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 threedimensional 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.
KUONA 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
www.cambridge.org
Information on this title: www.cambridge.org/9780521889162
C KuoNan Liou and Ping Yang 2016
vii
viii Contents
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 highlevel 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 fouryear 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
nonspherical 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 manmade 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
threedimensional 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 twodimensional cirrus cloud model to illustrate interactions of winds, ice
microphysics, and radiative transfer.
xii
Preface xiii
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 Aband 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 deltatwo
stream and deltafourstream approximations for efficient radiative flux transfer in non
homogeneous planeparallel 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 onedimensional 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 highflying
aircraft, a manmade 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 coauthor is grateful to KuoNan Liou, Warren Wiscombe, George Kattawar, and
Jerry North for longterm 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 parttime 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 threedimensional 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), SpringerVerlag (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
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.
∗
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
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 largescale 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 fairweather 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.cloudsonline
.com. Copyright information: http://www.cloudsonline.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 fairweather 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 highlevel clouds before dissipating.
Upperlevel 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 icesupersaturated 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 NOAA14 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 limbviewing 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
motherofpearl (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.
(b)
Figure 1.3 (a) An unusual spiral contrail, formed by a circling aircraft, surrounded by high clouds,
observed in a NOAA14 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.
Highlevel 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 spiralcommashaped 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 largescale rising motion are present under much of
this highcloud 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 northnorthwestward from an occluded front
anchored in an intense low centered at (57°S, 140°W). In addition, some spiralshaped
cirrus and middle clouds are shown in the vicinity of (31°S, 104°W); these are associated
with a lowpressure 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 northnorthwestward 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 welldefined 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 48hour period between Figures 1.4a and 1.4b, due to a largescale 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 largescale weather features and disturbances. In the tropics, these phenomena are
related to deepcumulus 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.
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
Top View
24 km
12 km
6 km
North
24 km
Side View
Figure 1.6 Threedimensional observations of cirrus clouds in the vertical direction and in a hori
zontal plane, from lidar backscattering (after E. Eloranta, personal communication 1995).
10
30
8
40
6 50
36.6 NO 1
2
98oW 97.5oW 97oW
(d) Pparameter results
36.6ON Cirrus
Clear
98oW 97.5oW 97oW
Figure 1.7 A case study from the SGPARM site on February 11, 2001 (1730 UTC). Plotted in (a)
is the 1hour mmwave 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 Pparameter 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
ATrain, discussed in Subsection 5.1.5.
Because water vapor concentration decreases rapidly with height, mmwave radar gen
erally operates at 35 GHz. Figure 1.7a shows an example of the 35 GHz mmwave radar
reflectivity time series with a onehour 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
twolayer 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 Pparameter
method, developed in the paper (Figure 1.7d).
14 Ice in the Earth’s atmosphere
(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
MODISCALIOP 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 crosssectional
elements are plotted (after Iwabuchi et al. 2012).
16 Ice in the Earth’s atmosphere
(a) Vertical Feature Mask UTC: 20070331 04:56:26.4 to 20070331 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
Latitude (oN)
Longitude (oE)
(d) τ3 (Land, DB) (e) re
Latitude (oN)
Latitude (oN)
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).
BTD1112 values are around 1 K, indicating mostly water clouds. The distributions of cirrus
and water clouds are consistent with cloudphase 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 nearIR channels and are displayed in Figure 1.9e.
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 caxis. The unit cell of the average
structure is marked ABCDEFGH, while the four oxygen atoms contained within this cell are shaded.
(b) A twodimensional 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 chairform hexameters, similar to
sheets lying on top of one another.
The threedimensional structure of ice Ih is complicated. Thus, a twodimensional
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 caxis, 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.
(a) Heterogeneous
Deposition
(b) Condensation
Followed by 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 submicronsized 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.
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 twodimensional 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 timedependent secondorder diffusion equation nt = D2 n, where
D is the molecular diffusion coefficient. Subject to the steadystate 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 nonspherical 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 platetype
ice crystals, C = 2rπ. Ice needles may be approximated by the formula for a prolate
spheroid of major and minor semiaxes 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
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 molecularkinetic 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 sixfold 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 caxis 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
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.
Flowers of plants and trees are generally fivepointed, but those of snow, which are called “Ying,”
are always sixpointed.
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 sixthcentury 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 snowflakes show forth their sixpetalled flowers.
There was no crystallographic explanation of the hexagonal nature of the crystals; however,
Chinese writers and poets attempted to define sixsided structure based on the principle
of the YinYang and WuShing (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 twelfthcentury 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 sixpointed.
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 sixpointed is because they are only sleet split open
by violent winds (and sleet being halffrozen rain, i.e. water) they must be sixpointed. Just so, if you
throw a lump of mud on the ground, it splashes into radiating angular petallike form. Now six is a
Yin number, and thaiyinhsuanchingshih (selenite crystal) is also sixpointed, 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 sixpointed 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 starshaped 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
closepacking 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.
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.
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).
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 quasispherical 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 Computergenerated ice crystal shapes: (a) quasispherical 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.
