close

Вход

Забыли?

вход по аккаунту

?

Classification and global distribution of ocean precipitation types based on satellite passive microwave signatures

код для вставкиСкачать
INFORMATION TO USERS
This manuscript has been reproduced from the microfilm m aster. UMI films the
text directly from the original or copy submitted.
Thus, som e thesis and
dissertation copies are in typewriter face, while others may be from any type of
computer printer.
The quality o f th is reproduction is d e p en d e n t up o n th e quality o f th e copy
su b m itted .
Broken or indistinct print, colored or poor quality illustrations and
photographs, print bleedthrough, substandard margins, and improper alignment
can adversely affect reproduction.
In the unlikely event that the author did not send UMI a com plete manuscript and
there a re missing pages, these will be noted.
Also, if unauthorized copyright
material had to be removed, a note will indicate the deletion.
Oversize materials (e.g., maps, drawings, charts) are reproduced by sectioning
the original, beginning at the upper left-hand com er and continuing from left to
right in equal sections with small overlaps. Each original is also photographed in
one exposure and is included in reduced form at the back of the book.
Photographs included in the original manuscript have been reproduced
xerographically in this copy. Higher quality 6” x 9” black and white photographic
prints are available for any photographs or illustrations appearing in this copy for
an additional charge. Contact UMI directly to order.
Bell & Howell Information and Learning
300 North Zeeb Road, Ann Arbor, Ml 48106-1346 USA
800-521-0600
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
NOTE TO USERS
Page(s) missing in number only; text follows.
Microfilmed as received.
93
This reproduction is the best copy available.
UMI
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Gnduu* School F«m 9
(Revised 7/94)
PURDUE UNIVERSITY
GRADUATE SCHOOL
Thesis Acceptance
This is to certify that the thesis prepared
B y _______ N i t i n Gautam ___________
Entitled
CLASSIFICATION AND GLOBAL DISTRIBUTION OF OCEAN PRECIPITATION
TYPES BASED ON SATELLITE PASSIVE MICROWAVE SIGNATURES
Complies with University regulations and meets the standards o f the Graduate School for originality
and quality
For the degree o f
D o c to r o f P h ilo s o p h y
Signed by the final^examining committee:
chair
Approved by: .
Department Head
^
Date
is
/C ^
This thesis KT is not to be regarded as confidential.
^
y-
c2u,
Major Professor
Form at Approved by:
or
Chair, Final Examining Committee
£ /S f
Thesis Format Adviser
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
j
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CLASSIFICATION AND GLOBAL DISTRIBUTION OF OCEAN
PRECIPITATION TYPES BASED ON SATELLITE PASSIVE MICROWAVE
SIGNATURES
A Thesis
Submitted to the Faculty
of
Purdue University
by
Nitin Gautam
In Partial Fulfillment of the
Requirements for the Degree
of
Doctor of Philosophy
December, 1998
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
UMI N um ber: 9939346
UMI Microform 9939346
Copyright 1999, by UMI Company. All rights reserved.
This microform edition is protected against unauthorized
copying under Title 17, United States Code.
UMI
300 North Zeeb Road
Ann Arbor, M I 48103
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
To my parents.
For their love, affection, and teachings...
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ACKNOWLEDGEMENTS
This thesis has been possible because of the support and encouragement of many
people. I would like to acknowledge the active involvement of my advisor, Prof. G. W.
Petty, throughout this research work. Without his valuable guidance and critical
comments it would not have been possible for me to accomplish this task. The comments
and suggestions provided by other committee members - Prof. D. G. Vincent, Prof. C. A.
Clayson, and Prof. L. W. Braile - were very useful, and are gratefully acknowledged. I
would also like to thank Prof. G. V. Rao of Saint Louis University, and Dr. P. C. Pandey
of Antarctica Research Center, India, for their help and moral support during my graduate
studies.
Thanks are due John Schrage, Bob Green, Ralph, and Ben Johnson for their
valuable suggestions during thesis writing. I would also like to thank my other
departmental colleagues - Jaya Ramaprasada, Doug Miller, Chen Shu-Hua, Chen Aidong,
and Chris Henon - for their friendship and company.
I was able to make it through the five years of graduate school because of the love
and affection provided to me by my family members. My parents and elder brothers,
Nalin and Navin, have been a constant source of guidance and encouragement throughout
my life. I would not have come this far without them. The love and affection of my
younger brother Namit, sisters-in-law Poonam and Richa, and niece Tanya have been a
constant source of cheer.
I have always been fortunate to be in the company of wonderful and caring
friends, who have been a great source of learning and joy for me. I would like to thank
Sarabjit Singh, Shane Mayor, Suvinay Sinha, Bharath Srinivasan, Krishnan Ramaswamy,
Shankar JayaGanapathy, Ishpal Rekhi, Sankaraguruswamy, Shubhasish Mitra, Tapendra
Sinha, Raktim Pal, and Rajesh Singh for all their help and companionship. The memories
of our shared experiences in school will always be cherished by me. I look forward to
more of such good times with them in the coming years.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
TABLE OF CONTENTS
Page
ABSTRACT....................................................................................................................vi
PARTI:
CLASSIFICATION OF OCEANIC PRECIPITATION TYPES BASED ON
MICROPHYSICAL CHARACTERISTICS USING SSM/I DATA...............................1
ABSTRACT....................................................................................................................2
1. INTRODUCTION......................................................................................................3
2. DATA AND PRECIPITATION CLASSES............................................................... 9
2.1 SSM/I...................................................................................................................9
2.2 SUBJECTIVE IDENTIFICATION OF PRECIPITATION CLASSES.............. 10
2.3 DATA COLLECTION AND METHODOLOGY................................................14
3. DESCRIPTIVE STATISTICS OF SUBJECTIVELY DEFINED...............................16
4. CLASSIFICATION METHODS................................................................................25
4.1 K-NEAREST NEIGHBORS (KNN)...................................................................26
4.2 NEURAL NETWORK (NN)...............................................................................27
5. PERFORMANCE RESULTS.....................................................................................30
6. MICROPHYSICAL RETRIEVALS OF PRECIPITATION CLASSES................... 36
6.1 RAIN-CLOUD MODEL.....................................................................................36
6.2 FORWARD MODEL..........................................................................................39
6.3 INVERSION SCHEME.......................................................................................40
6.4 MICROPHYSICAL PROFILES.........................................................................43
7. APPLICATION OF OBJECTIVE CLASSIFICATION SCHEME TO THE
INDEPENDENT DATA SET.....................................................................................54
8. CONCLUSIONS......................................................................................................... 66
9. REFERENCES...........................................................................................................68
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Page
PARTH:
GLOBAL CHARACTERISTICS OF OCEANIC PRECIPITATION FROM SSM/I
DATA..............................................................................................................................75
ABSTRACT.................................................................................................................... 76
1. INTRODUCTION...................................................................................................... 77
2. DATA AND METHODOLOGY................................................................................ 82
2.1 SSM/I.................................................................................................................... 82
2.2 RAIN-RATE ALGORITHM................................................................................ 84
2.3 PROCESSING...................................................................................................... 85
3. PRECIPITATION CLASSES..................................................................................... 87
4. RESULTS................................................................................................................... 89
4.1 GLOBAL STATISTICS....................................................................................... 89
4.2 REGIONAL FREQUENCIES OF PRECIPITATION......................................... 94
4.3 COMPARISON OF SSM/I DERIVED PRECIPITATION FREQUENCIES WITH
COADS DERIVED PRECIPITATION FREQUENCIES.....................................103
4.4 REGIONAL DISTRIBUTION OF PRECIPITATION TOTAL AND FRACTION
FOR EACH PRECIPITATION TYPE..................................................................107
5. CONCLUSIONS..........................................................................................................114
6. REFERENCES............................................................................................................117
VITA................................................................................................................................ 120
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ABSTRACT
Gautam, Nitin. Ph.D., Purdue University, December 1998. Classification and Global
Distribution of Ocean Precipitation Types Based on Satellite Passive Microwave
Signatures. Major Professor: Grant Petty.
The main objectives of this thesis are to develop a robust statistical method for the
classification of ocean precipitation based on physical properties to which the SSM/I is
sensitive and to examine how these properties vary globally and seasonally. A two step
approach is adopted for the classification of oceanic precipitation classes from
multispectral SSM/I data: (1) we subjectively define precipitation classes using a priori
information about the precipitating system and its possible distinct signature on SSM/I
data such as scattering by ice particles aloft in the precipitating cloud, emission by liquid
rain water below freezing level, the difference of polarization at 19 GHz - an indirect
measure of optical depth, etc.; (2) we then develop an objective classification scheme
which is found to reproduce the subjective classification with high accuracy. This hybrid
strategy allows us to use the characteristics of the data to define and encode classes and
helps retain the physical interpretation of classes. The classification methods based on knearest neighbor and neural network are developed to objectively classify six
precipitatoion classes. It is found that the classification method based neural network
yields high accuracy for all precipitation classes. An inversion method based on
minimum variance approach was used to retrieve gross microphysical properties of these
precipitation classes such as column integrated liquid water path, column integrated ice
water path, and column integrated rain water path. This classification method is then
applied to 2 years (1991-92) of SSM/I data to examine and document the seasonal and
global distribution of precipitation frequency corresponding to each of these objectively
defined six classes. The characteristics of the distribution are found to be consistent with
assumptions used in defining these six precipitation classes and also with well known
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
climatological patterns of precipitation regions. The seasonal and global distribution of
these six classes is also compared with the earlier results obtained from Comprehensive
Ocean Atmosphere Data Sets (COADS). It is found that the gross pattern of the
distributions obtained from SSM/I and COADS data match remarkably well with each
other.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1
PARTI
CLASSIFICATION OF OCEANIC PRECIPITATION TYPES BASED ON
MICROPHYSICAL CHARACTERISTICS USING SSM/I DATA
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2
ABSTRACT
A robust statistical method for the classification of ocean precipitation is developed using
satellite passive microwave signatures from Special Sensor Microwave/Imager (SSM/I).
A two step approach is developed for the classification of six precipitation classes from
multispectral SSM/I data. First, six precipitation classes are defined using a priori
information about the precipitating system and its possible distinctive signature on SSM/I
data. An objective classification scheme based k-nearest neighbors and neural network is
then developed to reproduce subjectively defined precipitation classes. It is found that
classification method based on neural network technique yields better results than the
classification method based on k-nearest neighbors approach. This classification approach
allows us to use the characteristics of the data to define classes and also helps retain the
physical interpretation of precipitation classes.
An inversion method based on a minimum variance approach is used to retrieve gross
microphysical properties of each precipitation class (except for TYPE-VI, which is non­
precipitating). The microphysical properties such as column integrated rain water path,
column integrated ice water path, etc. are used to examine relative changes in these
values for each precipitation class. Finally, four months of independent SSM/I data for all
raining pixels ( > 0.2 mm/Hr) are extracted to demonstrate the meaningfulness and
consistency of subjectively defined precipitation class.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3
1. INTRODUCTION
Knowledge of the global distribution of precipitation and of its microphysical and
dynamical properties is extremely important to the complete understanding of the
integrated earth-atmosphere system. Precipitation is an interactive component of the
highly non-linear global climate system. The spatial and temporal variability of the
vertical distribution of diabatic heating resulting from atmospheric precipitation is the
main driver of the atmospheric circulation over a broad range of scales. Therefore, the
accurate and continuous observation of precipitation is vital to the understanding,
monitoring, and modeling of short and long term climatic fluctuations. A large body of
observational and modeling evidence has accumulated during the past decade which
shows significant relationships, particularly in the Pacific Basin, between changes in
tropical sea surface temperatures (SST), tropical rainfall and global circulation, and short
term climate variability [Ramanathan et. al. 1989; Janowiak et. al. 1985; Stephens and
Greenwald 1991; Trenberth and Branstator 1992; Meehl 1992; Zhang 1993; Nakazawa
1996]. However, most of these observational studies used proxy variables like outgoing
longwave radiation (OLR) instead of more direct measurements of oceanic precipitation.
Another striking example is the role played by large scale time averaged precipitation
fields to monitor and understand the ENSO (El-Nino and Southern Oscillations)
phenomenon [Lau and Qian 1986; Weare 1987; Arkin and Meisner 1987; Arkin and
Ardanuy 1989; Spencer 1993]. It is also recognized that the spatial and temporal
variability of precipitation types (stratiform, convective, warm, etc.) and associated
microphysical properties (like liquid water, snow, and graupel amounts and their vertical
distributions), which can also be used to infer latent heat profile, are the building blocks
for a complete understanding of cloud feedback mechanisms (Wielicki et. al.1995).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4
Types of clouds (precipitating or non-precipitating) that have been defined and used by
the international weather community were classified visually on the basis of their shapes,
sizes, and altitudes. The physical properties (microphysical or dynamical) of these clouds
have been studied mostly by measurements of variables like liquid water content, drop
size and distribution, upward motion, temperature, etc., either from aircraft platforms or
from ground based experiments. Since these studies were very localized and difficult to
carry out on a large scale, especially over the vast oceanic region, a meaningful and
robust spatial and temporal distribution of cloud physical properties of various
precipitation classes over the entire globe is possible only through the use of satellite
remote sensing data.
The launch of the first meteorological satellite in the early sixties has paved the way to
study precipitating and non-precipitating clouds from the sky. Satellites provide a large
scale view of clouds that is impossible to get from any other source. The importance of
satellite observation in the study of cloud systems, especially precipitating cloud systems,
is recognized by the scientific community. Many satellites with advanced sensors have
been launched [Special Sensor Microwave/Imager (SSM/I), Advanced Very High
Resolution Radiometer (AVHRR), Tropical Rainfall Measuring Mission (TRMM)
Microwave Imager (TMI) etc.] and many more satellites are planned to be launched with
more sophisticated sensors [Advanced Microwave Sounding Radiometer (AMSR),
Advanced Visible and Infrared Sounder (AVIRS) etc.] in the next five years to study
clouds and precipitation over the globe.
A significant amount of work has been done in the classification of precipitating and non­
precipitating clouds using data from visible and infrared (VIS/IR) sensors onboard polar
and geostationary satellites [Inoue 1987; Gerald 1988; Bankert 1994; Barret et. al. 1994;
Baum et. al. 1997]. These sensors have an advantage in terms of high spatial resolution (a
few kilometers) and fine temporal coverage (every three or less hours in the case of
geostationary satellites). The classification methods, developed using VIS/IR sensor data,
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
5
typically use cloud thickness, inferred from visible data, and cloud top temperature,
derived from infrared data. Most of the techniques for cloud classification based on
VIS/IR data suffer from inherent limitation of their indirect interaction with clouds and
also saturation of VIS/IR radiation in deep convective clouds. Due to this indirectness, it
is very difficult to retrieve information about microphysical characteristics of the clouds
from VIS/IR data.
The ability of microwave radiation to penetrate through clouds makes this band uniquely
suited for the remote sensing of precipitation. Since the launch of the first advanced
passive microwave sensor Special Sensor Microwave/Imager (SSM/I) in 1987 on a series
of Defense Meteorological Satellite Program (DMSP) satellites, significant progress has
been made in the retrieval of rain rate using passive microwave observations [Wilheit et.
al. 1994; Petty 1995; Petty and Krajewski 1996]. Some efforts have also been made to
indirectly retrieve information about cloud microphysical properties from SSM/I data
using physically-based rain rate retrieval algorithms [Kummerow et. al. 1989; Adler et al.
1991; Smith et. al. 1992; Mugnai et. al. 1993; Kummerow and Giglio 1994a,b]. These
algorithms employ detailed cloud microphysical models that supply realistic hydrometeor
profiles for forward radiative transfer calculations. These algorithms are sophisticated and
detailed in their consideration of thermodynamical and microphysical aspects of clouds,
but must directly deal with the non-uniqueness of the solution and sensitivity to
numerous geometric, microphysical, and optical assumptions made in the implementation
of these physically based rain rate algorithms.
Observational studies were conducted by McGaughey et. al. (1996) and McGaughey and
Zipser (1996) using high resolution passive microwave observations, obtained from
Advanced Microwave Precipitation Radiometer (AMPR) to investigate the relationship
between strong emission and ice scattering signatures for convective and stratiform
regions of precipitation in the tropical Pacific ocean. Their study further strengthens the
outcome of earlier theoretical and empirical studies that microwave observations at lower
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
6
channels (10.0, 19.35, and 37.01 GHz ) are found to be sensitive to the vertical
distribution of rain water while microwave radiance at 85.0 GHz is strongly affected by
scattering from ice particles aloft in these systems.
Despite the progress which has been made so far in the retrieval of precipitation and its
characteristics, it is also recognized by the scientific community that there is room for
additional improvements. The issues such as identification of a precipitation phase at the
surface, and classification of precipitation type (according to dynamical and
microphysical criterion) etc. can be studied in detail using data from current sensors. A
recent World Climate Research Programme (WCRP) report [WCRP Informal Report No.
6, 1996] recommended more efforts in the direction of extracting additional precipitation
information from available data sources, either in the form of improved spatial and
temporal resolution rainfall products or in the form of distinctly new products related to
physical and statistical properties of precipitation and their global distribution. So far, the
precipitation estimated from satellites has been classified either as “convective” or as
“stratiform”, which does not do justice to the variety of precipitating clouds that exist in
nature. The more refined classification of precipitation, on the basis of dynamical and
microphysical processes from satellite techniques (microwave and visible/infrared), is
still in an experimental stage and yet to be validated on a global basis. A recent work of
Sheu et. al. (1997) combined microwave and VIS/IR observations to infer tropical cloud
properties on the basis of cloud depth and cloud top information from VIS/IR and cloud
liquid water content and microwave index from SSM/I. The microwave index used in
their work accounts for both emission and scattering of microwave signals from raining
pixels. Their study yields ten identifiable precipitating and non-precipitating classes, but
does not attempt a detailed microphysical and dynamical interpretation of the classes.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
7
The main objective of this study is to develop a robust statistical method for the
classification of ocean precipitation based on its physical properties to which the SSM/I is
sensitive and also to examine the global characteristics of these precipitation classes. It is
found that six subjectively defined precipitation types can be distinguished objectively on
the basis of their multichannel microwave signatures alone, which in turn are related to
microphysical properties such as atmospheric liquid water content, snow/graupel amount
and their vertical distribution. A theoretical radiative transfer model calculation is used as
guidance for inferring gross microphysical properties for each type of precipitation.
Since passive microwave sensors like SSM/I have coarse resolution, the precipitating
clouds that can generally be seen from these types of sensor are mesoscale to synoptic
scale in extent. Most such systems are composed of a mix of cloud types, which together
comprise a precipitating complex. The major types of dynamically distinct precipitating
cloud systems which can be viewed from SSM/I are: (1) convective and stratiform
precipitation in mesoscale convective systems (hereafter MCSs), tropical cyclones, and
extratropical (baroclinic) cyclones; (2) widespread stratiform precipitation in mid- and
high-latitude regions; and (3) widespread precipitation generated from warm-topped
clouds. The last two types can only be viewed from SSM/I if they are spatially extensive
and nearly fills the Instantaneous Field of View (DFoV). The details concerning the cloud
dynamics and visual appearance of these precipitating system can be found in Houze
(1993). The identification of various precipitation types in SSM/I brightness temperature
data is a challenging task and will be accomplished on the basis of distinct features of
various precipitation types which can be observed using SSM/I data.
