined in Eq. (3.3.58), such that
-
Va = n × 4πra3 3, (3.3.64b)

where πra2 = As . It is postulated that the correction parameter in Eq. (3.3.62) for non-
spherical particles may be defined by the ratio of two volumes as follows:

f = Vv /Va = (rv /ra )3 . (3.3.65)

It turns out that Vv (or rv ) is smaller than Va (or ra ), so that f is equal to or less than
1. For spheres, rv = ra , and f = 1; for elongated particles, rv  ra , f  0. These limits
provide a theoretical basis and boundary conditions for the application of Eq. (3.3.62) to
ice particles.
3.3 Generalized geometric optics for light scattering 173

3.3.7.4 Comparison with available FDTD and DDA results


The preceding theoretical development for non-spherical particles has been compared to
calculations of light scattering and absorption by column and plate crystals using the finite-
difference time domain (FDTD) method (Yang and Liou 1996a). Comparison has also been
made to calculations using the discrete dipole approximation (DDA) method originally
developed by Purcell and Pennypacker (1973) and later by Draine and Flatau (1994),
in which single-scattering results for small size parameters for column and plate can be
computed from existing computer programs. The DDA method calculates light scattering
and absorption by a target with an arbitrary geometry by defining the target as an array
of discrete dipoles located on a cubic lattice. Both the FDTD and DDA methods require a
substantial amount of computational effort for moderate size parameters, especially when
the 3-D orientation of the non-spherical particles is taken into account. For this reason,
it is computationally challenging to use FDTD and DDA for many practical applications
involving non-spherical particles with size parameters larger than about 20 (see Chapter 4
for further discussion).
For comparison purposes, wavelengths of 3.7 and 23 µm have been selected, which have
ice indices of refraction of 1.4005 − i 0.007 201 and 1.4424 − i 0.027 01, respectively.
Figure 3.21 compares the extinction efficiency, single-scattering albedo, and asymmetry
factor for randomly oriented columns (a) and plates (b) as a function of size parameter,
defined as xmax = krmax (= Lmax /2), where Lmax is the maximum dimension and k is the
wavenumber. Because of computational limitations, FDTD and DDA results are shown
for size parameters less than about 20. For columns, GOS results compare closely with
values computed from FDTD and DDA for both wavelengths for the three single-scattering
parameters. GOS results show slight systematic overestimates of extinction efficiency for
size parameters smaller than 2. For plates, GOS results compare closely with values
calculated from FDTD and DDA for single-scattering albedo as well as asymmetry factor.
Some deviations are shown in extinction efficiency curves; in particular, the GOS approach,
which is a linear combination of the geometric-optics and surface-wave approaches, does
not produce the fluctuations of the FDTD results at 23 µm.

3.3.8 Geometric-optics and surface-wave approach coupled with the


Rayleigh–Gans–Debye approximation
In the preceding presentations, we demonstrated that application of the improved geometric-
optics method (IGOM) can be extended to light scattering by ice particles with size parame-
ters of 20. Further, by incorporation of surface-wave contributions to account for the edge
effect, we have further extended the geometric-optics and surface-wave (GOS) approach
to size parameters on the order of 1. Results computed from FDTD and DDA numerical
methods have been used to calibrate those from GOS for light scattering by column and
plate crystals with size parameters from 1 to 20. Ice crystal sizes in the atmosphere
cover a range from a few µm to thousands of µm (see Figure 1.23) with size parameters
generally larger than 1 for visible and infrared wavelengths. In what follows, we present
(a) Extinction Efficiency Qext Single-Scattering Albedo ϖ Asymmetry Factor g
1.0 1.0
3
l = 3.7μm 0.8 0.8

2 0.6 0.6

Column
0.4 0.4
1
GOS
0.2 0.2
FDTD
DDA
0 0.0 0.0
1.0 1.0
3
0.8 0.8
l = 23 μm

2 0.6 0.6

0.4 0.4
1
0.2 0.2

0 0.0 0.0
1 10 100 1 10 100 1 10 100
Size Parameter xmax Size Parameter xmax Size Parameter xmax

(b) Extinction Efficiency Qext Single-Scattering Albedo ϖ Asymmetry Factor g


1.0 1.0
3
0.8 0.8
l = 3.7μm

2 0.6 0.6
Plate

0.4 0.4
GOS
1
FDTD
0.2 0.2
DDA

0 0.0 0.0
1.0 1.0
3
0.8 0.8
l = 23 μm

2 0.6 0.6

0.4 0.4
1
0.2 0.2

0 0.0 0.0
1 10 100 1 10 100 1 10 100
Size Parameter xmax Size Parameter xmax Size Parameter xmax

Figure 3.21 Comparison of extinction efficiency, single-scattering albedo, and asymmetry factor
for randomly oriented columns (a) and plates (b) as a function of size parameter xmax (= 2π rmax λ),
where rmax denotes one-half of the maximum dimension and λ is the wavelength, as computed from
the geometric-optics and surface-wave (GOS) approach for size parameters up to 100, and from the
finite-difference time domain (FDTD) and discrete dipole approximation (DDA) computer codes for
size parameters smaller than about 20 (after Liou et al. 2011).
3.3 Generalized geometric optics for light scattering 175

an approximation that is applicable to size parameters smaller than 1, as in the case of
aerosols, some of which can serve as effective condensation and ice nuclei for the formation
and growth of ice crystals in the Earth’s atmosphere.
We shall first consider Rayleigh–Gans scattering for a spherical particle. This is an
extension of Rayleigh scattering to particle sizes larger than molecules by taking into
account the interference effect from all elements at different positions in a sphere for the
case when the real refractive index mr is close to 1.
The original Rayleigh–Gans scattering intensity equation was derived by Gans (1925)
in the form


9k 4 V 2 4
I RG = (m r − 1) 2
I0 G2 (u)(1 + cos2 ), (3.3.66)
32π2 r 2 9
where V is the volume of a sphere, k = 2πλ, λ is the wavelength, r is the distance from
the scattering particle to an observation point, I0 is the incident intensity,  is the scattering
angle, and the factor G(u), which accounts for phase interferences, is given by
G (u) = 3 (sin u − u cos u) /u3 , (3.3.67)
where the parameter u = 2 ka sin (2). In Eq. (3.3.66), the polarizability α is approximated
by the term 2(mr −1)3.
In order to apply this to particles with a real index of refraction much larger than 1,
we may replace the factor (mr − 1)2 in the scattering intensity expression by the exact
polarizability formula given by
α 2 = (9/4)|(m2 − 1)/(m2 + 2)|2 , (3.3.68)
where m (= mr − imi ) is the complex refractive index. In this manner, the scattered intensity
can now be expressed in the form
 2
9k 4 V 2  m2 − 1 
I =
RG
I0 G2 (u)(1 + cos2 ). (3.3.69)
32π2 r 2  m2 + 2 
Integrating Eq. (3.3.69) over all directions, we obtain the scattering cross-section as follows:
  2  π
1 9k 4 V 2  m2 − 1 
Csca =
RG
I r d =
RG 2
G2 (u)(1 + cos2 ) sin  d. (3.3.70a)
I0 16π  m2 + 2  0
Additionally, the asymmetry factor is defined by
 π  π
g =
RG RG
I cos  sin  d I RG sin  d . (3.3.70b)
0 0

Because the interference effect does not affect absorption, the absorption cross-section
for Rayleigh–Gans scattering is similar to that for Rayleigh scattering, which is given by
2
m −1
RG
Cabs = −3kV Im , (3.3.71)
m2 + 2
where V is the volume and Im denotes the imaginary part of the bracket value.
176 Geometric optics for light scattering by ice crystals

For general application, we shall consider an aggregated particle (such as black carbon
or soot) consisting of Ns homogeneous spheres and define a form factor F as the ratio of
the scattered intensity for Ns spheres to that for a single sphere. We shall follow the form
factor developed by Debye (1915) in the form


Ns
2 
Ns 
Ns 
Ns 
Ns 
Ns 
Ns
F= |eiq·rj | = ei q·(rj −rk ) = cos(q · rj k ) = cos(|q||rj k | cos η),
j =1 j =1 k=1 j =1 k=1 j =1 k=1
(3.3.72)

where q is the vector corresponding to the scattered wave with a real value given by
|q| = 2k sin(/2), |rj k | = |rj − rk | = rj k , where rj and rk are position vectors of primary
spheres, and η is the angle between the q and rjk vectors. Each primary sphere is governed
by Rayleigh–Gans scattering, and to a good approximation it can be considered to be inde-
pendent of the other primary spheres. Further, we shall consider an ensemble of aggregates
randomly oriented in space so that
 1  1
I RGD = I RG F dcosη ∼ = I RG F dcosη = I RG F̄ , (3.3.73)
0 0

where the mean form factor is given by


 1  Ns 
Ns  1 
Ns 
Ns
sin[2krj k sin(/2)]
F̄ = F dcosη = cos(|q||rj k | cos η) dcosη = .
0 j =1 k=1 0 j =1 k=1
2krj k sin(/2)
(3.3.74)

Accordingly, the scattering cross-section is given by


 2  π
9k 4 V 2  m2 − 1 
RGD
Csca = G2 (u)(1 + cos2 )F̄ sin  d. (3.3.75)
16π  m2 + 2  0
The absorption cross-section is simply Ns times the Rayleigh absorption cross-section,
given by
2
m −1
Cabs = Ns Cabs = −3Ns kV Im
RGD RG
. (3.3.76)
m2 + 2
The preceding equations, (3.3.73)–(3.3.76) constitute the Rayleigh–Gans–Debye (RGD)
approximation, applicable to size parameters x (= ka) smaller than 1.
In Figure 3.22, we show a comparison of the scattering phase matrix for randomly
oriented soot aggregates of Ns = 105, corresponding to a size parameter of 1.39, between
the RGD approximation and the superposition T-matrix method (see Section 4.2 for the T-
matrix approach). The phase function P11 shows a peak at a scattering angle of 0°, resulting
from diffraction, and becomes flat at backscattering directions. The degree of linear polar-
ization −P12 P11 has a maximum value of 1 at a scattering angle of 90°, which is the same
3.3 Generalized geometric optics for light scattering 177

10 2 1.0
P22 /P11
0.8

RGD 0.6

T−matrix 0.4
−P12/P11
0.2
101
Phase Function P11

0.0
0 30 60 90 120 150 180
1.0

0.5
10 0

0.0
P43 /P11
P33 /P11
−0.5
P44/P11

10−1 −1.0
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Scattering Angle (deg.) Scattering Angle (deg.)

Figure 3.22 Comparison of scattering phase matrix elements for a soot aggregate with Ns = 105
(xv = 1.39) between the Rayleigh–Gans–Debye (RGD) approximation and the superposition T-matrix
method (after Takano et al. 2013 with modifications).

as the Rayleigh scattering result, and P21 P11 = P12 P11 under the condition of random
orientation. Values of the matrix element P22 P11 are close to 1, whereas P43 P11  0.
The element P44 P11 is essentially equal to the element P33 P11 at all scattering
angles. Results computed for all matrix elements from the two approaches reveal close
comparison.
Based on the RGD approach, the scattering phase matrix for size parameters less than
about 1 may be expressed in the form [also see Eq. (2.4.58b)]
⎡ ⎤
P11 P12 0 0
⎢ P12 P11 0 0 ⎥
P(ka < 1) ≈ ⎢
⎣ 0
⎥. (3.3.77)
0 P33 0 ⎦
0 0 0 P33

Thus, the scattering phase matrix for randomly oriented small ice particles with size param-
eters less than 1 contains only three independent elements, similar to the Rayleigh scattering
case (Rayleigh 1871, Chandrasekhar 1950).
178 Geometric optics for light scattering by ice crystals

3.4 A unified theory of light scattering by ice crystals based on generalized


geometric optics
3.4.1 The essence of a unified theory of light scattering by ice crystals
In recognition of the inherent limitations of various approaches, including those pre-
sented above and in Chapter 4 for light-scattering calculations, it is unlikely that one
specific method can be satisfactorily employed to resolve intricate and complex scattering–
absorption problems involving ice crystals and snowflakes covering all possible size param-
eters from 0.1 to 1000, the numerous shapes displayed in Section 1.5, as well as external
and internal mixings of aerosol particles serving as ice nuclei or condensation nuclei. For
this reason, we have presented the concept of “a” unified theory of light scattering by ice
crystals (Liou et al. 2000), which combines the IGOM (improved geometric-optics method)
covering size parameters larger than 20 and the FDTD covering size parameters smaller
than 20. In light of the development of generalized geometric-optics approaches discussed
in preceding sections, we present in the remainder of this subsection “a” unified theory of
light scattering by ice crystals covering essentially all size parameters and shapes. We must
hasten to add that such a theory could “in principle” be developed by a combination of
numerical approaches to light scattering solutions as presented in Chapter 4.
It is pertinent herein to define the meaning of “exact” solutions within the context of
light scattering and radiative transfer. In Subsection 2.2.3, we demonstrated that “exact”
solutions exist only for sphere, circular cylinder, and spheroid, for which known coordinate
systems can be imposed, leading to analytic solutions of the electric field in the far field
in terms of a number of mathematical functions. In this context, we have also shown that
“exact” solutions can be derived for diffraction by sphere, rectangular, hexagonal cylinder,
and spheroid on the basis of Babinet’s principle. Numerical solutions developed for the
FDTD, DDA, and T-matrix approaches presented in Chapter 4 for light scattering problems
require a large number of numerical matrix inversions, resulting in substantial computer
time to achieve convergence and significant efforts to circumvent numerical instability.
Therefore, these numerical approaches are not within the purview of our definition of
“exact.”
In essence, we built “a” unified theory of light scattering by ice crystals based on
physical rationale and postulations within the context of first principles, which is critical
for understanding of and insight into the scattering, absorption, and polarization features
that occur in theoretical and numerical results. To demonstrate the unified theory, we shall
first use the spherical case for which the analytic solutions for extinction efficiency, single-
scattering albedo, and asymmetry factor can be derived from the exact Lorenz–Mie (LM)
theory, followed by numerical computations.
In Figure 3.23a, we present a comparison of single-scattering parameters Qext , co-albedo
1−  , and g for a homogeneous ice sphere computed from LM and from the GOS/RGD
approximation as a function of size parameter x from 0.1 to 1000 associated with the
0.55 µm wavelength. In the Qext curve, we show the applicability of the CGOM, IGOM,
and GOS/RGD approaches in reference to exact LM results. In the context of CGOM
(a) (b)
5

3 IGOM

Qext
CGOM
2

1 GOS/RGD
0
10 −4 1

Ice l = 0.55 μm 0.8


10 −5
(1 - v) / v

GOS/RGD
LM 0.6 mcore = 2 - 1i, mshell = 1.5
10 −6 acore/ashell = 0.2
0.4
GOS/RGD
10 −7 0.2 LM
1-v v
10 −8 0
1

0.8

0.6
g

0.4

0.2

0
10 −1 100 10 1 10 2 10 3 10 −1 10 0 101 102 103
Size Parameter x Size Parameter x shell

Figure 3.23 Illustration of a unified theory of light scattering by ice crystals vis-à-vis GOS/RGD. (a) Comparison of the unified theory and exact
LM results for a homogeneous ice sphere in terms of the extinction efficiency Qext , the single-scattering co-albedo (1 −  ), and the asymmetry
factor g as functions of the size parameter from 0.1 to 1000, using a wavelength of 0.55 µm in the calculations. In the Qext curve, limitations of
CGOM and IGOM are also displayed. (b) Same as (a), but for a core–shell sphere. Also, the single-scattering albedo  is used in the comparison
and the size parameter is in reference to the spherical shell radius. The real and imaginary indices of refraction for the core and the shell employed
in the calculations are depicted in the figure, as is the core–shell ratio.
180 Geometric optics for light scattering by ice crystals

(conventional geometric-optics method), the extinction cross-section is a factor of 2 larger


than the geometric cross-section πr2 , as a result of the postulation of equal partition of
diffraction and geometric reflection/refraction processes. The IGOM is applicable to size
parameters 20–30. The results from GOS (for x 2) coupled with RGD (for x  2)
closely match those from LM, and differences cannot be distinguished between the two
curves presented. For x  2, Qsca is proportional to x4 , whereas Qabs is proportional to x
(van de Hulst 1957). For this reason, the pattern of the single-scattering co-albedo (1−  )
is proportional to x−3 in this size parameter range. The sinusoidal oscillation in (1 −  ) is
out of phase in comparison to Qext . When x 4, (1 −  ) computed from LM increases
monotonically with sharp peaks due to resonances produced from non-linear interference
between diffraction, surface waves, and geometric reflection/refraction absent from GOS
based on a linear combination of these three components [see Eq. (3.3.55)]. The asymmetry
factor has sinusoidal oscillations that are in phase with Qext and we see close agreement of
g between the results from the unified theory based on the GOS/RGD approach and from
exact LM calculations.
Figure 3.23b compares the scattering and absorption results for a core–shell sphere
determined from GOS/RGD and the LM computational program. In this figure, the curves
for Qext ,  , and g for a sphere with core–shell structure are similar to those for a homo-
geneous sphere, depicted in Figure 3.23a. Minor differences stem from deviation in the
real refractive index (1.311 versus 1.5). However, when x 2,  becomes almost constant
due to a large value of mi (core, assuming a value of 1) used in the calculations. This
constant behavior is quite different from the monotonic decrease for an equivalent sphere if
an average refractive index is derived from the Maxwell-Garnett mixing rule (Chýlek et al.
1984) and used to compute the single-scattering properties, suggesting that we must be
cautious in using an average refractive index as a surrogate for an inhomogeneous sphere
with a core–shell structure.
We wish to link the core–shell structure to black carbon (BC, or soot) in the atmosphere.
BC has been identified as the second most important anthropogenic global warming agent
in the atmosphere because it strongly absorbs solar radiation and concurrently acts as
cloud condensation nuclei (CCN) or ice nuclei for cloud formation. The two effects are
significantly influenced by its aging processes in the atmosphere, which transform BC from
external to internal mixing. Freshly emitted BC particles are mostly hydrophobic and shortly
after emission they form irregular aggregates. Hydrophobic BC can become hydrophilic
through coating by soluble materials during aging, including condensation of sulfate, nitrate,
and some organics, coagulation with pre-existing aerosols, and heterogeneous reactions
with gaseous oxidants. Aged BC particles would experience hygroscopic growth and could
be activated as CCN for the formation of ice particles in subfreezing temperatures. BC
particles in the atmosphere are mostly in the form of core (BC) and shell (mostly water).
Figure 3.24 illustrates a comparison of the single-scattering parameters Qext ,  , and g
for randomly oriented columns (a) and plates (b) as a function of size parameter among the
three approaches: GOS/RGD, FDTD, and DDA. The size parameter is defined as xmax =
kLmax 2, where Lmax denotes the maximum dimension and k is the wavenumber. Because
(a) (b)
5

Qext 2

0
1

0.8

0.6
Ice l = 3.7 μm Ice l = 3.7 μm
v

0.4
GOS/RGD
FDTD
0.2
DDA
0
1

0.8

0.6
g

0.4

0.2

0
10 −1 10 0 101 10 2 103 10 −1 10 0 101 10 2 103
Size Parameter xmax Size Parameter xmax

Figure 3.24 Illustration of a unified theory of light scattering by ice crystals vis-à-vis GOS/RGD. Comparison of the unified theory and FDTD and
DDA results for randomly oriented ice crystals with variable length-to-width ratios for (a) solid columns and (b) plates, in terms of the extinction
efficiency Qext , the single-scattering albedo  , and the asymmetry factor g as functions of the size parameter (with reference to the maximum
dimension) from 0.1 to 1000, using a wavelength of 3.7 µm in the calculations (see also Figure 3.21).
182 Geometric optics for light scattering by ice crystals

10 4 1.0
l = 0.63 μm 0.8
De = 24 μm
0.6
50% +30% +20%
10 3 0.4 P22 /P11
smooth surfaces
0.2 −P12 /P11
0.0
Phase Function P11

10 2 −0.2
0 30 60 90 120 150 180
1.0

10 1
P44 /P11
0.5

10 0 P43 /P11
0.0

P33 /P11
10 −1 −0.5
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Scattering Angle (deg.)