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 caxis; (c) tilt of the caxis from the zaxis by an angle θʹ in the xz plane; and
(d) rotation of the caxis about the zaxis by an angle φʹ.
where a and c are the lengths of the semiaxes, as shown in Figure 1.19a. The semiaxis
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 halfheight 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
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 caxis is parallel to the zaxis 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
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 caxis
as in Figure 1.20b, (2) tilting of the caxis from the zaxis by an angle θ ʹ in the xz plane
as in Figure 1.20c, and (3) rotation of the caxis about the zaxis 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 sixpointed 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 3D Koch snowflake contains two flat basal planes and 3 × 4n highly irregular
side planes (with 6sided 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
(c) (d) l3
l1
L
l2
bb a1 a2
O a3
O
O O
bt
a
Figure 1.22 Side and top views of (a) a 3D 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 2D area is given by
A0 = ( 34)a2 . The area generated by n iterations is then given by
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 caxis is along the zaxis. 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 fourbullet
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.
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 2DC and 2DP 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 2D 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 cloudtop (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 highflying 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 lookup 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 MixedPhase
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).
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 lightscattering 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 lightscattering 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 caxis.
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 twodimensional 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 highlevel 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 modelgenerated 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 singlecolumn model.
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 crosssectional
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 lightscattering 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 millimetersized 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 CRYSTALFACE 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 2DC 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 CRYSTALFACE
datasets were comprised of ice crystal measurements averaged over 30second intervals.
Due to the low sampling rate of CRYSTALFACE data, size distributions derived from
this experiment were averaged over 5 minutes. After selecting datasets having more than
fivechannel size measurements to ensure proper size average, only 261 CRYSTALFACE
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)
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 CRYSTALFACE (7/2002)
CEPEX (34/1993)
Best Fit
0
10
106 105 104 103 102 101 100 101
Figure 1.24c displays 11 293 data points in a 2D 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:
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 2D
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 2DC probe on board the UND Citation. Aircraft datasets were composed
of measurements averaged over a 5second interval. We selected size distributions that
had measurements greater than 5size 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 2DC measurements. The data points were averaged over
a 30second 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 highlatitude 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.
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 indepth 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 onedimensional
(1D) model, in which the vertical velocity is prescribed. The 1D 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 semimajor axes and the aspect ratio to define
hydrometers was developed by Chen et al. (1997) in a 1D model to study cirrus clouds.
Employing a 1D 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 1D
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 twodimensional (2D) 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
2D cirrus model that includes a secondorder 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 2D models for the simulation of ice
crystal size distributions.
The modeling of anvil cirrus clouds has normally been connected with 3D cumulonim
bus simulations. Tao and Moncrieff (2009) have undertaken a comprehensive review of the
development of the cloud resolving model (CRM) from 1D, 2D, and 3D perspectives
over the last four decades. The 3D 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 2D 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
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 righthand 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.
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
welldefined function of temperature. However, Meyers et al. (1992) found that Fletcher’s
equation tends to underpredict ice crystal concentrations for temperatures warmer than
−20°C, while overpredicting 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 nonspherical 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).
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 crosssection 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 crosssection 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 m3)
Figure 1.25 Vertical profile of the horizontally averaged IWC simulated from a 2D 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 crosssection 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 2D 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
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 highenergy 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 singlescattering 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 nonabsorbing 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
∇ · 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
(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.
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
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:
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 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 righthand 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
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).
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
d 2
+ 2 = 0. (2.2.39)
dφ 2
The singlevalue solution of Eq. (2.2.39), a secondorder differential equation, is given by
where a and b are arbitrary constants. Furthermore, Eq. (2.2.38) is the wellknown
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
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 halfintegralorder 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 wellknown 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
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:
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
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 zaxis is chosen as the axis of revolution. The incident plane contains the incident
direction and the zaxis. The xaxis 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 zaxis (after Asano and Yamamoto 1975).
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 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 nthorder 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 singlescattering 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, singlescattering
albedo, and asymmetry factor.
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 secondorder 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 halfwidth 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 halfwidth αN
is referred to as the normal or Lorentz profile (Lorentz 1906).
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 halfwidth, 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 caxis. 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
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 caxis of the crystal and the propagation direction of
the waves. As shown in Takano and Liou (1989a), the effect of ice birefringence on the
singlescattering 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.
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
102 101 100 101 102 103 104 105 106 107
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.
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
(b)
Spectral ice indices of refraction are fundamental parameters that determine the relative
scattering and absorption properties of ice particles due to a singlescattering event for
a given wavelength and ice crystal size. If there is no absorption, realpart 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.
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
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
It follows that
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
r
(al cosδ,ar)
(al,ar cosδ )
2ar l
(al,ar cosδ)
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.
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 righthand circular polarization are defined
according to the sign of sin δ.
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 semimajor and semiminor 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 lefthanded polarization, respectively. See also Figure 2.11.
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
Moreover, substituting Eqs. (2.4.17a)–(2.4.17d) into the expression for U = 2al ar cos δ,
we derive the following equations:
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)
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.
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.
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
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
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)
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, quarterwave plate, and
analyzer, respectively; and S denotes the center of scattering.
the modulator uses the linear electrooptic 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
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 singlescattering (light scattered just once) properties of
ice crystals.
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
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)
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
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 nonzero elements as follows:
⎡ ⎤
P11 P12 P13 P14
⎢ P21 P22 P23 P24 ⎥
P(general) = ⎢
⎣ P31 P32 P33 P34 ⎦ .