A two step approach is adopted in this study for the classification of multispectral SSM/I
data of various precipitation classes: (1) we subjectively define precipitation classes using
a priori information about the precipitating system and its possible distinct signature on
SSM/I data such as scattering by ice particles aloft in the precipitating cloud, emission by
liquid rain water below freezing level, difference of polarization at 19 GHz - an indirect
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8
measure of optical depth, etc.; and (2) we then develop an objective classification scheme
which is found to reproduce the subjective classification with high accuracy. This hybrid
strategy allows us to use the characteristics of the data to define and encode classes and
helps retain the physical interpretation of classes. In this study, neural network and Knearest neighbors are used as classifiers since these methods do not depend upon the
nature of the statistical distribution of data. It will be shown that six types of precipitation
classes can be objectively discriminated with high classification accuracy by applying
neural network classifiers on SSM/I data. The mean brightness temperatures of each
precipitation type are used to retrieve microphysical characteristics of each precipitation
type using a method developed by Petty and Gautam (1998). The microphysical
properties retrieved for each of the precipitation class are used to provide gross physical
characteristics of these precipitation classes such as column integrated liquid water path,
column integrated ice water path, column integrated rain water path, etc. They are also
compared with gross physical properties of the precipitating clouds classified by
conventional method.
The description of the SSM/I, data collection, subjective identification of precipitation
classes and a brief description of the characteristics of these classes are discussed in
section 2. Descriptive statistics (such as mean and standard deviation, etc.) of final
classified data sets are presented in section 3. Section 4 briefly describes the classification
methods explored in this analysis. Section 5 describes the performance results of the
classification methods. Section 6 outlines the method used to retrieve microphysical
properties of all precipitation classes and discusses the retrieved microphysical profiles
for all six precipitation classes and their closeness to the conventional precipitation
classes. The evaluation of the objective classification scheme for its meaningfulness and
consistency is presented in Section 7. Discussions and Conclusions appear in Section 8.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
9
2 DATA AND PRECIPITATION CLASSES
2.1 SSM/I
The SSM/I is a 7 channel passive microwave radiometer which measures radiances at
19.35, 37.0, and 85.5 GHz in the horizontal and vertical polarization, and at the water
vapor line at 22.235 GHz in the vertical polarization only. The spatial resolution of these
channels varies from 43x69 km at 19.35 GHz to 13x15 km at 85.5 GHz. The SSM/I has
flown on DMSP series of sun synchronous polar orbiting satellites since 1987. The swath
of SSM/I orbit is about 1400 km. A detailed description of SSM/I can be found in
Hollinger et. al. (1987). It is well documented that microwave frequencies below 30 GHz
are mainly affected by the thermal emission from liquid water below freezing level and
the response at higher microwave frequencies is dominated by scattering from ice
particles aloft [Adler et. al. 1991, Wang et. al. 1997]. The presence of the precipitating ice
(snow / graupel) particles aloft in the precipitating system is inferred from a scattering
index (S85) proposed by Petty (1994a,b). The rain rate retrieved by Petty’s (1994a,b)
algorithm is also used in this study.
Petty’s (1994a,b) rain rate retrieval technique over the ocean consists of two algorithms
based on two indices, normalized polarization index and scattering index, derived using
SSM/I brightness temperature data. The normalized polarization index is associated
basically with liquid water content in raining clouds while the scattering index is
controlled primarily by scattering from ice particles aloft in raining clouds. The first
algorithm yields an 85.5 GHz scattering index Sgs that selectively responds to brightness
temperature depressions associated with volume scattering by precipitation ice. The
second index performs a physical retrieval of the surface rain rate, based on observed
depolarization of the sea surface signal at 19.35 and 37.0 GHz and using Sg5 as the basis
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
10
for a first guess of a rain field. The detailed description of these algorithms can be found
in Petty (1994a,b). The noteworthy feature is the behavior of two algorithms towards
precipitation: S85 is sensitive primarily to size and concentration of large ice particles aloft
(e.g. graupel), which might be found in cold cloud (especially convective) precipitation,
whereas the rain rate algorithm is more specifically sensitive to path integrated rain water
between the surface and freezing level.
2.2 SUBJECTIVE IDENTIFICATION OF PRECIPITATION CLASSES
In the last three decades, visible and infrared images from many meteorological satellites
have been used to recognize and characterize primary precipitation-producing clouds on
the basis of texture, spectral signature, and spatial pattern. In this study, the subjective
classification of precipitation events was performed based on the appearance of the
precipitation in SSM/I imagery, including both radiometric properties and association
with recognizable mesoscale or synoptic weather scale weather patterns. More
specifically the classes in SSM/I data set were identified keeping the following issues
under consideration: (1) distinguishing features like indication of amount of ice present in
a precipitating system, retrieved rain rate at the surface (hereafter RR), atmospheric liquid
water content, transparency of the precipitating system, relative contribution of snow
/graupel and liquid rain in a given class of precipitation; (2) prior information about these
precipitating systems like tropical mesoscale convective system, cyclone, mid-latitude
fronts and extratropical cyclones; and (3) subjectively selecting the precipitating events
where the above physical characteristics (such as RR and S85, etc.) were spatially
widespread for each precipitation type. The raining pixels adjoining the boundaries of
raining and non-raining areas were not selected in this data set. This is done to minimize
the effect of the “beam-filling” problem (due to clear-sky pixels) on the data selected for
each types of precipitation.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
11
An attempt is made to postulate a relationship between these subjectively defined six
precipitation classes and the conventional description of precipitating clouds. A brief
description of the expected characteristic response of these precipitation types on SSM/I
brightness temperature and their assumed proximity to the conventional classification of
precipitation is given below:
•
TYPE-I: is chosen to represent deep convective precipitation associated with
organized tropical systems such as mesoscale convective systems, squall lines,
tropical cyclones, and hurricanes. These types of precipitating systems are
characterized by strong convective updrafts that support the production of large
amounts of liquid water content both above and below the freezing levels. The rapid
accretion of supercooled water by ice particles leads to graupel and hail formation
aloft and is a significant contributor to precipitation growth. These systems produce
large amounts of rain within a short period of time. The possible distinct signatures at
SSM/I frequencies due to TYPE-I precipitating systems could be: (1) a significant
depression on higher frequency channels due to significantly large amounts and sizes
of ice and graupel particles aloft in the precipitating system; (2) high values of
brightness temperature at lower frequency channels due to deep liquid rain column
and high liquid water content; and (3) strongly reduced polarization differences. The
selection criterion used for extracting SSM/I data for this precipitation class are: (1)
tropical systems; (2) areas characterized by high values of S8S and RR with strong
variability; and (3) low polarization difference at 19 GHz.
•
TYPE-II: is considered to represent convective regions of extratropical cyclones
(large frontal cyclones, comma cloud systems, and smaller polar lows). These
precipitating systems are mainly developed in response to synoptic-scale baroclinic
disturbances. The nimbostratus or cumulonimbus clouds in these precipitating
extratropical systems can be considered closest to TYPE-II. Significant amounts of
liquid water are present in these clouds. The growth of ice and graupel particles is
mainly characterized by riming and aggregation processes. These precipitating
systems produce significant amounts of precipitation in mid and higher latitudes.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
12
Since TYPE-H precipitation systems normally associate with clouds having weak to
moderate updraft , these systems do not produce large sized ice and graupel particles.
The drop in the brightness temperatures at higher frequencies is, therefore, not
expected to be as strong as for TYPE-I. Similarly, the brightness temperatures at
lower SSM/I channels are not expected to show a pronounced increase as for TYPE-I
because of relatively shallow depth for liquid rain in these precipitating systems. The
selection criterion used for extracting SSM/I data for TYPE-II precipitation class are:
(1) extratropical systems; (2) areas characterized by high values of S85 and moderate
values of RR where these variables also show reasonable variability within the
selected scene; and (3) relatively large polarization difference at 19 GHz.
•
TYPE-III: corresponds to precipitating clouds associated with the stratiform portion
of the organized tropical systems or widespread stratiform precipitation (with
moderate surface rain rate) in the tropics or in the midlatitudes. The precipitation
produced in these clouds are mostly from the collision-coalescence of liquid water
droplets. Some growth of ice particles aloft due to vapor deposition processes also
contributes to precipitation, especially in the case of the stratiform region of the
organized convection. These types of precipitating systems are expected to produce
relatively small depressions in brightness temperatures at higher SSM/I frequencies
and moderately high brightness temperatures for lower SSM/I channels because net
precipitation in these clouds is mainly from liquid rain. The selection criterion used
for extracting SSM/I data for this precipitation class are: (1) a homogeneous
precipitation region in the tropical systems or a widespread region of moderate rain
rate in the tropics or midlatitudes; (2) areas characterized by moderate values of S83
and RR with these variables showing relatively less variability in the scene; and (3)
low polarization difference at 85 GHz.
•
TYPE-IV: is chosen to represent a stratiform precipitation that is not associated with
convection. In these systems, the precipitation growth above the freezing level is
primarily by vapor deposition, with subsequent growth predominantly by aggregation
and/or light riming. Stratiform precipitation of this type is most often associated with
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
13
synoptic-scale baroclinic disturbances. These systems are more frequent in mid and
higher latitudes, but occasionally these systems can make intrusions equatorward of
the 40° latitude. Since the clouds associated with these precipitating systems are not
deep and do not have much liquid water in them, the brightness signatures at lower
SSM/I channels are not expected to produce a pronounced increase. High frequency
SSM/I channels are characterized by relatively small depressions due to small ice
particles present aloft in these precipitating systems. The selection criterion used for
extracting SSM/I data for this precipitation class are: (1) stratiform portion of the
extratropical systems or widespread precipitating clouds with low rain rates in
midlatitude or high latitude; (2) areas characterized by large values of the ratio of S85
and RR; and (3) relatively low RR values where the SSM/I scene is characterized by
less variability in RR values; and (4) large values of polarization difference at 85
GHz.
•
TYPE-V : is assumed to represent shallow, warm-topped convective showers or
marine stratocumulus in the tropical or midlatitude regions. In the absence of any
significant amount of ice, the precipitation growth in these systems is entirely
dominated by collision-coalescence of liquid water droplets. Recent studies have
indicated that warm-cloud precipitation could produce significant amounts of
precipitation in some geographical regions [Petty, 1997]. Since these precipitating
clouds have significant liquid water content but are characterized by shallow depths
of liquid rain, the signatures at higher SSM/I channels are expected to show
pronounced increases due to emission by liquid water and also due to no apparent
scattering by ice. The lower SSM/I frequencies are expected to be characterized by
low brightness temperatures due to shallow liquid rain and low surface rain rate. Also,
no apparent scattering due to ice is observed. The selection criterion used for
extracting SSM/I data for the TYPE-V precipitation class are very simple and straight
forward: all cases of precipitation events which are characterized by zero values of
S85 and non-zero RR values.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
14
•
TYPE-VI: is chosen to represent an widespread anvil cloud with large ice particles
aloft. These cloud systems are basically left over from large scale organized systems
in the tropics (mesoscale convective systems or tropical cyclones) or in the mid
latitudes (extratropical systems). These cloud systems do not produce any
precipitation, but can play a very crucial role in the radiation balance of the
atmosphere [Ramanathan and Collin, 1991]. These types of non-precipitating clouds
are found to have sufficiently large ice particles spread over large area which are
expected to produce significant depressions at higher SSM/I channels. The selection
criterion used for extracting SSM/I data for this precipitation class are very simple
and straight forward: all cases of precipitation events which are characterized by non­
zero values of S85 and zero RR values.
The general statistical characteristics of the six subjectively defined precipitating and
non-precipitating cloud systems, and as well as the retrieved microphysical properties of
these precipitating classes will be thoroughly examined in a later section of the paper.
Detailed information about the microphysical and dynamical properties of the clouds
associated with the above subjectively defined precipitating and non-precipitating classes
can be found in Houze (1993).
2.3 DATA COLLECTION AND METHODOLOGY
For this study about 100 precipitating events over the global oceans were selected from
two years of SSM/I data (1987 and 1991). Out of these 100 precipitating events, 55
percent were selected from the tropical regions (35'S - 35*N) and the remaining cases
were located in the mid-latitude and high latitude regions of the globe. These cases
represent most of the geographic regions and all seasons of the year. Once raw SSM/I
brightness temperature data for these precipitating events were selected from the SSM/I
tapes, they were quality controlled for unphysical brightness temperatures and other
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
15
artifacts. Petty’s (1994a,b) rain rate algorithm was then applied to these data to compute
rain rate at the surface and other products like scattering index at 85 GHz. The next step
was to subjectively classify the data set into the six classes, described previously, on the
basis of physical characteristics of various precipitation events inferred from SSM/I
brightness temperature data and the general appearance of these systems (i.e. whether
tropical cyclones or extratropical cyclones etc.). This classified data set was further
divided into training and testing data sets. Finally, the training data set was used to
develop an objective classification method, and the classification accuracy of the method
was validated using the independent testing data set. Figure 1 explains the steps involved
in the development of a method for the objective classification of precipitation types
using SSM/I data.
SUBJECTIVE CLASSIFICATION
QUALITY
CHECKED
SSM/I DATA
KNOWLEDGE OF DISTINCT
RESPONSE OF PRECIPITATION
CLASSES ON SSM/I DATA
RR and Su
FROM PETTY’S
RR ALGO
SUBJECTIVE
CLASSIFICATION OF
PRECIPITATION
OBJECTIVE CLASSIFICATION
TRAINING
DATA
TESTING
DATA
DEVELOPMENT OF
CLASSIFICATION
METHODS
EVALUATION OF NN
AND KNN METHODS
PRIOR KNOWLEDGE OF CHARACTERISTICS
OF VARIOUS PRECIPITATION TYPES
FINAL CLASSIFIER
Figure 1 Schematics of steps involved in the development of the classification method.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
16
3. DESCRIPTIVE STATISTICS OF SUBJECTIVELY DEFINED
PRECIPITATION CLASSES
A variety of statistical properties and graphical displays of the data set, comprising all
precipitation classes, has been explored to examine and understand the structure and
nature of the attributes (seven SSM/I brightness temperatures and derived products S85
and RR) variability for each precipitation class individually and collectively. The class
level statistics for each precipitation type are given in Table 1. A brief overview of the
variability of SSM/I for each precipitation class is furnished later in this section.
Table 1 : Class Level Statistiscs
Precipitation Classes
TYPE-I
TYPE-E
TYPE-m
TYPE-IV
TYPE-V
TYPE-VI
Number of Observations
1035
899
822
974
933
845
RR and Sg5 are used, respectively, to characterize precipitation intensity and the relative
concentration of ice and graupel particles present for each precipitation class. The
brightness temperature depolarization at 85 GHz and 19 GHz can be used to gain insight
regarding the shape and size of ice particles aloft in the precipitating system, and the
optical thickness of the cloud at the respective frequencies [Spencer et. al. 1989, Wu and
Weinmann 1989, and Turk and Vivekananden 1994]. It was found by Spencer et. al.
(1989), Wu and Weinmann (1989), and Turk and Vivekananden (1994) through
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
17
theoretical and empirical studies that non-spherical ice and graupel particles present in the
cloud, for which the contribution from the ocean surface is negligible, produced
significant polarization differences at higher microwave frequencies. Any significant
polarization difference at 85 GHz frequency for pixels having a rain rate greater than a
few mm/hr can be assumed to carry information about the shape (deviation from
spherical shapes) and size of ice particles aloft in the precipitating system. Similarly, the
polarization difference at 19 GHz frequency for pixels satisfying minimum rain rate
criterion (except for TYPE-VI), is related to optical thickness of the precipitating system.
The theoretical studies conducted earlier indicate that at lower microwave frequencies the
extinction is primarily dominated by the absorption due to liquid rain depth below the
freezing level. Figure 2 shows the variability in the median values of physically
meaningful quantities, such as RR, S85, TB19V-TB19H, and TB85V-TB85H. The
information obtained from these plots can be summarized and interpreted for each
precipitation class as follows:
•
TYPE-I precipitation is characterized by high RR at the surface; high scattering
index due to significant amounts of ice/graupel aloft; moderate weak values of the
polarization difference at 85 GHz, which could be due to depolarization of the signal
at this frequency due to non-spherical ice/graupel particles present in the
precipitating system [Spencer 1989, Turk and Vivekananden 1994]; and low values
of the polarization difference at 19GHz indicate high optical thickness due to a deep
layer of liquid precipitation.
•
TYPE-II precipitation is defined by a moderate RR and S85 values; high values for
the ratio of S85 and RR which indicates relatively high contributions of ice and
graupel to net precipitation at the surface; almost the same value for the difference of
brightness temperature at 85 GHz indicates the presence of aspherical ice/graupel
particles in this precipitation system; and a relatively high value of the polarization
difference at 19 GHz could be attributed to a comparatively shallow depth of liquid
precipitation.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
18
• TYPE-III precipitation is characterized by a moderate RR and S85; the difference of
brightness temperatures at 85 GHz shows low values indicating less deviation from
sphericity assumption for ice particles aloft in the precipitating system; the
polarization difference at 19 GHz shows higher values compared to TYPE-I and
TYPE-II, which could be assigned to shallower liquid precipitation depths.
• TYPE-IV precipitation indicates that the ratio of RR and S85 is significantly larger
than other precipitating classes. This could be attributed to the dominance of vapor
deposition and aggregation processes in precipitation growth.; the larger brightness
temperature difference at 85 GHz could be attributed to large amounts of aspherical
ice particles in these systems; the relatively large difference in brightness temperature
at 19 GHz indicates a very shallow depth of liquid precipitation and low amount of
liquid water content.
•
TYPE-V is distinctly characterized by zero values of S85 and thus indicates that net
precipitation is basically due to warm cloud microphysics; since TYPE-V
precipitating systems do not have any noticeable contribution from ice particles aloft,
the signatures at 85 GHz were found to be almost unpolarized; the large difference in
brightness temperature at 19 GHz could be explained in terms of shallow depth of
liquid precipitation and relatively lower RR values.
•
TYPE-VI clouds are characterized by zero rain rate values and by significant
contribution from scattering at 85 GHz; highly polarized signatures at 85 and 19 GHz
frequencies indicate significant contributions from the surface.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
19
o
TYPE-I
TYPE-H
TYPE-III
TYPE-IV
TYPE-V
TYPE-VI
Precipitation Types
Figure 2a Mean values of retrieved ocean surface rain rate for each precipitation class.