Figure 3.25 Scattering phase matrix elements computed from GOS for an effective size of 24 µm
composed of 50% bullet rosettes, 30% hollow columns, and 20% plates, all with smooth surfaces,
using a wavelength of 0.63 µm in the calculations.

of computational limitations, FDTD and DDA results are shown for size parameters less
than about 20. The curves for Qext ,  , and g are similar for columns and plates, although
minor differences can be seen. Comparison of the GOS/RGD results with those computed
from FDTD and DDA in the size parameter range 0.3–20 shows close agreement, providing
a validation of the GOS/RGD approach. Due to ice absorption at the 3.7 µm wavelength,
Qext ,  , and g values approach their asymptotic limits of 2, 0.5, and 1, respectively,
when the size parameter approaches 103 . In the following paragraphs, we present scattering
phase matrix elements for typical ice crystal sizes and shapes that occur in the Earth’s
atmosphere as computed from GOS.
Figure 3.25 illustrates phase matrix elements for light scattering by a group of ice crystals
having an effective size De of 24 µm, corresponding to cold cirrus comprised of 50% bullet
rosettes, 30% hollow columns, and 20% plates. In the calculations, we have assumed that
all ice particles have smooth surfaces and have used a visible wavelength of 0.63 µm. In
this case, the ice crystal size parameter is on the order of 120. In the phase function P11
we see inner and outer halo features at the scattering angles  = 22° and 46°. Also, we
see another peak at a scattering angle of 10°, which is produced by rays transmitted (or
3.4 A unified theory of scattering by ice crystals 183

10 4 1.0
l = 0.63 μm 0.8
De = 24 μm
0.6
5 0% +30% +20%
10 3 0.4
rough sufaces P22 /P11
0.2 −P12 /P11
0.0
Phase Function P11

10 2 −0.2
0 30 60 90 120 150 180
1.0

10 1
P44/P11
0.5

10 0 P43 /P11
0.0

P33 /P11

10 −1 −0.5
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Scattering Angle (deg.)

Figure 3.26 Same as Figure 3.25, except that all ice particles are assumed to have rough surfaces.
See Subsection 3.1.3 for the treatment of light scattering by roughed ice crystals.

refracted) through a pyramid face and a prism face involving bullet rosettes. A broad peak
at  = 154° results from rays undergoing one or two internal reflections. With reference
to the degree of linear polarization LP = −P12 P11 , two local minima are seen at around
22° and 46°, corresponding to the halo features in P11 . At  50°, LP is positive, except
in backward directions. P22 P11 is close to 1 at forward directions and deviates from 1
at  30°. Deviation from 1 in this phase matrix element is an indication of scattering
by non-spherical ice particles. The element P44 P11 is larger than P33 P11 , due also to
non-sphericity, as is in the case for spheroids illustrated in Figure 2.20. The non-diagonal
element P43 P11 is closer to 0 than the diagonal elements.
In Figure 3.26, we show phase matrix elements when intense roughness (see Subsection
3.1.3 for light scattering by roughed ice crystals) is added to all ice particles. The surface
roughness results in substantial reduction of the 22° halo peak in the phase function. Also,
the halo features located at  = 10° and 46° as well as the 170° peak largely disappear.
In the backscattering direction from 110° to 180°, an angular region important in remote
sensing applications, the phase function is essentially flat and featureless. The degree of
linear polarization −P12 P11 is closer to zero than in the smooth surface case. We also
note that matrix elements P22 P11 , P33 P11 , P44 P11 , and P43 P11 are smoothed out at
184 Geometric optics for light scattering by ice crystals

backward directions for  130°. In order to interpret the phase function results derived
from laboratory-generated ice crystal clouds, it is noted that a certain portion of rough ice
crystals must be included in theoretical calculations (see the following Subsection 3.4.2).
Also, the phase function curves determined by a number of researchers from cirrus clouds
in the atmosphere have illustrated that in general the 22° halo feature is relatively small and
that most of the fascinating halos and arcs have not been observed from aircraft platforms.
Finally, in Figure 3.27, we present the extinction efficiency Qext , single-scattering albedo
 , and asymmetry factor g as a function of wavelength from 0.2 to 100 µm for randomly
oriented Koch snowflakes, having a spherical volume-equivalent radius of 100 µm. For
comparison purposes, also shown are these parameters for hexagonal plates as a function
of wavelength from 0.2 to 5 µm (solar or shortwave regions) taken from Liou et al. (2014).
As shown, Qext is close to 2 for solar wavelengths. But for thermal infrared or longwave,
Qext is less than 2, partly because of substantial absorption. The single-scattering albedo for
Koch snowflakes is similar to that for hexagonal plates at wavelengths smaller than 5 µm
due to the same volume used in light absorption and scattering. For wavelengths between
5 and 100 µm, due to substantial absorption, the single-scattering albedo is similar to those
for spheres and circular cylinders, as shown in Figures 2.17 and 2.18. The asymmetry factor
for Koch snowflakes is smaller than that for hexagonal plates at visible and near-infrared
wavelengths, consistent with the value at λ = 0.55 µm presented in Liou et al. (2014).
Variation of the asymmetry factor for λ> 5 µm is similar to the values obtained for sphere
and circular cylinder.
In summary, we have demonstrated that the unified theory of light scattering by ice
crystals on the basis of the GOS/RGD approach can be employed for accurate calculations
of the extinction efficiency, single-scattering albedo, and asymmetry factor for hexagonal
and irregular ice particles as well as internal mixing of absorbing aerosols (e.g., black
carbon and dust) using spherical particles as a proxy. These single-scattering properties
are most relevant to application in radiation parameterizations involving ice crystals and
snowflakes in the Earth’s atmosphere and for incorporation into computer models used in
climate studies, as illustrated in Section 6.2.
Furthermore, we have shown that GOS/RGD can be used for the calculation of the
scattering phase matrix. The elements, which comprise the phase function and linear
polarization information for ice particles, have been used to develop numerical algorithms
for satellite remote sensing of cirrus clouds in terms of their optical depth and ice crystal
size and shape by means of bidirectional reflected sunlight as a function of satellite orbit
and sun position. In Sections 5.2 and 5.3, we will demonstrate the importance of ice crystal
phase functions for reliable retrieval of ice cloud parameters. In addition, backscattering
and polarization data for ice crystals with respect to their orientation state and in reference
to spherical water droplets are important to lidar detection of cirrus cloud microphysics
(Section 5.5).
In Subsection 3.4.2, we discuss the development of a separate but associated experi-
mental program which provides an independent phase function and polarization datasets
determined from laboratory scattering and ice microphysics experiments, for the specific
purposes of comparing with and cross-checking theoretical results.
3.4 A unified theory of scattering by ice crystals 185

Extinction Efficiency
3

0
1.0
rsnow = 100 μm Single-Scattering Albedo
0.8

0.6

0.4

0.2

0.0
1.2
Asymmetry Factor
1.0

0.8

0.6

0.4
10 −1 10 0 101 10 2
Wavelength (μm)
Figure 3.27 Extinction efficiency, single-scattering albedo, and asymmetry factor as a function of
wavelength from 0.2 to 100 µm for Koch snowflakes with a spherical volume-equivalent radius of
100 µm. Also shown for comparison purposes are results for hexagonal plates covering the region
from 0.2 to 5 µm.

3.4.2 Laboratory scattering and ice microphysics experiments with ice crystals
Measurements of the angular scattering and polarization patterns of ice crystals have
been conducted in laboratory cloud chambers, beginning with experiments performed
by Huffman and Thursby (1969) and Huffman (1970). Subsequently, Liou and Lahore
186 Geometric optics for light scattering by ice crystals

(1974), Liou et al. (1976), Dugin and Mirumyants (1976), Sassen and Liou (1979a, 1979b),
Volkovitsky et al. (1980), Barkey and Liou (2001, 2006, 2008), and Barkey et al. (1999,
2002) have reported laboratory scattering and ice microphysics experiments and presented
pertinent scattering results for ice particles. In addition to laboratory experiments, a number
of scattering measurements have been conducted in cirrus clouds (C.-Labonnote et al. 2000,
Oshchepkov et al. 2000, Auriol et al. 2001, Jourdan et al. 2003, Shcherbakov et al. 2006).
In this subsection, we shall confine our discussion to laboratory scattering and ice cloud
experiments in support of the unified theory of light scattering by ice crystals as presented
above.
Desirable ice crystal sizes and shapes are difficult to generate and sustain for a time
period sufficiently long to perform light-scattering experiments. For this reason, Barkey
et al. (1999) carried out an experimental light-scattering program using hexagonal ice-like
crystals measured in the analog manner so that optical experiments could be performed over
a relatively long period of time for complex-shaped particles. The experiment consisted of a
polarized laser beam at λ = 0.63 µm and an array of 36 highly sensitive photodiode detectors
arranged between the scattering angles of 2.8° and 177.2°, mounted in a linear array on
a half-dome which could be rotated to vary the azimuthal angle. After careful calibration
and signal acquisition, this system was first used to measure the phase functions of a glass
sphere and a glass fiber configured to scatter light like an infinite cylinder. The experimental
results closely match those computed from the Lorenz–Mie and Rayleigh–Wait theories
(see Subsection 2.2.3). Subsequently, sodium fluoride (NaF) crystal, which has a refractive
index of 1.33, a value close to that of ice in the visible wavelength, was adopted for the
light-scattering experiments. The NaF crystal could be machined to a desirable shape with
a size on the order of a millimeter. It was mounted on top of a small pedestal, and its
orientation was controlled by a rotator. Angular integrations in the experiment could follow
the theoretical and computational procedures that were carried out for randomly oriented ice
particles. The phase functions determined from laboratory measurements and theoretical
results derived from IGOM for an aggregate and other crystal shapes assembled from NaF
columns showed close agreement. Also illustrated was excellent agreement between the
measured phase functions and theoretical results for a rough-surface plate having all eight
sides sanded with small scratches evenly distributed across the crystal surfaces. In the latter
case, the 22° and 46° features vanished and the phase function was essentially without
noticeable features.
The light scattering experiment, shown in Figures 3.28a and 3.28b, was conducted in
the cloud chamber at the Desert Research Institute at the University of Nevada, Reno.
It consisted of an insulated and refrigerated column approximately 4 m tall and 0.3 m
in diameter positioned above a larger cold chamber. Two dedicated chillers were used
to control the growth column temperature above and below the nucleation temperature,
while several thermocouples were placed along the length of the growth column. Water
droplets from an ultrasonic humidifier fell and cooled to the desired temperature in the
upper column; some of these were nucleated by a wire spring, and continuously cooled by
liquid nitrogen and presented to the ice cloud by a system of pulleys and a slowly rotating
3.4 A unified theory of scattering by ice crystals 187

(b)
From ultrasonic
water droplet
generator Water droplets

(a) Data-acquisition
Photodiode card and PC Continuous
detector/amplifiers computer Refrigerated nucleator
and insulated
Fiber light growth
guides chamber Dewar with
liquid nitrogen

Mirror Nucleated ice crystals


Refrigerated box
(not to scale)

Nephelometer

Forward
Diode Cloud scope
beam dump
laser Sample
(670 nm) guide tubes

Figure 3.28 (a) The physical structure of the polar nephelometer and its interaction with the laser
beam and the falling ice crystals confined to the center of the fiber-optic coupled detector array along
with the sample tube with a 3 mm inside diameter (after Barkey and Liou 2001). (b) The structure
of a cloud chamber for the generation of ice crystals. Water droplets are injected into the top of the
growth column and nucleated near the center. The cloud scope and polar nephelometer are placed in
the cold chamber at the bottom of the growth column (after Barkey et al. 2002).

electric motor. The water-vapor pressure could be controlled to certain degree by the rate at
which water droplets were injected into the crystal chamber, but ice crystal habit was mainly
controlled by the temperature at which the crystals were allowed to grow after nucleation.
The nucleated ice crystals continued to grow at the expense of water droplets as they fell
through the growth column and into the lower cold chamber where the polar nephelometer
was placed. Because the system was sealed, colder hence denser air fell to the bottom of
the lower cold box, which produced a temperature stratification that retarded the ice crystal
free-fall. Also, because of the slow fall speeds of the generated ice crystals, a suction
pump was connected to the sample outlet of the polar nephelometer to pull ice crystals
into the scattering volume. Simultaneously, the ice crystal evolution was monitored by a
cloud scope (a video microscope that observed the ice particles’ size and shape) positioned
approximately 20 cm from the nephelometer sample inlet.
The polar nephelometer system’s accuracy was first verified by measuring the scatter-
ing properties of water droplets generated by an ultrasonic humidifier at room temperature.
Water droplets falling through the upper sample tube were seen as a collimated and continu-
ous white stream of particles. Because there were no direct measurements of the droplet-size
distribution, several mean effective sizes and variances are generated by Lorenz–Mie calcu-
lations and compared with experimental results, which were adjusted to fit the normalized
188 Geometric optics for light scattering by ice crystals

102

101 Theory

Phase Function Experiment

100

10-1

10-2
0 60 120 180
Scattering Angle (deg.)

Figure 3.29 Experimentally measured two-dimensional scattering intensity (normalized phase func-
tion) and the theoretically derived phase function (from Lorenz–Mie theory) for water droplets with
unpolarized incident light. The measurement error bar is shown at the scattering angle of 90° (after
Barkey and Liou 2001).

theoretical results by use of a multiplicative constant. A log-normal size distribution with a


mean effective radius of 3.75 µm and a variance of 0.1 produced the best match. The error
bar of 10% depicted in Figure 3.29 represents the system error. Overall, measured values
matched closely with theoretical results and reproduced the rainbow feature at a scattering
angle of 140°.
In the light scattering experiment involving ice crystals, the volume was defined by a
small cylinder 3 mm in diameter and 4 mm in length. Measured ice crystal concentration
was normally of the order of one particle per 2 mm3 , which corresponds to approximately
14 particles (20 µm sizes) in the scattering volume. The effect of multiple scattering on
nephelometer measurements can thus be neglected. The intensity of the light scattered into
the two-dimensional plane where the fiber light guides were positioned depends on the
number of crystals in the scattering volume and their size, shape, and orientation. These
experimental parameters vary considerably during experiments, which causes the intensity
of the measured results to vary widely from one measurement to the next. For this reason,
the scattering patterns over several hundred measurements were averaged, a process which
took approximately 2–4 minutes to complete. Ice crystals pulled into the scattering volume
were likely to be randomly oriented.
At least 190 measurements are needed to reduce the standard error of the mean to less
than 5%. In this manner, the averaged scattering pattern corresponds to an ice crystal with the
scattering properties of all the particles that existed during the time period of the experiment.
To obtain the phase function, we need the scattering cross-section in normalization processes
3.4 A unified theory of scattering by ice crystals 189

[see Eq. (2.4.34)]. For randomly√oriented simple hexagonal particles, the scattering cross-
section σ s is approximated by 3( 3a 2 + 2aL)/2 [Eq. (2.4.49)], where the ice crystal radius
a and length L were measured directly from the images, and absorption of ice at visible
wavelengths has been neglected. The phase functions presented in the following paragraphs
were determined from the sum P11 + P12 (see Subsection 2.4.3), which corresponds to
the two polarization states of the incident light for randomly oriented ice crystals in the
scattering volume.
Shown at the center of Figure 3.30 is a cloud-scope video image when the crystal
habit of the chamber cloud consisted predominantly of plates grown at a temperature of
approximately −6°C. The vertical white stripe in this image is due to glare from the cloud-
scope light source. The size distribution of the counted crystals is shown in the histogram.
The average aspect ratio L2a  0.8, where a is the radius of the hexagonal face and L is
the length of the column or the thickness of a plate. Fitted to the theoretical expectation
shown in the graph at the top, and based on the observed ice crystal microphysics, were
experimentally measured scattering properties from an average of 189 measurements taken
over a period of approximately 3 minutes. The error bars correspond to the 5% standard
error of the mean.
Figure 3.31 is similar to Figure 3.30, except that it involved a cloud event when mostly
hollow columns, grown between −5 and −7°C, were seen by the cloud scope. The 26
particles counted from this image consist of mostly hollow columns with an average aspect
ratio of approximately 2.1 and an average maximum dimension of approximately 36 µm.
However, regular columns, trigonal shapes, and plates can also be seen. The experimental
result was the average of 511 measurements taken over a period of approximately 4.5 min-
utes. The standard error of the mean for this result produces an approximate 3% error.
Normalized phase functions for simple hexagons and hollow columns calculated by IGOM
with various aspect ratios corresponding to the measured crystal sizes are combined by use
of the extinction cross-section determined from the measured crystal size and the counted
population to weight the scattering contributions of each particle. Approximately 80% of
the particles were assumed to have rough surfaces to account for the surface features that
can be seen on some of the larger plates in the cloud-scope image. Substantial surface
roughness was incorporated in the phase functions by random perturbations of the crystal
surface.
In Figure 3.31, approximately 10% of the phase functions used to produce theoretical
values were based on hollow column habit. The other crystal types were simple columns
and plates, of which 80% had rough surfaces. Overall, experimental results closely follow
theoretical values. Both the 22° and 46° halo intensity imprints are clearly displayed,
although less intense at 46° due to the presence of hollow columns; this is expected as the
46° halo feature is produced by two refractions through a 90° prism. For both results, a
large percentage of the theoretical results required crystals to have rough surfaces in order
for a reasonable fit to experimental results. This is most likely due to surface features on
the crystals that cannot be seen on the cloud-scope images because of the low resolution
of the video camera, or smaller particles not seen by the cloud scope. Also, we see that
190 Geometric optics for light scattering by ice crystals

102

Normalized Phase Function


Theory
101
Experiment
100

10-1

10-2
0 60 120 180
(deg.)
Scattering Angle (deg.)

Figure 3.30 Two-dimensional angular scattering intensities measured by the polar nephelometer
(top panel) when predominantly plates are seen, as shown in the video cloud-scope image (middle
panel), along with an ice crystal count summarized in the histogram (bottom panel). The white area
in the middle image is due to the cloud-scope light source (after Barkey et al. 2002).
3.4 A unified theory of scattering by ice crystals 191

102

Normalized Phase Function


Theory
101

Experiment
100

10-1

10-2
0 60 120 180
Scattering Angle (deg.)

Figure 3.31 Two-dimensional angular scattering intensities measured by the polar nephelometer
(top panel) when predominantly columns are observed, as shown in the video cloud-scope image
(middle panel), along with ice crystal count summarized in the histogram (bottom panel). The white
area seen in the middle of the image is due to the cloud-scope light source (after Barkey et al. 2002).
192 Geometric optics for light scattering by ice crystals

more measured light was scattered into angles in the ranges 10°–20° and 160°–175° than
in the theoretical results, which revealed that it is extremely difficult to completely match
the two results in view of numerous unknown factors involving ice crystal sizes and shapes
generated in cloud chambers and inherent limitations in theoretical calculations. Between
120° and 160°, the experimental results are not smooth in both cases due to systematic errors
in the calibration that are more noticeable at lower intensities. There is a slight difference
between the angular positions of both the 22° and the 46° intensity peaks because each fiber
optic sensed light scattered into an angle of approximately 2° and the angular positions of
the fiber optics were not at the points of maximum scattered intensity.
In summary, and in view of the preceding presentations, it is evident that the unified
theory of light scattering by ice crystals developed herein has been successfully used
to interpret phase functions produced from laboratory scattering and ice microphysics
experiments, thus providing new dimensions in the evolution of theoretical developments.
Furthermore, the well-known features of diffraction, 22° and 46° intensity maxima, and
large backscattering values that occur in the phase function measured in laboratory ice
cloud chambers match well to theoretical calculations.