⎥ (2.4.38)
P41 P42 P43 P44
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 crosssection in an arbitrarily
oriented nonspherical 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 farfield 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 crosssectional 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
It follows that the extinction crosssection 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 crosssection 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 crosssections for a non
spherical particle. For incident natural light, the extinction crosssection is the average of
the two linear polarization components:
σe = (σe,1 + σe,2 )/2. (2.4.45c)
For a group of randomly oriented nonspherical 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)
2π
σe ∼
= 2 Re[S1 (0) + S2 (0)]. (2.4.45d)
k
at maximum. Thus, we should have DL cos α cos(π/6 − β). The geometric crosssection
for the entire hexagonal crystal is the sum of the preceding mapped crosssections:
√
3 3 2
G(α, β) = D sin α + DL cos α cos(π/6 − β). (2.4.48)
8
Defining the average crosssection 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 crosssection,
which is twice the geometric crosssection, is related to the surface area by σe = 2Ḡ = S/2.
Thus, the average geometric crosssection for randomly oriented plate/column crystals Ḡ
is onefourth 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
where
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)
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 lightscattering 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) crosssection 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 nonspherical 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 .
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
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, singlescattering 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 gfactor is the first moment
of the phase function and represents the strength of forward scattering. The extinction
efficiency, singlescattering 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 singlescattering 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 gfactor 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 gfactor for
2.4 Definition of scattering properties for ice crystals 103
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, singlescattering 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.
0.8
(a) (b)
1x106
Cs (re = 20 μm) 0.6
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
crosspolarized to the total scattered intensity. The depolarization ratio for total intensity is
defined as = 1 − P22 P11 , which is a measure of nonsphericity 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
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 singlescattering 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
planeparallel 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).
(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.
where ρ is the density of the cloud and k denotes the mass extinction (scattering +
absorption) crosssection (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 selfemission 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
crosssection. Combining the two radiation sources defined in Eqs. (2.5.1) and (2.5.2), we
obtain
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 planeparallel 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 xaxis.
For convenience of analysis, we introduce the nondimensional 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)
dτ
where µ = cos θ . This is the basic equation for radiative transfer in planeparallel 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 planeparallel
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 )],
4π
where the second term on the righthand side represents multiple scattering contribu
tions from all directions (4π solid angle), where the singlescattering albedo has been
defined in Eq. (2.4.60a); the phase function P (μ, φ; μ , φ ) represents the normalized 3D
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)
dτ
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 singlescattering albedo for a cloud is made. The
transformation matrix for the Stokes vector (I, Q, U, V ) defined on the xy plane relative
to that on the lr 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
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)
4π
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 ,
4π
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 nonspherical 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 nonspherical 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.
The extinction and scattering crosssections, however, depend only on the direction of the
incident beam and the ice particle orientation.
For a sample of nonspherical ice particles randomly oriented in space, the average
singlescattering 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 crosssections for randomly oriented nonspherical particles
are independent of direction. If all nonspherical 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
nonspherical 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 diameterto
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 threedimensional 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 singlescattering parameters are
dependent on the direction of the incident light beam. Thus, the conventional formulation
for the multiplescattering 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 crosssections 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 singlescattering 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 nonspherical 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 singlescattering 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 singlescattering and optically thin approximations, and omit
the emission term. The phase function and singlescattering 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 solarzenith angledependent 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 nondimensional reflected light beam that has been
scattered for horizontally oriented (2D) and randomly oriented (3D) 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 Nondimensional reflected light at a wavelength of 0.63 µm for ice columns randomly
oriented in space (3D) and oriented parallel to the ground (2D 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 3D case, two features are seen: (a) the antisolar peak and (b) limb
brightening. For the 2D 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 3D 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 2D 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 3D 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 threedimensional
space defined by coordinate systems. Computations can then be performed using solution
equations referred to as the “exact” method, to obtain the singlescattering 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 nonspherical ice
crystals, however, cannot be resolved by the conventional approach to solutions in view
of their complex structure. This chapter presents the geometricoptics approach to light
scattering by ice crystals. We first review the essence of geometric optics from several
historical perspectives, including diffraction and surface waves.
115
116 Geometric optics for light scattering by ice crystals
White light
Sunlight
Sunlight
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 minimumdeviation rays. van
de Hulst (1947) suggested that glories are caused by grazingincident 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 geometricoptics 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 geometricoptics results visà
vis the values determined from lightscattering and icemicrophysics 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.
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
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
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
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
).
normal of interface
incident wave reflected wave
er
ei
qi qr
Air
Ice
qt
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
equiamplitude 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
raytracing.
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
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 nonabsorptive cases, kα is zero. The corresponding
122 Geometric optics for light scattering by ice crystals
where Ai , Ar , and At denote amplitude vectors. Furthermore, we may define the following
normalized parameters:
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
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)]
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:
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
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 highorder reflection/refraction events when the ray transmission
is from an ice crystal to air. Figure 3.4a depicts a 2D 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 firstorder
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
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
higherorder 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 higherorder ray
is much lower than that carried by a lowerorder 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 hitandmiss approach
to ice crystals
Use of the Monte Carlo method in connection with geometric raytracing 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 hitandmiss 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 computergenerated ice crystal sizes and shapes.