50
^
&
40
8
30
•o
G
M
c 20
*8
£ io
o
C/3
0
I
TYPE-I
I ■ ■
TYPE-II
TYPE-IU
TYPE-IV
TYPE-V
Precipitation Types
Figure 2b Mean values of retrieved scattering index at 85 GHz for each precipitation class.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
20
m
40
TYPE-I
TYPE-II
TYPE-III
TYPE-IV
TYPE-V
TYPE-VI
Precipitation Types
Figure 2c Mean values of the difference of the polarization at 19 GHz for each precipitation
class.
40
& 30
X
<n
oo
OQ 20
>
m
oo
CQ 10
H
1
TYPE-I
TYPE-II
TYPE-III
TYPE-IV
TYPE-V
TYPE-VI
Precipitation Types
Figure 2d Mean values of the difference of the polarization at 85 GHz for each
precipitation class.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
21
Since S85 and RR carry physical information about the precipitating events that is easily
available from SSM/I, we further examined the variability of these two attributes for all
six precipitation classes. Figure 3 depicts scatter plots between S85 and RR for all six
precipitation classes. Each of the precipitation classes are represented by different colors
in this scatter plot. This plot indicates significant overlapping of information in twodimensional description space among various classes, especially for TYPE-III and TYPEIV precipitation classes. It can be assumed from this scatter plot that simple classification
methods based on these two attributes would not be able to provide good accuracy for the
classification for all four precipitation classes (TYPE-I to TYPE-IV). This point is further
explored in Section S by developing a classification scheme based on S8S and RR scatter
plots and comparing it with two other more sophisticated schemes.
The interactive dynamic graphics program (XGOBI) has been used for the exploration
and visualization of multivariate data. It allows a smooth randomized sequence of a twodimensional projection in order to explore the variability of higher-dimensional data
points. This program was used to explore the variability of nine attributes for six
precipitation classes. Figure 4 shows a two-dimensional projection of nine-dimensional
data points representing, in this study, all six precipitation classes. A two-dimensional
projection of all nine attributes in Figure 4 indicates that these six precipitation classes
can be discriminated on higher dimensional space. It is evident from these scatter plots
that the SSM/I data, for these six precipitation classes, is characterized by some degree of
non-linearity and on higher dimension description space the boundaries among various
classes are more complex than simple hyperplanes.
Although the non-normality of SSM/I data is obvious from earlier scatter plots, formal
Univariate and multivariate normality tests were performed on this data set to examine
the nature of the distribution of the data within each precipitation class. Rencher (1993)
explained a few methods which can be used to examine the multivariate normality of a
data set. It was found that the SSM/I data do deviate significantly from the multivariate
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
normality distribution criterion. The domain analysis performed on SSM/I data set for all
precipitation classes forms the basis for choosing particular types of classifiers for the
classification of precipitation type using SSM/I data.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
23
TYPE-VI
4 0 .0 -
+ ♦ ■ TYPE-V
Rain Rate (mm/Hr)
TYPE-IV
20 .0 -
TYPE-III
TYPE-II
TYPE-I
-1------------------------ 1----50.0
0
50.0
— p—
100.0
Scattering index (K)
Figure 3 Scatter plot of S85 and RR representing all six precipitation classes.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
24
TYPE-VI
TYPE-V
TYPE-IV
v*
*TB8SH
TYPE-III
TYPE-II
p H
♦*+*
\
) * *
+ ■+
v-+
TYPE-I
I
++
♦
♦
*
+
♦ ♦
* ♦ +
♦♦
Figure 4 Two-dimensional projection of multidimensional data. All nine attributes are
used in this display.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
25
4. CLASSIFICATION METHODS
Classification approaches can typically be divided into two types - supervised and
unsupervised approaches. Unsupervised techniques identify features that are separable in
the measurement space; for example, methods using clustering based on K-mean or
minimum variance. In this approach a priori information about the classes is not required
and the main task is to partition the data into subsets in some appropriate way. The
physical interpretation of the resultant classes becomes difficult in unsupervised
classification and, therefore, the cluster classes determined in this way are found to be
sensitive to the characteristics of the observing system and the choice of clustering
variables. On the other hand if a priori information about classes is available then
classification can be done by supervised learning. Supervised techniques attempt to
define classes known to be associated with particular physical processes. In the present
context, examples could include distinctly different dynamical or microphysical
properties of each precipitation type. However, the success of this approach depends upon
the choice o f method (e.g., discriminant analysis, k-nearest neighbors, neural network,
etc.) for a given data set.
Because the SSM/I data do not follow multivariate normality, it is essential to consider
only those classification techniques which are non-parametric and, hence, do not assume
anything about the nature of the distribution of data. It is also necessary for the
classification method to be able to discriminate classes for data exhibiting some degree of
non-linearity in brightness temperature data and some amount of overlapping at lower
dimensional space among various precipitation classes. In this paper two classification
methods based on KNN and NN are explored. These two non-parametric methods have
been used extensively for developing classification schemes for various applications
using remotely sensed data. Brief descriptions of these methods are given below:
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
26
4,L K-Nearest-Neighbors (KNN)
The KNN, also known as the instance based method, represents each instance using a
collection {Y„ Y2,........ Y„} of properties or attributes. Each of these instances also
belongs to one of a fixed set of mutually exclusive classes c„ c2,...c4. Given a training set
Y of c labeled classes, a nearest-neighbor procedure decides that some new observation,
X, belongs to the same category as do its closest neighbors in Y. More precisely, a KNN
method assigns a new pattern, X, to the category to which the plurality of its k closest
neighbors belong. The KNN method can be thought of as estimating the values of the
probabilities of the classes given X. The central questions in this process are as follows:
(1) how should closeness in the description space be measured; (2) which training
attributes should be retained; and (3) how many neighbors should be used when making a
prediction.
KNN procedure is a non-parametric discriminant method which can be used to generate
non-parametric density estimates in each group and to produce a classification criterion
based either on a generalized distance (Mahalanobis or Euclidian) function or posterior
probability of membership in each group to determine proximity. The mathematical
expression for these two criterion are:
D2(X,Y) = (X -Y ) tCOV'‘(X -Y ) - Mahalanobis Distance (1)
or
D2(X,Y) = (X -Y )T(X -Y ) - Euclidian Distance
(2)
Pr(j|X) = nij (X)Priorj / X (mk(X)Priork)
it
(3)
and
where, mk(X) = Proportion of observation in group K in nearest neighbors of X,
and Priork = prior known probability o f occurrence of each precipitation type.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
27
The KNN classifiers are found to perform best when the training data is essentially noise
free. The improvement in performance can be achieved by discarding noisy data. Most
KNN approaches use a fixed number of neighbors when classifying a new instance. The
size of neighborhood is important for good classification performance. If it is too small,
predictions will be unduly sensitive to the pressure of unclassified training instances,
whereas too large a value will cause regions of the description space containing fewer
examples to be merged with surrounding regions. A review of this subject can be found in
Dasarathy (1991).
4.2 Neural-Network fNNl
An artificial neural network (NN) is a network of many processors (“units”), each
possibly having a small amount of local memory. The units are connected by
unidirectional (or bi-directional) communication channels (“connections”), which carry
numeric (as opposed to symbolic) data. The units operate only on their local data and on
inputs they receive via connections. Most neural networks have some sort of “training”
rule whereby the weights of connections are adjusted on the basis of presented patterns.
NN can be grouped into categories on the basis of their connection architecture: (1) feed
forward networks which have no loops, and; (2) recurrent (or feedback) networks, in
which loops occur because of feedback connections. A learning process in the NN
context can be viewed as the problem of updating network architecture and connection
weights so that a network can efficiently perform a specific task. The network usually
must leam the connection weights from available training patterns. Performance is
improved by iteratively updating the weights in the network. An excellent review of NN
methods is provided by Jain and Mao (1996).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
28
In the NN approach to pattern recognition, the neural network operates as a black box
which receives a set of input vectors x (observed signals) and produces responses y, from
its output units ( i = 1.2....1, where 1 depends on the number of classes). The most
important factors to the performance of any NN method are the network architecture and
the learning rule. Figure 5 shows a schematic of the network architecture for a feed
forward NN.
Back-propagation is the most common algorithm used in NN method to update weights.
The back-propagation algorithm is based on error-correction principles. In this supervised
learning paradigm, the network is given a desired output for each input pattern. During
the learning process, the actual output y generated by the network may not equal the
desired output d. The basic principle of back-propagation is to use the error signal to
modify the connection weights so as to gradually reduce this error. The error at the output
layer is given by:
ej- = g '(h j-)[d y -y J-]
(4)
where h,Lrepresents the net output to the i1" unit in the Ith layer, and g' is the derivative of
the activation function g (could be tanh or sigmoid). The errors of the preceding layers of
the network are computed by propagating the errors backward using the following
expression:
=! - g 'O O Z . w f e r 1, fori
= ( L - l) ..........1
(5)
The weights of the network are updated using the error at that level and output of the
preceding level.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
29
W|
*i
INPUT
LAYER
HIDDEN
LAYER
yk
OUTPUT
LAYER
Figure 5 Schematic of network architecture for feed-forward neural network.
The NN method is able to overcome the shortcomings of the traditional statistical
classification method. NN classifiers are non-parametric classifiers and hence do not
require any a priori knowledge about the nature of the distribution of data. The NN
approach is preferable in multiclass analysis where the influence of each class on the
classification is not known and boundaries between classes on description surfaces are
more complex than simple hyperplanes.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
30
5. PERFORMANCE RESULTS
The final data set, comprising nine attributes (seven SSM/I brightness temperatures, RR
and S85), was used for the development of a classifier based on the KNN and NN method
for the classification of the following four precipitation classes: TYPE-I, TYPE-II,
TYPE-III, and TYPE-IV. The class level information about the “testing” and “training”
data set is provided in Table 4. It is evident from this table that all four precipitation types
have almost equal representation in the training and testing data set. It is also important to
note that all classes have been given equal a priori probability in the classification
scheme.
Table 4
Class
1
2
3
4
Class
1
2
3
4
Class Level Information
TRAINING TEST
Frequency
Proportion
723
0.28
635
0.24
567
0.22
675
0.26
TESTING TEST
Frequency
Proportion
312
0.28
264
0.23
255
0.23
299
0.26
Prior Probability
0.25
0.25
0.25
0.25
Prior Probability
0.25
0.25
0.25
0.25
The final performance of the classification schemes was evaluated using the classification
summary table (also known as the confusion matrix). The confusion matrix provides
information about the correct classification and misclassification in the form of a matrix,
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
31
where diagonal elements represent correct classification and non-diagonal elements
represent misclassification. The Heidke skill score was also used to compute the overall
accuracy of the classification method based on its confusion matrix. The Heidke skill
score can be computed as follows:
Skill Score = (NC - E)/(Xtt - E)
where,
W C = i> „
I-1
(6)
£ =
(7)
f*I
■ r.= E 2 > s
<
(8)
i
Before going into the details of results obtained from the KNN and NN methods, it is
useful to have the skill score or confusion matrix from a simple classification scheme
developed using linear intercepts on scatter plots generated from S85 and RR values for
all four precipitation classes. These results provide a basis to compare the results of the
KNN and NN methods and also help us to further believe that, besides S85 and RR
information, other physical properties of the precipitation in the form of the variability in
7 SSM/I brightness temperatures played an important role in the discrimination of these
four precipitation classes. Table 4 summarizes the results for a simple linear classifier for
the overall data set. The rows in the classification summary table represent “true” values
and the “retrieved" values are shown in the columns. The Heidke skill score for the whole
data set is 0.72. It is interesting to note that this simple method yields relatively high
accuracy for TYPE-I (91%) but the accuracy for other precipitation classes drops
significantly with the lowest accuracy observed for TYPE-III (58%). It is also observed
that TYPE-m and TYPE-IV precipitation classes show significant overlapping in the
classification distribution summary table.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
32
Table 4 Confusion Matrix - Simple Intercept Method
Classes
1
2
3
4
1
2
3
4
914
115
0
0
94
737
173
26
0
40
481
152
0
7
168
796
Table 5 summarizes the results of KNN classifier for training and testing data sets
respectively. The Heidke skill score for the training data set is 0.92 and it is found to
decrease by about 1 percent for testing data set. This is an improvement of about 20
percent in skill score over the simple linear classifier. An overall improvement in
accuracy for all four precipitation classes is observed. It is evident from the testing data
set that the KNN method yields high accuracy for TYPE-I (97%), TYPE-III (97%), and
TYPE-IV (98%) but this method was not able to achieve relatively high accuracy for
TYPE-II (86%). All misclassified data points of TYPE-I precipitation class were found to
get classified as a TYPE-II precipitation type. In case of TYPE-II precipitation class, a
majority of misclassified observations were classified as TYPE-III precipitation class.
Misclassified data points for TYPE-m and TYPE-IV precipitation classes were found
distributed among TYTE-H and TYPE-IV in case of TYPE-III precipitation class and
TYPE-II and TYPE-III for TYPE-IV precipitation class.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
33
Table 5 Confusion Matrix - KNN Method
1
690
15
1
0
1
302
4
0
0
TRAINING DATASET
2
3
33
540
13
6
0
75
550
17
TESTING DATASET
2
3
10
228
8
1
0
27
245
7
4
0
5
0
652
4
0
5
2
291
The data set with z-normalized attributes (normalization was done using the overall mean
and standard deviations of the data set) and unnormalized attributes for each precipitation
class was used separately to develop a NN classifier. Figure 6 shows that normalized data
yields a significantly better result (about 97% accuracy) compared to the result obtained
with the unnormalized data set (about 75% accuracy). Normalization of the data provides
equal weight to all classes in the data set and thus helps the NN classifier to map
boundaries among the four precipitation classes more accurately. Table 6 summarizes the
classification accuracy for each precipitation type for training and testing data sets. The
Heidke skill scores for the training and testing data sets indicate a substantial
improvement (about 4%) in overall classification accuracy over the KNN method. The
distribution of classification accuracy among all four classes is: 99% accuracy observed
in the case of TYPE-I; the lowest accuracy ( 93%) is observed for TYPE-II precipitation
class; a high classification accuracy of 97% is found for TYPE-III precipitation class; and
TYPE-IV is also classified with high accuracy (98%). The distribution of misclassified
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
34
observations for each precipitation class tends to follow the same trend as observed in the
case of the KNN method.
Classification Accuracy Neural Network
100
A Normalized Variables
4-»
ua
au
(X
_c
>*
Unnormalized Variables
§o
u
<
1
2
3
4
5
6
7
8
9
10
11
Number of Iterations (Multiplied by 50)
Figure 6
Classification accuracy of neural network method for normalized and
unnormalized data sets.
Finally, the NN classifier has been chosen on the basis of its performance based on the
testing data set. Although both classification methods, KNN and NN, are non-parametric
and are expected to handle the non-linearity of the remote sensing data far more
efficiently than a parametric method (such as linear or quadratic discriminant methods), it
was found that the NN classifier was able to incorporate the non-linearity of the SSM/I
data more accurately. The classification method based on neural networks also effectively
handles the mapping of description surfaces, which are more complex in nature and are
defined by irregular boundaries for this problem.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 6 Confusion Matrix - NN Method
Classes
1
1
2
3
4
710
11
1
0
Classes
1
1
2
3
4
310
12
0
0
TRAINING DATASET
2
3
13
571
3
2
0
53
560
9
TESTING DATASET
2
3
2
244
3
0
0
5
248
5
4
0
0
4
664
4
0
3
5
294
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
36
6. MICROPHYSICAL RETRIEVALS OF PRECIPITATION CLASSES
The microphysical properties of five precipitation classes were retrieved from the mean
brightness temperatures of all 7 SSM/I channels, in order to examine and demonstrate the
differences in the microphysical properties such as Ice Water Path (IWP), Snow Water
Path (SWP), Graupel Water Path (GWP), Rain Liquid Water Path (RLWP), and Cloud
Liquid Water Path (CLWP) among these precipitation classes. This section is divided into
four parts: (1) the first part deals with the description of the rain-cloud model used in the
retrievals; (2) the second sub-section briefly describes the forward model used to compute
brightness temperatures at SSM/I frequencies; (3) the third part deals with the inversion
scheme; and (4) the last sub-section describes the retrievals of microphysical properties
and examines the reasonableness of these retrievals.
6.1 RAIN CLOUD MODEL
The rain cloud model used in this study is developed with the objective of creating
simplified profiles of snow, graupel, liquid rain, cloud liquid water content using a
minimum set o f user defined input parameters. The basic features of the rain cloud model
are a parameterized scheme for liquid water (cloud liquid and rain water) and ice phase
(snow/ice and graupel). The schematic of the rain cloud model is shown in Figure 7. This
model requires 11 user-specified parameters to generate profiles of temperature, water
vapor, cloud liquid water, snow/ice, graupel and liquid rain. Table 7 lists the parameters
that are used to describe the rain cloud model.
Cloud water is taken to be zero at and above the Z jOP and at and below ZBASE. Within
cloud, it follows a quadratic profile which is determined from Z j0P, ZBASE, and LWP. The
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
37
snow precipitation profile is bi-linear. It is allowed to increase linearly from ZCTOP to Z^OP
while it is allowed to increase/decrease linearly from Z j0P to ZL. The formation of snow
precipitation is controlled by two pre-defmed parameters a , and a 2. Snow precipitation is
assumed to be zero below ZL and above ZCT0P. Graupel formation is assumed to be zero
below ZL and above Zt0P, and it follows a linear profile with zero value at Z j0? and
maximum value at ZL. Graupel formation is considered to be a function of cloud liquid
water, snow precipitation at Zr0P, the depth of the layer between Zt0P and ZL, and a
constant p which determines the rate of graupel formation in this layer.
Z-rop
Zl
Z
(II)
(I)
^STOP
STOP
so
S O _________® _
7
t 'L
G ^SO /O
RO
©
©
b a se
ro
R = S(ZL), CONST (
^7BASE
©
O
©
RO
©
r
(HI)
‘ -STOP
Z top
(IV)
sO
gOs
Z sT O P
S O
O/O
Z top
^BASE
G & S = CONST
R O
e)
Z top
©
(5)
- Value increases with Z
gO
sO
/ ' l)
Z base
zL
G & S = CONST
©
■O' Value decreases with Z
Figure 7 The schematic of rain cloud model under four different atmospheric conditions
based on ZBASE, ZL, and Z j0P heights. (I) ZLis above ZBASEand below ZtOP (All types of
hydrometeors are present in this case); (II) ZLis considered above Zr0P (No graupel); (III)
ZL is assumed below ZBASE ; (TV) ZL is at or below the surface (All precipitation is in
snow or graupel form - no liquid rain).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
38
The liquid precipitation profile is assumed to be bi-linear. It is allowed to increase
linearly from ZL to ZbASE and then it decreases linearly with a constant evaporation rate
from ZBASE to surface. The production of liquid precipitation depends upon the total
amount of precipitation at ZL (graupel and snow), the depth o f layer between ZL and
ZBASE, and a constant y which determines the growth of liquid precipitation in this layer.