3.4.3 Application to snow grains contaminated by black carbon (BC) and dust
In this subsection we apply the unified theory of light scattering by ice crystals, as presented
above, to absorption and scattering by snow grains with sizes on the order of 100–1000 µm
contaminated by BC (0.1–1 µm) and dust (1–10 µm). Liou et al. (2011) demonstrated
that a small BC particle on the order of 1 µm internally mixed with snow grains could
effectively reduce visible snow albedo by as much as 5–10%. Subsequently, Liou et al.
(2013) applied this approach to contaminated ice particles in contrail cirrus and investigated
the effects of internal and external mixing of BC in ice particles on the radiative forcings
and heating rates of contrail cirrus. Also presented was a parameterization of the spectral
extinction coefficient, single-scattering co-albedo, and asymmetry factor for contaminated
ice particles, for efficient application to radiative transfer calculations on the basis of the
comprehensive spectral database of the single-scattering properties of pure ice particles of
various shapes and sizes developed by Yang et al. (2013). In conjunction with this study, we
noted observations by Petzold et al. (1998, 1999) relative to plural numbers of BC particles
inside and outside of contrail ice crystals.
Multiple internal mixing of BC/dust in snow grains can be produced by nucleation,
accretion, riming, aggregation, and sintering during aerosol–cloud–precipitation processes
known as wet deposition. Internal mixing can also occur via the direct or dry deposition of
aerosols onto high-elevation snow, followed by successive snow events. Mountain glaciers
and polar ice sheets are excellent archives and imprints of atmospheric BC/dust particles
associated with wet and dry depositions. Ming et al. (2008) measured BC concentrations
in a 40 m ice core over the East Rongbuk Glacier in the northeast saddle of Mt. Everest,
which provided an historical record of BC deposition in the high Himalayas over the past
3.4 A unified theory of scattering by ice crystals 193

50 years. Additionally, Ming et al. (2009) collected BC concentrations from snow samples
taken at a number of selected glaciers in west China during 2004–2006 and found that BC
concentrations are higher at lower-elevation sites (Tienshan Mountains) than comparable
samples obtained on the Tibetan Plateau, likely because of the topography effect. Xu
et al. (2009) measured aerosol concentrations including BC in ice cores spanning five
Tibetan glaciers, showing an increasing deposition of anthropogenic aerosols during the
period 1998–2005. Hadley et al. (2010) reported one of the first direct measurements
documenting the efficient deposition of BC on Sierra Nevada snow packs while revealing
high BC concentrations in falling snow. Sterle et al. (2013) showed that BC concentrations
were enhanced seven-fold in surface snow compared to snowpack bulk values in the eastern
Sierra Nevada from February to May 2009. Ice cores extracted from high elevations in the
European Alps also displayed increasing BC concentrations from the era of the Industrial
Revolution to the twentieth century (Jenk et al. 2006, Thevenon et al. 2009). Using a single
particle soot photometer, Schwarz et al. (2013) showed that the modal volume-equivalent
radius for BC particles in snow is about 0.1 µm.
With reference to dust, Kumai (1977) investigated the long-term variation of atmospheric
aerosols in snow and deep ice core samples taken from Camp Century, Greenland, and
showed that a majority of these aerosols were silicate mineral dust with a mean size on
the order of 1 µm. Hammer et al. (1985) found dust in snow grains with a size on the
order of 0.1–2 µm in Greenland ice cores. Zdanowicz et al. (1998) measured and compared
the concentration, deposition rate, and size distribution of dust aerosols in snow pits on
the Penny Ice Cap, Baffin Island, Arctic Canada with earlier studies. Also, Zdanowicz
et al. (2006) measured the dustfall over snow packs in the St. Elias Mountains, Canada,
finding high dust concentrations in snow in April 2001. Kang et al. (2010) used ice cores to
reconstruct dust load history over the central Tibetan Plateau from 1940 to 2005 and found
that dust loading in this region has increased since the 1990s.
From the instant that snow hits the ground, it begins an endless process of metamorphism,
a term used in reference to physical changes in snow grains within a snowpack resulting from
differences in temperature and pressure. Using scanning electron microscopy in February
2002, Dominé et al. (2003) studied the natural metamorphism of snow samples collected
near Chamrousse, a ski resort town in the French Alps. The fresh snow was collected within
minutes of its fall and immersed immediately in liquid nitrogen to preserve its shape and
size. The sample pictures illustrate the wide variety of shapes and habits, including plates,
columns, column rosettes, stellar and dendritic crystals, and various irregular crystals, all of
which exhibit sharp angles. After 14 days of isothermal metamorphism, extensive changes
occurred such that the original shapes were substantially modified with no sharp edges
between two flat faces, leading to rounded crystals with different shapes. These snow
grains are referred to as old snow.
In view of these and other observations (e.g., Aoki et al. 2000), the shapes of hexag-
onal plate/column/bullet rosette and spheroid/sphere were used to model fresh and old
snow grains, respectively, for light scattering and absorption calculations (Figure 3.32).
Liou et al. (2014) have developed a stochastic approach to model the positions of BC/dust
194 Geometric optics for light scattering by ice crystals

(a) 100 μm 100 μm (b)

100 μm

100 μm
100 μm
100 μm 100 μm 100 μm

Soot Dust

(c) (d)

100 μm
100 μm

100 μm
30 μm

30 μm 30 μm 30 μm

Figure 3.32 The images on the left side are observed grain shapes for (a) fresh snow and (c) old snow
(Dominé et al. 2003). On the right side are corresponding parameterized grain shapes for (b) fresh
snow (plate, column, and bullet rosette) and (d) old snow (prolate and oblate spheroids and sphere)
with multiple internal and external mixing, using a stochastic procedure. The red dots represent either
soot or dust.
3.4 A unified theory of scattering by ice crystals 195

1 2
Two-Layer Model (rsnow = 100 μ m, rsnow = 1000 μ m), μ 0 = 0.5
(a)
BC (rBC = 0.1 μm)

1.0 Pure Snow


0.01 ppm 1.0
0.1 ppm
0.8
1 ppm
0.9
0.6
VIS NIR
0.4
0.8
0.2 0.4 0.6 0.8 1.0
0.2
Snow Albedo

0.0
(b)
Dust (rdust = 1 μm)

1.0 Pure Snow


2 ppm 1.0
10 ppm
0.8
100 ppm

0.9
0.6

VIS NIR
0.4
0.8
0.2 0.4 0.6 0.8 1.0
0.2

0.0
0.2 0.8 1.4 2.0 2.6 3.2 3.8 4.4 5.0
Wavelength (μm)

Figure 3.33 Snow albedo for a two-snow layer model (new snow on top of old snow) as a function
of wavelength for three BC/dust concentrations. New snow internally mixed with (a) BC or (b) dust
has an optical depth of 200, while pure old snow is considered to be semi-infinite. The insets display
snow albedo in the wavelength region 0.2–1 µm where substantial BC/dust absorption occurs (after
Liou et al. 2014).

internally mixed with three snow-grain types: hexagonal plate/column and spheroid (con-
vex) and bullet rosette (concave). Subsequently, light absorption and scattering analysis
can be carried out using the unified theory of light scattering by ice crystals, the theme of
Subsection 3.4.1, which includes the geometric-optics and surface-wave (GOS) approach
coupled with Monte Carlo photon tracing, to determine BC/dust single-scattering properties.
196 Geometric optics for light scattering by ice crystals

Figure 3.32 shows examples of external and internal positions of BC (or dust) particles for
these particles; the action of internal mixing absorbs substantially more light than external
mixing.
Figure 3.33 depicts spectral albedo (0.2–5 µm) by snow grains internally mixed with
BC/dust for a two-layer spectral snow model involving contaminated fresh snow on top
of old snow to investigate and understand the climatic impact of multiple BC/dust inter-
nal mixing associated with snow grain metamorphism, particularly over mountain/snow
topography. Concentrations of BC (Figure 3.33a) and dust (Figure 3.33b) are 0.01–1 ppm
(parts per million by mass) and 2–100 ppm, respectively. The impact of BC/dust absorption
is confined to wavelengths shorter than about 1.4 µm, beyond which ice absorption pre-
dominates. Based on the single-scattering properties determined from stochastic and light
absorption parameterizations and using the adding/doubling method for spectral radiative
transfer, the effect of snow-grain shape on absorption is relatively small, but its effect on for-
ward scattering is substantial. Due to a greater probability of intercepting photons, multiple
inclusions of BC/dust exhibit a larger absorption than an equal-volume single inclusion, and
the snow-grain shape plays a critical role in snow albedo calculations through its forward
scattering strength.
4
Other useful approaches to light scattering
by ice particles

During the past four decades, numerous computational methods and numerical approaches
have been developed to solve light scattering by non-spherical and inhomogeneous particles
for application to atmospheric and geophysical sciences, applied physics, computer science,
electrical engineering, and bioengineering. It is especially evident that the applicability and
accuracy of these approaches in terms of particle size range and morphology, established
by comparison with the “exact” Lorenz–Mie theory, has been enhanced as a result of
the availability of supercomputers as well as clever technical innovation. Most of these
developments have been reviewed from various perspectives by Liou and Takano (1994),
Wriedt (1998), Kokhanovsky (1999), Mishchenko et al. (2000b), Liou (2002), Kahnert
(2003), and Yang and Liou (2006). In particular, the finite-difference time domain (FDTD)
method, the T-matrix method, and the discrete dipole approximation (DDA) have been
extensively used for a number of practical applications. The following discussions will
be confined to these three methodologies within the context of their applications to light
scattering by ice crystals and in association with the geometric-optics, diffraction, and
surface-wave approaches presented in Chapter 3.

4.1 Finite-difference time domain method


4.1.1 A brief review
It appears that Yee (1966) pioneered the development of the FDTD method for application
to electromagnetic wave propagation and scattering problems. He demonstrated in a concise
paper the basic FDTD concept for simulating the propagation of an electromagnetic wave in
the near-field zone. However, several important aspects related to the application of FDTD
to light scattering problems, including the boundary condition imposed on spatial-domain
truncation and mapping of the near field to the far field, were not properly addressed.
From an historical standpoint, the FDTD approach had received insufficient recognition
until substantial efforts were made by electrical engineers and computational physicists in
the late 1970s and early 1980s (e.g., Taflove and Brodwin 1975, Holland 1977, Kunz and
Lee 1978, Taflove 1980, Kunz and Simpson 1981, Umashankar and Taflove 1982). After
their pioneering contributions, FDTD has been applied to a wide range of electromagnetic

197
198 Other useful approaches to light scattering

Artificial Absorbing Layer

Connecting Surface

Scattering Particle

Total-field Region

Scattered-field Region

Figure 4.1 A conceptual diagram for the computation, by means of the finite-difference time domain
(FDTD) method, of the near field for electromagnetic scattering by a non-spherical particle illuminated
by plane waves. The connecting surface (or the Huygens surface) divides the computational domain
into total-field and scattered-field regions.

problems, including antenna scattering, numerical modeling of microstrip structures, elec-


tromagnetic absorption by human tissues such as red blood cells, and scattering of light
by atmospheric particles (e.g., Taflove and Umashankar 1990, Taflove 1995, Yang and
Liou 1996a and 1996b, Sun et al. 1999, Yang et al. 2000b, Lu et al. 2005). In particular,
with the development of the perfectly matched layer (PML) absorbing boundary condition
(Berenger 1994) and further improvements, such as the uniaxial anisotropic PML (UPML)
(Gedney 1996) and convolution PML (CPML) boundary conditions (Roden and Gedney
2000), the FDTD approach has become a powerful and robust numerical tool to facilitate
simulation of the interaction of an arbitrarily oriented structure or particle with incident
electromagnetic waves.
The most computationally efficient implementation of the FDTD technique can be real-
ized through the use of a Cartesian grid mesh, as schematically illustrated in Figure 4.1.
Specifically, the spatial region comprised of a scattering particle is truncated and subse-
quently discretized in terms of a number of rectangular cells. Appropriate values of permit-
tivity, permeability, and conductivity are assigned to the grid cells to specify electromag-
netic characteristics of the particle and the surrounding medium. A direct implementation
4.1 Finite-difference time domain method 199

of Maxwell’s curl equations is invoked to solve the temporal variation of electromag-


netic waves in the truncated spatial domain within which the scatterer is imbedded. More
specifically, the finite-difference discretization is performed for Maxwell’s curl equations
in both time and space. The propagation of excited waves (i.e., scattered waves) in the time
domain can be simulated with the finite-difference analog of Maxwell’s equations through
a time-marching iterative procedure.
In numerical computation, the scattering of an incident electromagnetic wave by a
particle must be confined to a finite spatial domain. In order to apply the FDTD tech-
nique, an appropriate boundary condition must be imposed such that the simulated field
within the truncated spatial domain would be the same as that for the unbounded case.
As a matter of fact, the implementation of an optimal boundary condition to suppress
spurious reflections is critical to practical computations based on the FDTD method in
terms of their numerical accuracy and stability. Substantial technical efforts have been
expended to minimize artificial reflections from boundaries of the truncated FDTD compu-
tational domain (e.g., Moore et al. 1988, Blaschak and Kriegsmann 1988, Yang and Liou
1998b). However, it appears that the use of an artificial absorbing medium, known as the
PML medium (Berenger 1994), may be the most efficient approach to implementing an
absorbing boundary condition to effectively truncate the spatial domain. The absorbing
boundary condition must be imposed only for outgoing waves. For this reason, a connect-
ing surface (Taflove 1995), or the so-called Huygens surface (Merewether et al. 1980),
is defined in the computational domain, as shown in Figure 4.1. Inside the connecting
surface, the total field is simulated, while the scattered (outgoing) waves are simulated
outside the connecting surface, along with initiation of the incident wave at the connecting
surface. Also, a scattered-field FDTD scheme (i.e., the incident field is subtracted from
the total field), which uses the scattered field within the entire computational domain,
has been reported by Britt (1989) and Yang and Liou (1995). This technique, however,
appears to be less computationally efficient compared with the counterpart depicted in
Figure 4.1.
The solution of the finite-difference analog of Maxwell’s curl equations is in the time
domain. To obtain the frequency response of the scattering particle to illumination by the
incident wave, an appropriate transformation of the relevant electromagnetic field values
from the time domain to the frequency domain is required. The discrete Fourier transform
technique can be employed to obtain the frequency spectrum of time-dependent signals
if a pulse is used as an initial excitation. In practice, selection of an appropriate pulse is
required to avoid numerical aliasing and dispersion. Furthermore, a mapping of the near-
field result to the far field must be performed in order to derive the particle scattering and
polarization properties, namely, the scattering phase matrix, extinction cross-section, and
single-scattering albedo. A surface integration or a volume integration technique can be
employed to obtain the far-field solution. Several computational issues exist in performing
FDTD numerical calculations, including the staircasing effect in approximating the particle
shape as well as the determination of appropriate white space between the particle and the
artificial boundary to balance computational demand and numerical accuracy.
200 Other useful approaches to light scattering

4.1.2 Numerical discretization of Maxwell’s curl equations


To present the basic principle of the FDTD technique, we shall consider the source-
dependent form of Maxwell’s curl equations in the time domain for a non-ferromagnetic
medium with μ = 1, a valid simplification for ice crystals, in the forms [see Eqs. (2.2.1a),
(2.2.1.b), and (2.2.4)]
ε ∂E(r, t) 4π
∇ × H(r, t) = + σ E(r, t), (4.1.1a)
c ∂t c
1 ∂H(r, t)
∇ × E(r, t) = − , (4.1.1b)
c ∂t
where ε and σ are, respectively, the permittivity and conductivity and both are real quanti-
ties. To construct a computationally efficient finite-difference scheme corresponding to the
preceding two equations, Yang and Liou (1996b) introduced a new variable, τ = 4πσ/ε,
based on which Eq. (4.1.1a) can be expressed in the form


ε ∂E(r, t)
∇ × H(r, t) = + τ E(r, t) . (4.1.2)
c ∂t
We can use Eqs. (4.1.1b) and (4.1.2) to construct the finite-difference analog of
Maxwell’s curl equations to circumvent the use of complex quantities even if the scat-
tering particle is absorptive, i.e., the corresponding complex refractive index has a non-zero
imaginary part. To discretize Eq. (4.1.2) in time, we multiply both sides by a factor exp(τ t)
and rearrange the resultant terms in the form
∂E(r, t) c
exp(τ t) + τ exp(τ t)E(r, t) = exp(τ t)∇ × H(r, t). (4.1.3)
∂t ε
Upon using the chain rule for derivatives, we can rewrite this in the form
∂[exp(τ t)E(r, t)] c
= exp(τ t) ∇ × H(r, t). (4.1.4)
∂t ε
Furthermore, consider the second-order difference scheme, referred to as the leap-frog
scheme and usually used in finite-difference analysis in computational mathematics, for
the derivative of a function f with its argument t, in the form

df (t)  f [(n + 1)t] − f (nt)
 ≈ + O(t 2 ) = (f n+1 − f n )/t + O(t 2 ).
d t t=(n+1/2)t t
(4.1.5)

In the preceding equation, the superscripts n + 1 and n indicate that the associated quantities
are evaluated at the time steps t = (n + 1)t and t = nt. The term O(t 2 ) indicates that
the truncation error of this leap-frog scheme is of the second order. Applying this scheme
to Eq. (4.1.4) at time step t = (n + 1/2)t, we obtain
 
∂[exp(τ t)E(r, t)]  c 
 = exp(τ t) ∇ × H(r, t) . (4.1.6a)
∂t t=(n+1/2)t ε t=(n+1/2)t
4.1 Finite-difference time domain method 201

We may rewrite this equation in a more explicit form as follows:


exp[τ (n + 1)t]En+1 (r) − exp(τ nt)En (r) c
= exp[τ (n + 1/2)t] ∇ × Hn+1/2 (r).
t ε
(4.1.6b)
Simplifying and rearranging Eq. (4.1.6b), we obtain an explicit scheme for the finite-
difference analog of Eq. (4.1.4) for simulating the temporal variation of the electric field,
in the form
ct
En+1 (r) = exp(−τ t)En (r) + exp(−τ t/2) ∇ × Hn+1/2 (r). (4.1.7)
ε
Similarly, the finite-difference analog of Maxwell’s curl equation for the magnetic field can
be expressed as

Hn+1/2 (r) = Hn−1/2 (r) − ct∇ × En (r). (4.1.8)

Note that Eqs. (4.1.7) and (4.1.8) can also be derived by directly integrating the dif-
ferential forms of Eqs. (4.1.4) and (4.1.1b). For example, if we integrate both sides of
Eq. (4.1.1b) from t = (n − 1/2)t to t = (n + 1/2)t, we obtain
 (n+1/2)t 
1 (n+1/2)t ∂H(r, t)
∇ × E(r, t)dt = − dt, (4.1.9)
(n−1/2)t c (n−1/2)t ∂t
which can be approximated in the form
1
∇ × En (r)t ≈ − [Hn+1/2 (r) − Hn−1/2 (r)]. (4.1.10)
c
Note that rearranging the terms in Eq. (4.1.10) leads to Eq. (4.1.8).
For numerical FDTD simulations, the vector forms of Eqs. (4.1.7) and (4.1.8) must be
decomposed into their scalar components. For example, the governing equations for the
z-components of the electric and magnetic fields can be obtained from these two equations
as follows:
 n+1/2 n+1/2

ct ∂Hy (r) ∂Hx (r)
Ez (r) = exp(−τ t)Ez (r) + exp(−τ t/2)
n+1 n
− ,
ε ∂x ∂y
(4.1.11)


∂Eyn (r) ∂Exn (r)
Hzn+1/2 (r) = Hzn−1/2 (r) − ct − . (4.1.12)
∂x ∂y
To discretize the spatial derivatives in the preceding equations, Yee (1966) suggested a
configuration for the electromagnetic field components on a rectangular cell, as schemat-
ically illustrated in Figure 4.2. The cell center is located at (I x, J y, Kz) in which
x, y, and z are cell dimensions along the x-, y-, and z-axes, respectively. It turns
out that the staggered configuration, shown in Figure 4.2, for electromagnetic field vec-
tor components inherently guarantees that electromagnetic boundary conditions at cell
202 Other useful approaches to light scattering

(I -
/ 2) 1/
2,
K+1 J+
2, Ey 1/2
1/ ,K
J- Hz +1
2, /2
1/ )
(I- Ex Ex
Ez z
Ez Ey

Ez Hy
(I, J, K)
Ez y
Hx Ey
x
(I + Ex
1/ )
2, Ey Ex 1 /2
J- K-
1/ 2,
2,
K- 1/
1 J+
/2 2,
) 1/
(I +

Figure 4.2 Locations of various field components at staggered locations on a cubic cell, following a
finite-difference scheme suggested by Yee (1966). The E field components are tangential to the cell
edges, while the H field components are normal to the cell faces.

interfaces are such that the tangential components of E-field and the normal components
of H-field are continuous at interfaces.
In the FDTD formalism, the spatial location (I x, J y, Kz) is concisely represented
in terms of indices (I, J, K), and any variable, such as the z-component of the electric field
vector as a function of space and time, is denoted in the form