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 bodyframed
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)
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
hitandmiss 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 icecrystal 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 reenter
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 reenters 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 singlescattering 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 raytracing approach.
The surfaces of ice crystals may not be exactly smooth, particularly if they undergo
collision and riming processes. Also, a careful electronmicroscopic 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 raytracing 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 twodimensional 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).
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., finitedifference time domain method). The fascinating optical phenomena produced
by defined ice crystals are numerous, and are presented in the following subsections.
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.
On the basis of the geometry of the incident (skew ray) and refracted rays defined in
Figure 3.7a, we can further show that
Eliminating θ i and θ t from Eq. (3.1.30) and the Snell equation, we obtain
Set mr (cos εt /cos εi ) = mr on the lefthand 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 3D 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
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
Thus, we obtain
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
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 2D columns (rotational
orientation about the caxis 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 2D
columns is similar to that for randomly oriented columns.
The physical causes of the preceding optical features generated by “smooth” 2D 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 raytracing 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
x
y
Plane waves
O' r
P
φ
r0
θ
z
O
r0  r
A
u0 eikz
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 xy
plane. See text for notation definition.
3.2.1 Sphere
Let the zaxis 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 lightwave 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 zaxis, 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
(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 semimajor axis with the semimajor 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 threeaxis 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 zaxes 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 bodyframed 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
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)
2π
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
The solution of diffraction due to the lightwave 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
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
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
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 zeroorder 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
firstorder 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 semiminor
axis a and the semimajor 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 threeaxis ellipsoid, which is governed by the
following geometric relationship:
x2 y2 z2
2
+ 2 + 2 = 1. (3.2.35)
a b c
When a threeaxis 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 semiaxis lengths p and q, and the angle η between the paxis and x axis
as well as its diffracted intensity can then be computed.
Finally, for randomly oriented nonspherical 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.
2 2 2
θ θ
4 4 4
210 210
6θ
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
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
6θ
330
6θ
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 threeaxis 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
firstorder 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 circlelike
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 parallelogramlike pattern
occurs around the incident direction and is surrounded by four fans. Figure 3.12i depicts
the diffraction pattern generated by a threeaxis ellipsoid whose projected shadow is an
ellipse with a semi paxis length of 6.6 µm, semi qaxis length of 4.0 µm, and an angle η
between the paxis and xʹaxis of 29.0°. In Figure 3.12b, we show the diffraction pattern
for an ellipse with its long axis along the qaxis.
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 crosssectional areas are
3.2 Fraunhofer diffraction by ice crystals 145
(a) (b)
108 108
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 crosssectional 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 nonspherical particle with a specific orientation
relative to the incident light beam, the scattering light produced by diffraction displays
significant fluctuations in 3D space. However, when an ensemble of nonspherical particles
are randomly oriented in 3D 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 semiexact solution for the diffraction of irregular ice particles (bullet
rosette, aggregate, snowflake, and roughsurfaced 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 crosssectional 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
(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 raytracing technique, including
diffraction, for computing the singlescattering 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 geometricoptics integralequation
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 crosssection
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 crosssection 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:
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
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 nearinfrared 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 firstorder refracted
rays or internally reflected rays, as follows:
With some vector algebraic manipulations on the basis of Snell’s law, it can be shown that
the propagating directions of pthorder reflected and refracted rays are given by
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 raytracing 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 raytracing 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
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 pth order rays (p>1)
Figure 3.15 Schematic diagrams for the direction of the incident, reflected, and refracted rays for
the firstorder ray (p = 1) and pthorder 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.
the pthorder 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 pthorder 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 firstorder 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
Because the field components on the lefthand 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
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 firstorder 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 crosssection 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 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
If the sign of the rotation angle φp is changed, then the signs of the nondiagonal 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 righthand 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, nondiagonal 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 raytracing computations.
The followon 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 singlescattering 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 pthorder 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 pthorder 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
crosssection of the jth initial ray. To speed up the computation, the preceding raytracing
algorithm can be augmented with the Monte Carlo hitandmiss 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 raytracing 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 nonzero, total absorption can be accounted for by considering the
absorption of individual rays. In general, the absorption crosssection of a particle depends
on the polarization configuration of the incident light. However, for randomly oriented
particles, the absorption crosssection is the average of absorption crosssections 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
firstorder 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 crosssection 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 firstorder refracted rays to the absorption crosssection
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 crosssection 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
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 higherorder rays is relatively insignificant.
Equations (3.3.27a) and (3.3.28) provide explicit formulations for absorption crosssection
within the framework of the raytracing technique in which the polarization configuration
is fully accounted for. In the conventional raytracing 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 raytracing method is the production of
deltatransmission associated with the refraction of rays through two parallel faces of
pristine ice crystals. The deltatransmission phenomenon has been identified and discussed
in detail by Takano and Liou (1989a) and Mishchenko and Macke (1998). Let the portion of
the scattering crosssection associated with the deltatransmission be σδ . Then the scattering
crosssection can be separated into three terms as follows:
The first term corresponds to contributions from externally reflected rays and various
transmitted rays excluding deltatransmitted rays, the second term denotes contributions
from deltatransmitted rays, and the third term is associated with diffraction.