The user-defined parameters a , a 2 P, and y are assumed to parameterize cloud
microphysical processes in this simplified one-dimensional rain-cloud model. The vapor
deposition process in the top “snow-only” layer represented by a „ and a 2 is assumed to
be parameterizing any of these physical processes: vapor deposition or aggregation of
snow particles, or melting of snow particles in a layer where snow and supercooled water
exists together. The graupel formation through accretion processes is considered to be
parameterized by the free parameter P in the layer above ZL and below ZtOP. The liquid
rain formation by collision/coalescence process is assumed to be represented by y in the
layer below ZLand above ZbASE.
Snow/cloud ice particle size distribution is assumed to follow a modified gamma
distribution of the form:
N ( r ) = a r s“ e x p ( - b r )
(9)
where a and b are computed from:(l) the snow precipitation; (2) the user-specified mode
volume snow particle radius; and (3) the user-specified value of Sa. Graupel particle and
rain drop size distribution is assumed to follow the well known Marshall-Palmer
distribution.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
39
Table 7 List of parameters used in rain-cloud model
Input to Rain-Cloud Model
O011)
ZL(km)
Zjgp (km)
^STOP O®1)
LWP (kg/m2)
r0(mm)
Sa (dimensionless)
a, (mm/Hr m)
^BASE
a 2 (mm/Hr m)
(3 (mm/Hr Kg m'2)
y (mm/Hr Kg m‘2)
Description
Cloud base
Freezing level
Cloud top
Snow layer top
Liquid water path
Mode volume particle radius
Snow particles size distribution
parameterizes vapor deposition
process in “snow-only” layer
parameterizes vapor deposition
process in “snow and graupel” layer
parameterizes accretion/riming
process in “snow and graupel” layer
parameterizes collision/coalescence
process below ZLand above ZBASE
6.2 FORWARD MODEL
The radiative transfer code used for simulation at microwave frequencies is that of Evans
and Stephens (1991). The observed SSM/I signatures can be compared with brightness
temperatures predicted by polarized-multistream radiative transfer calculations for a plane
parallel atmosphere containing spherical hydrometeors. The shape of hydrometeors are
considered to be spherical. The polarized 4x4 phase matrices for these hydrometeors were
calculated as described by Turk and Vivekananden (1995). The radiative transfer model is
applied to the temperature, humidity, and hydrometeor profiles provided by the rain cloud
model to compute brightness temperatures at given microwave frequencies emerging
from the top of the atmosphere.
The model allows for a detailed structure of the atmosphere. A separate treatment is given
to different classes of hydrometeors (such as snow/ice, graupel, and raindrop etc.). In this
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
40
model simulation, the atmosphere is divided into plane parallel layers. For each layer
mean microphysical properties are determined using the profiles generated by the raincloud model. Bulk radiative parameters are obtained by integrating over various classes
of hydrometeors in each region. The Rayleigh approximation is used to model cloud
water.
6.3 INVERSION SCHEME
Figure 8 schematically depicts the approach adopted to retrieve microphysical properties
of each precipitation type. This is an iterative approach of finding a solution
(microphysical variables), using the minimum variance method, for a given set of
brightness temperatures corresponding to a particular precipitation class - a well known
inverse problem. This is an optimal estimation problem for which a priori information is
of a statistical nature. The main principle behind this approach is to find a linear predictor
D, such that the variance of the error in the estimate is a minimum. A solution is sought
of the form:
x= Dy
(10)
where x is a vector representing microphysical variables to be retrieved, and y is a set of
observed variables (SSM/I TB’s). The matrix D for which the error variance in x is a
minimum may be derived either empirically from a large set of coincident satellite and
microphysical profile measurements, or theoretically from knowledge of the statistical
covariance S, of the microphysical variables, sensor channel noise covariance Se, and the
sensitivity matrix K (sensitivity of SSM/I channels to microphysical variables used in
rain-cloud model). In the latter case, optimal D is given by
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
41
D = S*K^(&rK^ + Se)",
(11)
The vector y in this case is mean brightness temperature of all seven SSM/I channels.
Table -8 provides these TBs with associated standard deviations for all six precipitating
classes. These brightness temperatures are used as target brightness temperatures in
Figure 8. Our objective is to find a set of values x, using the above defined inversion
scheme, for which the computed brightness temperatures are close enough (within a pre­
defined value) to the target brightness temperatures of a given precipitation class. This
final set of vector x will be treated as the retrieved microphysical properties of that
precipitation class.
Table 8 Mean and Standard Deviations of 7 Channel SSM/I Brightness
Precipitation Class
TB19V
TB19H
TB22V
TB37V
TB37H
TB85V
TB85H
TYPE-I
264.2
(8.8)
252.6
(15.2)
270.6
(3.2)
260.1
(5.1)
254.9
(7.9)
228.3
(13.8)
221.0
(14.1)
TYPE-II
231.2
(11.5)
201.1
(19.5)
249.1
(10.1)
250.9
(7.9)
238.6
(14.2)
243.7
(8.4)
236.2
(10.1)
TYPE-m
233.8
(7.2)
199.1
(12.2)
258.5
(8.2)
253.9
(8.7)
237.1
(14.6)
258.6
(4.4)
253.6
(5.6)
TYPE-IV
198.5
(6.5)
143.3
(11.5)
215.7
(11.5)
225.2
(7-0)
187.4
(14.3)
249.8
(5.0)
235.3
(8.9)
TYPE-V
236.9
(7.7)
203.9
(11.7)
264.2
(6.8)
257.8
(7.3)
240.4
(12.8)
277.3
(2.6)
275.8
(2.8)
TYPE-VI
190.7
(8.9)
127.1
(13.5)
211.3
(16.1)
209.8
(6.2)
153.4
(10.1)
239.8
(11.4)
208.9
(17.1)
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
42
Since the problem of retrieving microphysical variables (x) from 7 SSM/I brightness
temperatures (y) is ill-posed, the solution is under-constrained. In this situation, the real
challenge is not only to find a physically plausible solution from an infinite number of
solutions available, but also to examine its uniqueness and noise sensitivity. These
aspects of the inversion method are studied in depth by Petty and Gautam (1998). It is
also important to note that problem under consideration is highly non-linear and we have
tried to tackle this non-linearity by finding a solution iteratively by using a gradient
descent approach. The initial guess of input parameters to rain-cloud model is
intelligently chosen to represent approximate values of microphysical properties for a
given precipitation class (from earlier published average values of cloud microphysical
variables for that precipitation class). The change in x (dx) is computed by a minimum
variance method using the difference between the target brightness temperature and
current brightness temperature computed from the theoretical radiative transfer model for
a given set of microphysical parameters. This computed dx is added to the initial guess
values to move the iteration closer to the solution. This step is repeated until computed
brightness temperatures for a given raining cloud are close enough to the target brightness
temperatures of a given precipitation class. The input parameters to rain-cloud model are
constrained to change within physically plausible values.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
43
"DATASET FOR
GENERATE FIRST GUESS PARAMETERS
EACH
PRECIPITATION
CLASS
INPUT PARAMETERS TO
RAIN-CLOUD MODEL
GENERATE MEAN
SSM/I TB VECTOR
FOR EACH
PRECIPITATION CLASS
GENERATE VERTICAL PROFILES
OF HYDROMETEORS
INPUT PROFILES TO RADIATIVE
TRANSFER MODEL
GENERATE TB VECTOR FOR SSM/I
CHANNELS :CURRENT-TB
TARGET-TB
ADD dx TO INITIAL
VALUES OF INPUT
PARAMETERS
COMP ARECURRENT-TB WITH
TARGET-TB
FF <= EPSILO
COMPUTE dx USING
MINIMUM VARIANCE
METHOD
STOP
Figure 8 Schematic of the method employed for the retrieval of microphysical properties
of a given class of precipitation.
6.4 MICROPHYSICAL PROFILES
In this study our objective is to demonstrate the relative changes in the vertically
integrated values of microphysical quantities for all precipitating classes. It is not
desirable to attach too much importance to the absolute values of the retrieved
microphysical variables themselves as there can be a number of choices for microphysical
properties that can yield approximately close solutions to a given set of brightness
temperatures. However, we have tried to constrain the retrievals by only allowing each
input parameter to the rain-cloud model to vary within physically plausible upper and
lower limits of each parameter. Figure 9 represents the microphysical profiles of
hydrometeors for TYPE-I to TYPE-V precipitation classes. These profiles were retrieved
from the mean SSM/I brightness temperatures obtained from the data set of each of these
five precipitation classes. The retrieval for TYPE-VI precipitation class was not
attempted as this class was characterized by a cloud type which is almost transparent to
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
44
all SSM/I frequencies. A dominant contribution of ocean surface to overall signature at
SSM/I frequencies for this precipitation class was expected. Under this situation the
contribution of clouds to microwave signature at lower SSM/I frequencies will be at a
minimum compared to other atmospheric contributions to the microwave signature. Yet
the depression at 85 GHz is strong enough to allow pixels to be selected for this
precipitation class. The remaining five precipitation classes were characterized by clouds
producing significant rain and, hence, it can be assumed that in these cases the
contribution of raining clouds to SSM/I channels will be sufficiently larger than ocean
surface contributions to SSM/I channels. The effects of non-raining pixels were also
reduced by not considering boundary pixels. However, these pixels may include clouds
with varying RR and microphysical properties. The impact of extreme values of RR and
microphysical properties on a pixel corresponding to a particular precipitation class was
minimized by assuming spatial homogeneity in SSM/I images for a given precipitation
event. These retrieved microphysical profiles can be assumed to represent the typical or at
least physically plausible profiles of hydrometeor in each precipitation class. It is noticed
that the brightness temperature data (Table-8) of five precipitation classes (TYPE-I to
TYPE-V) show a certain degree of variability within each precipitation class, which
could be related to the variability in microphysical properties corresponding to each
precipitation class. A detailed study is conducted by Petty and Gautam (1998) to examine
the variability associated with spatially extensive tropical precipitation using empirical
orthogonal function analysis on selected SSM/I TB data sets. The microphysical
properties for the mean brightness temperature and the first three eigenvectors were
retrieved. These three eigenvectors explain more than 95 percent of the variability in the
data set. The variance associated with each of these three eigenvectors is examined and
explained in terms of the variability in gross features of the retrieved microphysical
properties. The variability in microphysical properties of each precipitating class can be
studied separately by collecting a sufficiently large SSM/I data set and applying the
methodology developed by Petty and Gautam (1998) to that data set. However, in this
study we have not attempted to examine the variability in brightness temperature data for
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
45
each precipitation class. It is essential not to attach too much importance to the absolute
values of the retrieved microphysical profiles because of the uncertainties associated
with the nature of the inverse problem and significant overlapping of information in the
seven SSM/I brightness temperatures. However, the gross features of the retrievals and
their relative changes for each precipitation class can always be used to document the
distinctly different microphysical properties of subjectively defined classes. The salient
gross features of the retrieved microphysical profiles for each precipitation class (shown
in figures 9 a to e) are also summarized in Table 9 and are explained as follows:
TYPE-I: As stated earlier, the TYPE-I precipitation class was chosen to represent deep
convection (MCS, squall line, and tropical cyclone etc.) in the tropical region. The
retrieved microphysical profile of this precipitation class (figure 9a) is characterized by a
large atmospheric liquid water path and ice water path. Deep layers of atmospheric liquid
water above and below freezing level contribute to a significant growth of graupel and
liquid rain in this precipitation class. The average characteristics of retrieved
microphysical profiles of hydrometeors such as large amounts of atmospheric liquid
water, graupel formed in the layer of supercooled liquid water, and sufficiently large
liquid precipitation grown by collision/coalescence process, is in general agreement with
the mean characteristics of hydrometeor profiles of cumulonimbus clouds prepared from
the data obtained from 90 research aircraft [Chapter 8, Houze (1993)]. It can be assumed
that the cloud microphysics of these clouds is basically dominated by riming and
accretion processes, above freezing level collision and coalescence processes below the
freezing level.
TYPE-II: This precipitation class was chosen to represent convective precipitation in
extratropical systems in mid-latitude and high latitude regions. A large atmospheric liquid
water path and ice water path were also retrieved for this precipitation class (figure 9b).
The atmospheric liquid water path values were found to be relatively low when compared
with the TYPE-I precipitation class. The retrieved microphysical profiles indicate that
more than fifty percent of net precipitation above the cloud base comes from graupel
growth in this precipitation class. The freezing level exists at a relatively lower heights
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
46
and cloud below freezing level is characterized by a noticeable liquid rain growth. The
precipitation growth above the freezing level is primarily due to the vapor deposition
process in the topmost layer of clouds (where no supercooled water exists) and from
riming and accretion processes in the region where supercooled water and snow droplets
exist together. The riming and accretion processes are believed to contribute to a
significant growth of graupel which is a dominant contributor to the overall precipitation
growth in these clouds. The precipitation growth due to collision and coalescence below
the freezing level can also be important for this precipitation class, but it is not as
dominant as it is in the case of the TYPE-I precipitation class. A typical microphysical
structure o f clouds in the convective region of an extratropical cyclone [Chapter 11,
Houze (1993), Chapter 6, Wallace and Hobbs (1990)] shows similar broad characteristics
with retrieved mean microphysical profiles for this precipitation class such as a relatively
low freezing level, dominant growth of precipitation above the freezing level by vapor
deposition and accretion processes, and a large atmospheric liquid water paths.
TYPE-III: TYPE-HI precipitation class represents stratiform regions of organized
convection or widespread shallow convective precipitation in the tropical region. The
microphysical profile of this precipitation class (figure 9c) was characterized by a smaller
atmospheric liquid water path and relatively low ice water path. Deep layers of
atmospheric liquid water contribute to a significant growth of liquid rain below freezing
level in this precipitation class. The clouds above freezing level are basically dominated
by snow growth. No significant graupel formation was observed in the retrieved profiles.
The retrieved microphysical profiles of this precipitation class indicate a dominant
contribution from collision and coalescence processes to overall precipitation below the
freezing level. The vapor deposition process is found to be a primary source of
precipitation growth above the freezing level. The general characteristics of mean
hydrometeor profiles retrieved for this precipitation class is in broad agreement with
characteristics of known average microphysical profiles (such as dominant growth of
precipitation by collision/coalescence processes, non-significant growth of precipitation
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
47
due to graupel particle, etc.) of cumulus congestus or clouds in stratiform portions of
organized convection [Houze 1993].
TYPE-IV: The stratiform precipitation emanating from extratropical cyclone or
widespread stratocumulus clouds were chosen to represent this precipitation class. The
retrieved microphysical profiles o f this precipitation class were characterized by low
cloud liquid water path and a relatively high amount of atmospheric ice water path (figure
9d). More than fifty percent of net precipitation above cloud base comes from snow
growth in this precipitation class. The freezing level exists at relatively lower heights and
clouds below the freezing level are characterized by a low amount of liquid rain growth.
The clouds above freezing level characterized by a large contribution of snow in the
retrieved profiles. The retrieved profile of hydrometeors show that the precipitation
growth is mainly due to vapor deposition of snow particles above freezing level.
TYPE-V: The deep layers of atmospheric liquid water with a large atmospheric liquid
water path contribute to the growth of liquid rain below the freezing level in this
precipitation class (figure 9e). The clouds above the freezing level are basically
dominated by cloud water particles and some liquid rain growth. This precipitation class
was defined by an absence of detectable scattering by ice and, therefore, no graupel and
snow formation was observed in the retrieved microphysical profiles. This precipitation
class was assumed to represent clouds with relatively warm tops. The precipitation
growth in these types of cloud are primarily dominated by warm microphysics [Houze
(1993), Rogers and Yau (1990)]. The mean microphysical properties retrieved for this
precipitation class indicates only liquid rain growth and an absence of any graupel or
snow. It can be assumed that collision and coalescence processes lead to precipitation
growth in this precipitation class.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
48
Table 9 Summarized microphysical properties of precipitation classes
Precipitation Class
TYPE-I
TYPE-H
TY PE-m
TYPE-IV
TYPE-V
SWP GWP IWP
1.31
0.81
0.30
2.59
0.0
0.5 7
0.59
0.0 1
0.00
0.0
1.88
1.40
0.31
2.59
0.0
RLWP CLWP TLWP
1.7
0.8
0.5
0.2
0.42
3.30
2.45
1.43
0.50
1.65
5.0
3.25
1.93
0.70
2.07
R0
0.12
0.06
0.09
0.06
0.0
zl
:SFCRR
4.0
2.1
3.8
2.0
4.5
6.2
4.6
1.5
0.8
1.2
SWP : Snow water path in kg/m2; GWP : Graupel water path in kg/m2;
RLWP : Rain liquid water path in kg/m2; CLWP : Cloud liquid water path in kg/m2;
IWP : Ice water path in kg/m2; TLWP : Total Liquid water path in kg/m2;
RO : Snow peak radius in mm; Z L : Freezing Level in km ;
S F C R R : Rain rate in mm/Hr
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
49
12r
TYPE-I
LW P = 3.30 kg/mA2
RW P = 1.66 kg/mA2
SW P = 1.31 kg/mA2
G W P - 0.57 kg/m A2
SNOW R0 = 0.12 mm
I
10 -t
Height (Km)
I
1
I
I
8h i
tiraupel m in i Hr)
Snow (mm/Hr)
Liquid Rain (m m /H r)
CLOUD LIQUID TOP
TN
I \
•H
I
I
I
I
\
FREEZING LEVEL
CLOUD BASE
10
Figure 9a Retrieved microphysical profiles o f hydrometeors for TYPE-I precipitation
class.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
50
12r
TY PE-H
LW P = 2.45 kg/mA2
RW P = 0.77 kg/m A2
SW P = 0 .8 1 kg/mA2
G W P = 0.59 kg/m A2
SN O W R0 = 0.05 mm
Height (Km)
10 -
IU .II
■(
I
W
I'_
m
)
(jraupol ( mm Hr)
S n o w (m m /H r)
L iquid R ain (m m /H r)
Y
\
\
I
1
'hi
4NL
CLOUD LIQUID TOP
/
/
t
FREEZING LEVEL
/
CLOUD BASE
10
Figure 9b Retrieved microphysical profiles of hydrometeors for TYPE-H precipitation
class
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
51
12r
10h
TYPE-01
LW P = 1.43 kg/m A2
RW P = 0.49 kg/m A2
SWP = 0.30 kg/m A2
GW P = 0.01 kg/m A2
SNOW R0 = 0.09 m m
1(1 ' ) ' (
i
'A
f L!