Ezn (I + 1/2, J + 1/2, K) = Ez [(I + 1/2)x, (J + 1/2)y, Kz, nt]. (4.1.13)

With the preceding notation and the field component configuration depicted in Figure 4.2,
application of the second-order central difference scheme to the spatial derivatives defined
in Eqs. (4.1.11) and (4.1.12) leads to

Ezn+1 (I + 1/2, J + 1/2, K)


= exp[−τ (I + 1/2, J + 1/2, K)t]Ezn (I + 1/2, J + 1/2, K)
ct
+ exp[−τ (I + 1/2, J + 1/2, K)t/2]
ε(I + 1/2, J + 1/2, K)

1 ) n+1/2 *
× Hy (I + 1, J + 1/2, K) − Hyn+1/2 (I, J + 1/2, K)
x

1 ) n+1/2 *
− H (I + 1/2, J + 1, K) − Hx n+1/2
(I + 1/2, J, K) , (4.1.14)
y x
4.1 Finite-difference time domain method 203

Hzn+1/2 (I, J, K + 1/2)



1 ) n
= Hzn−1/2 (I, J, K + 1/2) − ct Ey (I + 1/2, J, K + 1/2)
x
* 1 ) n
− Eyn (I − 1/2, J, K + 1/2) − E (I, J + 1/2, K + 1/2)
y x

*
− Exn (I, J − 1/2, K + 1/2) . (4.1.15)

Equations (4.1.14) and (4.1.15) represent an explicit finite-difference scheme; that is, the
temporal variations of the electromagnetic field values at the grid points can be simulated via
a straightforward time-marching iterative procedure without invoking the matrix inversion
required by an implicit finite-difference scheme.
For a given complex refractive index m = mr − imi , where mr and mi are, respectively,
the real and imaginary parts of the refractive index [both are positive real numbers; see
Eq. (2.3.14)], we must evaluate the corresponding permittivity and conductivity. To do so,
we shall consider a time-harmonic electromagnetic wave in the form
E(r, t) = eikct E(r), (4.1.16a)
H(r, t) = eikct H(r). (4.1.16b)
In the preceding equations, k is the circular wavenumber associated with the incident
wave, given by k = 2π/λ, and λ is the incident wavelength. The assumption of the time-
dependence factor exp(ikct) leads to a negative value for the imaginary part of the refractive
index. Substituting Eqs. (4.1.16a) and (4.1.16b) into Eq. (4.1.2) and simplifying the resultant
expression, we obtain
 τε
∇ × H(r) = ik ε − i E(r) = ik ε̃E(r), (4.1.17)
kc
where the effective complex permittivity ε̃ is given by
ε̃ = ε − iτ ε/(kc). (4.1.18)
It is clear that, for a given pair of permittivity ε and conductivity σ , the effective
complex permittivity ε̃ is dependent on the frequency, ω (= kc), of the electromagnetic
waves. Furthermore, for a non-ferromagnetic medium, we have the following relationship:
 
ε̃ = (mr − imi )2 = m2r − m2i − i2mr mi . (4.1.19)
Comparing Eqs. (4.1.18) and (4.1.19), we obtain
ε = m2r − m2i . (4.1.20a)
2mr mi kc
τ= . (4.1.20b)
m2r − m2i
The dielectric properties of the scattering particle and surrounding medium are given in
discrete form through τ (I + 1/2, J + 1/2, K) and ε(I + 1/2, J + 1/2, K) in Eq. (4.1.14).
204 Other useful approaches to light scattering

Because of the discretization of the parameters τ and ε, a pronounced shortcoming of the


FDTD technique is the staircasing approximation of the scattering particle in a Cartesian
grid mesh. To alleviate the staircasing effect, various techniques have been suggested to
evaluate permittivity values involved in FDTD equations (e.g., Yang and Liou 2000, Yang
et al. 2001), including the use of effective permittivity values for grid cells based on the
Maxwell-Garnett (1904) theory.
In terms of truncation errors, Eqs. (4.1.14) and (4.1.15) represent a second-order
scheme in both time and space. Furthermore, to avoid numerical instability, the choices
of x, y, z, and t values cannot violate the Courant–Friedrichs–Lewy (CFL) condi-
tion (Courant et al. 1928). Specifically, it was shown by Taflove and Brodwin (1975) that
the following condition is mandatory in order to achieve numerical stability of the FDTD
numerical scheme:
1
ct ≤  . (4.1.21)
1/x 2 + 1/y 2 + 1/z2
It should be pointed out that the finite-difference expressions in Eqs. (4.1.14) and (4.1.15)
along with their counterparts for the x and y components of electromagnetic fields constitute
an explicit numerical form; that is, the propagation of an electromagnetic wave can be
simulated through a time-marching iterative procedure without invoking the solution of a
set of linear equations involved in an implicit numerical scheme. For practical simulations,
the spatial increment needs to be smaller than approximately 115  120 of the incident
wavelength such that spatial variations of the field components are negligible over the
lengths of cell dimensions.

4.1.3 Absorbing boundary conditions and electromagnetic fields


in the frequency domain
An appropriate boundary condition for the truncation of the spatial domain is essential to the
successful implementation of the FDTD technique. Among various techniques developed
for the imposition of boundary conditions in conjunction with application of the FDTD
technique to electromagnetic scattering problems, the innovative perfectly matched layer
(PML) boundary condition developed by Berenger (1994) has been an effective and robust
approach to the suppressing of artificial reflections from boundaries of the truncated FDTD
computational domain. Technically, the absorption of outgoing waves by the PML method
is based on absorption by an artificial absorbing medium located at boundary layers in
the computational domain, backed by a perfectly conducting surface that is numerically
realized by setting the field values at outermost boundary-layer grid points to zero. The
conventional absorbing boundary condition is to specifically define the wave impedance
of the medium so that it matches that of free space. Such a simple matching approach
produces substantial non-zero reflections when a scattered wave impinges obliquely on an
absorbing medium. To overcome the disadvantage of the conventional method, the PML
absorbing medium can be selected such that the wave decay due to absorption is imposed
4.1 Finite-difference time domain method 205

on field components parallel to their boundary layers. To achieve this goal, each Cartesian
component of the electromagnetic field is split into two parts as follows:

(Ex , Ey , Ez ) = [(Ex2 + Ex3 ), (Ey1 + Ey3 ), (Ez1 + Ez2 )], (4.1.22a)

(Hx , Hy , Hz ) = [(Hx2 + Hx3 ), (Hy1 + Hy3 ), (Hz1 + Hz2 )], (4.1.22b)

where the subscripts 1, 2, and 3 denote the components of the electric (or magnetic) field,
which are respectively associated with the spatial derivatives of magnetic (or electric) field
components along the x, y, and z directions. The splitting of field components leads to 12
scalar equations which govern the propagation of electromagnetic waves within boundary
layers. For example, the PML boundary condition equations for the Ex and Hx components
at a boundary perpendicular to the z-axis are given by
exp[−τz (z)t] ∂ ∂(Hy1 + Hy3 )
{exp[τz (z)t]Ex3 } = − , (4.1.23a)
c ∂t ∂z
1 ∂Ex2 ∂(Hz1 + Hz2 )
= , (4.1.23b)
c ∂t ∂y
exp[−τz (z)t] ∂ ∂(Ey1 + Ey3 )
{exp[τz (z)t]Hx3 } = , (4.1.23c)
c ∂t ∂z
1 ∂Hx2 ∂(Ez1 + Ez2 )
=− . (4.1.23d)
c ∂t ∂y
In Eqs. (4.1.23a) and (4.1.23b), τz (z) is defined at boundary layers near z = 0 and
z = zmax , where zmax represents the computational domain dimension along the z-axis, as
follows:

τmax |(z − LPML )/LPML |p , for 0 ≤ z ≤ LPML ,
τz (z) = (4.1.24)
τmax |[LPML − (zmax − z)]/LPML | , for zmax − LPML ≤ z ≤ zmax ,
p

where LPML denotes the PML medium thickness, p is usually selected between 2 and 3
(Lazzi and Gandhi 1997) and τ max denotes the maximum absorption at z = 0 and z =
zmax , which can be determined by specifying the reflectance of boundary layers at a normal
incidence. From Eq. (4.1.24), the absorbing strength of the PML medium can be seen
to gradually increase from the interior PML medium boundary (e.g., z = LPML ) to the
exterior PML medium boundary (e.g., z = 0) in the case of the boundary near z = 0.
Various numerical experiments and sensitivity studies reported in the literature (e.g., Katz
et al. 1994) have proven that PML boundary conditions are superior to other absorbing or
transmitting boundary conditions such as the one derived from the one-way wave equation
(Mur 1981).
Values of the near field in the time domain can be computed using a time-marching
iterative procedure based on the finite-difference analog of Maxwell’s curl equations
and boundary condition equations. To obtain the single-scattering properties that are the
inherent optical characteristics of the particle in the frequency domain, time-dependent
206 Other useful approaches to light scattering

field values must be transformed to their corresponding counterparts in the frequency


domain. In practice, we may select an incident pulse as the initial excitation in the com-
putation and employ the discrete Fourier transform to obtain frequency-domain counter-
parts of the simulated field. Let Ez be the z-component of the electric field at location
[(I + 1/2)x, (J + 1/2)y, Kz] and its value at the time step n be Ezn . The time varia-
tion of Ez can then be written as

Ez [(I + 1/2)x, (J + 1/2)y, Kz, t]


N
= Ezn [(I + 1/2)x, (J + 1/2)y, Kz]δ(t − nt), (4.1.25)
n=0

where δ is the Dirac delta function and the maximum time step N is chosen such that
the field in the time domain is reduced to a small value. Accordingly the corresponding
spectrum in the frequency domain is given by

Ez [(I + 1/2)x, (J + 1/2)y, Kz, k]


 ∞ N

= Ez [(I + 1/2)x, (J + 1/2)y, Kz]δ(t − nt) exp(ikct)dt
n
−∞ n=0


N
= Ezn [(I + 1/2)x, (J + 1/2)y, Kz] exp(ikcnt). (4.1.26)
n=0

To avoid aliasing and numerical dispersion and to obtain a correct frequency spectrum,
the maximum wavenumber or the minimum wavelength, for the region in which the fre-
quency response of the scattering is evaluated, is bounded. In any finite-difference equation,
it is required that the wavelength of a simulated wave be larger than the grid size. In addi-
tion, field values in the frequency domain obtained by this procedure must be normalized
by the counterpart from Fourier transform of the incident wave at the center of the grid
mesh such that the frequency response of the scattering particle is scaled with reference to
an incident harmonic wave with unit amplitude.
To initialize the FDTD simulation using time-marching iterations, a Gaussian pulse can
be used as the incident wave, for example, the incident electric field polarized along the x
direction, given by

n
Ex,inc = exp[−(n/w − 5)2 ]. (4.1.27)

Using the trial and error method, Yang and Liou (1995) found w = 30 to be an optimal
value to avoid the numerical dispersion that can occur in the pulse propagation involved
in the simulation of light scattering by non-spherical ice crystals. Note that in Eq. (4.1.27)
the pulse peak is shifted by 5w such that the incident signal begins at a small value on the
order of ∼ 10−11 .
4.1 Finite-difference time domain method 207

4.1.4 Near-to-far-field mapping and optical property calculations


The frequency-domain near-field values computed from the FDTD algorithm and the dis-
crete Fourier transform must be mapped to their far-field counterparts. Either a surface-
integral method (SIM) or a volume-integral method (VIM) can be employed to map elec-
tromagnetic signals from the near-field regime to the far-field regime. For large indices
of refraction, the SIM allows a much coarser grid mesh resolution than the VIM in terms
of similar numerical accuracy (Zhai et al. 2007). These two approaches are in principle
equivalent; however, with reference to the optical properties of ice crystals in solar and
thermal infrared spectral regions, we found that the VIM technique is most efficient within
the framework of FDTD, particularly in terms of the accuracy of simulating backscatter
and phase matrix elements associated with polarization. Following is a brief summary of
the basic physical principle and mathematical framework for both SIM and VIM. In the
far field (or radiation zone), kr → ∞, the scattered far field for the electric vector may be
written in terms of a volume integral or a surface integral equation in the forms

k 2 exp(ikr)
VIM: Es (r) =
4πr

) 2  *
× mr (r ) − m2i (r ) − 1 + i2mr (r )mi (r ) {E(r ) − η[η · E(r )]}
V

× exp(−ikη · r )d 3 r , (4.1.28)


k 2 exp(ikr)
SIM: Es (r) = η ×  {ns (r ) × E(r ) − η × [ns (r ) × H(r )]}
−ik4πr s

× exp(−ikη · r )d 2 r , (4.1.29)

where η = r/r denotes the scattering direction. In Eq. (4.1.28) the integration is over the
particle interior volume, whereas in Eq. (4.1.29) the integration is performed over a surface
enclosing the scattering particle, and ns is a unit vector that is locally normal to the surface.
Note that the scattered field given by Eq. (4.1.28) or (4.1.29) is exact if the near field within
the particle interior or on the aforementioned surface is precisely known. To compute the
scattering phase matrix, the scattered field must be decomposed into components parallel
and perpendicular to the scattering plane in the form

Es (r) = X// Es,// (r) + X⊥ Es,⊥ (r), (4.1.30)

where X// and X⊥ are, respectively, the unit vectors parallel and perpendicular to the
scattering plane and satisfy the relationship X⊥ × X// = η. We can express the scat-
tered E-field in terms of a 2 × 2 amplitude scattering matrix containing four elements
Si (i = 1, 2, 3, 4) in the form




Es,// (r] exp(ikr) S2 S3 Eo,//
= , (4.1.31)
Es,⊥ (r) −ikr S4 S1 Eo,⊥
208 Other useful approaches to light scattering

where Eo,// and Eo,⊥ are the incident E-field components defined with respect to the
scattering plane. In the FDTD method, the incident wave is defined with respect to the
incident coordinate system whose z-axis is aligned with the incident direction, and is
represented by Eo,x and Eo,y . To map the incident E-field vector with respect to the incident
coordinate system to that with respect to the scattering plane, the following transformation
is required:




Eo,// X⊥ · x −X⊥ · y Eo,y
= . (4.1.32)
Eo,⊥ X⊥ · y X⊥ · x Eo,x
In this equation, x and y are unit vectors along the x- and y-axes of the incident coordinate
system. We can then perform an inversion to obtain the equivalent form, as follows:




Eo,y X⊥ · x X⊥ · y Eo,//
= . (4.1.33)
Eo,x −X⊥ · y X⊥ · x Eo,⊥
To compute the scattering amplitude matrix, we must consider two polarization states
for the incident wave, specifically with polarization configurations of (Eo,x = 1, Eo,y = 0)
and (Eo,x = 0, Eo,y = 1). Using the SIM or VIM scheme in conjunction with the two
polarization cases and performing a number of vector algebraic calculations, we can obtain
the amplitude scattering matrix. Specifically, we may define the following equations in
conjunction with either SIM or VIM in the following forms:
SIM scheme:



F//,x k2 X⊥ · [ns × E(r )] + X// · [ns × H(r )]
= 
F⊥,x 4π s X⊥ · [ns × H(r )] − X// · [ns × E(r )]


× exp(−ikη · r )d 2 r  , (4.1.34a)
Eo,x =1,Eo,y =0



F//,y k2 X⊥ · [ns × E(r )] + X// · [ns × H(r )]
= 
F⊥,y 4π s X⊥ · [ns × H(r )] − X// · [ns × E(r )]


× exp(−ikη · r )d 2 r  , (4.1.34b)
Eo,x =0,Eo,y =1

VIM scheme:



F//,x ik 3 ) 2  * X// · E(r )
= mr (r ) − m2i (r ) − 1 + i2mr (r )mi (r )
F⊥,x 4π V X⊥ · E(r )


× exp(−ikη · r )d 3 r  , (4.1.35a)
Eo,x =1,Eo,y =0

3 

F//,y ik ) 2  * X// · E(r )
= mr (r ) − m2i (r ) − 1 + i2mr (r )mi (r )
F⊥,y 4π V X⊥ · E(r )


× exp(−ikη · r )d 3 r  . (4.1.35b)
Eo,x =0,Eo,y =1
4.1 Finite-difference time domain method 209

From Eqs. (4.1.28), (4.1.29), (4.1.33), (4.1.34a), (4.1.34b), (4.1.35a), and (4.1.35b) and
after analysis, we obtain




Es,// exp(ikr) F//,y F//,x Eo,y
=
Es,⊥ −ikr F⊥,y F⊥,x Eo,x




exp(ikr) F//,y F//,x X⊥ · x X⊥ · y Eo,//
= . (4.1.36)
−ikr F⊥,y F⊥,x −X⊥ · y X⊥ · x Eo,⊥

Comparing Eqs. (4.1.31) and (4.1.36), we obtain the amplitude scattering matrix in the
form




S2 S3 F//,y F//,x X⊥ · x X⊥ · y
= . (4.1.37)
S4 S1 F⊥,y F⊥,x −X⊥ · y X⊥ · x

With the scattering amplitude matrix given in Eq. (4.1.37), it is straightforward to


evaluate the scattering phase matrix using the formalism outlined in Subsection 2.4.3,
where we showed that, for ice particles oriented randomly with an equal number of mirror-
imaging positions in space, the scattering phase matrix has a block-diagonal structure with
eight non-zero elements of which only six are independent. Moreover, the absorption and
extinction cross-sections σabs and σext , respectively, can be computed from the near field
within the interior of ice particles via the following equations:

k
σabs = 2mr (r )mi (r )E(r ) · E∗ (r )d 3 r , (4.1.38)
Einc · E∗inc V
 
k ) 2  2 
*  ∗  3 
σext = Im mr (r ) − mi (r ) − 1 E(r ) · Einc (r )d r , (4.1.39)
Einc · E∗inc V

where an asterisk denotes complex conjugation and Im{ } in Eq. (4.1.39) signifies the
imaginary part of the related variables. Equations (4.1.38) and (4.1.39) can be derived from
the electromagnetic energy conservation principle involving the Poynting vector associated
with light scattering processes, as illustrated in Yang and Liou (1996b).

4.1.5 Validation of the method and some pertinent results


The accuracy of the FDTD method can be checked using a number of canonical cases,
for instance, infinite circular cylinders and spheres (see Subsection 2.4.6) for which exact
solutions are available. In general, when the size of grid cells is on the order of 120
of the incident wavelength, solutions are in excellent agreement with their corresponding
analytical counterparts for cases involving ice indices of refraction at visible and infrared
wavelengths. Accuracy improvement could be achieved by decreasing the ratio of the grid
size to the incident wavelength; however, this improvement would be at the expense of
increasing computational effort. In what follows, we present several representative results
to illustrate the numerical performance of the FDTD method and its general applicability
to light scattering by ice particles.
210 Other useful approaches to light scattering

103 1

Absolute Error -P43/ P11 Absolute Error P33/ P11 Absolute Error -P12/ P11
Lorenz-Mie 0.5
FDTD 0.0
102
Phase Function P11

-0.5
x = 10 -1
0.3
10 1 l/Δs = 30
0.15
m = 1.5015 - i 0.067
0.0
100
-0.15
1
-1 0.5
10
0.0
-0.5
102 -1
0.3 0.3
Relative Error (%) Absolute Error

0.15
0.15 0.0
0.0 -0.15
-0.3
-0.15 1
0.5
-0.3
0.0
30 -0.5
15 -1
0.4
0.0 0.2
0.0
-15
-0.2
-30 -0.4
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180
Scattering Angle (deg.) Scattering Angle (deg.)

Figure 4.3 Non-zero phase matrix elements P11 in terms of absolute and relative (%) errors, and
−P12 , P33 , and −P43 normalized by P11 in terms of absolute errors, computed from the FDTD method
in comparison to the Lorenz–Mie solution for an ice sphere at a wavelength of 25 µm (after Yang
et al. 2004).

