Analytically Derived Relations for Rain Estimation Using Polarimetric Radar Measurements

Page created by Darrell Salinas
 
CONTINUE READING
Analytically Derived Relations for Rain Estimation
                            Using Polarimetric Radar Measurements
                             J. Vivekanandan*, Guifu Zhang and Edward Brandes
                                   National Center for Atmospheric Research
                                               P. O. Box 3000
                                           Boulder, CO 80307-3000

Abstract
                                                            parameters, and derived a µ-Λ relation from the video-
Polarimetric radar has been successful in                   disdrometer measurements collected during a special
characterizing cloud and precipitation. Polarization        field experiment in east-central Florida evaluating the
parameters such as radar reflectivity (Z), differential     potential for polarimetric radar to estimate rainfall in a
reflectivity (ZDR), linear reflectivity difference (ZDP),   subtropical environment.
specific differential phase shift (KDP), linear de-
polarization ratio (LDR) as well as the correlation         1. Introduction
coefficient (ρHV) have been successfully measured.
Polarimetric measurements provide more information          The accuracy of rain rate estimation by well-calibrated
about precipitation and allow better characterization of    radar is limited by the lack of detail knowledge of drop
hydrometeors, accurate rain rate estimation, and            size distribution (DSD). Rain rate is usually estimated
retrieval of rain drop size distribution (DSD). In the      from radar reflectivity using a Z-R relation based on
past, rain rate (R) estimation from S-Pol was based on      convective or stratiform rain. The Z-R relation was
empirical models such as Z-R, R (Z, ZDR) and R (KDP)        obtained by fitting gauge measurements and radar
relations, which were derived from regression analyses      reflectivity. It is known that the Z-R relation changes
of radar and rain gauge measurements or numerical           from location to location and time to time depending
simulations. These fixed empirical relations, however,      on changes in DSD. Therefore a fixed empirical Z-R
cannot give accurate estimation results for various         relation cannot provide accurate rain estimation for
types of rain and are very sensitive to selection of        various types of rain because it cannot handle variation
dataset for fitting, threshold, sampling effects,           in drop size information. The relation between radar
measurement errors and so forth.                            reflectivity and rain rate is almost completely
                                                            quantified only if the drop size distribution is
Accurate rain rate estimation requires understanding of     specified because they are proportional to moments of
rain microphysics and knowledge of raindrop size            DSD; namely, reflectivity is the 6th moment and rain
distribution (DSD), shape and canting angle, while the      rate is proportional to the 3.67th moment of the drop
information about rain DSD is important in                  spectrum. Accurate rain rate estimation requires
understanding rain development and evolution. In the        detailed knowledge of rain DSD and hence various
past, rain DSD was commonly assumed to be an                rain rate estimators are derived using polarimetric
exponential distribution (this may be true for a long       radar observation that includes reflectivity, differential
sample period) but recent observations indicate that        reflectivity and propagation phase [Doviak and Zrnic,
the instant rain DSD is better characterized by a           1993].
Gamma distribution. The three Gamma DSD
parameters are usually difficult to retrieve. During the    The polarimetric radar technique has attracted great
analysis of data measured by a video-disdrometer, we        attention because most of the hydrometeors are non-
found there is a high correlation between the shape (µ)     spherical. Particularly in the case of raindrops there is
and slope (Λ)                                               a well-defined relation between size and shape [Seliga
                                                            and Bringi, 1976; Oguchi, 1983; Vivekanandan,
_________________________________________                   1999a]. The idea of using differential reflectivity in
                                                            rain estimation was first proposed by Seliga and
* Corresponding author address:                             Bringi in 1976. Much progress has been made in
         J. Vivekanandan                                    implementation of a polarization radar measurement
         NCAR/RAP, P. O. Box 3000                           system and a microphysical retrieval algorithm [Bringi
         Boulder, CO 80307 – 3000                           and Hendry, 1990]. Polarization parameters such as
         USA                                                radar reflectivity (ZHH), differential reflectivity (ZDR),
         Email: vivek@ucar.edu                              linear reflectivity difference (ZDP), specific differential
phase shift (KDP), linear de-polarization ratio (LDR),      distribution with µ = 0. It has been found that the three
as well as the correlation coefficient (ρHV) have been      parameters are not mutually independent [Ulbrich,
successfully measured. These multi-parameter                1983; Haddad et al. 1997]. Haddad et al (1997)
measurements provide additional information about           parameterized rain DSD with transformed parameters
precipitation and allow better microphysical                which are uniformly random. The effort has
characterization of hydrometeors. In general, ZHH,          concentrated on generating Gamma DSD from
ZDR, and KDP are used for estimating rain rate and drop     independent random variables.
spectrum, since they depend mainly on drop size and
shape [Jameson, 1983a&b]. LDR and the covariances           Since the three DSD parameters do not correspond to
are used for retrieving canting angles because they are     physical parameters such as liquid water content or
sensitive to particle orientation [Ryzhkov et al., 1999;    median volume diameter, various normalization
Vivekanandan et al., 1999b].                                techniques were used [Willis, 1984; Dou et al., 1999].
                                                            Since the dimension of N0 is ill-defined, Chandrasekar
Usefulness of the fixed power-law rain rate                 and Bringi (1987) proposed to use the total number
estimators is limited by various factors. Although it is    concentration Nt instead of N0. Furthermore, a
well known that the polarimetric radar measurements         normalized Gamma distribution was first proposed by
contain information regarding rain DSD, the study of        Willis and recently adopted by Illingworth and
retrieving rain DSD from polarimetric radar                 Blackman to eliminate the dependence between N0
measurements is limited to retrieving all three             and µ [Willis, 1984; Illingworth and Blackman, 1999;
parameters of a Gamma function [Seliga and Bringi,          Testud, 2001]. They recommended using physically
1978; Richter and Hagen, 1997]. An estimate of KDP          meaningful parameters to characterize a Gamma DSD.
involves the averaging of differential phase (φDP) over     Nevertheless, the number of parameters is the same,
range, hence R(KDP) may overestimate or                     and the DSD expression becomes more complicated.
underestimate rain rate due to the range averaging of       In practice, there is no simplification of the DSD
φDP [Doviak and Zrnic, 1993]. R(ZHH, ZDR) may give          function except that DSD parameters are expressed
acceptable rain estimation in some cases, but does not      using NT, LWC, and D0.
provide rain drop size distribution (DSD).
                                                             In this paper, we study constrained Gamma rain DSD
For a standard sampling time such as 1 minute,              and its application to rain estimation from polarimetric
however, some observations indicate that natural rain       radar measurements. The paper is organized as
DSD contains fewer of both very large and very small        follows: In section 2, video disdrometer measurements
drops than an exponential distribution [Ulbrich, 1983;      are analyzed. The three parameters of Gamma DSD
Tokay and Short, 1996]. Estimated rain rate may be          are obtained from three moments of the measured
comparable to the actual rain rate of the measured          DSD. A constrained relation between µ and Λ is
spectrum using either exponential distribution or           derived from disdrometer observations. In section 3,
Gamma distribution when the 3rd or 4th moment is            we derive closed-form expressions for calculating rain
included, because rain rate is proportional to the 3.67th   rate and characteristic size of raindrops. Rain rate
moment that is close to the central moments. But, the       estimation using the closed-form expression is
problem is the assumed exponential distribution would       compared with in situ observation and fixed-power
not be able to provide moments other than the central       law-based estimates are presented in Section 4.
moments, such as reflectivity (Z), linear reflectivity      Specific propagation phase is derived using Z and ZDR
difference (ZDP), and specific differential phase (KDP).    variables. The estimated propagation phase using Z
Thus, an accurate mathematical description of DSD is        and ZDR measurements is compared with the actual
essential for estimating rain as well as multi-             measured propagation phase for verifying self-
parameter radar observations such as Z, ZDR, and KDP.       consistency between Z, ZDR and KDP and also the
                                                            applicability of analytically derived relations using a
Ulbrich (1983) suggested the use of the Gamma               Λ-µ relation. A summary of the results are given in
distribution for representing a rain drop spectrum as       section 5.