Ill
1 I
Ciraupcl ( mm-Hr)
S n o w (m m /H r)
L iquid R ain (m m /H r)
Height (Km)
CLOUD LIQUID TOP
FREEZING LEVEL
CLOUD BASE
/
/
10
Figure 9c Retrieved microphysical profiles of hydrometeors for TYPE-in precipitation
class.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
52
12
10
Height (Km)
I M I I H ! A \ I - lit
TYPE-IV
LW P = 0.50 kg/mA2
RW P = 0.18 Kg/mA2
SWP = 2.59 kg/mA2
GW P = 0.001 kg/mA2
SNOW R0 = 0.08 mm
<I
Graupel ( m m Hr)
Snow (mm/Hr)
Liquid Rain (m m /H r)
\
h\
\
\
\
\
h
\
CLOUD LIQUID TOP
\
V',
-
/
FREEZING LEVEL
I
CLOUD BASE
/
0
t
0
0
0
1
1
_______ L ____________ 1---------------------
10
Figure 9d Retrieved microphysical profiles of hydrometeors for TYPE-IV precipitation
class.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
53
12
—
TY PE-V
LW P = 1.65 kg/m A2
RW P = 0.41 kg/m A2
IW P = 0.0 kg/m A2
G W P = 0.0 kg/m A2
SN O W R0 = 0.0 mm
—
i o n •( | \V i 2 m ; i
( i r a u p d ( mm Hr)
S n o w (m m /H r)
L iquid Rain (m m /H r)
Height (Km)
10
T
T
CLOUD LIQUID TOP
FREEZING LEVEL
CLOUD BASE
8
10
Figure 9e Retrieved microphysical profiles of hydrometeors for TYPE-V precipitation
class.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
54
7. APPLICATION OF THE OBJECTIVE CLASSIFICATION SCHEME TO
THE INDENPENDENT DATA SET
This section is devoted to demonstrating the meaningfulness of the objective
classification scheme and also to show consistency in the logic used in subjectively
defining precipitation classes. The evaluation of the qualitative consistency of the
objective classification scheme was done using independent data collected from SSM/I
for all raining pixels ( RR greater then 0.5 mm/hr) for four representative months
(January, May, July, and October) and for the year, 1988. The geographical distribution
of data points of all four months are shown in Figure 10. It is evident from figure 10 that
all major ocean areas are covered with a sufficiently large number of data points.
A qualitative evaluation of the first five (TYPE-I to TYPE-V) retrieved precipitation
classes is done by examining the zonal sum of occurrence of each precipitation class and,
also, zonally averaged percentage of each precipitation class to the total precipitation
amount. Figures 11 and 12 show the distribution of the zonal sum of occurrence and
fraction (in percent) of each precipitation class to total precipitation . It is evident from
Figure 11 that the latitudinal distribution of frequency and percentage of each
precipitation class to the overall precipitation is consistent with earlier assumptions about
the geographical location of the precipitation types. The salient features of zonal
frequency are: the frequency of the TYPE-I precipitation class peaks around 10° N and
10° S with almost no presence beyond 40° in southern and northern latitudes; the highest
frequency for the TYPE-II precipitation class is observed at 45° N and 35° S, with some
presence in tropical latitudes; the primary peak for the TYPE-m precipitation class is
observed around 10° N and secondary peaks occur around 40° in northern and southern
latitudes; the frequency distribution for the TYPE-IV precipitation class shows high
values around 55° north and south o f the equator, with almost no presence in the tropical
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
55
latitudes; the frequency distribution plot for the TYPE-V precipitation class shows high
values around 10° N and then relatively small values between 45° N and 45° S. As was
stated earlier in Section 3, TYPE-I was predominantly a tropical type of precipitation,
TYPE-III and TYPE-V were basically assumed to be more frequent in the tropical and
midlatitude regions, TYPE-II was considered to be a midlatitude type precipitation, and
TYPE-IV was considered to be frequent in the extratropical regions. It is interesting to
observe from Figure 12 that TYPE-I and TYPE-II (convective type precipitation)
contribute more than 60% of the total precipitation in the tropical belt (30° N to 30° S).
The stratiform type precipitation (TYPE-III and TYPE-IV)contributes more than 60% in
midlatitudes and high latitudes. The contribution of TYPE-V precipitation to the total
precipitation is about 5 % in the regions between 30° N and 30° S.
Two precipitating systems, one in the tropical region and the other in the extratropical
region, have also been analyzed to further strengthen our confidence in the objective
classification scheme. The data from these two precipitating systems were not included in
the data set used for the development of the objective classification scheme. Figures 13 a
and b show images for RR and S85 for tropical cyclone “Kelly” observed on 14 October,
1987. The characteristic structure of a cloud system associated with a tropical cyclone can
be seen in the form of spiral bands of high RR and S85; a trailing region of moderate RR
and S85 is also evident in these images. The eye of the cyclone can also be seen in both
images, with almost no rain left to the high RR band in figure 13a and very low S85 left
to very high S85 band in figure 13b. Figure 13c depicts the image generated for the
classification of precipitation types obtained from the NN classifier by feeding SSM/I
data (9 attributes as discussed earlier). It is interesting to note that the TYPE-I
precipitating class matches very well with the spiral band of high S85 and RR in figures
13 a and b, and that a trailing region of the tropical system is classified as a TYPE-m
precipitation class. It is also noteworthy that a significant portion of raining pixels in the
tropical cyclone, which lies outside the core of the system, is classified as TYPE-V. The
TYPE-VI precipitation class is not evident for tropical cyclone “Kelly” near the outer
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
56
spiral band of high S85 values. However, some portion between the strong convection
and a trailing stratiform region of this system is classified as a TYPE-II precipitation
class. On the other hand, Figures 14 a and b show the RR and S85 for the extratropical
system observed on 11 September, 1987. The frontal band structure, which is typical of
extratropical cyclones, can be clearly seen in the images o f RR and S85. The convective
region (high RR and S85) of this extratropical system is classified as a TYPE-II
precipitation class. The trailing region of low RR and relatively high S85 is classified as
TYPE-IV. The area characterized by relatively low RR and comparatively high S85
values near the convective region of the extratropical system is classified as a TYPE-III
precipitation class. In general, results of the classification of these two systems are
consistent with the earlier definition of the subjectively defined classes.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
57
Geographical distribution of points selected from four months of SSM/I data
.
h z
V -
, ;,V=. ■
7',/'
•• .
_
i*
>o
7
:' *
' ' r\. - " ' Z
>• \
rii
Iif
Figure 10 Geographical distribution of data points over the global ocean using four
months (January, May, July and October) of SSM/I data for the year 1988.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
58
2000
1500
TYPE-I
_ 4 _ TYPE-II
4 TYPE-IH
_ 4 _ TYPE-IV
TYPE-V
uc0
g
1
u
1000
o
"c3
c
o
N
500
o
On
o
o
NO
O
CO
o
NO
Latitude (Deg)
Figure 11 Latitudinal distribution of zonally averaged occurrences of TYPE-I to TYPEV precipitation classes.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
59
100
TYPE-I
TYPE-II
A TYPE-III
TYPE-IV
-s—TYPE-V
u
a
o
vo
! i 'll
vn o
^
ci
o
o
VO
Latitude (Deg)
Figure 12 Latitudinal distribution of zonally averaged percentage of TYPE-I to TYPE-V
precipitation classes.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
60
SSM/I IMAGE OF TROPICAL CYCLONE KELLY
14 October, 19871
Rain Rate (nmm/Hr)
52
Figure 13a Rain rate (mm/Hr) retrieved from SSM/I using the Petty (1994a,b) rain rate
algorithm for Tropical Cyclone Kelly observed on 14th October, 1987.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
61
SSM/I IMAGE OF TROPICAL CYCLONE KELLY
October, 198”^
Figure 13b Scattering index (K) retrieved from SSM/I using the Petty (1994a,b) rain rate
algorithm for Tropical Cyclone Kelly observed on 14th October, 1987.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
62
SSM/I IMAGE OF TROPICAL CYCLONE KELLY
14 Octobei
TYPE-VI
TYPE-V
TYPE-IV
TYPE-III
TYPE-U
TYPE-I
I
Figure 13c Objective classification of precipitation types from the neural network
classifier using SSM/I data for Tropical Cyclone Kelly observed on 14th October, 1987.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
63
SSM/I IMAGE OF AN EXTRATROPICAL CYCLONE
jl September, 1987
Figure 14a Rain rate (mm/Hr) retrieved from SSM/I using the Petty (1994a,b) rain rate
algorithm for an extratropical cyclone observed on 11th September, 1987.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
64
SSM/I IMAGE OF AN EXTRATROPICAL CYCLONE
i
00
CO
xu
~a
Jg
00
c
•C
Is
CO
11‘September, 1987
*
Figure 14b Scattering index (K) retrieved from SSM/I using Petty (1994a,b) rain rate
algorithm for an extratropical cyclone observed on 11th September, 1987.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
65
SSM/I IMAGE OF EXTRATROPICAL CYCLONE
TYPE-IV
TYPE-I
II September, 1987
Figure 14c Objective classification of precipitation types from the neural network
classifier using SSM/I data for an extratropical cyclone observed on 11th September,
1987.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
66
8. CONCLUSIONS
It is concluded, on the basis o f extensive analysis of the SSM/I data selected for more
than 100 precipitating events, that passive microwave signatures at SSM/I frequencies
carries sufficient information to discriminate precipitating events into six broad classes
characterized by significantly different microphysical properties. It is also evident from
the analysis that classifiers based on the neural network classification methods yield
better results than classifiers based on the K nearest neighbor approach and the simple
linear intercept method based on S85 and RR scatter plot. An overall accuracy of about
97 percent is achieved for the classification of four precipitation types using the NN
classifier method compared to the overall accuracy of 93 percent in the case of K nearest
neighbors methods. The microphysical properties of five precipitation classes (TYPE-I to
TYPE-V) retrieved from mean brightness temperatures show a broad resemblance with
the characteristics of microphysical properties of conventional clouds. The relative
changes in the retrieved microphysical properties of five precipitation classes appear
reasonable in terms of the general physical characteristics of the clouds associated with
these precipitation classes.
The application of an objective classification scheme to the independent data set shows
promising results and strengthens our confidence in the scheme and also demonstrates the
meaningfulness of the subjectively defined precipitation classes. In Part II, the global
characteristics (spatial and temporal distribution) of all these precipitation classes will be
investigated over the ocean using SSM/I data for longer time period and will also be
compared with precipitation climatologies of earlier research work. The spatial and
temporal distribution of precipitation classes, which are found to be characterized by
distinctly different microphysical properties, would help us gain a better insight about
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
67
the distribution of convective, stratiform, and warm precipitation over the vast oceanic
region.
With TRMM (Tropical Rainfall Measuring Mission) operating and providing data in
various electromagnetic spectra and other advanced microwave and VIS/IR sensors to be
launched in next few years, it is proposed to further investigate the classification of
precipitating and non-precipitating cloud on the basis of their microphysical and
dynamical properties using complimentary information obtained from VIS/IR and
microwave sensors onboard these satellites. TRMM with a variety of sensors like TMI,
VISSR ( Visible Infrared self scanning radiometer), and PR (Precipitation Radar) has
been providing a wealth of information about the tropical precipitation. It would be
interesting to apply the methodology developed in this study to TMI data and also using
textural and spatial information obtained from VIS/IR to develop a robust classification
scheme for tropical precipitation. The availability of PR data from TRMM can serve as
in-situ information to indirectly validate some microphysical properties of classes, such
as freezing level.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
68
REFERENCES
Adler, R.F., Hwa-Young, M.Y., N. Prasad, W.K. Tao, and Joanne Simpson, 1991:
Microwave simulations of tropical rainfall system with a three-dimensional cloud
model. J. Appl. Meteor ., 30,924-953.
Arkin, P.A., and P.E. Ardanuy, 1989: Estimating climatic-scale precipitation from space:
A review. J. Climate, 2, 1229-1238.
Arkin, P.A., and B.N. Meisner,1987: The relationship between large-scale convective
rainfall and cloud cover over the western hemisphere during 1982-1984. Mon.
Wea. Rev., 115, 51-74.
Bankert, R.L., 1994: Cloud classification of AVHRR imagery in maritime regions using a
probabilistic neural network. J. A p p l Meteor., 33,909-918.
Barret, E.C., J. Dodge, M. Goodman, J. Janowik, and E. Smith, 1994: The First WetNet
Intercompariosn Project (PIP-1), Remote Sens. Rev., 11,49-60.
Baum, A.B., V. Tovinkere, J. Titlow, and R. M. Welch, 1997: Automated cloud
classification of global AVHRR data using a Fuzzy logic approach. J. Appl.
Meteor, 36, 1519-1540.
Chahine, M.T., 1992: The hydrological cycle and its influence on climate. Nature, 359,
373-379.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
69
Dasarathy, B.V., 1991: Nearest Neighbors Pattern Classification Techniques. IEEE
Computer Society Press.
Evans, K.F., and G.L. Stephens, 1991: A new polarized atmospheric radiative transfer
model. J. Quant. Spectroscopy and Radiative Transfer , 46,413-423.
Gerald, L., 1988: Automated recognition technique of oceanic cloud patterns. Part I:
Methodology and application of cloud climatology. J. Climate, 1,20-39.
Houze, R.A., Jr., 1993: Cloud dynamics. Academic Press, San Diego,573 p.
Hollinger, J.P. P.G. Lo, R. Savage, and J. Pierce, 1987: Special Sensor MicrowaveI
Imager u ser’s Guide, naval Research Laboratory, 120p.
Inoue, T., 1987: A cloud type classification with NOAA 7 split -window measurements.
J. Geophys. Res., 92,3991-4000.
Jain, A.K., and Jianchang Mao, 1996: Artificial Neural Networks: A Tutorial, Computer.
31-44.
Janowiak, J.E., A.F. Krueger, and P.A. Arkin, 1985: Atlas of outgoing longwave
radiation
derived from NOAA satellite data. NOAA Atlas No. 6, U.S. Dept, o f Commerce.
Kummerow, C.D., R.A. Mack, and I. M. Hakkarinen, 1989: A self-consistency approach
to improve microwave rainfall estimates from space. J. Appl. Meteor., 28, 869884.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
70
Kummerrow, C.D., and L. Giglio, 1994a: A passive microwave technique for estimating
rainfall and vertical structure information from space. Part-I Algorithm
description. J. Appl. Meteor., 33,3-18.
Kummerrow, C.D., and L. Giglio, 1994a: A passive microwave technique for estimating
rainfall and vertical structure information from space. Part-I Application to SSM/I
data. J. Appl. Meteor ., 33, 19-34.
Lau, K.M., and P.H. Chan, 1986: The El Nino Southern Oscillation and 40-50 day
oscillation: A new perspective. Bull Am. Meteor. Soc., 67, 533-540.
McGaughey, G., E.J. Zipser, R.W. Spencer, and R. E. Hood, 1996: High resolution
passive microwave observations of the convective systems over the tropical
pacific oceanic. J. Appl. Meteor., 35,1921-1948.
McGaughey, G., and E. J. Zipser, 1996: Passive microwave observations of the stratiform
regions of two tropical oceanic mesoscale convective systems. J. Appl. Meteor.,
35, 1949-1962.
Meehl, G. A., 1992: A coupled air-sea biennial mechanism in the tropical Indian and
Pacific regions: role of the ocean. J. Climate, 6,31-41.
Mugnai, A., E.A. Smith, and G. J. Tripoli, 1993: Foundations for statistical-physical
precipitation retrieval from passive microwave satellite measurements. Part-II:
Emission source and generalized weighting function properties of a time
dependent cloud-radiation model. J. Appl. Meteor., 32, 17-39.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
71
Nakazawa, T., 1996: Tropical super clusters within intraseasonal variations over the
western pacific. J. Meteor. Soc., 66, 823-839.
Petty, G.W., 1994a: Physical retrievals of over ocean rain rate from multichannel
microwave imagery. Part-I: Theoretical characteristics of normalized polarization
and scattering indices. Meteor. Atmos. Phys., 54, 79-100.
Petty, G.W., 1994b: Physical retrievals of over ocean rain rate from multichannel
microwave imagery. Part-I: Algorithm implementation. Meteor. Atmos. Phys., 54,
101- 122 .
Petty, G.W., and W.F. Krajewski, 1996: Satellite estimation of precipitation over land.
Hydro. Sciences, 41(4).
Petty, G.W.,1995: The status of satellite-based rainfall estimation of precipitation over
land Remote Sens. Environ., 51, 125-137.
Petty, G.W., 1998: On the prevalence of precipitation from warm-topped clouds over
eastern Asia and the western pacific. J. Climate (in press).
Petty, G.W., and N. Gautam, 1998: Multichannel microwave signatures and effective
hydrometeor structure of spatially extensive tropical precipitation. J. Appl.
Meteor, (submitted)
Ramanathan, V., and R.D. Cess, E.F. Harrison, P. Minnis, B.R. Barkstorm, E. Ahmad,
and D. Hartmann, 1989: Cloud-radiative forcing and climate: Results from the
earth radiation budget experiment. Science, 243,57-63.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
72
Ramanathan, V., and W. Collins, 1991: Thermodynamic regulation of the ocean warming
by cirrus clouds deduced from observations of 1987 El-Nino. Nature, 351,27-32.
Rogers, K, and W., Yau, 1990: Cloud Microphysics. Butterworth and Heinemann
Publication. 290p.
Sheu, R. S., J.A. Curry, and Guosheng Liu, 1997: Vertical stratification o f tropical cloud
properties as determined from satellite. J. Geophys. Res., 102,4231-4245.
Smith, E.A., A. Mugnai, H.J. Cooper, G.J. Tripoli, and X. Xiang, 1992: Foundations for
statistical-physical precipitation retrieval from passive microwave satellite
measurements. Part-I: Brightness temperature properties of a time dependent
cloud radiation model. J. Appl. Meteor., 31, 506-531.
Spencer, R.W., H.W., Goodman, and R.E. Hood, 1989: Precipitation retrieval over land
and ocean with SSM /I: Identification and characteristics of the scattering signal.
J. Atmos. Ocean. Tech., 6,254-273.
Spencer, R.W., 1993: Global oceanic precipitation from MSU during 1979-1991 and
comparison to other climatologies. J. Climate, 6,1301-1326.
Stephans, G.L., and T.J. Greenwald, 1991: The earth’s radiation budget and its relation to
atmospheric hydrology. Part-II: Observations of cloud effects. J. Geophys. Res.,
96,15325-15340.
Trenberth, K.E., and G.W. Branstator, 1992: Issues in establishing causes of the 1988
drought over north America. J. Climate, 5,159-172.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
73
Turk, J. and J. Vivekanandan, 1994: Effects of hyrdometeor shape and orientation upon
passive microwave brightness temperature measurements, 1994 Specialists
Meeting on Microwave Radiometery and Remote Sensing o f the Environment,
U niversitadi Roma " Tor Verata", Rome, Italy, 14-17 Feb., 1994.
Wallace, J.M., and P.V. Hobbs, 1996: Atmospheric Science - An Introductory Survey.
Academic Press. 461 p.
Wang, J.R., Zhan, J., and P. Racette, 1997: Storm-associated microwave radiometric
signatures in the frequency range of 90-220 GHz. J. Atmos. And Ocean. Tech, 14,
13-31.
WCRP Informal Report No. 6/1996, May 1996: International workshop on research
issues in the identification of precipitation type and rates in global data sets (
Washington, D.C., 6-8 December, 1995).