Figure 4.3 shows a comparison of the non-zero phase matrix elements computed with
the FDTD method and the Lorenz–Mie results for an ice sphere at a far-infrared wavelength
λ = 25 µm with a refractive index of m = 1.5015 − i 0.067. In the past two decades, far-
infrared spectral information has received substantial attention in the atmospheric research
community (e.g., Mlynczak et al. 2002). In particular, its signatures have been demon-
strated to effectively facilitate the retrieval of microphysical and optical properties of thin
cirrus clouds (Yang et al. 2003b). For the FDTD simulation shown in Figure 4.3, 5872
time-marching iterations were performed to map the near field from the time domain to
the frequency domain through the discrete Fourier transform. The left-hand column of
Figure 4.3 shows FDTD and Lorenz–Mie phase functions along with both absolute and rel-
ative errors of the FDTD results. For the FDTD phase function displayed here, the largest
absolute errors are found in forward directions with scattering angles smaller than 20°,
whereas the largest relative errors are noted at scattering angles near 160°. The right-hand
column of Figure 4.3 depicts a comparison between FDTD and Lorenz–Mie solutions for
4.1 Finite-difference time domain method 211

102 1

Absolute Error -P43/ P11 Absolute Error P33/ P11 Absolute Error -P12/ P11
Lorenz-Mie 0.5
FDTD 0.0
Phase Function P11

-0.5
-1
101 x=5 0.1
l/Δs =165 0.05
m = 8.2252 - i 1.6808 0.0
-0.05
100 -0.1
1
0.5
0.0
-0.5
-1
10 -1
0.25 0.1
Relative Error (%) Absolute Error

0.05
0.0 0.0
-0.05
-0.25 -0.1
1
0.5
-0.5
0.0
10 -0.5
5 -1
0.1
0.0 0.05
0.0
-5
-0.05
-10 -0.1
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180
Scattering Angle (deg.) Scattering Angle (deg.)

Figure 4.4 Comparison between FDTD and Lorenz–Mie solutions for a sphere with a large refractive
index. The presentation in this figure has the same format as in Figure 4.3 (after Yang et al. 2004).

the other phase matrix elements related to the polarization state of scattered light. Because
these phase matrix element values can be zero, errors in the FDTD results are only quan-
tified in terms of absolute errors in the right-hand column of Figure 4.3. Overall, FDTD
results for phase matrix elements are quite accurate with respect to exact Lorenz–Mie
values.
To demonstrate the applicability of the FDTD method for an optically dense medium
with a large refractive index, Figure 4.4 shows a comparison between FDTD and Lorenz–
Mie solutions for the scattering phase matrix of a sphere with a size parameter of 5 and
a refractive index of 8.2252 − i1.6808, which represents the value for liquid water at a
temperature of 300 K at a microwave wavelength of 3.2 cm. Because an optically dense
medium is involved in this case, a very fine grid-mesh resolution (s = λ165) was
used and, in this case, 7500 time-marching steps were required to obtain a convergent
solution. In terms of the relative error, deviations of FDTD phase function results from
exact Lorenz–Mie values are less than 5% across the entire scattering angle domain. For
other phase matrix elements, FDTD solutions closely match Lorenz–Mie counterparts.
The comparisons shown in Figures 4.3 and 4.4 clearly demonstrate the applicability and
212 Other useful approaches to light scattering

1
102 ADDA 0.8 P22/P11
FDTD 0.6
0.4
0.2
101 kL = 30, l = 0.66 μm
1
m = 1.3078 - i1.66x10-8 P33/P11
0.5
P11

0
100
-0.5
-1
0.4
10-1 0.2 P34/P11
0
-0.2
10-2 -0.4
1
P44/P11
P12/P11

0.5 0.5
0 0
-0.5 -0.5
1
102 ADDA 0.99
FDTD 0.98
0.97
P22/P11
0.96
101
kL = 30, l = 12 μm 1
m = 1.28 - i0.41332 0.5
P33/P11
P11

0
100
-0.5
-1
0.5 P34/P11
10-1
0

-0.5
10-2 1
1
P12/P11

0.5
0 0 P44/P11
-0.5
-1 -1
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Scattering Angle (deg.) Scattering Angle (deg.)

Figure 4.5 Comparison between phase matrix elements calculated from FDTD and DDA at 0.66
and 12 µm wavelengths. The Amsterdam DDA (ADDA) code developed by Yurkin et al. (2007) was
used for the present DDA simulation (after Yang et al. 2013).

accuracy of FDTD for the scattering of light by a dielectric sphere with a moderate or large
refractive index.
Applicability of the FDTD method to light scattering by hexagonal particles is illustrated
in Figure 4.5, in which a comparison between FDTD and DDA solutions was made for the
scattering phase matrix of randomly oriented hexagonal ice crystals at 0.66 µm (visible) and
4.1 Finite-difference time domain method 213

Extinction Efficiency Qext 4

l = 0.55 μm
3 L/a = 6

FDTD
1
Ray-by-ray

0
6 10 100 1000

4 1

Single-Scattering Albedo
Extinction Efficiency Qext

l = 3.7 μm
l = 3.7 μm
3

2 0.75

0 0.5
6 10 100 1000 6 10 100 1000

Size Parameter kL Size Parameter kL

Figure 4.6 Comparison of extinction efficiency and single-scattering albedo computed by FDTD
and by the ray-by-ray geometric-optics approach for randomly oriented column crystals with an
aspect ratio of 6 using 0.55 µm (essentially without absorption) and 3.7 µm wavelengths (after Yang
and Liou 1997).

12 µm (thermal IR) wavelengths. Herein, we note that FDTD and DDA are substantially
different in terms of their fundamental approaches: The former is a time-domain method,
whereas the latter is a frequency-domain method. Figure 4.5 illustrates that FDTD results
closely match their DDA counterparts, a confirmation of the applicability and accuracy of
the FDTD approach to light scattering by ice crystals from the perspective of another well
validated and numerically rigorous method.
Figure 4.6 illustrates the extinction efficiency and single-scattering albedo computed
from FDTD and the ray-by-ray geometric-optics algorithm (Yang and Liou 1997) dis-
cussed in Chapter 3. At a size parameter of approximately 20, results computed with the
two methods converge. The ray-by-ray algorithm is much more computationally efficient
than FDTD, particularly for computing extinction and absorption cross-sections. Thus,
results shown in Figure 4.6 provide an approximate accuracy guideline in terms of deter-
mining extinction and absorption characteristics; that is, FDTD can be effectively applied
to size parameters smaller than 20, whereas the ray-by-ray approach can be employed
214 Other useful approaches to light scattering

2
10
1 branch
3 branches
4 branches
1 5 branches
10

Phase Function
6 branches
8 branches
9 branches
0
10 10 branches
12 branches

-1
10

x = 10, l = 11 μm
-2
10
0 60 120 180
Scattering Angle (deg.)

Figure 4.7 Phase functions for randomly oriented single-bullet and bullet-rosette ice crystals as a
function of the scattering angle at a wavelength of 11 µm with an index of refraction m = 1.0925 −
i 0.248. The size parameter is defined with respect to the length of a bullet element (after Yang et al.
2004).

for larger size parameters. The convergence of ray-by-ray and FDTD approaches at a size
parameter of 20 and an aspect ratio of La = 6 for ice columns illustrates that the edge
effect (see Subsections 3.3.5–3.3.7), which was not included in the former, appears to be
insignificant.
The FDTD method has been employed to study the optical properties of complex ice
crystals. For example, Figure 4.7 shows phase functions of randomly oriented single-bullet
and bullet-rosette ice crystals at an infrared wavelength of 11 µm. The angle between the
pyramidal faces of a bullet element and the element axis is 26.5°. The phase function
is quite sensitive to the number of ice bullet elements. In particular, the phase function
of a multi-branched rosette with more than eight branches displays maximal features in
the side-scattering direction at 60°. Although the radiative properties of ice crystals are
dominated by absorption at λ = 11 µm, the scattering effect is not negligible (Chou et al.
1999). Thus, using an appropriate scattering phase function for multi-branched rosette ice
crystals, representative of the morphology in some cirrus clouds, may enhance the accuracy
in downstream remote sensing applications (Heymsfield et al. 2002).
Figure 4.8 depicts phase functions for six ice crystal habits with a small size parameter kD
of 20, where D is the maximum dimension, to illustrate ice crystal shape effect. The imprint
of halo peaks in the phase function pattern largely disappears for small ice crystals. The
phase function patterns for small plates and columns are rather smooth, particularly in the
scattering region from 90° to 180°. Furthermore, the phase function of droxtal ice crystals
displays substantial variation in the scattering angle domain due to phase interference
effects that have not been smoothed out in random orientation average.
4.1 Finite-difference time domain method 215

102 102 102

101 101 101

100 kD = 20 100 kD = 20 100 kD = 20

10-1 10-1
10-1
Phase Function

10-2 10-2
10-2
0 30 60 90 120 150 180 0 30 60 90 120 150 180 0 30 60 90 120 150 180
102 102 102

101 101 101

ka = 10 kL = 20 kL = 20
100 L/a = 0.5 100 100
L/a = 8 L/a = 8

10-1 10-1 10-1

10-2 10-2 10-2


0 30 60 90 120 150 180 0 30 60 90 120 150 180 0 30 60 90 120 150 180
Scattering Angle (deg.) Scattering Angle (deg.) Scattering Angle (deg.)

Figure 4.8 Comparison of phase functions computed from the FDTD method for ice crystal shapes
commonly observed in ice clouds. The parameter D is the maximum dimension for droxtal, bullet
rosette, or aggregate. For plate and column, a denotes the half-width and L is the length (column) or
depth (plate), k = 2π/λ, and λ is the wavelength (after Yang and Liou 2006).

4.1.6 Pseudo-spectral time domain method


Stemming from the traditional FDTD, Liu (1997) employed the pseudo-spectral method,
instead of the finite-difference method, to approximate the spatial derivatives in Maxwell’s
equations; this is frequently referred to as the pseudo-spectral time domain (PSTD) method.
A comprehensive review of the PSTD method can be found in Panetta et al. (2013). In this
subsection, we shall capture the essence of the PSTD approximation in terms of spatial
derivatives of the form
⎡ ⎤
1  2πI l i ⎣ 
N/2−1 N/2−1
∂F n (I, J, K)
F n (m, J, K)e− N i ⎦,
2πlm
= e N 2πkl i (4.1.40)
∂x N l=−N/2 m=−N/2

where N is the total number of grid points in the x-axis direction and kl is the Fourier spectral
variable given by kl = l(Nx). Unlike FDTD, which uses the “Yee cell” (Figure 4.2),
PSTD employs a centered grid scheme in which all the field components are specified at the
216 Other useful approaches to light scattering

centers of grid cells. PSTD provides a higher degree of accuracy and smaller numerical
dispersion errors, leading to the enhancement of computational applicability and efficiency
due to the use of a relatively coarse grid mesh and fast Fourier transform. Furthermore, the
CFL condition for PSTD, which differs from the FDTD counterpart, is given by (Liu 1997)

2
ct ≤  . (4.1.41)
π 1/x + 1/y 2 + 1/z2
2

Chen et al. (2008) applied PSTD to compute the single-scattering properties of ice
crystals with small to medium size parameters by alleviating the Gibbs phenomenon and
parallelizing PSTD implementations. As an example, Figure 4.9 depicts phase functions
computed by the PSTD method compared with those from the Lorenz–Mie theory for size
parameters 30 and 50 in an optically soft case (m = 1.05 − i 0) and in a case involving a
moderate refractive index (m = 1.6 − i 0). For the optically soft case, a grid spatial resolution
of 10 was used, whereas the grid resolution was increased to 16 for the moderate case.
It is clear from Figure 4.9 that an excellent agreement between PSTD and Lorenz–Mie
results has been achieved, although slight differences are observed at scattering angles
of 120° and 160° at which phase function minima occur. Also, PSTD phase functions
deviate slightly from their Lorenz–Mie counterparts at scattering angles near 180°.
Liu et al. (2012a) further improved the PSTD method using an empirical approach to
mitigate the effect associated with the Gibbs phenomenon caused by medium discontinuity.
An illustrative explanation of the Gibbs phenomenon can be found in Panetta et al. (2013).
Furthermore, Liu et al. (2012b) applied PSTD to moderately large size parameters. For
example, Figure 4.10 shows the PSTD solution for the optical properties of a sphere with
a size parameter of 200 in comparison with Lorenz–Mie results. The refractive index of
the sphere used was 1.312 − i 1.489 × 10−9 . The relative error in the normalized phase
function and the absolute errors in the ratios P12 P11 , P33 P11 , and P34 P11 are illustrated
in Figure 4.10. Overall, PSTD and Lorenz–Mie results are in close agreement.
Figure 4.11 depicts a comparison between PSTD and T-matrix solutions for the phase
function associated with an oriented spheroid with a refractive index of 1.381 − i 3.9 × 10−9
for finite circular cylinder and spheroid cases. The cylinder size is specified by the radius of
its cross-section r and by its length L, while the spheroid size is determined in terms of its
polar radius b and equatorial radius a. For T-matrix results, the T-matrix code developed by
Mishchenko and Travis (1998) was used, while for PSTD simulations, a grid resolution of
12 was employed. Additionally, the phase functions in Figure 4.11 are defined with respect
to the plane containing the incident direction and the particle’s symmetry axis. PSTD results
for finite cylinders and spheroids closely match T-matrix results.
Figure 4.12 illustrates the phase functions of ice crystals defined by two shapes, hollow
hexagonal column and aggregate, computed by the PSTD method. The orientation of the
ice crystals in the laboratory coordinate system is shown in the upper diagrams. The phase
function of ice crystals with a size parameter of 20 was calculated using the PSTD method
with a spatial resolution of 14 at a wavelength of 3.7 µm. In the simulation, the incident
4.1 Finite-difference time domain method 217

106
(a) 2πr/l = 30 (b) 2πr/l = 30
m = 1.05 m = 1.6
105
Δs = l /10 Δs = l /16
104
Phase Function

103 Lorenz-Mie
PSTD
102

101

100

10-1

10-2

107
(c) 2πr/l = 50 (d) 2πr/l = 50
106 m = 1.05 m = 1.6
Δs = l /10 Δs = l /16
105
Phase Function

104

103

102

101

100

10-1

10-2
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Scattering Angle (deg.) Scattering Angle (deg.)

Figure 4.9 Comparison of phase functions computed by the PSTD method and by the Lorenz–Mie
theory for spheres with size parameters of 30 and 50 (after Chen et al. 2008).

direction was defined by θi = 45◦ and φi = 0◦ , where θi and φi are the incident zenith
and azimuthal angles specified with respect to the laboratory coordinate system. Phase
functions were computed for θ s = 0°–180° and φ s = 0°. PSTD solutions are compared
with the corresponding results computed by the FDTD method. The spatial resolutions
for PSTD and FDTD simulations were 12 and 32, respectively. The phase functions com-
puted from the PSTD and FDTD methods for each geometry are compared in the graphs,
Figure 4.12a for a hollow column ice crystal and Figure 4.12b for an aggregate. The aspect
218 Other useful approaches to light scattering

2
4
10
Phase Function, P11

Lorenz−Mie

Relative Error
2 1
10 PSTD

10 0
0

−2
10

−1
1.0 1.0
P12 /P11

0.0 0.0

−1.0 −1.0
1.0 1.0
Absolute Error
P33 /P11

0.0 0.0

−1.0 −1.0

1.0 1.0
P34 /P11

0.0 0.0

−1.0 −1.0
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Scattering Angle (deg.) Scattering Angle (deg.)

Figure 4.10 Non-zero phase matrix elements computed using PSTD for a sphere with a size param-
eter of 200 and a refractive index of 1.312 − i 1.489 × 10−9 . The relative errors in the phase function
(P11 ) and absolute errors in the terms P12 P11 , P33 P11 , and P34 P11 are displayed in the right panels
(after Liu et al. 2012b).

ratio of the hollow column is 2aL = 0.7, where a and L are the semi-width and length of
the column, respectively. The depth of cavities at each end of the column is given by d =
0.25L. The aggregate geometry definition for a given maximum dimension D of an ice crys-
tal follows that of Yang and Liou (1998a). From Figure 4.12a, the PSTD results agree well
with their FDTD counterparts in all scattering directions, except for slight differences at
several scattering angles. However, the FDTD method requires a much finer grid resolution
than the PSTD method in order to achieve similar accuracies.
4.1 Finite-difference time domain method 219

104 106
(a) 2πb/l = 20 (b) 2πb/l = 50
b/a = 2 b/a = 2
103 105

102 104
Phase Function

T-matrix
PSTD
101 103

100 102

10-1 101

10-2 100

105 107
(c) 2πr/l = 20 (d) 2πr/l = 50
L/r = 1 106 L/r = 1
104
105 o
o 30
Phase Function

30
103
104

103
102

102
101
101

100 100
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Scattering Angle (deg.) Scattering Angle (deg.)

Figure 4.11 Comparison of PSTD and T-matrix results for the phase functions of spheroids and
circular cylinders. The index of refraction used is m = 1.381 − i 3.9 × 10−9 and the size parameters
are 20 and 50 with an aspect ratio of 2 for the spheroid and 1 for the circular cylinder (after Chen
et al. 2008).

Figure 4.13 shows the non-zero phase matrix elements for randomly oriented hexagonal
columns with a size parameter of 100. With moderate size parameters, the PSTD results
can only be compared with those given by geometric-optics methods. Here, the solutions
based on the improved geometric-optics method (IGOM) (Yang and Liou 1996a) are used
220 Other useful approaches to light scattering

Z Z
(a) (b)

Y
Y

X
X
103 103
Lπ/l = 20 Dπ/l = 20
102 l = 3.7 μm 102 l = 3.7 μm
m = 1.4005 - i 7.2x10-3 m = 1.4005 - i 7.2x10-3
Phase Function

101 101
FDTD
PSTD
100 100

10-1 10-1

10-2 10-2

10-3 10-3
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Scattering Angle (deg.) Scattering Angle (deg.)

Figure 4.12 Comparison of FDTD and PSTD results for the phase function as a function of scattering
angle for hollow column and aggregate ice crystals. The input parameters are shown in the figure
(after Chen et al. 2008).

for the comparison, and the IGOM results closely agree with their PSTD counterparts.
The two methods give almost the same overall variation patterns, particularly, for P11 ;
however, the PSTD solutions for P12 P11 , P22 P11 , P33 P11 , P34 P11 , and P44 P11 show
pronounced variations versus scattering angle, and the IGOM results are relatively smooth.
The differences occur because the PSTD simulation rigorously takes into account the phase
interference of the electromagnetic waves. Moreover, Figure 4.13 clearly identifies strong
scattering peaks at 22° and 46° in the phase function computed from both PSTD and IGOM
methods for a size parameter of 100.
The single-scattering properties (extinction efficiency, single-scattering albedo, and
asymmetry factor) of randomly oriented hexagonal columns and plates computed by FDTD
and PSTD are illustrated in Figure 4.14. The indices of refraction of ice at wavelengths
of 3.7 µm and 23 µm are 1.4005 − i 0.007 201 and 1.4424 − i 0.027 01, respectively. The
results obtained by FDTD and PSTD are shown to agree with each other quite well.
Due to its flexibility in handling particle geometry and inhomogeneity, the FDTD method
has been widely used to compute the single-scattering properties of non-spherical ice
4.1 Finite-difference time domain method 221

4
10 0.4
PSTD
Phase Function, P11
IGOM 0.2
2
10 x = 100

12 11
P /P
0
0
10
−0.2

−2
10 −0.4
1 1

0.5
0.5

33 11
22 11
P /P

P /P
0

0
−0.5

−0.5 −1
0.2 1

0 0.5
34 11

44 11
P /P

P /P

−0.2 0

−0.4 −0.5

−0.6 −1
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Scattering Angle (deg.) Scattering Angle (deg.)