Let fδ be the ratio of deltatransmitted energy to the total scattered energy, defined by
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 singlescattering 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 raytracing 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 crosssection 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 raytracing technique leads to deltatransmission (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 geometricoptics approaches, described in the following Subsections 3.3.3 and
3.3.4.
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 farfield 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 farfield solution can also
be determined by a volume integral involving the internal field.
By means of geometric raytracing, 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 3D 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 raytracing,
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 nearfield 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 raybyray 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 crosssections. 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)
4π
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 outwardpointing 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 crosssection 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 crosssection 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)
4π
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 )
4π
c
= [H∗ · (∇ × E0 ) − E0 · (∇ × H∗ ) + H0 · (∇ × E∗ ) − E∗ · (∇ × H0 )].
4π
(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 )]
4π
∗ (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
4 λ = 0.63 µm
L /a = 6
3
IGOM CGOM
2
FDTD
1
0
Figure 3.16 Comparison of the extinction efficiency computed from conventional (CGOM) and
improved (IGOM) geometricoptics approaches and from the finitedifference 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 raytracing 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 crosssection 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
crosssection is on the order of k −1 so that the phase change over this crosssection 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
crosssection can be neglected, numerical results are not sensitive to the ray’s crosssectional
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
geometricoptics method (CGOM) and that computed from the improved geometricoptics
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 finitedifference 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 crosssection 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 crosssection 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 hollowcolumn 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).
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
101 101
FDTD FDTD
102 102
IGOM IGOM
103 103
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
101 101
FDTD
FDTD
102 102
IGOM
IGOM
103 103
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).
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
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 crosssections.
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 crosssection is defined by the sum of the two diagonal elements in the
forward direction as follows [see also Eq. (2.4.45d)]:
2π
σe = Re[S1 (e0 ) + S2 (e0 )], (3.3.45a)
k2
where e0 denotes the incident direction and the absorption crosssection 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 twicerefracted
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.
reflected wave
e1
e3
Q2
reflected
wave
e2
diffracted wave
Q3
Figure 3.18 A conceptual raybyray geometricoptics approach which includes surfacewave 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 surfacewave 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 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).
where w denotes ext (extinction), abs (absorption), or pr (radiation pressure). In this equa
tion, linearity between the geometricoptics (GO) and surfacewave () 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:
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 geometricoptics and surfacewave 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 righthand side of Eq. (3.3.46). To investigate applicability of the
surfacewave 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 surfacewave 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
raybyray geometricoptics approach outlined above, without including the surfacewave
effect, match closely with those determined from the exact LM theory when ρ < 2.86, so
surfacewave adjustments are not required in this region.
3.3 Generalized geometric optics for light scattering 169
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, singlescattering albedo, and asymmetry factor
as a function of size parameter between the results computed from Lorenz–Mie (LM) theory and
the geometricoptics and surfacewave (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
singlescattering 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 singlescattering patterns
are seen to be progressively damped when mi or absorption increases.
Surface wave
Reflected
photons
Incident photons
Figure 3.20 A conceptual depiction of the geometricoptics and surfacewave approach to light
scattering by an ensemble of core–shell spheres (aggregates), representing general nonspherical
particles, including ice crystals. The geometricoptics components include the hitandmiss Monte
Carlo photon tracing associated with external and internal reflections and refractions, and diffraction
based on Babinet’s optical principle for randomly oriented nonspherical 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 crosssectional area in
accordance with Babinet’s principle and passing around this area. In geometric reflec
tion/refraction, we have employed the hitandmiss 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 raybyray approach described in Subsection 3.3.4.
where α and β are angles that denote the orientation of an aggregate in a twodimensional
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
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 nonspherical 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 nonspherical particles as follows:
where GO stands for geometric optics and the extinction and absorption efficiencies Qext
and Qabs have been defined in Eq. (3.3.59).
where Qw is the efficiency factor for extinction (ext), absorption (abs), or radiation pressure
(pr), and Qw is the surfacewave adjustment. Additionally, a parameter f has been added
into this equation for the case of nonspherical 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
where the singlescattering 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 crosssectional 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 nonspherical 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 surfacewave propagation. Therefore, the parameter
f introduced in Eq. (3.3.62) must be less than 1 for nonspherical 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 (threedimensional)
must also play a key role in the effectiveness of surfacewave 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 crosssectional 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:
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
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
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, singlescattering 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 onehalf of the maximum dimension and λ is the wavelength, as computed from
the geometricoptics and surfacewave (GOS) approach for size parameters up to 100, and from the
finitedifference 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 crosssection 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 crosssection
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(qrj 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
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 Tmatrix
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 IGOM
Qext
CGOM
2
1 GOS/RGD
0
10 −4 1
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
1v 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 visàvis 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 singlescattering coalbedo (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 singlescattering 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
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 visàvis GOS/RGD. Comparison of the unified theory and FDTD and
DDA results for randomly oriented ice crystals with variable lengthtowidth ratios for (a) solid columns and (b) plates, in terms of the extinction
efficiency Qext , the singlescattering 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 nonspherical ice particles. The element P44 P11 is larger than P33 P11 , due also to
nonsphericity, as is in the case for spheroids illustrated in Figure 2.20. The nondiagonal
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 laboratorygenerated 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 , singlescattering albedo
, and asymmetry factor g as a function of wavelength from 0.2 to 100 µm for randomly
oriented Koch snowflakes, having a spherical volumeequivalent 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 singlescattering 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 singlescattering 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 nearinfrared
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, singlescattering 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 singlescattering 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 crosschecking theoretical results.