Weare, B.C., 1987: relationships between monthly precipitation and SST variations in the
tropical region. Mon We. Rev., 115,2687-2698.
Wilheit, T.T., R. Adler, S. Avery, E. Barret, P. Bauer, W.Berg, A. Chang, J. Ferriday, N.
Grody, S. Goodman, C. Kidd, D. Kniveton, C. Kummerrow, A. Mugnai, W.
Olson, G.W. Petty, A. Shibata, and E. Smith, 1994: Algorithms for the retrieval of
rainfall from passive microwave measurements. Remote Sens. Rev., 11, 163-194.
Wielicki, B. A., R.D. Cess, M.D. King, D.A. Randall, and E. F. Harrison, 1995: Misson
to planet earth :Role of clouds and radiation in climate. Bull. Amer. Meteor. Soc.,
76,11,2125-2153.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
74
Wu, R., and J.A. Weinman, 1984: Microwave radiances from precipitating clouds
containing aspherical ice, combined phase and liquid hydrometeors. J. Geophys.
Res., 89, 7170-7178.
Zhang, C., 1993: Large-scale variability of atmospheric deep convection in relation to sea
surface temperature in the tropics. J. o f Climate, 6,1898-1913.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
75
PART II
GLOBAL CHARACTERISTICS OF OCEANIC PRECIPITATION FROM SSM/I
DATA
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
76
ABSTRACT
This part examines the global characteristics of the oceanic precipitation derived from
SSM/I data in terms of seasonal and regional distributions of precipitation frequency
among the various precipitation types obtained from the objective classification scheme
discussed in Part I. This part also compares precipitation frequencies derived using
synoptic weather reports from Comprehensive Ocean Atmosphere Data Sets. The
seasonal and regional distribution of precipitation amount and the fractional contribution
of each precipitation classes to the total precipitation amount is also examined.
It is observed that the global characteristics of each precipitation class show consistency
with known climatological patterns of precipitation. The comparison of SSM/I derived
precipitation frequencies with COADS derived frequencies yields remarkably good
results.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
77
1. INTRODUCTION
Precipitation is one of the most difficult processes to model and predict because of its
highly variable nature in space and time. Cloud microphysics, thermodynamics of the
atmosphere and oceans control the spatial and temporal distribution of the rainfall.
Prediction of rainfall amount from various numerical models show significant amount of
uncertainty, partly because the initial distribution of rain is not well known. Satellites
offer a great deal of hope of adequately estimating global rainfall and studying its spatial
and temporal distribution over the vast oceanic region. Since the launch of the first
satellites in the early sixties, significant progress has been made on the estimation of
rainfall from visible/infrared and microwave sensors onboard various satellites.
Nevertheless, it is recognized by the scientific community that more efforts are needed to:
(1) further enhance the understanding of spatial and temporal distribution of rainfall over
the globe; (2) intercompare the rainfall estimates from various data sources and
techniques; and (3) develop a reliable satellite-derived precipitation climatology.
Efforts to develop precipitation climatologies started well before the satellite era after
recognizing the necessity to have some form of global distribution of the precipitation
amount and its frequency over the ocean. These studies involved indirect estimates based
on extrapolation from nearby continents and/or observations of sea surface salinity.
Tucker (1961) has provided excellent review of these studies. More direct estimates of
rainfall from shipboard present-weather observations were prepared by Tucker (1961). He
developed precipitation climatology using 5 years of ship data over the north Atlantic.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
78
These efforts were further extended by many authors [Reed 1979; Reed and Elliott 1979;
Dorman and Bourke 1979, 1981] and methodologies were improved to incorporate
corrections for systematic regional differences in the average precipitation intensity
associated with various reports from the present weather code. Jaeger (1983) developed a
precipitation climatology using more than 124 years (1854-1978) of shipboard presentweather reports. He also improved the technique to estimate monthly precipitation by
using precipitation frequency. Legates and Willmott (1991) developed a precipitation
climatology over the ocean using a synthesis of techniques and data from Dorman and
Bourke (1979, 1981) and Jaeger (1983).
These precipitation climatologies developed using shipboard present-weather reports
suffered from the uncertainties due to human intervention in reporting and also due to the
qualitative nature of synoptic precipitation codes. These studies were further hampered by
the poor sampling density over most of the ocean, except in major shipping lanes. Mainly
restricted by the above limitations of conventional reports over the ocean, an interest in
utilizing satellite data for extracting surface rainfall rate and its properties has grown
tremendously in the last three decades. Significant effort has been put into developing a
reasonable estimate of rainfall over a variety of temporal and spatial scales using data
from visible/infrared and passive microwave sensors on board various geostationary and
polar orbiting satellites [Arkin and Xie 1994; Barrett et. al. 1994a].
Early efforts in the development of a satellite derived precipitation climatology used the
GOES Precipitation Index (GPI), developed by Richards and Arkin (1981), for both polar
and geostationary orbiting satellites [Arkin and Ardanuy 1989, Janowiak and Arkin
1990]. These precipitation climatologies were able to provide a reasonably credible
global distribution of the rainfall amount and captured large scale precipitation features
such as the Inter-Tropical Convergence Zone (ITCZ), South Atlantic Convergence Zone
(SACZ), and the most wet and dry regions with good accuracy. Spencer (1994)
conducted an extensive study to develop a global oceanic precipitation climatology using
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
79
Microwave Sounding Unit (MSU) data from the polar orbiting TIROS-N series for the
years, 1979-1991. Comparison of the MSU derived oceanic precipitation climatology
with those of Jaeger (1983), Legates and Wilmont (1991), and the GPI of Janowiak and
Arkin (1990) indicates several differences. The precipitation climatology developed by
Jaeger(1983) did not show pronounced extratropical storm tracks, nor the intensity of the
ITCZ in the eastern Pacific and western Atlantic region. Legates and Wilmott (1991)
global distribution of precipitation was able to capture the features missed by Jaeger
(1983) but with low intensity. Significant differences were also observed in the
precipitation climatologies developed using MSU and GPI. It was found that GPI shows
higher values of rainfall over the eastern Indian ocean than MSU, and much less than the
MSU over the tropical eastern Pacific, tropical western Atlantic, and the western Pacific
extratropical storm track. The reason for these differences could be attributed to the
inherent nature of the rain rate algorithms and the interaction of electromagnetic radiation
with rain at these wavelengths.
These efforts to develop rainfall climatologies using satellite data were propelled by the
hope that only satellites can provide a relatively fine temporal and spatial coverage over
the vast oceanic areas at a reasonable cost. These studies also indicated that concerted
efforts are needed to calibrate and validate satellite derived rainfall products derived from
variety of algorithms. Many organized efforts under the Global Precipitation Climatology
Project (GPCP) were initiated by the World Meteorological Organization (WMO) to
intercompare satellite estimates from a wide variety of rain-rate algorithms and satellite
data (such as passive microwave data from polar orbiting satellites or visible infrared data
from polar/geostationary satellites). A detailed intercomparison was conducted under
many algorithm intercomparison projects [like Algorithms Intercomparison Project
(AIP)-1,2,3 and Precipitation Intercomparison Projects (PEP)-1,2,3] to examine the
performance of various classes of rain rate retrieval algorithms using VIS/IR and MW
data. The details about the outcome of some of these projects can be found in review
articles by Wilheit et. al. (1994), and Barrett et. al. (1993).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
80
These projects highlighted the need for a detailed global climatology of the frequency of
occurrence of precipitation and its general physical properties to be compared with the
global estimates of rainfall derived from satellites. Petty (1995) developed a detailed
global climatology using synoptic weather information from the Comprehensive Ocean
and Atmosphere Data Sets (COADS) to document some aspects of global macro-physical
properties of oceanic precipitation and limitations of satellite remote sensing techniques
to sense some of these characteristics. This study provided an independent source of
information about oceanic precipitation (frequency, phase, character and intensities) from
surface observations to compare with satellite derived oceanic rainfall climatologies.
A World Climate Research Programme (WCRP) report [WCRP Informal Report No. 6,
1996] recognized the need to further explore the possibility of the classification of
precipitation type according to dynamical/microphysical criteria using current sensor
data. The necessity of developing climatologies of new products (such as occurrence of
crude precipitation types classified from current sensor data, or contribution of a
particular precipitation type to monthly/seasonal total rainfall over a grid) was also
emphasized. It was also recognized that the potential of contributing algorithms to
overlook or underestimate certain types of precipitation ( e.g., warm precipitation or
highly convective rain etc.) should also be carefully studied and documented.
The objective of this paper is to develop a global climatology of precipitation amount and
its frequency over the ocean for six different precipitation types. These six precipitation
types were retrieved from objective classification scheme based on neural network
classifier using two years (1991-92) of SSM/I data. These six precipitation classes were
chosen to represent microphysically different rain clouds. The primary goal of this part of
the study is to document the gross features of the global and seasonal characteristics of
these six precipitation classes. The qualitative consistency of the objective classification
scheme will also be examined by looking at the climatology of these precipitation types
over the global ocean. Some of the statistics generated using SSM/I data for six
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
81
precipitation classes will be compared with global climatology o f precipitation derived
using the COADS data by Petty (1995). The contribution of stratiform and convective
type precipitation to total precipitation amounts in a given grid box will also be examined
over the ocean on the basis of the contribution of six precipitation classes to total
precipitation amounts and their closeness to conventional precipitation types.
The description of SSM/I and details about the Defense Meteorological Satellite Program
(DMSP) satellite orbital passes, Petty’s (1994a,b) over ocean rain rate algorithm, and
processing of data are given in Section 2. The characteristics of the six precipitation
classes are briefly summarized in Section 3. Section 4 deals with the results of the global
characteristics of the distribution of valid raining SSM/I pixels and their distribution
among all six precipitation classes, the regional distribution of fraction of valid raining
pixels, and pixels belonging to each precipitation class over the season. Some of these
results will be compared with Petty’s (1995) results based on Comprehensive Ocean
Atmosphere Data Set (COADS). This section will also deal with the regional variability
of total precipitation amount over the season and the seasonal and regional distributions
of each of these six precipitation classes to total precipitation amount. Conclusions and
future work are given in Section 5.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
82
2. DATA AND METHODOLOGY
2.1 SSM/I
The first SSM/I was launched on June 1987 aboard the Defense Meteorological Satellite
Program (DMSP) F-8 satellite. Since then 5 more DMSP satellites with SSM/I have been
successfully launched. These satellites were placed into a sun-synchronous near polar
orbit at an altitude of 833 km with an inclination of 98.8° and an orbital period of 102
minute. This orbital period corresponds to 14.12 complete orbits per 24 hour period. Its
equator crossing local times are at nearly 0600 and 1800 corresponding to ascending
(northbound) and descending (southbound) portions of the orbit respectively. The antenna
of the SSM/I consists of an offset parabolic dish of diameter 61 by 66 cm. The angular
resolution of the antenna is limited primarily by diffraction and is thus proportional to the
wavelength. Hence, the SSM/I is a 7 channel passive microwave radiometer operating at
19.35, 22.235, 37, and 85.5 GHz and at both vertical and horizontal polarization except
22.235 GHz, where only the vertically polarized radiation is measured. The spatial
resolution of these channels varies from 43x69 km2 at 19.35 GHz to 13x15 km2 at 85.5
GHz.
The scan geometry of SSM/I is shown in Figure 15. The entire SSM/I antenna/receiver
assembly rotates about the spacecraft vertical axis, reflecting radiation from a spacecraftrelative boresight angle of 44.8° into a feedhom which is upwardly directed and centered
on this axis. Although SMM/I instrument rotates continuously through 360° with a period
o f 1.9 second, only an arc segment of 102.4° aft of the spacecraft and centered on the
subtrack is actually used to observe the earth. This corresponds to a surface data swath
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
83
which is about 1400 km wide. Geographic coverage by the SSM/I during a typical 24 hr
period is indicated in Figure 16. Further details about SSM/I instrument can be found in
Hollinger et. al. (1987).
vat
Figure 15 Scan geometry of the Special Sensor Microwave/Imager (after Hollinger et.
al. 1987).
Figure 16 Global coverage of the SSM/I in a 24-hour period (after Hollinger et.
al.1987).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
84
The SSM/I data obtained from F-10 satellite for the year 1991 and F-l 1 satellite for the
year 1992 is analyzed for this study.
2.2 RAIN RATE ALGORITHM
Petty’s (1994a,b) over water rain-rate algorithm is used to retrieve rain rate and scattering
index (S85). This is an iterative, physical-inversion based algorithm that seeks a high
resolution rain rate field that is physically consistent with normalized polarization
differences at 19 and 37 GHz. A high resolution “no rain/possible rain” mask is
determined based on liquid water estimates compiled from the normalized polarization
difference at 85 GHz. A first guess rain rate is obtained from a scattering index computed
at 85 GHz. A final rain rate is obtained by doing forward calculations of normalized
polarization, based on a simple analytic relationship between depolarization of the sea
surface signal and rain cloud transmittance. These normalized polarization differences at
19 and 37 GHz are then compared with observed polarizations differences at these
frequencies, and adjustment made based on these differences.
This algorithm has been submitted to many algorithm intercomparison projects (PIP1,2,3) and it has emerged as one of the best algorithms among the variety of algorithms
submitted to these projects in terms of its correlation and bias with data obtained from
atoll stations. The comparison of monthly mean rainfall derived using this algorithm with
monthly mean rainfall of atoll station yields correlation of 0.74. The details about the
results of this intercomparison can be seen in PIP-3 Intercomparison Results (1996).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
85
2.3 PROCESSING
The quality checked seven channel SSM/I brightness temperatures are supplied to Petty’s
over ocean rain-rate algorithm to compute surface rain rate (RR) and scattering index at
85 GHz (S85). The neural network (NN) classifier is then applied to seven brightness
temperatures, RR, and S85 to retrieve the precipitation types. The simple statistics such
as total number of valid pixels, total number of raining pixels and corresponding
precipitation amount, total number of pixels and precipitation amount for each
precipitation class were stored for each month over a 5 degree latitude/longitude grid for
two years (1991-92). The coarse grid size of 5 degrees is chosen, since our objective is to
examine and document the large scale features of global precipitation characteristics of
each precipitation type. It is difficult with one satellite to have a sufficient number of data
points on smaller grids to represent large scale features of global precipitation
characteristics with a statistically significant number of data points over the grid on
monthly time scales. These monthly statistics were then used to generate seasonal and
annual averages over the global oceans. Figure 17 is the schematic representing the steps
involved in the processing of SSM/I data and finally generating global statistics on
various temporal scales.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
86
QUALITY CHECKED
7 SSM/I TBs
NN CLASSIFIER
(SIX PRECIPITATION
CLASSES)
PETTY’s RAIN-RATE
ALGORITHM
(S85 and RR)
MONTHLY STATISTICS
OF TOTAL PRECIPITATION
AMOUNT AND OCCURENCE
MONTHLY STATISTICS
OF TOTAL PRECIPITATION
AMOUNT AND OCCURRENCE
FOR EACH PRECIPITATION
CLASS
GENERATE GLOBAL, SEASONAL, ANNUAL STATISTICS BASED ON
MONTHLY PRODUCTS
Figure 17 Schematic representing the steps involved in the processing of SSM/I data
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
87
3. PRECIPITATION CLASSES
As mentioned in Part I, six precipitation classes were subjectively defined and chosen to
represent various types of precipitating system with different microphysical properties. It
was also shown that the objective classification scheme developed on neural network
method was able to classify these six subjectively-defined precipitation classes with high
accuracy. The inversion scheme based on the minimum variance approach was also
developed to retrieve gross microphysical properties of these six precipitation types. A
gross summary of microphysical properties of these six precipitation classes and their
chosen closeness to conventional precipitating system is shown in Table-1. On the basis
of the broad microphysical characteristics retrieved for each precipitation class, it can be
assumed that TYPE-I and TYPE-II precipitation classes are more close to convective type
precipitation, and TYPE-in and TYPE-IV can be roughly assumed to represent stratiform
precipitation. This assumption is solely made on the basis of the gross features of
retrieved microphysical properties of these precipitation classes. Spatial characteristics
(such as homogeneity or non-homogeneity of microphysical properties) of these
precipitation types are not taken into consideration for generating global statistics. The
spatial characteristics are one of the very important considerations for discriminating
convective and stratiform precipitation. Nevertheless, the broad features of the global
distribution of the ratio of the contributions made by convective and stratiform
precipitation, can be seen with the above defined precipitation classes based on their
gross microphysical properties.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
88
Table 1 Summarized Description of Precipitation Gasses
General Characteristics
Conventional Gouds
TYPE-I
Large liquid water path
Large rain water path
Large ice water path
Larger ice particles
Deep clouds
Tropical deep convective
systems
(Convective portion of
Goud clusters, Tropical
cyclones etc.)
TYPE-n
Moderate liquid water path
Moderate rain water path
Large ice water path
Larger ice particles
Moderately deep clouds
Midlatitude convective
systems
(Extratropical Cyclones)
TYPE-01
Moderate to low liquid water path
Low rain water path
Low ice water path
No large ice particles
Shallow clouds
Straiform region of
of the tropical convective
system
TYPE-IV
Very Low liquid water path
Low rain water path
Large ice water path
No largeice particles
Shallow clouds
Straiform region of
of the extratropical cyclone
TYPErV
Moderate to low liquid water path
Low rain water path
No snow
No large ice particles
Shallow clouds
warm-top clouds
(with non-zero rain
rate)
TYPE-VI
No rain
(Non-Precipitation Gass)
Gouds with large ice
particles
(left over of convective
system or
widespread cirrus)
Precipitation Gasses
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
89
4. RESULTS
4.1 GLOBAL STATISTICS
Two years of SSM/I data over the ocean were analyzed and the final global statistics were
presented for the Southern and Northern Hemisphere separately. Figure 18 shows total
counts of valid SSM/I pixels (raining and non-raining) and raining pixels over the
Northern and Southern Hemisphere. The interesting features of this plot are: (1) high
values of total number of valid pixels for the year 1992 when compared with the values
for the year 1991 - this was mainly due to significant number of days of SSM/I data either
found completely missing or partial missing (approximately 40-50% records found
missing in the tape when compared with number of records for the year 1992)for the year
1991; (2) significantly larger counts of valid pixels were observed over the Southern
Hemisphere than over the Northern hemisphere - primarily because significantly large
areas are covered by oceans in the Southern Hemisphere compared to Northern
Hemisphere. The SMM/I data used for this study have come from two different SSM/I
instruments (with exactly the same specifications) on board two different DMSP satellites
(F10 and FI 1) with almost the same specifications.
Figure 19 depicts the variability in the fraction of SSM/I pixels reporting rain each month
over the southern and northern ocean. It is evident from this figure that the fraction of
raining pixels during the months from January to May is higher over the southern ocean
than over the northern ocean. The northern ocean is characterized by high values of
raining pixels during the months from June to December. The presence of monsoons,
tropical cyclones and hurricanes, and extratropical storm tracks during the second half of
the year could attribute to the high values of raining pixels over the northern oceans. The
high values of fraction of SSM/I pixels reporting rain over the southern ocean during first
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
90
half of the year could be related to the presence of the extratropical South Pacific
Convergence Zone (SPCZ), the South Atlantic Convergence Zone (SACZ), and the
persistent transient weather patterns over the polar regions of the southern ocean.