Figure 4.13 Non-zero phase matrix elements of randomly oriented hexagonal columns as a function
of scattering angle using a size parameter of 100, simulated by PSTD and IGOM. The hexagonal
column has a diameter-to-length ratio of 1.0 and a refractive index of 1.312 − i 1.489 × 10−9 (after
Liu et al. 2012b).

crystals and aerosols (Yang and Liou 1996b, 2000; Fu et al. 1998; Sun et al. 1999; Yang
et al. 1997, 2000b, 2005; Baran 2003; Baum et al. 2007; Ishimoto et al. 2012). Application
of the more powerful PSTD method to complex ice crystals and aerosols appears to be
quite promising, as shown in the work of Chen et al. (2008) and Liu et al. (2012b). With
continued numerical and technical advances in FDTD and PSTD as well as an increase in
supercomputer power, it is anticipated that these two methods will play an important role
in solutions for electromagnetic scattering and absorption by atmospheric particles which
exhibit complex geometry and internal inhomogeneity.
222 Other useful approaches to light scattering

(a ) (b)
Single-Scattering Albedo Extinction Efficiency l = 3.7μm l = 23μm l = 3.7μm l = 23μm
3
2.5
2
1.5
1 FDTD FDTD
PSTD PSTD
0.5
0
1

0.8

0.6

0.4

0.2

0
1
Asymmetry Factor

0.8

0.6

0.4

0.2

0
100 101 102 100 101 102 100 101 102 100 101 102
Size Parameter xmax Size Parameter xmax

Figure 4.14 The extinction efficiency, single-scattering albedo, and asymmetry factor for (a) ran-
domly oriented hexagonal columns and (b) plates as a function of size parameter xmax at a wavelength
of 3.7 µm (m = 1.4005 − i 0.007 201) and 23 µm (m = 1.4424 − i 0.027 01), where xmax is defined
as 2πrmax λ, rmax denotes one-half of the maximum dimension, and λ is the wavelength (courtesy of
Chao Liu).

4.2 T-matrix numerical method


4.2.1 Fundamentals of the T-matrix approach
The incident and scattered electric fields can be expanded in terms of suitable vector
wave functions. The expansion coefficients associated with the scattered wave can be
related to their counterparts for the incident wave through a transmitting matrix, referred
to as the T-matrix, which contains all the inherent single-scattering characteristics of the
corresponding scattering particle. Waterman (1971) developed a technique known as the
extended boundary condition method (EBCM) to solve the T-matrix. The EBCM has been
thoroughly explained by Barber and Yeh (1975). Mishchenko and Travis (1998) have
implemented EBCM to efficiently compute the single-scattering properties of a number
4.2 T-matrix numerical method 223

of axially symmetric non-spherical particles, including spheroids, circular cylinders, two-


sphere clusters, and Chebyshev particles.
The EBCM technique has been found to be numerically unstable in calculations for
particles with large size parameters, and the maximum convergent size parameter has
been shown to be sensitively dependent on the refractive index and particle geometry.
In the early development of EBCM, Iskander et al. (1983) further improved Waterman’s
EBCM in terms of the numerical stability for spheroidal particles. Additionally, Mugnai
and Wiscombe (1986, 1989) and Wiscombe and Mugnai (1988) applied EBCM to non-
spherical shapes with smooth surfaces that could be defined by Chebyshev polynomials.
More efficient approaches that overcome the numerical instability problem in computing the
T-matrix for highly elongated particles have subsequently been developed. With computer
coding improvements, in particular, of the analytical orientation-average algorithm and
the computation of relevant special functions, the T-matrix method can be applied to
axially symmetric homogeneous non-spherical particles with size parameters as large as 180
(Mishchenko et al. 2000a). However, EBCM has not been frequently used for application
to non-symmetric and inhomogeneous particles due to inadequate numerical performance.
The T-matrix method commonly begins with the expansion of the incident and scattered
fields with respect to vector spherical wave functions in the forms
∞ 
 n
Ei (R) = [amn RgMmn (kR) + bmn RgNmn (kR)], (4.2.1a)
n=1 m=−n
∞ 
 n
Es (R) = [pmn Mmn (kR) + qmn Nmn (kR)], (4.2.1b)
n=1 m=−n

where RgMmn , RgNmn , Mmn , and Nmn are vector spherical wave functions that satisfy the
vector Helmholtz equation; Mmn and Nmn involve spherical Hankel functions and form a
basis for expanding the scattered field satisfying the Somerfield radiation condition; and
RgMmn and RgNmn are regular spherical wave functions obtained from spherical Bessel
functions. Because of the linearity of Maxwell’s equations and boundary conditions, the
relationship between the scattered coefficients (pmn and qmn ) and incident coefficients (amn
and bmn ) must be linear and can be expressed by a T-matrix as

∞ 
 n
) 11 *
pmn = Tmnm n am n + T12
mnm n bm n , (4.2.2a)
n =1 m =−n

∞ 
 n
) 21 *
qmn = Tmnm n am n + T22
mnm n bm n . (4.2.2b)
n =1 m =−n

In compact matrix form, we can write





11

p a T T12 a
=T = . (4.2.3)
q b T21 T22 b
224 Other useful approaches to light scattering

The preceding equation can also be written in a more detailed matrix form as
⎡ ⎤ ⎡ ⎤⎡ ⎤
p1 T11 11 T11 12 ... ... T11 1,lmax T12 1,lmax a1
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢ q1 ⎥ ⎢ T11 21 T11 22 ... ... T21 1,lmax T22 1,lmax ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ b1 ⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢ ... ⎥ ⎢ ... ... ... ... ... ... ⎥⎢ ... ⎥
⎢ ⎥=⎢ ⎥⎢ ⎥ . (4.2.4)
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢ ... ⎥ ⎢ ... ... ... ... ... ... ⎥⎢ ... ⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢ p ⎥ ⎢ T11 T12 T11 T12 ⎥⎢ ⎥
⎣ lmax ⎦ ⎣ lmax ,1 lmax ,1 ... ... lmax ,lmax lmax ,lmax ⎦ ⎣ almax ⎦

qlmax T21
lmax ,1 T22
lmax ,1 ... ... T21
lmax ,lmax T22
lmax ,lmax blmax

In Eq. (4.2.4), double indices (mn) are combined into a single index l defined as
n(n + 1) + m. Equation (4.2.3) or (4.2.4) forms the foundation of the T-matrix approach.
If the T-matrix for a given scattering particle is known, the scattering amplitude matrix can
be calculated in a straightforward manner. To illustrate the computational procedure, we
shall consider the amplitude scattering matrix of the Lorenz–Mie type, given by


S2 0
S= , (4.2.5)
0 S1

where the two diagonal elements are defined as follows:

∞
2n + 1
S2 = [an τn (cos θ ) + bn πn (cos θ )], (4.2.6)
n=1
n(n + 1)

∞
2n + 1
S1 = [an πn (cos θ ) + bn τn (cos θ )]. (4.2.7)
n=1
n(n + 1)

In Eqs. (4.2.6) and (4.2.7), θ is the scattering angle, and the two coefficients are related to
the T-matrix elements by
∞ 5
 2n + 1 n −n ) 21 *
an = − i T1n1n + T1n1n
22
 , (4.2.8)
n =1
2n + 1
∞ 5
 2n + 1 n −n ) 11 *
bn = − i T1n1n + T1n1n
12
 . (4.2.9)
n =1
2n + 1

When the coefficients an and bn in Eqs. (4.2.8) and (4.2.9) are known, the amplitude
scattering matrix and, subsequently, the phase matrix can be computed.
The T-matrix contains the complete inherent optical properties of the scattering parti-
cle and is independent of the orientation of the particle with respect to the direction of
the incident light. This unique feature of the T-matrix discovered by Mishchenko (1990)
provides a computationally efficient and exact approach to calculating single-scattering
properties averaged over random orientations. For example, the extinction and scattering
4.2 T-matrix numerical method 225

cross-sections can be obtained for T-matrix elements through explicit summations given by
Mishchenko et al. (2002) as follows:
2π 
lmax 
lmax
 11 2  12 2  21 2  22 2 
Csca  = T   + T   + T   + T   , (4.2.10)
k l=1 l  =1 ll
2 ll ll ll

2π 
lmax
 11 
Cext  = − 2
Re Tll + Tll22 . (4.2.11)
k l=1

The EBCM approach to determining the T-matrix is to first find the field scattered
by an object bounded by a closed surface in terms of an integral equation that involves
the unknown surface field on the exterior of the closed surface, which is related to the
internal field through boundary conditions. The internal field within the particle defined by
a refractive index m relative to the surrounding medium is defined by
∞ 
 n
E (R) =
int
[cmn RgMmn (mkR) + dmn RgNmn (mkR)] , (4.2.12)
n=1 m=−n

where cmn and dmn are unknown coefficients (the symbol for the refractive index should not
be confused with the subscript m). By using boundary conditions at the scatterer surface,
which requires the continuity of tangential components of electric and magnetic fields, we
obtain the following matrix equation that relates scattered and internal fields:




p RgQ11 RgQ12 c
=− . (4.2.13)
q RgQ21 RgQ22 d
The elements in the RgQ matrix are surface integrals of the products of vector spherical
wave functions that depend only on the particle’s size, shape, and refractive index. Moreover,
from the boundary condition and the cancelation of incident and scattered fields within the
particle with reference to equivalent surface currents, we can show that


11

a Q Q12 c
= , (4.2.14)
b Q21 Q22 d
where the Q elements are also given by integrals over the surface of the particle and depend
only on its geometric characteristics. The unknown expansion coefficients of the internal
field, c and d, can be expressed in terms of known coefficients, a and b, by inverting the
matrix equation (4.2.14). From Eqs. (4.2.3), (4.2.13), and (4.2.14), we obtain the T-matrix
in the form
T = −Rg Q Q−1 , (4.2.15)
where the T-matrix is independent of the incident field represented by coefficients a and b.
We present Figure 4.15 to show the phase function P11 and the degree of linear polar-
ization −P12 P11 computed for a sphere and for three spheroids with aspect ratios 1.5, 2.0,
and 2.5. The same size parameter was used for the sphere and spheroids, defined in terms
of a volume-equivalent sphere of 35. The simulation was conducted by using the EBCM
T-matrix computational program developed by Mishchenko and Travis (1998). Note that
226 Other useful approaches to light scattering

104 1
sphere
3
10 a/b = 1.5

a/b = 2.0 0.5


102 a/b = 2.5
xv = 35

-P12/P11
P11

1 0
10

100

-0.5
-1
10

10-2 -1
0 45 90 135 180 0 45 90 135 180
Scattering Angle (deg.) Scattering Angle (deg.)

Figure 4.15 T-matrix computations of the phase function P11 and the degree of linear polariza-
tion −P12 P11 as a function of the scattering angle for a sphere and three spheroids defined by a
volume equivalent spherical size parameter, xv . The T-matrix computational program developed by
Mishchenko and Travis (1998) was used in these simulations. The refractive index employed in the
calculation for these four particles was 1.53 − i 0.008.

−P12 P11 represents the degree of linear polarization in the case involving an unpolar-
ized incident light. Figure 4.15 clearly illustrates deviations between the scattering and
polarization properties of spheroids and a sphere.

4.2.2 Invariant imbedding T-matrix method


In this subsection, we present an alternative to the conventional T-matrix to light scattering
computations which follows the invariant imbedding principle. This principle has been
used to facilitate radiative transfer computations for external emergent quantities associ-
ated with reflection and transmission of a system (e.g., Chandrasekhar 1950). It appears
that application of the invariant imbedding principle to the single-scattering of light by a
particle has been accomplished by Johnson (1988), who used this principle to compute
T-matrix elements associated with the single-scattering properties of dielectric particles in
a number of axially rotational symmetry cases for small size parameters. Johnson’s work,
however, has not been noted and cited by a number of excellent books which discuss the
T-matrix approach (Barber and Hill 1990, Mishchenko et al. 2002, Doicu et al. 2006), nor
by the comprehensive T-matrix reference database (Mishchenko et al. 2010a, Zakharova
4.2 T-matrix numerical method 227

(a) (b) (c)

(d) (e) (f)

Figure 4.16 A conceptual representation of a 2-D plate as an example of the invariant imbedding
T-matrix (II-TM) method. (a) The 2-D plate shape. (b) This is circumscribed by an imaginary
inhomogeneous sphere; the portion of the sphere exterior to the 2-D plate is empty. (c) Discretization
of this inhomogeneous sphere into multi-layers. (d) Computation of the T-matrix using an internal
sphere whose T-matrix can be obtained from the Lorenz–Mie theory. (e) An intermediate step in
computing the T-matrix via iterative procedures based on the II-TM approach. (f) The final particle
geometry, a 2-D plate, after completing II-TM computations.

et al. 2012). Also, according to the ISI Web of Knowledge, Johnson’s work in 1988 was
cited only five times by other researchers as of February 8, 2013. We further note that John-
son’s exploration of the invariant imbedding principle for incorporation into the T-matrix
approach can be traced back to his study of quantum scattering (Johnson and Secrest
1968).
Figure 4.16 illustrates a conceptual procedure for application of the invariant imbedding
principle to T-matrix calculations (hereafter referred to as the II-TM approach) for scattering
by an arbitrarily shaped particle: (a) A 2-D plate is used as an example. (b) Scattering by
this plate can be thought of as scattering by an inhomogeneous circumscribed sphere
that is partially empty, such that the refractive index of the empty portion is 1. (c) This
inhomogeneous sphere is then discretized into a multi-layered sphere, so the problem
of light scattering by a non-spherical particle is transformed into that of light scattering
by a multi-layered inhomogeneous sphere. (d) The T-matrix is computed for an internal
sphere using Lorenz–Mie theory; for computational efficiency, we may select a sphere that
is inscribed within the 2-D plate. (e) An intermediate step is to compute the T-matrix by
means of an iterative procedure based on the II-TM approach. (f) The final particle geometry
when iterative II-TM computations are completed would be the original 2-D plate.
The principle of II-TM is to obtain the T-matrix of an inhomogeneous sphere with p
layers based on the T-matrix of an inhomogeneous sphere with p − 1 layers. As an initial
point for iterative computations, the T-matrix at the origin is zero. The II-TM recursive
228 Other useful approaches to light scattering

4
10 0.5

11
II-TM

-P /P
3
EBCM 0

12
10
c
2 -0.5
10 a
1
0.5

11
P /P
11

1
10 0
P

22
m = 1.311 -0.5
10
0
kc = 2ka = 95 -1
1
0.5

11
-1

P /P
10 0

33
-0.5
-2
10 -1
0.5 1
0.5
11
11

P /P
P /P

0 0
44
43

-0.5
-0.5 -1
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Scattering Angle (deg.) Scattering Angle (deg.)

Figure 4.17 Six independent scattering phase matrix elements simulated by the EBCM method
(Mishchenko and Travis 1998) and the new II-TM approach for randomly oriented spheroids as a
function of the scattering angle (after Bi et al. 2013).

equation is given as follows:


T(rp+1 ) = Q11 (rp+1 ) + [I + Q12 (rp+1 )][I − T(rp )Q22 (rp+1 )]−1 T(rp )[I + Q21 (rp+1 )],
(4.2.16)

where T(rp ) is the T-matrix composed of p layers, I is the identity matrix, and Qij are the
matrices involving integration over inhomogeneous spherical layers. To speed the II-TM
computation, the method of separation of variables (Bohren and Huffman 1983) can be
employed to compute the T-matrix of a sphere inscribed within the non-spherical particle
to reduce the number of iterations of Eq. (4.2.16); this equation can be derived from an
electromagnetic volume integral equation as follows (Johnson 1988, Bi et al. 2013):

) 2 *↔
E(r) = E (r) + k
inc 2
m (r) − 1 G(r − r ) · E(r )d 3 r , (4.2.17)
V
inc
where E is the total field, E is the incident electric field, m is the refractive index, V is

the particle volume, and G(r − r ) is the dyadic Green function.
Since Johnson’s work in 1988, numerical techniques and computing power have sig-
nificantly advanced. We have revisited the II-TM method for single-scattering computa-
tions. Based on a contemporary numerical implementation (Bi and Yang 2014), the II-TM
approach can apply to size parameters substantially larger than its EBCM-TM counter-
part. To validate the II-TM implementation, Figure 4.17 compares the scattering phase
4.2 T-matrix numerical method 229

105 1 1
II-TM
c IGOM -P12 /P11
a 0.5 P22 /P11
0.5
P11

m = 1.311 0
ka = 150, kc = 300
0
100 -0.5

-1 -0.5
0 45 90 135 180 0 45 90 135 180 0 45 90 135 180
Scattering Angle (deg.) Scattering Angle (deg.) Scattering Angle (deg.)

Figure 4.18 Comparison of phase matrix elements simulated using the II-TM method and using an
improved geometric-optics method (IGOM) (Yang and Liou 1996b, Yang et al. 2005) for randomly
oriented spheroids with a size parameter of 300 (after Bi et al. 2013).

matrices computed from II-TM and EBCM-TM computational codes for randomly ori-
ented spheroids with a size parameter of 95, specified in terms of the semi-major axis, and
an aspect ratio of 2. It is evident that the results for six scattering phase matrix independent
elements computed from these two numerical codes are indistinguishable.
The II-TM program was further used to compute randomly oriented spheroids with an
unprecedentedly large size parameter of 300. In this case, the EBCM-TM code developed
by Mishchenko and Travis (1998) was unable to achieve a convergent solution. Figure 4.18
shows a comparison of the results of three scattering phase matrix elements computed from
the II-TM technique with the results of an improved geometric-optics method (IGOM)
developed by Yang and Liou (1996a), which is primarily applicable to large size parameters
(see Subsection 3.3.3). The comparison depicted in Figure 4.18 illustrates the potential
capability of the II-TM technique for application to large non-spherical particles of various
shapes.
Bi and Yang (2014) further improved the II-TM method for computing the single-
scattering properties of a sample of randomly oriented particles such that the principle
of reciprocity can be applied. In this case, the scattering phase matrix comprises ten
independent parameters [see Eq. (2.4.55)]. To simplify the following presentation, we will
use the notation of van de Hulst (1957) for the presentation of scattering phase matrix
elements in the form

⎡ ⎤
a1 b1 b3 b5
⎢ b1 a2 b4 b6 ⎥
P =⎢
⎣ −b3
⎥. (4.2.18)
−b4 a3 b2 ⎦
b5 b6 −b2 a4

In terms of the circular-polarization representation, the scattering phase matrix for a


macroscopically isotropic and mirror-symmetric medium may be expressed in the form
230 Other useful approaches to light scattering

(Mishchenko et al. 2002)


⎡ ⎤
(a2 + a3 ) − 2ib4 (b1 + b6 ) + i(b2 − b3 ) (b1 − b6 ) − i(b2 + b3 ) a2 − a3
⎢(b + b ) + i(b − b ) a1 + a4 + 2b5 a1 − a4 (b1 + b6 ) − i(b2 − b3 )⎥
⎢ 1 6 2 3 ⎥
P = 1/2 ⎢
c
⎥.
⎣(b1 − b6 ) − i(b2 + b3 ) a 1 − a4 a1 + a4 − 2b5 (b1 − b6 ) + i(b2 + b3 )⎦
a2 − a3 (b1 + b6 ) − i(b2 − b3 ) (b1 − b6 ) + i(b2 + b3 ) (a2 + a3 ) + 2ib4
(4.2.19)

In Eqs. (4.2.18) and (4.2.19), the notations a1−4 and b1−6 should be distinguished from the
expansion coefficients defined in Eqs. (4.2.6) and (4.2.7).
Bi and Yang (2014) further showed that scattering phase matrix elements can be
expressed in the following forms:


a1 (θ ) = α1s d00
s
(cos θ ), (4.2.20a)
s=0


  s  s
a2 (θ ) = α2 + α3s d22 (cos θ ), (4.2.20b)
s=0


  s  s
a3 (θ ) = α2 − α3s d2−2 (cos θ ), (4.2.20c)
s=0



a4 (θ ) = α4s d00
s
(cos θ ), (4.2.20d)
s=0



b1 (θ ) = β1s d02
s
(cos θ ), (4.2.20e)
s=0



b2 (θ ) = β2s d02
s
(cos θ ), (4.2.20f)
s=0



b3 (θ ) = β3s d02
s
(cos θ ), (4.2.20g)
s=0



b4 (θ ) = β4s d22
s
(cos θ ), (4.2.20h)
s=0



b5 (θ ) = β5s d00
s
(cos θ ), (4.2.20i)
s=0



b6 (θ ) = β6s d20
s
(cos θ ), (4.2.20j)
s=0
s
where di,j (i, j = 0, 2, −2) denote the amplitude functions as functions of cos θ , and
4.2 T-matrix numerical method 231