3.4 A unified theory of scattering by ice crystals 185
Extinction Efficiency
3
0
1.0
rsnow = 100 μm SingleScattering 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, singlescattering albedo, and asymmetry factor as a function of
wavelength from 0.2 to 100 µm for Koch snowflakes with a spherical volumeequivalent 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 lightscattering experiments. For this reason, Barkey
et al. (1999) carried out an experimental lightscattering program using hexagonal icelike
crystals measured in the analog manner so that optical experiments could be performed over
a relatively long period of time for complexshaped 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 halfdome 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
lightscattering 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 roughsurface 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) Dataacquisition
Photodiode card and PC Continuous
detector/amplifiers computer Refrigerated nucleator
and insulated
Fiber light growth
guides chamber Dewar with
liquid nitrogen
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 fiberoptic 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 watervapor 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
freefall. 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 dropletsize
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
100
101
102
0 60 120 180
Scattering Angle (deg.)
Figure 3.29 Experimentally measured twodimensional 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).
[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 cloudscope 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 crosssection 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 cloudscope 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 cloudscope 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
101
102
0 60 120 180
(deg.)
Scattering Angle (deg.)
Figure 3.30 Twodimensional angular scattering intensities measured by the polar nephelometer
(top panel) when predominantly plates are seen, as shown in the video cloudscope 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 cloudscope light source (after Barkey et al. 2002).
3.4 A unified theory of scattering by ice crystals 191
102
Experiment
100
101
102
0 60 120 180
Scattering Angle (deg.)
Figure 3.31 Twodimensional angular scattering intensities measured by the polar nephelometer
(top panel) when predominantly columns are observed, as shown in the video cloudscope 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 cloudscope 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 wellknown 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, singlescattering coalbedo, and asymmetry factor for contaminated
ice particles, for efficient application to radiative transfer calculations on the basis of the
comprehensive spectral database of the singlescattering 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 highelevation 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 lowerelevation 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 sevenfold 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 volumeequivalent
radius for BC particles in snow is about 0.1 µm.
With reference to dust, Kumai (1977) investigated the longterm 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
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
TwoLayer Model (rsnow = 100 μ m, rsnow = 1000 μ m), μ 0 = 0.5
(a)
BC (rBC = 0.1 μm)
0.0
(b)
Dust (rdust = 1 μm)
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 twosnow 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 semiinfinite. 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 snowgrain 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 geometricoptics and surfacewave (GOS) approach
coupled with Monte Carlo photon tracing, to determine BC/dust singlescattering 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 twolayer 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 singlescattering properties determined from stochastic and light
absorption parameterizations and using the adding/doubling method for spectral radiative
transfer, the effect of snowgrain 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 equalvolume single inclusion, and
the snowgrain 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 nonspherical 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 finitedifference time domain (FDTD)
method, the Tmatrix 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 geometricoptics, diffraction, and
surfacewave approaches presented in Chapter 3.
197
198 Other useful approaches to light scattering
Connecting Surface
Scattering Particle
Totalfield Region
Scatteredfield Region
Figure 4.1 A conceptual diagram for the computation, by means of the finitedifference time domain
(FDTD) method, of the near field for electromagnetic scattering by a nonspherical particle illuminated
by plane waves. The connecting surface (or the Huygens surface) divides the computational domain
into totalfield and scatteredfield regions.
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 leapfrog 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 Finitedifference time domain method 201
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
zcomponents 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 zaxes, 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
finitedifference 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 Efield and the normal components
of Hfield 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 zcomponent of the electric field
vector as a function of space and time, is denoted in the form
With the preceding notation and the field component configuration depicted in Figure 4.2,
application of the secondorder central difference scheme to the spatial derivatives defined
in Eqs. (4.1.11) and (4.1.12) leads to
Equations (4.1.14) and (4.1.15) represent an explicit finitedifference scheme; that is, the
temporal variations of the electromagnetic field values at the grid points can be simulated via
a straightforward timemarching iterative procedure without invoking the matrix inversion
required by an implicit finitedifference 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 timeharmonic 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 nonferromagnetic 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
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:
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 zaxis 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 zaxis, 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 oneway wave equation
(Mur 1981).
Values of the near field in the time domain can be computed using a timemarching
iterative procedure based on the finitedifference analog of Maxwell’s curl equations
and boundary condition equations. To obtain the singlescattering properties that are the
inherent optical characteristics of the particle in the frequency domain, timedependent
206 Other useful approaches to light scattering
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
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 finitedifference 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 timemarching 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 nonspherical 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 Finitedifference time domain method 207
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
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 Efield 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 Efield 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 zaxis is aligned with the incident direction, and is
represented by Eo,x and Eo,y . To map the incident Efield 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 yaxes 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 Finitedifference 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
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).