SSM/I SAMPLE OVER THE GLOBAL OCEAN
1.0E+8
| 1.0E+7
o
U
>*
5
§ l.OE+6
S
Valid Pixels - SH
Raining Pixels - SH
Valid Pixels - NH
Raining Pixels - NH
1.0E+5
3
5
7
1991
9
11
13
15
Months
17
19
21
23
1992
Figure 18 Monthly counts of valid SSM/I pixels and raining pixels over the ocean in the
southern and northern oceans.
The fraction of pixels corresponding to each precipitation class with respect to raining
pixels is shown in Figure 20 and Figure 21 over the southern and northern oceans. The
striking features of these figures are: (1) the percentage occurrence of TYPE-IV
precipitation type is significantly larger than other precipitation types over the southern
ocean; (2) TYPE-I and TYPE-II precipitation classes found to be out of phase with each
other over both oceans; (3) TYPE-IV precipitation class found to be in out of phase with
TYPE-in and TYPE-V precipitation classes over both oceans; and (4) the annual cycle
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
91
can easily be noticed in all precipitation types over both ocean and the occurrence of
TYPE-III, TYPE-IV and TYPE-V precipitation types is relatively larger than the
occurrence of TYPE-I and TYPE-II precipitation types. The presence of monsoons, a
very active Inter Tropical Convergence Zone (ITCZ), and late Fall and early Spring
presence of extratropical cyclones over the northern ocean could lead to a dominant semi­
annual cycle in all precipitation types. By contrast, the annual cycle observed in the
occurrence of all precipitation types over the southern ocean could be due to the SPCZ,
the SACZ, the South African Convergence Zone. The regional and seasonal variability of
all precipitation types over the globe will be analyzed in detail in the next few sections.
GLOBAL PRECIP REPORT FRACTION
0.12
Percent Raining Pixels - NH
Percent Raining Pixels - SH
oe
O 0.08
2
0.06 ..
0.04
1
3
5
7
1991
9
11
13 15
Months
17
19
1992
21
Figure 19 Fraction of SSM/I pixels reporting rain over the ocean.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
23
92
FRACTION OCCURENCE OF PRECIP TYPES I TO V - SH
0.8
TYPE-!
TYPE-III
TYPE-V
0.6
oc
TYPE-II
TYPE-IV
0.4
§
0.2
1
3
5
7
9
1991
11
13
15
Months
17
19
21
23
1992
Figure 20 Fraction of the monthly counts of each precipitation class over the ocean in
the southern ocean.
FRACTION OCCURRENCE OF PRECIP TYPES I TO V - NH
TYPE-I
TYPE-II
TYPE-IV
TYPE-III
TYPE-V
0.4
oc
o
2
S 0.2
1
3
5
1991
7
9
11
13
15
17
19
21
23
Months
Figure 21 Fraction of the monthly counts of each precipitation class over the ocean in
the northern ocean.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
NOTE TO USERS
Page(s) missing in number only; text follows.
Microfilmed as received.
93
This reproduction is the best copy available.
UMI
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
94
4.2-_REGIONAL FREQUENCIES OF PRECIPITATION
This subsection investigates global maps of the absolute frequencies of precipitation and
the frequencies of each precipitation type with respect to raining pixels except for TYPEVI (non-raining cloud with sufficient 85 GHz scattering by ice particles), which is
represented as an absolute frequency. These frequencies were tabulated for each 5° grid
over the four seasons [March-April-May (MAM); June-July-August (JJA); SeptemberOctober-November (SON); and December-January-February (DJF)] using monthly
counts for valid SSM/I pixels, raining pixels, and pixels for each precipitation class. The
large scale features of the regional and seasonal distribution of precipitation frequency of
all raining pixels, and pixels corresponding to each precipitation class individually will be
documented in this section.
Figure 22 shows the global distribution of an absolute frequency of raining pixels seen
from SSM/I for all four seasons. These frequencies range from less than 1% in certain
portions of the subtropical dry zones to more than 16% in the portions of the ITCZ, the
SPCZ, monsoon regions, and the area of extratropical storm tracks in the Atlantic Ocean
and the Western Pacific Ocean. The monsoonal cycle is clearly evident in the Arabian
Sea and the Bay of Bengal. The increase in frequency is also observed over the northern
Pacific and Atlantic Oceans, and the SPCZ region during northern winter months. A
broad belt of high frequency (16%) is noticed in southern ocean throughout the year
except for JJA season, which could be attributed to transient weather systems over this
region.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
95
t± w n - y a r w t t « fL O C A L P B B C fft3 » i)to M A M tm H W 2 >
fcHNlTAKT HUM p fL O C A L r K B C g |5 d N )fc r llA (l» > .IW n
OOI
002
OM
OOO
0.1*
«) tNOTAKTPIlOa ofLOCAL n tflC V (3 * * ) Or ION (I M M OK)
OOI
i t tNTTANT m m .
JOB
MB
0.01
MB
I2DB
002
MOB
tO
000
I30W 12DW
0 00
WW
010
MW
MW
MB
008
001
MB
002
004
OOO
LOCAL PRBCff (3 * 0 ) t o O ff ( t U M M )
13*
002
1906
ID
004
130W I2DW
001
0 to
WW
MW
MW
0 10
Figure 22 Fraction of SSM/I pixels indicating rain greater than 0.5 mm/Hr over the
ocean surface.
The global distribution of the frequency of TYPE-I precipitation type (presented as a ratio
of total number of counts corresponding to a particular precipitation type to the total
number of raining pixels) for all four seasons is shown in Figure 23. It is quite interesting
to observe that no significant occurrence of TYPE-I precipitation is seen beyond the
tropical belt (30°N - 30°S). This observed feature is consistent with the assumption made
in the subjective classification of this precipitation type. The presence of high frequency
of TYPE-I precipitation class is noticed in the portions of the ITCZ in the Indian Ocean
and Eastern Pacific Ocean, the SPCZ in south central Pacific Ocean, and near the
Mexican coast in the western Pacific Ocean. The existence of the well-known tropical
cloud clusters and monsoons over the Indian and eastern Pacific regions can explain the
observed pattern of high frequency of TYPE-I precipitation type. The migration of high
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
96
frequency portion can also be seen, with its maximum northward migration in MAM and
JJA seasons and maximum southward migration in DJF season. This migration of the
high frequency region of TYPE-I precipitation type can be related to the well known
migration of the ITCZ due to the seasonal march of thermal equator. The high frequency
of TYPE-I precipitation is also observed in the Gulf of Mexico during JJA and SON,
which can be attributed to the hurricane season during these months over this region.
» )T Y T B - i/A U .r* a n p ( 3 * in f a ri;A ( tw M « a
a) TYFE-I / ALLPIIfiCIFO *■.> fer MAM (IM M M 3)
St
90S
OOI
006
012
O il
<) W E - I ALL WUtClP(3 d»| ) fer tOH (1991-1992)
0 24
026
006
012
0.11
OJA
026
I MB
ID
IMW
001
006
012
O il
SI TYFfi-l - A U . PBBCff (3 * 1 ) ter O ff (1991-1993
M
0.01
I2D6
«
001
«
128
006
IM
012
D
IMW
&0W
MW
»W
OOW
MW
0 24
I20W I2DW
O il
90W
024
026
Figure 23 Fraction of raining SSM/I pixels reporting TYPE-I precipitation class.
Figure 24 depicts the global distribution of the frequency of TYPE-II precipitation type
over the ocean for all four seasons. The salient features of this figure are: (1) absence of
any significant occurrence of this precipitation type in the tropical regions over most of
the oceans, except some intrusion of this precipitation type in the tropical belt seen over
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
97
the Atlantic and western Pacific Ocean; (2) winter months of southern and northern ocean
are characterized by high frequency of occurrence in the mid-latitudinal belt; (3) winter
months over the northern ocean shows portion of high values of occurrence in the regions
of extratropical storm belts; and (4) the region of high frequency of occurrence of this
precipitation type over south ocean lies in the area of the extratropical portion of the
SPCZ or the South African Convergence Zone or the South Atlantic Convergence Zone.
It is postulated that TYPE-II precipitation type represents convective precipitation
emanating from extratropical storms and the geographical and seasonal distribution of
this precipitation type appears to be more or less consistent with our hypothesis.
•1 rVTt-ll ALL PRECIP <3
ME
«06
90E
1206
] toa MAM (»W |.tW 2l
I ME
ID
IJOW I20W
0.0S
0.06
012
a il
0 rypE -n / a l l m b c t (3 4—-) e> k h o w t . n w )
ME
ME
aoi
90S
1206
ao6
IM S
ai2
ID
0 24
IJOW I30W
a it
»W
024
oOW
MW
0.36
SOW
(SOW
om
ME
606
ME
1206
aOI
006
<0TY PS-n/A iJ. m a n y
WW
ME
606
aot
906
1206
aot
IMS
ID
JJOW IJ0W
012
O lt
d ip ( m i - i w n
IME
012
ID
l» W
an
SOW
024
I20W
024
SOW
60W
MW
OM
60W
MW
03*
Figure 24 Fraction of raining SSM/I pixels reporting TYPE-II precipitation class.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
98
The global distribution of the frequency of occurrence of TYPE-HI precipitation type is
shown in Figure 25. The eastern Pacific Ocean, Indian Ocean, and South China Sea show
the migration of the high frequency belt of TYPE-III precipitation type over the course of
the year. The maximum northward migration is noticed during JJA and SON, while
maximum southward migration of this belt is observed during DJF and MAM. High
values of frequency are observed throughout the year in the central Indian, eastern
Pacific, South China Sea, and Gulf of Mexico regions. Main convergence regions of the
Indian Ocean (ITCZ and South African Convergence Zone), Pacific Ocean (ITCZ, South
Pacific Convergence Zone - Tropical and Extratropical portion), and Atlantic Ocean
(SACZ) are characterized by relatively large values of frequency of occurrence of this
type of precipitation class. The region off the United States coast in the north central
Atlantic Ocean also shows high values for this precipitation type throughout the year,
being most widespread region during SON. The lowest values of frequency of occurrence
for this precipitation type are noticed in the northwestern Pacific and extreme North
Atlantic Ocean in DJF and MAM. The northwestern Pacific Ocean show higher
frequency off the California coast in JJA and SON seasons. This precipitation type was
chosen to represent stratiform precipitation of the organized convection or the widespread
stratocumulus clouds. The spatial and seasonal distribution of this precipitation type
appears to be more or less consistent with the assumption made for the subjective
classification of this precipitation type.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
99
•) TYFE.HI / ALL P M C T 0 * » . ) fcr MAM (W M W g )
b > T Y > 8 » C T /A U .P M C I ftf^ |fc r iiA C W > .im )
0 03
a 10
09
030
t ) TYPE-III' AU. fRBCIP (3 P tf.) fcr 0 0N (IW H fW )
0 03
0.10
0 .9
030
040
030
qq 3
010
0 9
0.30
040
0 30
Figure 25 Fraction of raining SSM/I pixels reporting TYPE-III precipitation class.
The striking feature of the distribution of frequency of TYPE-IV precipitation over the
ocean for all four seasons is the total absence of any occurrence in the tropical regions
from 30° N to 30° S (Figure 26) . The high latitude regions of the southern ocean are
mainly represented by this type of precipitation class with more than 80% of occurrence
of this precipitation type in this region. The effect of extratropical storm tracks in the
north Pacific Ocean and north Atlantic Ocean during winter months (DJF and MAM) can
be seen with relative high frequency of occurrence for this precipitation type. This
precipitation type is considered to represent stratiform precipitation associated with
extratropical storms or widespread stratocumulus clouds in high latitudes with relatively
low precipitation amount in the form of rain or snow. This assumption is broadly
consistent with the expected seasonal and global distribution of the frequency of
occurrence of this precipitation class.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
100
» )T V F B ^ V /A IX P M C T (S ^ )fcr M A M Q W |.|fW >
W T V F H V /A U .P K B C I F (S ^ |f c > iiA llW |.|W »
Q T Y FB 4V /A U P H B C tP(3d»f )(W D ^ (1W ».|9W )
Figure 26 Fraction of raining SSM/I pixels reporting TYPE-IV precipitation class.
Figure 27 depicts the global distribution of regional frequencies of TYPE-V precipitation
type over the ocean for all four seasons. In general, the subtropical belt is characterized
by high frequency of occurrence of this precipitation type throughout the year. The low
values of frequency of occurrence of TYPE-V precipitation class is observed in the
eastern Pacific region of the ITCZ throughout the year. The tropical belt of the Atlantic
Ocean is characterized by high frequency of occurrence of this precipitation type. The
geographical distribution of this precipitation type with high values in the eastern Pacific
ocean, the south China Sea, the east of Australia and low values in the western Pacific
ocean region of ITCZ are consistent with earlier results of Petty (1998).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
101
« )T Y y B .V /A U .P lB C T (3 * ^ > fe rM A M (1 9 » l-IW B
WN
o r m -v
»6
00)
0 10
0 20
0 )0
AIJ. P R K W f) 4 » f ) far ION ( I f t l >1992)
“
*--------------- “ -------------
406
00)
906
1206
110
1)06
0J0
ID
» T Y W .V /A U .F M a F 0 ^ ) * r ilA O < 9 |.|» W )
040
•
DOW I20W
0J0
0 )0
»W
0 40
0QW
0 )0
MW
OOJ
0 10
020
0 )0
flTYFfi.V A U.P*EC IP() * t ) farOJF(l«9M «*S)
V S ------------------------------------------- --------------------------------
M8
406
00)
906
1306
010
1)08
020
ID
040
DOW I2DW
0 )0
0 40
0 )0
«W
UW
)0W
0 )0
Figure 27 Fraction of raining SSM/I pixels reporting TYPE-V precipitation class.
The geographical distribution of regional frequencies (presented as a ratio of total number
of pixels corresponding to this precipitation class to the total number of valid pixels) of
TYPE-VI precipitation class shows relatively low values over the majority of the ocean
areas (Figure 28). The areas in north Pacific and north Atlantic Ocean are characterized
by relatively high values during winter months (DJF and MAM). These high values of
frequency o f occurrence in these regions can be attributed to the frequent occurrence of
extratropical storms during these seasons. The interesting feature of relatively high values
o f this type of precipitation class over the western Pacific ocean and non-existence over
the eastern Pacific Ocean is observed. This could be attributed to the relative high
occurrence of deep convective precipitation in the western Pacific region and almost no
presence of convection in the eastern Pacific Ocean (Figure 23). A belt o f relatively high
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
102
frequency of occurrence is observed in the high latitude region of the southern ocean
throughout the year, which could be due to sea-ice contamination.
•JT Y W -yi. ALL VALID PIXELS ( i d t i ) fcr MAM M M M 9B1
0001
om
om
am
0012
am
003
JOB
006
0.001
90S
1208
aoo)
1308
am
D
ISOW I20W
o.m
aoii
9QW
tOW
0.03
am
aoos
oooa
0012
003
d) TYHLVl / ALL VALID PIXELS Q On. )fcr O ff (1W H W 2)_______________
«> TYPB-Vl /ALL VALID PIXELS (3 O ^ U M jO N O W I -im ) ____________
SOW
0001
om
om
om
0012
00s
Figure 28 Fraction of SSM/I pixels reporting TYPE-VI precipitation class.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
103
4.3 COMPARISON OF SSM/I DERIVED PRECIPITATION FREQUENCIES
WITH COADS DERIVED PRECIPITTION FREQUENCIES
This subsection deals with the comparison of SSM/I products with those of presentweather ship reports of COADS data (Petty 1995). This is done to examine to what extent
the SSM/I results can be related to precipitation characteristics inferred from surface data.
We have compared the precipitation frequencies derived using SSM.I observations
(Figure 22) with the precipitation frequencies seen in COADS data (Figure 22 of Petty
1995). Figure 29 shows the recreated plot of Petty (1995) and displays the fraction of
present-weather reports indicating some form of surface precipitation occurring locally
and at the actual time of the observation. It is noteworthy that all major features of this
plot match pretty well with Figure 22 - such as: (1) high frequency regions in the eastern
Pacific Ocean, extratropical storm track regions in the northwestern, north central Pacific
Ocean, and mid Atlantic Ocean, semi-persistent regions of transient weather system in
high latitude regions; (2) very low precipitation frequency in the subtropical region to the
west of major continents. In general, values of precipitation frequency are higher in
COADS data than those obtained from SSM/I data. Earlier comparisons of SSM/I
precipitation frequencies with precipitation frequencies obtained from COADS data have
also shown high values for COADS data (PIP-3 Intercomparison Results 1996).
Nevertheless, these comparison have given sufficient confidence that SSM/I is able to
capture large scale precipitation features reasonably well.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 29 Fraction of present weather reports indicating some form of surface
precipitation occurring locally and at the actual time of the observation, (taken from Petty
1995).
Since TYPE-I and TYPE-II precipitation classes are chosen to represent some form of
strong convective precipitation, we have generated a plot representing the overall
frequency of convective precipitation seen from SSM/I by summing the frequency of
occurrence of TYPE-I and TYPE-II precipitation classes. This is done to compare the
frequency o f strong convective precipitation seen from SSM/I with those of the COADS
data (Figure 20 of Petty 1995). Figures 30 and 31 show the frequency of convective
precipitation seen from SSM/I and present-weather reports of COADS data.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
105
•)T V f& l» T Y W » g /A L L m C T ( 3 * t ) f e r U A M < m U i m )
joe
«ob
«s
ooi
line
aot
iiob
a> isow isw n v
01:
a is
on
. k)TYV&-t*TYPB*Jl /A U .P R E C IP < 3*«.)ferJJA (lttM «92)
aw nw
306
o.m
OOI
908
1306
000
1306
0.12
ID
020
006
1306
ID
012
ISOW I20W
O il
90W
024
MW
0 )0
406[
ISOW IJBW
O il
1206
4)TYTW . TYFB41 / A U .n u n P ( S * t .) f e r D I P ( l9 * l- l9 9 2 )
J
MB
«6
OOI
«> TYFE-l ♦ TYTE-n / A U . f t £ C g ( 3 * t - ) fcr< C N (IW > .|fB )_____________
308
006
306
OM
006
001
906
1206
000
1306
012
ID
ISOW I2DW
O il
024
90W
MW
30W
0 )0
Figure 30 Fraction of raining SSM/I pixels reporting TYPE-I and TYPE-II precipitation
classes.