θ (= ) is the scattering angle. The coefficients α i (i = 1–4) and β j (j = 1–6) are given by
α1 = (g0,0 + g−0,−0 + 2g0,−0 )/2, (4.2.21a)
α2 = Re{g2,2 } + g2,−2 , (4.2.21b)
α3 = Re{g2,2 } − g2,−2 , (4.2.21c)
α4 = (g0,0 + g−0,−0 − 2g0,−0 )/2, (4.2.21d)
β1 = Re{g0,2 } + Re{g−0,2 }, (4.2.21e)
β2 = Im{g0,2 } − Im{g−0,2 }, (4.2.21f)
β3 = −Im{g0,2 } − Im{g−0,2 }, (4.2.21g)
β4 = −Im{g2,2 }, (4.2.21h)
β5 = (g0,0 − g−0,−0 )/2, (4.2.21i)
β6 = Re{g0,2 } − Re{g−0,2 }. (4.2.21j)
The terms gi,j (i, j = 0, −0, 2, −2) are expressed in the forms

 
n+s 
min(n,i)
s
g00 = i1
hsni Cn1s0 im
Cnms0 00
Dmni , (4.2.22a)
n=1 i=max(1,|n−s|) −min(n,i)

 
n+s 
min(n,i)
s
g0−0 = hsni (−1)n+i+s Cn1s0
i1 im
Cnms0 0−0
Dmni , (4.2.22b)
n=1 i=max(1,|n−s|) −min(n,i)

 
n+s 
min(n,i)
−0−0
s
g−0−0 = i1
hsni Cn1s0 im
Cnms0 Dmni , (4.2.22c)
n=1 i=max(1,|n−s|) −min(n,i)

 
n+s 
min(n,i)
s
g22 = i1
hsni Cn−1s2 i2−m
Cn−ms2 22
Dmni , (4.2.22d)
n=1 i=max(1,|n−s|) −min(n,i)

 
n+s 
min(n,i)
s
g2−2 = hsni (−1)n+i+s Cn−1s2
i1 i2−m
Cn−ms2 2−2
Dmni , (4.2.22e)
n=1 i=max(1,|n−s|) −min(n,i)

 
n+s 
min(n,i)
−2−2
s
g−2−2 = i1
hsni Cn−1s2 i2−m
Cn−ms2 Dmni , (4.2.22f)
n=1 i=max(1,|n−s|) −min(n,i)

 
n+s 
min(n,i)
s
g02 =− i1
hsni Cn1s0 i2−m
Cn−ms2 02
Dmni , (4.2.22g)
n=1 i=max(1,|n−s|) −min(n,i)

 
n+s 
min(n,i)
−02
s
g−02 =− hsni (−1) n+i+s i1
Cn1s0 i2−m
Cn−ms2 Dmni , (4.2.22h)
n=1 i=max(1,|n−s|) −min(n,i)

where hsni = (2n + 1)/(2i + 1)(2s + 1)π/(k Csca ), k = 2πλ, and λ is the wavelength,
2

Csca  is defined in Eq. (4.2.10), Cnms1−m


i1
denote Clebsch–Gordan coefficients (Mishchenko
232 Other useful approaches to light scattering

et al. 2002). The D-terms are defined as



 
n1

00
Dmni = (2n1 + 1) B 3 kmnn1 (B 3 kmin1 ) , (4.2.23a)
n1 =|m−1| k=−n1


 
n1

0−0
Dmni = (2n1 + 1) B 2 kmnn1 (B 2 kmin1 ) , (4.2.23b)
n1 =|m−1| k=−n1


 
n1
−0−0 ∗
Dmni = (2n1 + 1) B 1 kmnn1 (B 1 kmin1 ) , (4.2.23c)
n1 =|m−1| k=−n1


 
n1

22
Dmni = (2n1 + 1) B 1 kmnn1 (B 3 −k2−min1 ) , (4.2.23d)
n1 =|m−1| k=−n1


 
n1

2−2
Dmni = (2n1 + 1) B 4 kmnn1 (B 2 −k2−min1 ) , (4.2.23e)
n1 =|m−1| k=−n1


 
n1
−2−2 ∗
Dmni = (2n1 + 1) B 3 kmnn1 (B 1 −k2−min1 ) , (4.2.23f)
n1 =|m−1| k=−n1


 
n1

02
Dmni = (2n1 + 1) B 2 kmnn1 (B 3 −k2−min1 ) , (4.2.23g)
n1 =|m−1| k=−n1


 
n1
−02 ∗
Dmni = (2n1 + 1) B 1 kmnn1 (B 4 −k2−min1 ) . (4.2.23h)
n1 =|m−1| k=−n1

j
The terms Bkmnn1 (j = 1, 2, 3, 4) can be expressed in the forms

j
1
n+n

n ,1 j
Bkmnn1 = Cn,m,n1 ,1−m
Aknn n1 , j = 1, 2, 3, 4, (4.2.24)
n =max(1,|n−n1 |)

where

j

i n −n  n
n ,m1 +k j
Aknn n1 = √ Cn,m T
1 ,n1 ,k m1 ,n,m1 +k,n
, (4.2.25a)

2n + 1 m1 =−n

Tm11 ,n,m1 +k,n = Tm111 ,n,m1 +k,n + Tm121 ,n,m1 +k,n + Tm211 ,n,m1 +k,n + Tm221 ,n,m1 +k,n , (4.2.25b)
Tm21 ,n,m1 +k,n = Tm111 ,n,m1 +k,n + Tm121 ,n,m1 +k,n − Tm211 ,n,m1 +k,n − Tm221 ,n,m1 +k,n , (4.2.25c)

Tm31 ,n,m1 +k,n = T−m


11
1 ,n,−m1 −k,n
 − T−m ,n,−m −k,n − T−m ,n,−m −k,n + T−m ,n,m −k,n ,
12
1 1
21
1 1
22
1 1

(4.2.25d)
Tm41 ,n,m1 +k,n = 11
T−m 1 ,n,−m1 −k,n
 − 12
T−m 1 ,n,−m1 −k,n
 + 21
T−m 1 ,n,−m1 −k,n
 − 22
T−m 1 ,n,m1 −k,n
.

(4.2.25e)
4.2 T-matrix numerical method 233

(a) (b)
3 4
10 10
II-TM II-TM
DDA IGOM
3 D
10
l = 0.66 μm
2
10
-8 L
m = 1.3078 - i 1.66 x 10
Phase Function

2πL /l = 2πD/l = 30 2
10 2πL/l = 2πD/l = 150
1
10
1
10

0
10
0
10

-1 -1
10 10
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Scattering Angle (deg.) Scattering Angle (deg.)

Figure 4.19 Comparison of phase functions computed from the II-TM and DDA methods for (a)
a short column with a size parameter of 30, and (b) the II-TM and IGOM methods for a large size
parameter of 150. Both calculations use a wavelength of 0.66 µm and a refractive index of m =
1.3078 − i 1.66 × 10−8 (after Bi and Yang 2014).

s s s s
In Eq. (4.2.22a)–(4.2.22h), the terms g00 , g0,−0 , g−0,−0 , and g2,−2 are generally real numbers,
s s s s
while the terms g2,2 , g−2,−2 , g0,2 , and g−0,2 are generally complex numbers.
In Figure 4.18, we illustrated the applicability of the II-TM method to the computation
of scattering phase matrix elements for a large size parameter of 300. To check the II-TM
method’s numerical implementation, we have compared the phase function computed from
II-TM and from DDA (Purcell and Pennypacker 1973, Draine and Flatau 1994, Yurkin
et al. 2007; also see the paragraphs below for further discussion) for a short column with
a size parameter of 30 using a wavelength of 0.66 µm, as shown in Figure 4.19a. The
phase functions computed from II-TM and a volume-integral based IGOM (Yang and Liou
1997, Bi et al. 2011b) for a short column with a size parameter of 150 are compared in
Figure 4.19b. It is evident that the II-TM result agrees well with the DDA counterpart for
a size parameter of 30, and with IGOM for a large size parameter of 150. In the latter case,
geometric-optics features of 22° and 46° halos and a maximum near the scattering angle of
150° are clearly produced by both approaches.
For randomly oriented particles without a plane of symmetry, 16 scattering phase matrix
elements contain ten independent parameters, as indicated previously. Shown in Figure 4.20
are ten scattering phase matrix elements of an aggregate of eight hexagonal columns with
different sizes and aspect ratios. Although each hexagonal column has a plane symmetry,
234 Other useful approaches to light scattering

II-TM ADDA

3
10 1 0.01
0.8 P22/P11 0.005
P
13
/P
11
2
10 0.6
Phase Function

l = 0.66 μm 0
0.4
10
1
D = 7 μm 0.2 -0.005

1 0.02
10
0 P /P P /P
0.5 33 11 14 11
0.01
-1 0
10
0
-0.5
-2 -0.8
10 -0.01
0.15 1
P /P P /P
44 11 0.01 23 11
0.1 0.5
0
0.05
0
-0.01
11

0.0
/P

-0.4
0.2
12

-0.05 P /P P /P
P

0.1 43 11 0.01 24 11
-0.1
0 0
-0.15 -0.1 -0.01
-0.2 -0.2 -0.02
0 45 90 135 180 0 45 90 135 180 0 45 90 135 180
Scattering Angle (deg.) Scattering Angle (deg.) Scattering Angle (deg.)

Figure 4.20 Comparison of ten independent elements of the scattering phase matrix for an aggregate
of eight columns, computed from the II-TM and ADDA methods (after Bi and Yang 2014).

aggregating processes lead to the destruction of symmetry, resulting in non-zero values


for P13 , P14 , P23 , and P24 (the third column in Figure 4.20). We also note that P13 P11 ,
P14 P11 , P23 P11 , and P24 P11 are relatively small. Based on a comparison between II-TM
and ADDA results in Figure 4.20, we find that the agreement between the two is much
closer for P11 , P22 , P33 , and P44 due to their relatively larger values than for the other six
non-zero elements.
The II-TM method can effectively apply to arbitrarily shaped non-spherical particles,
so particle shapes can be incorporated into the computational program without introducing
cumbersome procedures. For example, Figure 4.21 compares phase functions computed
from II-TM and ADDA for six ice crystal habits: droxtals, bullet rosettes, aggregates of
eight columns, hollow columns, hollow bullet rosettes, and aggregates of ten plates. The
wavelength of the incident light used in the calculations is 0.66 µm and the maximum
dimension of the ice crystals is 3 µm, resulting in a size parameter of 28.56.
Using the II-TM method, accuracies of the single-scattering properties of hexagonal
ice crystals with different sizes and aspect ratios can be investigated. The upper panels of
4.2 T-matrix numerical method 235

2 3 2
10 10 10

2
1 10 1
10 10
Phase Function

1
10
0 0
10 10
0
10
-1 -1
10 -1 10
ADDA 10
II-TM
-2 -2 -2
10 10 10
0 45 90 135 180 0 45 90 135 180 0 45 90 135 180
2 2 2
10 10 10

1 1
10 10
1
Phase Function

10

0 0
10 10

0
10
-1 -1
10 10

-2 -2 -1
10 10 10
0 45 90 135 180 0 45 90 135 180 0 45 90 135 180
Scattering Angle (deg.) Scattering Angle (deg.) Scattering Angle (deg.)

Figure 4.21 Comparison of the phase functions of six ice habits (droxtal, solid bullet rosette, aggre-
gate of eight columns, hollow column, hollow bullet rosette, and aggregate of ten plates) computed by
the ADDA and II-TM methods. The wavelength used was 0.66 µm and the particle size was defined
in terms of the maximum dimension of 3 µm (after Bi and Yang 2014).

Figure 4.22 shows the phase functions for hexagonal ice crystals with different sizes and
aspect ratios. For clarity, the phase functions of ice crystal of sizes 5 and 10 µm are shifted
by a factor of 10. To show the enhancement of ice halos when the ice crystal increases in
size, we define an indicator such that ξ = P11 (22◦ )/Pref (22◦ ), where the reference value
Pref (22◦ ) is calculated from interpolation of the phase function at scattering angles of 15°
and 40°. When the crystal size is small, the indicator fluctuates near unity, produced by
Fraunhofer diffraction. When the crystal size becomes larger than some threshold value,
the indicator is always larger than unity and increases with crystal size, indicating the
enhancement of ice halos associated with the refraction of light. As shown, halos can be
more easily observed for short hexagonal ice particles (DL = 1.0).
The II-TM approach can be employed to accurately model lidar polarimetric measure-
ments. As an example, Figure 4.23 shows the depolarization ratio of a linearly polarized
236 Other useful approaches to light scattering

6 6 6
10 10 10
D = 5 μm
D = 10 μm
D = 20 μm D
4 4 4
10 10 L 10

D/L = 0.5 D/L = 1.0 D/L = 2.0


P11

2 2 2
10 10 10

0 0 0
10 10 10

0 30 60 90 120 150 180 0 30 60 90 120 150 180 0 30 60 90 120 150 180


Scattering Angle (deg.) Scattering Angle (deg.) Scattering Angle (deg.)

4 4 4

3.5 3.5 3.5


P11(22o)/<P11,ref>

3 3 3

2.5 2.5 2.5

2 2 2

1.5 1.5 1.5

1 1 1

0.5 0.5 0.5

0 0 0
0 4 8 12 16 20 0 4 8 12 16 20 0 4 8 12 16 20
Size (μm) Size (μm) Size (μm)

Figure 4.22 Phase functions of hexagonal ice crystals with various sizes and aspect ratios (upper
panels) as a function of the scattering angle. The lower panels depict the enhancement of the 22° halo
using an indicator as a function of size, as discussed in the text.

light beam backscattered from randomly oriented hexagonal ice crystals with particle sizes
ranging from 0.1 to 20 µm. Three particle aspect ratios were selected for the simulations in
order to show the shape dependence of the depolarization ratios. The particles are assumed
to be randomly oriented in space. The incident wavelength was 0.532 µm, at which the
refractive index of ice is 1.3116 − i 1.48 × 10−9 (Warren and Brandt 2008). The depolar-
ization ratio was found to increase to a maximum, then decrease and become more stable
as the crystal size increased.
4.2 T-matrix numerical method 237

0.8

D/L = 0.5
D
0.6 1.0 L
Depolarization Ratio
2.0

0.4

0.2

0
0 5 10 15 20
Ice Crystal Size (μm)

Figure 4.23 Linear depolarization ratio of a hexagonal ice crystal with three different aspect ratios
at 0.532 µm, as a function of ice crystal size.

Figure 4.24 shows the asymmetry factor, the extinction efficiency, and the single-
scattering albedo computed from II-TM for randomly oriented hexagonal ice crystals with
size parameters (k1 L) ranging from 1 to 150. In the simulation, the aspect ratio of the hexag-
onal particle was assumed to be unity (i.e., L = 2a). As the size parameter increases, differ-
ences between the asymmetry factor computed from II-TM and from IGOM are smaller,
as IGOM becomes increasingly accurate. Because ice crystals are almost non-absorptive
at 0.66 µm, the interference between diffraction and forward scattering causes oscillation
of the extinction efficiency with respect to ice crystal size parameter. The behavior of the
extinction efficiency in terms of its asymptotic value and oscillation pattern computed from
IGOM was similar to T-matrix results. For absorptive particles, when the ice crystal size
parameter is larger than a certain value, internal reflection of geometric-optics waves can be
properly neglected. In this case, the contribution to scattering was essentially from external
reflection and diffraction. The asymptotic extinction efficiency can be explicitly proven to
be 2 in the framework of geometric optics (equal partition of geometric reflection/refraction
and diffraction; see Subsection 3.3.2).
To illustrate applicability of the II-TM method to extremely complicated particle geome-
tries, Figure 4.25 shows solutions for the phase function and the degree of linear polarization
of two types of oceanic particles, coccolith and coccolithophore, which are important in
the discussion of the global carbon cycle (Bi and Yang 2015). For comparison, the DDA
solutions are also shown. The size parameter defined in terms of the radius of a volume-
equivalent sphere is 7.23 for a coccolith and 34.25 for the coccolithophore. The refractive
index of coccolith relative to water is 1.20 and the relative refractive index of the spherical
238 Other useful approaches to light scattering

1 4
II-TM 1
0.8 IGOM
3 0.8
0.6 a 0.6
2
L
0.4 l = 0.66 μm 0.4
L = 2a 1
0.2 0.2

0 0
0
1 10 100 500 1 10 100 500

Single-scattering Albedo
1 4 0.8

Extinction Efficiency
Asymmetry Factor

0.8
3 0.6
0.6 l = 8.50 μm
2 0.4
0.4
1 0.2
0.2
0
0 0
1 10 100 500 1 10 100 500
1 2 1

0.8 0.8

0.6 l = 11.00 μm 0.6


1
0.4 0.4

0.2 0.2

0
0 01
1 10 100 500 10 100 500
Size Parameter k1L Size Parameter k1L

Figure 4.24 The asymmetry factor, the extinction efficiency, and the single-scattering albedo for a
short column as a function of size parameter, computed using the II-TM and IGOM methods for
wavelengths of 0.66, 8.50, and 11 µm (after Bi and Yang 2014).

coccolithophore cell is 1.03 − i 0.005. As shown in Figure 4.25, the results from two fun-
damentally different numerical methods agree closely with differences produced by shape
definition and orientation-average process of the DDA method. In the DDA orientation-
average procedure, 64 α, 65 β, and 32 γ angle values were used, where α, β, and γ are the
three Euler angles defined according to the x-y-z convention.

4.3 Discrete dipole approximation


4.3.1 Fundamentals of the DDA approach
The discrete dipole approximation (DDA) is a numerical technique for the calculation of
scattering and absorption properties of arbitrarily shaped particles. A particle is modeled
by an array of N-point dipoles at positions ri with polarizabilities α i . The method has been
widely employed for studying the optical properties of various particles, particularly those
4.3 Discrete dipole approximation 239

103 10
4

xv = 7.23 DDA xv = 34.25


II-TM
102
2
10
Phase Function

1
10
0
10
0
10

-2
-1
10
10

-2 -4
10 10
0 30 60 90 120 150 180 0 30 60 90 120 150 180
1 1
P12/P11

0 0

-1 -1
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Scattering Angle (deg.) Scattering Angle (deg.)