103 1
Absolute Error P43/ P11 Absolute Error P33/ P11 Absolute Error P12/ P11
LorenzMie 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 Nonzero 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 nonzero phase matrix elements computed with
the FDTD method and the Lorenz–Mie results for an ice sphere at a farinfrared 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
timemarching iterations were performed to map the near field from the time domain to
the frequency domain through the discrete Fourier transform. The lefthand 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 righthand
column of Figure 4.3 depicts a comparison between FDTD and Lorenz–Mie solutions for
4.1 Finitedifference time domain method 211
102 1
Absolute Error P43/ P11 Absolute Error P33/ P11 Absolute Error P12/ P11
LorenzMie 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 righthand 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 gridmesh resolution (s = λ165) was
used and, in this case, 7500 timemarching 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  i1.66x108 P33/P11
0.5
P11
0
100
0.5
1
0.4
101 0.2 P34/P11
0
0.2
102 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
101
0
0.5
102 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 Finitedifference time domain method 213
l = 0.55 μm
3 L/a = 6
FDTD
1
Raybyray
0
6 10 100 1000
4 1
SingleScattering 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
Figure 4.6 Comparison of extinction efficiency and singlescattering albedo computed by FDTD
and by the raybyray geometricoptics 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 timedomain method,
whereas the latter is a frequencydomain 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 singlescattering albedo computed
from FDTD and the raybyray geometricoptics 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 raybyray algorithm is much more computationally efficient
than FDTD, particularly for computing extinction and absorption crosssections. 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 raybyray 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 singlebullet and bulletrosette 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 raybyray 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 singlebullet
and bulletrosette 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 multibranched rosette with more than eight branches displays maximal features in
the sidescattering 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 multibranched 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 Finitedifference time domain method 215
101 101
101
Phase Function
102 102
102
0 30 60 90 120 150 180 0 30 60 90 120 150 180 0 30 60 90 120 150 180
102 102 102
ka = 10 kL = 20 kL = 20
100 L/a = 0.5 100 100
L/a = 8 L/a = 8
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 halfwidth and L is the length (column) or
depth (plate), k = 2π/λ, and λ is the wavelength (after Yang and Liou 2006).
where N is the total number of grid points in the xaxis 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 singlescattering 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 Tmatrix 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 crosssection r and by its length L, while the spheroid size is determined in terms of its
polar radius b and equatorial radius a. For Tmatrix results, the Tmatrix 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 Tmatrix 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 Finitedifference 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 LorenzMie
PSTD
102
101
100
101
102
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
101
102
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 Nonzero 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 semiwidth 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 Finitedifference 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
Tmatrix
PSTD
101 103
100 102
101 101
102 100
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 Tmatrix 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 nonzero 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 geometricoptics methods. Here, the solutions
based on the improved geometricoptics 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 7.2x103 m = 1.4005  i 7.2x103
Phase Function
101 101
FDTD
PSTD
100 100
101 101
102 102
103 103
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 singlescattering properties (extinction efficiency, singlescattering 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 singlescattering properties of nonspherical ice
4.1 Finitedifference 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 Nonzero 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 diametertolength 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)
SingleScattering 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, singlescattering 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 onehalf of the maximum dimension, and λ is the wavelength (courtesy of
Chao Liu).
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 Tmatrix 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
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 Tmatrix approach.
If the Tmatrix 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
∞
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 Tmatrix 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 Tmatrix 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 Tmatrix discovered by Mishchenko (1990)
provides a computationally efficient and exact approach to calculating singlescattering
properties averaged over random orientations. For example, the extinction and scattering
4.2 Tmatrix numerical method 225
crosssections can be obtained for Tmatrix 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 Tmatrix 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 Tmatrix
in the form
T = −Rg Q Q−1 , (4.2.15)
where the Tmatrix 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 volumeequivalent sphere of 35. The simulation was conducted by using the EBCM
Tmatrix 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
P12/P11
P11
1 0
10
100
0.5
1
10
102 1
0 45 90 135 180 0 45 90 135 180
Scattering Angle (deg.) Scattering Angle (deg.)
Figure 4.15 Tmatrix 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 Tmatrix 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.
Figure 4.16 A conceptual representation of a 2D plate as an example of the invariant imbedding
Tmatrix (IITM) method. (a) The 2D plate shape. (b) This is circumscribed by an imaginary
inhomogeneous sphere; the portion of the sphere exterior to the 2D plate is empty. (c) Discretization
of this inhomogeneous sphere into multilayers. (d) Computation of the Tmatrix using an internal
sphere whose Tmatrix can be obtained from the Lorenz–Mie theory. (e) An intermediate step in
computing the Tmatrix via iterative procedures based on the IITM approach. (f) The final particle
geometry, a 2D plate, after completing IITM 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 Tmatrix
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 Tmatrix calculations (hereafter referred to as the IITM approach) for scattering
by an arbitrarily shaped particle: (a) A 2D 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 multilayered sphere, so the problem
of light scattering by a nonspherical particle is transformed into that of light scattering
by a multilayered inhomogeneous sphere. (d) The Tmatrix is computed for an internal
sphere using Lorenz–Mie theory; for computational efficiency, we may select a sphere that
is inscribed within the 2D plate. (e) An intermediate step is to compute the Tmatrix by
means of an iterative procedure based on the IITM approach. (f) The final particle geometry
when iterative IITM computations are completed would be the original 2D plate.