These regional and seasonal distributions of frequency of convective precipitation seen
from SSM/I and shipboard present-weather reports agree remarkably well, given the
vastly different origin of the data and the different time periods covered. It is observed
that all high and low frequency bands seen from SSM/I match reasonably well in their
location with those in the COADS data. However, it is observed that shipboard presentweather reports show high magnitudes in some regions of high frequency band. For
example, areas near the Arabian Sea, Bay of Bengal, east coast of China, near the Gulf of
Mexico, and west of Africa continent show relatively high frequency of occurrence in
present-ship weather reports when compared with those of SSM/I.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
106
Figure 31 Fraction of present weather reports indicating some form of strong convective
precipitation occurring locally and at the actual time of the observation, (taken from Petty
1995).
The global distribution of TYPE-IV precipitation class for all four seasons is found to
show striking similarities with Petty’s (1995) Figure 19 (not shown here), which depicts
the global distribution of snow observed in shipboard present-weather reports.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
107
4.4 REGIONAL DISTRIBUTION OF PRECIPITATION TOTAL AND
FRACTION FOR EACH PRECIPITATION TYPE
This subsection examines the regional distribution of total precipitation amounts and the
contribution of each precipitation class to the overall seasonal precipitation amounts. The
two years of SSM/I data are used to generate seasonal precipitation fields over a 5°x5°
grid over the global ocean. Figure 32 shows the regional distribution of total precipitation
amounts for all four seasons. The persistent location of minimum values (< 30 mm) of
seasonal precipitation amount west of all major continents in the subtropical area of the
southern ocean is observed throughout the year. During DJF, maxima of seasonal
precipitation amounts ( > 500 mm) exist over the central Pacific, south of Indian Ocean ,
the SPCZ, the northwestern Pacific and the Atlantic Ocean. The elongated area of high
precipitation amount along the Pacific ITCZ is observed during MAM. The subtropical
belt in the northern ocean is characterized by very low values of seasonal precipitation
amounts during this season. During JJA, the characteristic increase in seasonal
precipitation can be seen over the Bay of Bengal, western Arabian Sea, off the Mexican
coast in the Pacific ocean, and in the Pacific ITCZ region. The SON months are
characterized by relatively large values of seasonal precipitation in the southeast Indian
ocean, the SPCZ, and the Pacific ITCZ. These patterns are found to be in broad
agreement with 13-yr climatology of seasonal precipitation derived using Microwave
Sounding Unit (MSU) data by Spencer (1993).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
108
*» SEASONAL n S C W TOT (3 * * ) fcr MAM (1W M 9H )
100
MO
«> SEASONAL PWaCtP TOT
100
M0
1000
2D00
MSBAICNALPMClPTCrr. (3 4 u .)fe r U A (1991'19921
3000
10Q0O
) Ibr SON (199M99?)
1000
M 00
100
MO
1000
3000
3000
10000
4) SEASONAL FftfiCP TOT 0< fc |)fcrD J F (I9 9 W 992)
3000
10000
10.0
JO.O
1000
2000
3Q00
IQOOO
Figure 32 Two years averaged seasonal precipitation total derived from SSM/I data.
The next five figures deal with the ratio of the seasonal precipitation amount
corresponding to a particular precipitation type to the total seasonal precipitation amount
over the 5° degree latitude/longitude grid. They can also be used to estimate the
contribution of convective versus stratiform precipitation in a given region based on the
assumed closeness to these precipitation types to conventional precipitation types and
also the gross features of their retrieved microphysical structure. Figure 33 depicts the
regional distribution of the TYPE-I contribution to the total precipitation amount. This
figure reflects that TYPE-I precipitation type contribute more than 60% to the over all
precipitation amount in the Pacific ITCZ region, tropical Indian ocean, South China Sea,
the SPCZ, the eastern and western coast of the southern part of North America. These are
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
109
basically regions of convective precipitation from a climatological point of view. The
north central Pacific and north central Atlantic Ocean are found to get relatively less
amounts of precipitation from this precipitation class.
• I TYTB4/TPT PKBCtT (3 4 ^ .) hrM AM (IW M 992)
02
0)
04
02
4) TYW»I / TOT H tB C g (S 4 » » .)h > D ff (IW M W 2)
Figure 33 Fraction of total precipitation amount corresponding to TYPE-I precipitation
class.
The global distribution for TYPE-II precipitation class for all four season is shown in
Figure 34. The majority of tropical areas do not show any significant contribution of
TYPE-Q precipitation class to the total precipitation amount. The midlatitude region of
the northern ocean is characterized by a 40-60% contribution of this precipitation type to
the overall precipitation amount during DJF, SON and MAM, with winter months
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
110
showing maximum values. Less than a 30% contribution of this precipitation type is
observed during the summer season (JJA). The southern ocean midlatitude belt is
characterized by 40-60 % throughout the year, except for DJF. The maximumvalues for
the contribution o f this precipitation type are observed east of Australia and Africa in the
southern ocean during JJA season.
a) TYH UI 'TOT P1UsCtP(J 4 ^ .) far MAM (1991*1992)
HTVWM1/TOT n tB C IP (3 fciH b rIM (lW M 9 W )
0)
04
03
00
07
«) rYTEH TOT H IBC g<34a»H br«C W IH tM »W 1_____________________
JOE
WE
TOE
12DE
OJ
I30E
04
ID
I30W I37W
03
06
«W
w
02
®-J
04
03
06
07
4 )TTFS-tl/ TOT n t£ C t> (3 # » .)l» r D ff (IW |.|f9 2 )_____________________
JOE
07
ME
02
TOE
I20E
OJ
130E
04
ID
130V
03
I3JW
06
90W
40W
JOW
07
Figure 34 Fraction of total precipitation amount corresponding to TYPE-II precipitation
class.
The global distribution of the fraction of TYPE-III precipitation type (Figure 35) is more
noisy compared to the global distribution of the earlier two precipitation types. High
values for the contribution of this precipitation type (about 50%) are observed over the
northwestern Pacific during JJA and the north central Pacific during MAM. Mid Atlantic
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Ill
Ocean is also characterized by moderately high values (30-40 %) during JJA and SON
seasons. During DJF season, a 30-40% contribution by this precipitation type to the total
seasonal precipitation is observed in mid-latitudinal belt over the ocean in the southern
hemisphere. The southern Indian Ocean is also observed to contribute 30-40% from this
precipitation type during JJA and SON. On an average, the contribution of this
precipitation type is observed to be 20-30% over most of the oceanic regions.
tiT Y PtH B'TQ T
W
(0 6
«
l »
MB
ID
IMW I3DW
M W M WJ O W
ME
«) TYWE-Ul/ TOT PK BCtr(3d»E)fcr<CN <IW I.tW a____________________________________^
406
TO
1206
4 ) 'H'TB-IP/ TOT FKflCIP(
3
I90B
ID
|M W
IS W
MW
flOW
MW
fcrDJF(IMI«>f»2)_____________________________
Figure 35 Fraction of total precipitation amount corresponding to TYPE-m precipitation
class
Figure 36 shows the global distribution of the contribution of TYPE-IV precipitation
class to the total seasonal precipitation amounts for all four seasons. The distribution does
not reflect any meridional pattern. It is observed that more than 80% of the precipitation
amount in high latitudes is found to come primarily from this type of precipitation class.
The midlatitude region of north Pacific and north Atlantic Ocean shows a relatively low
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
112
contribution (20-40%) of this precipitation class to over all seasonal precipitation
amounts.
•I TYFB jV
TOT r * £ O P ( 3 * H fcr MAM (IW M W )
0 .2
OJ
0.4
OS
»T Y H H V 'T P T W taC iy (3 r ftf)fc > iM <1991-199:)
00
0.7
02
0J
0.4
OS
00
07
Figure 36 Fraction of total precipitation amount corresponding to TYPE-IV
precipitation class.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
113
>)TYWUV'TOT
«> t w
003
v >t o t ,
01
02
0.3
mar <3 *».)>>oon o w i - i p w )
0.03
Ot
02
03
rf)T m > V /T O T f*ECTT (3 4 p far D t t Q W M m )
Figure 37 Fraction of total precipitation amount corresponding to TYPE-V precipitation
class.
The seasonal maps of the global distribution of the contribution of TYPE-V precipitation
class (Figure 37) indicate that majority of the global ocean is found to be characterized by
relatively low values (less than 10%). The areas which show a significant contribution of
this precipitation type to overall seasonal precipitation amount are north central Pacific
ocean during MAM, south central Pacific Ocean during JJA, southwest Indian Ocean
during JJA and SON, and the SACZ during JJA. An elongated region of high values of
contribution of this precipitation type is observed in the subtropical belt of the Pacific
ocean during DJF. Petty (1998) used 18 months of Geostationary Meteorological
Satellites (GMS-S) data to analyze the occurrence of warm-topped precipitation over the
Eastern pacific ocean and also found a relatively significant occurrence o f warm-topped
precipitation over south China and Indochina during DJF.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
114
5. CONCLUSIONS
The primary objective of this study was to demonstrate the meaningfulness of the
objective classification scheme developed to classify six precipitation types, and also to
document the global characteristics of the distribution of the frequency of each
precipitation class over the ocean. Two years of SSM/I data were analyzed to generate
statistics for each precipitation class. The results of the analysis reflect the broad
consistency in the logic used in subjectively defining precipitation classes and their actual
global characteristics.
The global distribution of frequencies of each class of precipitation yields the following
conclusions: (1) TYPE-I precipitation class is found to be more frequent in the tropical
seas and typically represents a tropical convective type precipitation; (2) the global
distribution shows that TYPE-II is more frequent in midlatitude regions and represents
the typical distribution of extratropical convective precipitation; (3) TYPE-m is found to
be prevalent over all major oceanic areas in tropical, as well as midlatitude regions; (4)
TYPE-IV is clearly a mid- and high latitude type precipitation; (5) TYPE-V precipitation
is found prevalent over all the major oceans in tropical and mid-latitude regions; and (6)
TYPE-VI, which is a non-raining type precipitation class, is found to be more frequent in
high latitude or in the extratropical storm track regions of midlatitude.
The global maps of the precipitation frequency derived from SSM/I and shipboard
present-weather reports indicate a fairly good agreement between them for all four
seasons. This comparison yields good confidence in SSM/I products. Maps of the
frequency o f convective precipitation seen from SSM/I match reasonably well with the
maps of convective precipitation seen from shipboard present-weather reports. All major
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
115
high and low frequency bands show consistent patterns in both maps. The frequency of
occurrence of other precipitation classes cannot be compared with present-ship weather
reports as these classes do not directly match with the classes seen in present-ship
weather reports.
The contribution of each precipitation type to the overall precipitation amount was also
examined and the outcome of this analysis reflects some known and expected patterns.
TYPE-I precipitation class is observed to contribute more than 60% in the majority of
well known convective areas of the tropics. The regions west of all major continents
show relatively less contributions from this precipitation type. TYPE-II precipitation
class contributes more than 50% in some of the extratropical storm tracks region of the
North Pacific and the North Atlantic Ocean. The midlatitude regions of the southern
ocean also gets more than 30-40% (zonal average) of their precipitation from TYPE-II
precipitation class. Although TYPE-in precipitation type is very frequent in the tropics,
as well as some of the midlatitude regions but the contribution by this precipitation class
to the total precipitation amounts is not very high in all major oceanic areas. It contributes
20-30% to the total precipitation amounts in some portion of the tropical and midlatitude
oceanic areas. The region west of California coast gets more than 50% of its rainfall from
this precipitation type during JJA. The high latitude regions are found to get a majority of
their precipitation (which itself is not more than 100 mm) from TYPE-IV precipitation
class. The TYPE-V precipitation class is again found to contribute meagerly to the overall
precipitation amounts in the tropics and midlatitude regions. Some areas in the
subtropical belt get more than 30-40% of their rain from this precipitation class. It is
interesting to note that although TYPE-I and TYPE-II are relatively less frequent than
TYPE-in in tropics and midlatitude regions, their contributions to the overall
precipitation amount are quite high in these regions. It also justifies that these
precipitation types could be convective in nature and produce heavy rainfall in short
period of time in these oceanic regions. In contrast, TYPE-lH and TYPE-IV precipitation
classes are characterized by very high frequency of occurrence in many of the oceanic
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
116
areas, but their contribution to total precipitation amount is not very high, which can be
explained by assuming the stratiform nature of precipitation types. It can be inferred that
TYPE-I and TYPE-II precipitation classes (which are more like a convective
precipitation) contribute more than 70% percent to the total precipitation amount in the
majority of the tropical and midlatitude regions, and TYPE-in and TYPE-IV
precipitation classes contribute more than 70% in a majority of high latitudes and some of
the midlatitude regions. It is important to note that the inference made about the
convective and stratiform nature of precipitation also requires spatial pattern of the
precipitation event which is not considered in generating the global statistics of all
precipitation classes. The indirect inference made here about the convective and
stratiform nature of precipitation type is purely on the basis of crude physical
characteristics of these precipitation types.
The analysis performed in this study used only two years of SSM/I data and it would be
interesting to generate statistics for frequency corresponding to each precipitation class
for the entire SSM/I data set since 1987. The regional patterns of frequency of each
precipitation class should also be examined for El-Nino and La-Nina years separately.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
117
REFERENCES
Adler, R., C. Kidd, M. Goodman, A. Ritchie, R. Schudalla, G. Petty, M. Morrissey, and
S. Greene, 1996: PIP-3 Intercompariosn Results. Prepared for the PIP-3
Workshop, 18-20 November, 1996, College Park, MD, USA.
Arkin, P. A., and J. E. Janowiak, 1993: Tropical and subtropical precipitation. Atlas o f
Satellite Observations Related to Global Change, R.J. Gurney, J.L. Foster, and
C.L. Parkinsons, Eds., Cambridge University Press, 165-180.
Arkin, P. A., and P. Xie, 1994: The Global Precipitation Climatology Project: First
Algorithm Intercomparison Project. Bull. Amer. Meteor. Soc., 75,401-419.
Arkin, P.A., and P.E. Ardanuy, 1989: Estimating climatic-scale precipitation from space:
A review. J. Climate , 2,1229-1238.
Barret, E.C., J. Dodge, M. Goodman, J. Janowiak, and E. Smith, 1994a: The First WetNet
Intercompariosn Project (PIP-1), Remote Sens. Rev., 11,49-60.
Barret, E. C., and Coauthors, 1994b: The first Wet-Net Precipitation Intercomparison
Project: Interpretation of results. Remote Sens. Rev., 11, 303-373.
Dorman, C. E., and R.H. Bourke, 1979: Precipitation over the Pacific Ocean, 30°S to
60°N. Mon. Wea. Rev., 107,896-910.
Dorman, C. E., and R.H. Bourke, 1981: Precipitation over the Atlantic Ocean, 30°S to
70°N. Mon. Wea. Rev., 109,554-563.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
118
Hollinger, J.P. P.G. Lo, R. Savage, and J. Pierce, 1987: Special Sensor Microwave/
Imager user’s Guide, naval Research Laboratory, 120p.
Janowiak, J.E., and P.A. Arkin, 1990: Rainfall variations in the tropics during 1968-1989
as estimated from observations of cloud temperatures. J. Geophys. Res., 96, 33593373.
Jaeger, L., 1983: Monthly and areal patterns o f mean global precipitation. Variations in
the Global Water Budget, A Street-Perrott, M. Beran, and R. Ratccliffe, Eds., D.
Reidel, 129-140.
Legates, D. R., and C. J. Willmott, 1990: Mean seasonal and spatial variability in gauge
corrected, global precipitation. Int. J. Climatol., 10, 11-127.
Petty, G. W., 1990: On the response of Special Sensor Microwave/Imager to the marine
environment - Implications for Geophysical parameter retrievals. Ph.D.
dissertation, University of Washinton, 291 pp. [Available from university
Microfilms International, Ann Arbor, MI 48106.]
Petty, G.W., 1994a: Physical retrievals of over ocean rain rate from multichannel
microwave imagery. Part-I: Theoretical characteristics of normalized polarization
and scattering indices. Meteor. Atmos. Phys., 54, 79-100.
Petty, G.W., 1994b: Physical retrievals of over ocean rain rate from multichannel
microwave imagery. Part-II: Algorithm implementation. Meteor. Atmos. Phys.,
54,101-122.
Petty, G.W., and W.F. Krajewski, 1996: Satellite estimation of precipitation over land.
Hydro. Sciences, 41(4).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
119
Petty, G.W.,1995: The status of satellite-based rainfall estimation of precipitation over
land. Remote Sens. Environ., 51,125-137.
Petty, G.W., 1998: On the prevalence of precipitation from warm-topped clouds over
eastern Asia and the western pacific. J. Climate (in press)
Reed, R. K., 1979: On the relationship between the amount and frequency of
precipitation over the ocean. J. Appl. Meteor., 18,692-696.
Reed, R. K., and W.P. Elliott 1979: New precipitation maps for the North Atlantic and
North Pacific Oceans. J. Geophys. Res., 84, 7839-7846.
Richards, F., and P. Arkin, 1981: On the relationship between satellite-observed cloud
cover and precipitation. Mon Wea Rev., 109, 1081-1093.
Spencer, R. W., 1993: Global oceanic precipitation from the MSU during 1979-91 and
comparisons to other climatologies. J. Climate, 6, 1301-1326.
Tucker, G. B., 1961: Precipitation over the North Atlantic Ocean. Quart. J. Roy. Meteor.
Soc., 87,147-158.
WCRP Informal Report No. 6/1996, May 1996: International workshop on research
issues in the identification of precipitation type and rates in global data sets
( Washington, D.C., 6-8 December, 1995).
Wilheit, T.T., R. Adler, S. Avery, E. Barret, P. Bauer, W. Berg, A. Chang, J. Ferriday, N.
Grody, S. Goodman, C. Kidd, D. Kniveton, C. Kummerrow, A. Mugnai, W.
Olson, G.W. Petty, A. Shibata, and E. Smith, 1994: Algorithms for the retrieval of
rainfall from passive microwave measurements. Remote Sens. Rev., 11,163-194.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
V IT A
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
120
VITA
Nitin Gautam was bom to Krishna and B. P. Gautam on May 24, 1965 in Roorkee (U.P.),
India. After finishing his high school from K. V. S, Fatehgarh in 1982, he entered Agra
College, Agra University where he received a B.Sc. Degree in science in 1984, and a
M.Sc. Degree in Physics in 1986. He spent one more year in finishing Post Graduate
Diploma in Space Sciences and their Applications from Gujarat University, Ahmedabad.
He started working as a scientist in a premier scientific organization, Indian Space
Research Organization (ISRO) in 1987. He spent for about six years in ISRO doing
research in the area of satellite remote sensing of atmosphere and oceans. During this
period in ISRO, he published and presented many papers in reputed scientific journals
and conferences. He then decided to pursue higher education towards Ph.D. in
Atmospheric Sciences in United States. He spent one year in Saint Louis University
taking graduate courses in Meteorology. He then moved to the Department of Earth and
Atmospheric Sciences at Purdue University in August 1994 for pursuing graduate studies
leading to Ph.D. Diploma. He finished his Ph.D. in December, 1998.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Документ
Категория
Без категории
Просмотров
0
Размер файла
4 826 Кб
Теги
sdewsdweddes
1/--страниц
Пожаловаться на содержимое документа