Figure 4.25 Comparison of phase function and P12 P11 for a sample of randomly oriented coccolith
(left column) and coccolithophore (right column) as a function of the scattering angle, computed
using the DDA and II-TM methods (courtesy of Lei Bi).

lacking rotational symmetry, having inhomogeneous chemical composition, or composed


of anisotropic material where an analytical or semi-analytical solution does not exist.
The DDA was originally introduced by Purcell and Pennypacker (1973) to study the
optical properties of interstellar dust grains. The basic idea of DDA appears to have been
developed by DeVoe (1964), who studied the optical properties of molecular aggregates.
Simply stated, the particle in this method is assumed to be a finite array of polarization
points (see the example of a hexagonal particle shown in Figure 4.26a), with each point
responding to the local electric field, the sum of the incident field and the field arising from
the other dipoles, as shown in Figure 4.26b. As a result of this arrangement, the amplitudes
of all dipole moments are self-consistent.
The electric dipole moment is proportional to the total electric field E(r) and the local
exciting field Eexc (r), defined as the superposition of the incident field and the fields
produced by other dipoles in the form

P(r) = χ (r)E(r) = αEexc (r), (4.3.1)


240 Other useful approaches to light scattering

(a) (b)

k E
0,k

j E
0, j

i E
0,i

Figure 4.26 (a) A schematic geometry to illustrate the approximation of a hexagonal aggregate in
terms of discretized dipoles. (b) An example of a three-dipole system (red dots) define by i, j, k indices
and associated incident electric vectors.

where the susceptibility χ (r) is given by

m2 − 1
χ (r) = d 3 (4.3.2)

and the polarizability α is related to the refractive index m through the Clausius–Mossotti
(or Lorentz–Lorenz) relation (Lorentz 1880, Lorenz 1880) as follows:
3 m2 − 1
α = d3 , (4.3.3)
4π m2 + 2
where d is the dipole length. The electric dipole field at the position of the jth dipole,
radiated by the ith electric dipole moment Pi , is given by (Jackson 1975):
% & % &
k 2 exp(ikR) R̂ R̂ 1 − ikR 3R̂ R̂
E(R = rj − ri ) = Aij · Pi = I− 2 − I− · Pi ,
R R (kR)2 (kR)2
(4.3.4)

where k is the wavenumber. It follows that the fundamental equation for the induced dipoles
is given by
% &

Pi = αi E0,i + Aij · Pj , (4.3.5)
i
=j

where E0,i is the incident field and Aij is defined in Eq. (4.3.4).
Once the amplitudes of dipole moments in Eq. (4.3.5) are solved, it can be shown
that the extinction and absorption cross-sections are given by the following equations
4.3 Discrete dipole approximation 241

[see Eqs. (4.1.38) and (4.1.39)]:




4πk   ∗ 
N
k ∗   3 
σext = Im (ε − 1) E0 (r ) · E(r ) d r = Im E0,j · Pj , (4.3.6a)
|E0 |2 V |E0 | j =1
2


4πk   ∗ 
N
k
σabs = Im(ε)E∗ (r ) · E(r )d 3 r  = Im Ej · Pj , (4.3.6b)
|E0 |2 V |E0 | j =1
2

where ε is the permittivity and Eq. (4.3.1) is used in these derivations. By neglecting higher-
order terms in Eq. (4.3.1) or Eq. (4.3.4), the scattered far field can be determined and is
given by
exp(ikr) 
Esca (r) |r→∞ = (−ik 3 )(Ī − r̂ r̂) · Pj exp(ikrj · r̂). (4.3.7)
−ikr j

To compute the scattering amplitude matrix and subsequently the phase matrix, the
scattered field must be decomposed into components parallel and perpendicular to the
scattering plane:

Esca (r) = α̂Eα + β̂Eβ . (4.3.8)

In matrix form, we have





Eα exp(ikr) E0,α
= S , (4.3.9)
Eβ −ikr E0,β
where



fx,α fy,α β̂ · êx β̂ · êy
S= , (4.3.10)
fx,β fy,β−β̂ · êy β̂ · êx


Pj · α̂
fy,α
= −ik 3
exp(ikrj · r̂)|E0,x =0,E0,y =1 , (4.3.11)
fy,β Pj · β̂
j =1


Pj · α
fx,α
= −ik 3
exp(ikrj · r̂)|E0,x =1,E0,y =0 . (4.3.12)
fx,β Pj · β
j =1

From Eqs. (4.3.11) and (4.3.12), we note that Eq. (4.3.1) must be solved twice, correspond-
ing to two cases: (a) E0,y = 1, E0,x = 0 and (b) E0,y = 0, E0,x = 1.
Equation (4.3.2), which was established by Purcell and Pennypacker (1973), is to a
certain degree based on physical rationale. In later developments, DDA was rigorously
refined on the basis of mathematical analysis starting from an electric volume integral
equation in the form

E(r) = E0 (r) + k 2 (ε − 1)Ḡ(r − r ) · E(r)dv, (4.3.13)
V
242 Other useful approaches to light scattering

where E0 is the incident field and Ḡ is the dyadic Green function satisfying the following
differential equation:

∇ × ∇ × G(r − r ) − k 2 G(r − r ) = Īδ(r − r ), (4.3.14)

where δ(r − r ) is the Dirac delta function. Direct computation of the integral in Eq. (4.3.13)
is problematic because of the inherent singularity problem associated with the dyadic Green
function, given by

1 exp(ikR) Ā(R = r − r )
Ḡ(R = r − r ) = [k 2
Ī + ∇∇] = , (4.3.15)
4πk 2 R 4πk 2

which contains a 1/R 3 dependence term in the form




1 Ī 3
Ḡs (R = r − r ) = − + R̂ R̂ . (4.3.16)
4πk 2 R3 R3

To treat the singularity problem for appropriate numerical computations, the volumetric
equation for R → 0 can be transformed to a surface integration equation. Based on the
assumption that the electric field is constant within a small volume, Eq. (4.3.13) can then
be rewritten for numerical calculations as follows:

E(r) ≈ E0 (r) + k 2 (ε − 1)Ḡ(r − r ) · E(r )dv + (ε − 1)[M̄(r) + L̄(r)] · E(r),
V −Vε
(4.3.17)

where Vε is an elementary volume and M̄(r) and L̄(r) are integrals defined by
⎧ ⎫
⎨ ⎬
M̄(r) = k 2 [Ḡ(r − r ) − Ḡs (r − r )]d 3 r  , (4.3.18)
⎩ ⎭

⎧ ⎫
⎨ ⎬
L̄(r) = k 2 Ḡs (r − r )d 3 r  . (4.3.19)
⎩ ⎭

From the divergence theorem (Arfken 1985), Eq. (4.3.19) is equivalent to



¯  
1 I 3 1 R̂ 1 n̂R̂
L̄ = − + R̂ R̂ dvε = − ∇ dvε = − ds. (4.3.20)
4π R3 R3 4π R2 4π s R 2
Vε Vε

If the differential volume is a cube or sphere, it can be proven that L̄ = −I/3 (Yaghjian
1980). Based on Eq. (4.3.17), the source of the field (or the exciting field) at the position
r, i.e., the contribution from the incident field and the scattered field from the remaining
4.3 Discrete dipole approximation 243

volume, is given by

Eexc (r) = [Ī − (ε − 1)(M̄ + L̄)] · E(r). (4.3.21)

With Eq. (4.3.1), the polarizability can now be defined in the form
V (ε − 1)
α= . (4.3.22)
4π[Ī − (ε − 1)(M̄ + L̄)]
Strong and weak forms of DDA have been proposed by Lakhtakia (1992) to distinguish
the treatment of the singularity problem associated with the dyadic Green function depicted
in Eq. (4.3.15). The weak-form DDA corresponds to the Clausius–Mossotti relation (i.e.,
M̄ = 0̄), while the strong-form DDA indicates non-zero M̄ stemming from the finite ele-
mentary volume used in the calculation. In the case of the strong-form DDA, several
polarizability relations (see Yurkin et al. 2007) are formulated instead of the well-known
Clausius–Mossotti polarizability based on different approximations to calculate M̄. A few
polarizability relations based on different mathematical analysis have also been discussed
by Yurkin et al. (2007). A number of DDA implementations have recently been presented
for the solution of light scattering by dielectric particles, such as the DDSCAT code devel-
oped by Draine (1988) and Draine and Flatau (1994) and the ADDA method developed by
Yurkin and Hoekstra (2011).
An apparent advantage of DDA is that it is flexible in handling the complex geometry of
a particle because it is not necessary to impose electromagnetic boundary conditions at the
particle surface. The major limitation is associated with size parameter, which determines
the number of dipoles and the associated computer memory requirement. With state-of-
the-art computers, DDA can efficiently obtain scattering and absorption properties when
particle size parameters are smaller than 20. For large size parameters, computational
resources such as memory and the number of processors increase dramatically with particle
size parameter. However, with the development of computer clusters, the light scattering
properties of large size parameters can be obtained by the use of parallel computation. For
example, the single-scattering properties of a sphere with a size parameter of 320 and a
refractive index of 1.05 have been obtained using DDA with 512 processors and 698 Gb of
memory (Yurkin and Hoekstra 2011).
Due to its inherent errors in defining shape and the nature of approximating polarizability
relations, DDA can be viewed as an approximate or “semi-exact” method. The accuracy of
DDA has been examined by comparison of its results with those computed from analytical or
semi-analytical methods such as the Lorenz–Mie theory for spheres and the EBCM numer-
ical method for spheroids and circular cylinders. Figure 4.27 compares the phase function
of a sphere computed from the Lorenz–Mie theory and from DDA in which the ADDA
code was employed in the calculation. Comparisons of the phase functions of spheroids and
cylinders computed by T-matrix and DDA methods are also given in this figure. The size
parameter, defined in terms of the semi-maximum dimension, was 20, while the refractive
index used was 1.33. Relative errors are displayed in the panels below the corresponding
phase functions to illustrate the accuracy of DDA results.
244 Other useful approaches to light scattering

4 2 3
10 10 10
Lorenz-Mie T-matrix T-matrix
ADDA ADDA ADDA
2 1 2
Phase Function

10 20 10 20 10
20
10 10

0 0 1
10 10 10

-2 -1 0
10 10 10

-4 -2 -1
10 10 10
0 45 90 135 180 0 45 90 135 180 0 45 90 135 180
0.5 0.15 0.15
Relative Error

0.1 0.1
0.25
0.05 0.05
0
0.0 0.0

-0.25 -0.05 -0.05


0 45 90 135 180 0 45 90 135 180 0 45 90 135 180
Scattering Angle (deg.) Scattering Angle (deg.) Scattering Angle (deg.)

Figure 4.27 Comparison of the phase function of a sphere computed from the Lorenz–Mie theory
and from ADDA, and of the phase functions of a spheroid and a circular cylinder computed from
T-matrix and from ADDA. In ADDA simulations, the number of dipoles per wavelength used was
13.8203.

4.3.2 Comparison among DDA, FDTD, and PSTD


We may carry out a comparison between the performances of DDA and FDTD because
these two numerical methods have similar applicability with reference to DDA in terms of
shape-independence and comparable computer memory requirements. Such a comparison
study can provide the best approach in selecting the optimized numerical method for
practical calculations. Yurkin et al. (2007) reported a comparison between DDA and FDTD
for the case of spheres with indices of refraction ranging from 1.02 to 2 and size parameters
up to 80. The comparison exercise revealed that DDA appears to be faster than FDTD when
the refractive index was smaller than 1.4.
The PSTD technique, discussed in Subsection 4.1.6, is more advantageous than FDTD
because the coarse spatial resolution can be used to discretize particle domain. Liu et al.
(2012a) reported a comparison between the numerical performance of the DDA and PSTD
techniques and found DDA to be faster than PSTD when the indices of refraction were
smaller than a certain value for a fixed size parameter (Figure 4.28). As shown, PSTD is
more efficient for particles with large indices of refraction, whereas DDA outperforms for
cases involving small indices of refraction. Furthermore, we notice that in the literature
DDA has seldom been employed in calculations when the refractive index is larger than
2 because of slow convergence or failure to converge. Some progress in extending the
application of DDA to indices of refraction larger than 2 has been made by Yurkin et al.
(2010). However, the FDTD/PSTD method has no limitation concerning the use of the
4.3 Discrete dipole approximation 245

2.0

1.8

PSTD
Refractive Index m

1.6

1.4

DDA
1.2

0 20 40 60 80 100
Size Parameter x

Figure 4.28 The relative performance of PSTD and DDA on the (x, m) domain. The green area
indicates that DDA is computationally more efficient than PSTD, whereas the blue region shows that
PSTD outperforms DDA (after Liu et al. 2012a).

refractive index in numerical calculations, although a fine grid resolution is required for
large indices of refraction.
Examples of DDA computations for the six phase matrix elements of hexagonal particles
are presented in Figure 4.29. The refractive index used was 1.33, along with a size parameter
of 50 (defined in terms of height), while the aspect ratio was kept at unity. We have compared
DDA computations with their counterparts simulated by the PSTD method and the efficient
geometric-optics hybrid (EGOH) method developed by Bi et al. (2011b), which is an
improved form of the ray-by-ray integration (RBRI) technique (Yang and Liou 1997).
In this method, beams with polygon-shaped cross-sections are traced within a scattering
particle, while the near field associated with the beams is analytically mapped to their
far-field counterparts based on a rigorous volume-integral electromagnetic relation. It is
computationally more efficient than RBRI and can represent a “benchmark” within the
framework of geometric optics. The two numerical methods (DDA and PSTD) produce
very close results. The approximate results determined from EGOH demonstrate similar
results for phase matrix elements as compared with those obtained from DDA and PSTD
numerical methods.
246 Other useful approaches to light scattering

4
10 0.2
ADDA
PSTD 0.1
2 EGOH
10

P12 /P11
0
P11

0 -0.1
10
-0.2
-2
10 -0.3
0 45 90 135 180 0 45 90 135 180
1 1

0.8
0.5

P33 /P11
P22 /P11

0.6
0
0.4
-0.5
0.2

0 -1
0 45 90 135 180 0 45 90 135 180
1
0.4

0.5
P44 /P11

P43 /P11

0.2

0 0

-0.5 -0.2
0 45 90 135 180 0 45 90 135 180
Scattering Angle (deg.) Scattering Angle (deg.)

Figure 4.29 Comparison of phase matrix elements for randomly oriented hexagonal ice columns as
a function of the scattering angle, computed using ADDA, PSTD, and EGOH methods.

In summary, various combinations of DDA, FDTD, PSTD, and EGOH methods can be
employed in principle to study the single-scattering properties, particularly the scattering
phase matrix, of non-spherical particles covering the size parameter range from Rayleigh
to geometric-optics domains.
5
Application of light scattering by ice crystals
to remote sensing

Remote sensing is a big interdisciplinary field involving science, engineering, and tech-
nology. However, we shall confine our presentations to the book’s theme, namely light
scattering by ice crystals. The subject of the remote sensing of ice clouds in the Earth’s
atmosphere and associated supporting material will be the focus of this chapter.
Remote sensing is differentiated from in situ measurements in that specific observations
are made outside of the medium. It involves the interpretation and inversion of radiation
signals measured some distance away, which are characterized by specific wavelengths
sensitive to some physical properties of the medium. In the case of ice clouds, the interpre-
tation and inversion processes require the use of fundamental light scattering and radiative
transfer theories. The general principle of the inversion theory is that the pieces of informa-
tion content that can be recovered from a target, such as clouds, cannot be larger than the
number of channels (wavelengths) employed for remote sensing. Simply put, initial efforts
must be greater than expectations.
Remote sensing can use the radiation sources readily available in nature – radiation
emitted from the sun and from the Earth–atmosphere system; this is referred to as passive
remote sensing. Many sensors in current satellites carry spectral radiometers operating
at visible and near-IR wavelengths to detect clouds and aerosols and make inferences
about their physical and radiative properties based on the sunlight reflected from the
cloud and aerosol particles. Because air (N2 and O2 ), trace gases, aerosols, and clouds
coexist in the atmosphere, along with reflecting surfaces, direct inversions from the basic
radiative transfer principle are not feasible and many numerical procedures, referred to as
algorithms, are required for the determination of these properties from suitable radiometers
or spectrometers. Independent validations become essential to the success of space remote
sensing.
Active remote sensing uses radiation generated by artificial means such as lasers and
microwaves, leading to the development of lidar and radar systems. This type of remote
sensing is primarily concerned with backscattering involving a system of collimated trans-
mitter and detector. In recent years, significant advances in the development of high-energy
lasers and mm-wave sources have led to the successful deployment of optical lidar and
94 GHz systems in space. These systems are collecting important information on clouds in
general and high clouds in particular based on the backscattering principle.

247
248 Application of light scattering to remote sensing

In this chapter, we introduce the remote sensing of ice clouds in the Earth’s atmosphere,
including discussions on atmospheric composition and structure, atmospheric absorption
spectra, the sun–satellite geometric configuration and radiative transfer, the bidirectional
reflectance spectra of ice clouds, and the A-Train satellite constellation. From this base, we
then present the subject of retrieving the optical depth and ice crystal size of cirrus clouds
using reflected visible and near-IR radiation, and illustrate the importance of the phase
function (scattering pattern) of ice crystals. A discussion follows on detecting thin cirrus
and vertical sizing of cirrus cloud layers. We next tackle the subjects of remote sensing of
ice clouds using reflected polarization and the principle of backscattering depolarization
based on geometric optics to differentiate non-spherical ice crystals and spherical water
droplets. Lastly, we present reflected line spectra at the 1.38 µm band and the oxygen
A-band for the potential determination of the composition and optical properties of high
clouds.

5.1 Introduction to remote sensing of ice clouds in the Earth’s atmosphere


5.1.1 Atmospheric composition and structure
Before proceeding to a discussion about the remote sensing of ice clouds imbedded in
the Earth’s atmosphere, we shall discuss atmospheric composition and thermal structure.
The present atmosphere of the Earth is composed of two groups of gases, one with nearly
permanent concentrations and the other with variable concentrations. Permanent gases such
as nitrogen molecules (N2 , 78.084%), oxygen molecules (O2 , 20.948%), and argon (Ar,
0.934%) account for more than 99.96% of the atmosphere by volume, and have virtually
constant volume ratios up to an altitude of 60 km.
Another permanent constituent, carbon dioxide (CO2 ), with a concentration of
400 ppm in 2014, has been increasing by about 0.4% per year as a result of fossil
fuel combustion, absorption and release of CO2 by the oceans, and photosynthesis. Obser-
vations of the atmospheric methane (CH4 ) concentration, with a present value of 1.7 ppm,
indicate that this gas is increasing by 1–2% per year and may have been increasing for
a long period of time due to greater biogenic emissions associated with rising human pop-
ulations and rice paddies. Changes in carbon monoxide (CO) concentration appear to be
closely associated with deforestation, biomass burning, and modification of CH4 sources.
There is also evidence of an increase in nitrous oxide (N2 O) concentrations by 0.2% per
year, attributable to fossil fuel combustion and fertilizer denitrification. CO2 , CH4 , and N2 O
have been identified as key greenhouse gases and man-made perturbations dating from the
era of the Industrial Revolution. These gases play an essential role in global warming and
climate change.
Water vapor (H2 O), a highly variable gas, is the major radiative and dynamic element in
the Earth’s atmosphere. Its concentration varies significantly with both space and time and is
governed by the local hydrological cycle via evaporation, condensation, and precipitation,
as well as by large-scale transport processes. H2 O concentrations decrease rapidly with
5.1 Remote sensing of ice clouds in the atmosphere 249

pressure and, to a lesser degree, with latitude. More than 50% of H2 O is concentrated below
850 hPa, while more than 90% is confined to the layers below 500 hPa. Variability
of the vertical H2 O concentration shows a bimodal distribution with a maximum in the
subtropics of both hemispheres below 700 hPa, and is very small in the equatorial region
and poleward of 60°. The H2 O concentration is relatively small, 3–4 ppm, in the lower
stratosphere, and is controlled by the tropical tropopause temperature and the formation
and dissipation of cirrus anvils due to outflows from cumulonimbus clouds in the tropics.
Ozone (O3 ) concentrations also vary significantly with space and time, and are detected
principally at altitudes from 15 to 30 km in an area known as the ozone layer. The
maximum ozone concentration occurs at 20–25 km, depending on latitude and season.
O3 is created by three-body collisions involving O2 , O, and a third body (N2 or O2 ) and
governed by numerous photochemical processes associated with solar ultraviolet radiation.
The absorption of deadly solar ultraviolet radiation by the ozone layer is essential to the
survival of life on Earth. Many photochemical reactions associated with O3 involve H2 O,
CH4 , and CO. The total O3 concentration is a function of latitude and season, with maximum
concentrations occurring during the polar night.
Nitrogen oxides (NOx ; NO or NO2 ) are important in the determination of both tropo-
spheric and stratospheric O3 concentrations. Atmospheric NOx are emitted by transportation
and combustion processes at the surface and by high-flying aircraft in the upper troposphere
and lower stratosphere. In the stratosphere, the major source of NOx is the dissociation
of N2 O via an excited oxygen atom. In the lower atmosphere, the major source of NOx
is the anthropogenic combustion of fossil fuels and biomass burning. Chlorofluorocarbons
(CFCs) have also been recognized as presenting a potential threat to the ozone layer. Large
amounts of these chemicals are produced by industry and are used in solvents, refrigerants,
and spray-can propellants. Principal chlorofluorocarbons include CFCl3 (CFC-11), CF2 Cl2
(CFC-12), and CF3 Cl (CFC-13). Sulfur dioxide (SO2 ) in the stratosphere is largely pro-
duced by volcanic eruptions; this along with other sulfur-based gases are primary precursors
of stratospheric aerosols. Emissions of SO2 from the surface are important in the formation
of tropospheric aerosols.
The atmosphere also contains various kinds of aerosols, clouds, and precipitation, which
are highly variable in space and time. In Sections 1.1 and 1.2, we introduced the vertical
and horizontal structures of clouds in general and cirrus clouds in particular. Size and shape
distributions of ice crystals were also presented. Fortunate clouds can produce precipitation
in the form of raindrops, various types of snowflakes, and hailstones with sizes on the
order of mm or cm. We further pointed out that the key to trigger the formation of clouds
in moist atmospheres is the existence of condensation and ice nuclei, which are a part
of ubiquitous aerosol particles ranging in size from 10−3 to 20 µm. These aerosols
are known to be produced by natural processes as well as by human activity. Natural
aerosols include volcanic dust, smoke from forest fires, particles from sea spray, wind-
blown dust, and small particles produced by the chemical reactions of natural gases. Primary
man-made aerosols include particles directly emitted during combustion processes and
particles formed from gases emitted during combustion. Aerosol concentration varies with
250 Application of light scattering to remote sensing

locality; the largest concentrations generally occur in urban and desert areas. Under normal
conditions, the background aerosol conce