The principle of IITM is to obtain the Tmatrix of an inhomogeneous sphere with p
layers based on the Tmatrix of an inhomogeneous sphere with p − 1 layers. As an initial
point for iterative computations, the Tmatrix at the origin is zero. The IITM recursive
228 Other useful approaches to light scattering
4
10 0.5
11
IITM
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 IITM approach for randomly oriented spheroids as a
function of the scattering angle (after Bi et al. 2013).
where T(rp ) is the Tmatrix composed of p layers, I is the identity matrix, and Qij are the
matrices involving integration over inhomogeneous spherical layers. To speed the IITM
computation, the method of separation of variables (Bohren and Huffman 1983) can be
employed to compute the Tmatrix of a sphere inscribed within the nonspherical 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 IITM method for singlescattering computa
tions. Based on a contemporary numerical implementation (Bi and Yang 2014), the IITM
approach can apply to size parameters substantially larger than its EBCMTM counter
part. To validate the IITM implementation, Figure 4.17 compares the scattering phase
4.2 Tmatrix numerical method 229
105 1 1
IITM
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 IITM method and using an
improved geometricoptics 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 IITM and EBCMTM computational codes for randomly ori
ented spheroids with a size parameter of 95, specified in terms of the semimajor 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 IITM program was further used to compute randomly oriented spheroids with an
unprecedentedly large size parameter of 300. In this case, the EBCMTM 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 IITM technique with the results of an improved geometricoptics 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 IITM technique for application to large nonspherical particles of various
shapes.
Bi and Yang (2014) further improved the IITM 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 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 Tmatrix 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
∞
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)
(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 Tmatrix numerical method 233
(a) (b)
3 4
10 10
IITM IITM
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 IITM and DDA methods for (a)
a short column with a size parameter of 30, and (b) the IITM 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 IITM method to the computation
of scattering phase matrix elements for a large size parameter of 300. To check the IITM
method’s numerical implementation, we have compared the phase function computed from
IITM 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 IITM and a volumeintegral 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 IITM 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,
geometricoptics 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
IITM 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 IITM and ADDA methods (after Bi and Yang 2014).
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
IITM
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 IITM 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 IITM 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
2 2 2
10 10 10
0 0 0
10 10 10
4 4 4
3 3 3
2 2 2
1 1 1
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 Tmatrix 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 IITM 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 IITM and from IGOM are smaller,
as IGOM becomes increasingly accurate. Because ice crystals are almost nonabsorptive
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 Tmatrix results. For absorptive particles, when the ice crystal size
parameter is larger than a certain value, internal reflection of geometricoptics 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 IITM 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
IITM 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
Singlescattering 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
0.2 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 singlescattering albedo for a
short column as a function of size parameter, computed using the IITM 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 orientationaverage 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 xyz convention.
103 10
4
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 IITM methods (courtesy of Lei Bi).
(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 threedipole system (red dots) define by i, j, k indices
and associated incident electric vectors.
m2 − 1
χ (r) = d 3 (4.3.2)
4π
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 crosssections are given by the following equations
4.3 Discrete dipole approximation 241
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:
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:
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
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)
⎩ ⎭
Vε
⎧ ⎫
⎨ ⎬
L̄(r) = k 2 Ḡs (r − r )d 3 r . (4.3.19)
⎩ ⎭
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
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 weakform DDA corresponds to the Clausius–Mossotti relation (i.e.,
M̄ = 0̄), while the strongform DDA indicates nonzero M̄ stemming from the finite ele
mentary volume used in the calculation. In the case of the strongform DDA, several
polarizability relations (see Yurkin et al. 2007) are formulated instead of the wellknown
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 stateof
theart 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 singlescattering 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 “semiexact” method. The accuracy of
DDA has been examined by comparison of its results with those computed from analytical or
semianalytical 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 Tmatrix and DDA methods are also given in this figure. The size
parameter, defined in terms of the semimaximum 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
LorenzMie Tmatrix Tmatrix
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
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
Tmatrix and from ADDA. In ADDA simulations, the number of dipoles per wavelength used was
13.8203.
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
geometricoptics hybrid (EGOH) method developed by Bi et al. (2011b), which is an
improved form of the raybyray integration (RBRI) technique (Yang and Liou 1997).
In this method, beams with polygonshaped crosssections are traced within a scattering
particle, while the near field associated with the beams is analytically mapped to their
farfield counterparts based on a rigorous volumeintegral 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 singlescattering properties, particularly the scattering
phase matrix, of nonspherical particles covering the size parameter range from Rayleigh
to geometricoptics 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 nearIR 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 highenergy
lasers and mmwave 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 ATrain 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 nearIR 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 nonspherical ice crystals and spherical water
droplets. Lastly, we present reflected line spectra at the 1.38 µm band and the oxygen
Aband for the potential determination of the composition and optical properties of high
clouds.
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 threebody 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 highflying 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 spraycan propellants. Principal chlorofluorocarbons include CFCl3 (CFC11), CF2 Cl2
(CFC12), and CF3 Cl (CFC13). Sulfur dioxide (SO2 ) in the stratosphere is largely pro
duced by volcanic eruptions; this along with other sulfurbased 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
manmade 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