         n( D ) = N 0 D µ exp( − ΛD )       .    (1)        2. DSD parameters using un-truncated and truncated
                                                            moment methods
 The Gamma DSD with three parameters (N0, µ, and            Raindrop size distribution can be measured using
Λ) is capable of describing a broader variation in          various instruments such as a momentum impact
raindrop size distribution than an exponential              disdrometer, wind profiler, particle measuring probe
distribution which is a special case of Gamma               (PMS) and video disdrometer [Yuter, 1997; Williams,
2000]. In this study, the video disdrometer
measurements collected in PRECIP’98 are analyzed.                                    ∞
The video disdrometer was operated by Iowa State                     < D n >= ∫ D n n( D )dD = N 0 Λ − ( µ + n +1)
                                                                                     0                                    .   (2)
University in east-central Florida during the summer
of 1998 when NCAR's S-Pol radar was also deployed                    Γ( µ + n + 1)
to evaluate the potential of polarimetric radar for
estimating rain in a tropical environment. Following is              In general, the three parameters (N0, µ, and Λ) can be
a brief review of the method of fitting the measured                 solved from any three moments such as the 2nd, 4th
DSDs to Gamma distribution and finding the relations                 and 6th moment. To eliminate Λ and find µ, a ratio is
among the DSD parameters.                                            defined as

The moment method has been widely accepted in the                               < D4 >2              ( µ + 3)( µ + 4)
meteorology community because of its robustness in                   η=                          =                    .       (3)
obtaining rain rate [Kozu and Nakamura, 1991; Tokay                        < D >< D >
                                                                                 2           6       ( µ + 5)( µ + 6)
and Short, 1996; Ulbrich, 1998]. In the past, the
integration of most moment calculations is usually                   Then, µ can be easily solved from (3) as
performed from zero to infinite size range as
__________________________________________

                                            (7 − 11η ) − [(7 − 11η )2 − 4(η − 1)(30η − 12)
                                       µ=                                                  ;                                   (4)
                                                                2(η − 1)

Λ can be calculated from

                                                             1/ 2                                      1/ 2
                                        <   D > Γ( µ + 5)          <   D > ( µ + 4)( µ + 3) 
                                               2                            2
                                   Λ   =                          =                                       .                (5)
                                        <   D > Γ( µ + 3)                     
                                               4                                         4
                                                                                                     

N0 can be calculated from any of the three moments
for specified µ and Λ.                                               to observe a raindrop larger than 8 mm). The typical
                                                                     range of raindrop size estimated by the Joss
It should be noted that the integration in (2) is                    disdrometer is between 0.3 mm and 5 mm while a
performed from 0 to infinite; i.e. un-truncated size                 video disdrometer can measure raindrop size between
distribution. Raindrop distribution is measured over a               0.1 and 8mm. However, the above-described method
finite sample volume and time, hence only a finite                   for estimating DSD parameters is applicable only for
number of raindrops were counted within a finite size                un-truncated DSD. For a Gamma distribution with
range [Dmin, Dmax] because of practical and sampling                 truncated size range, the statistical moments are
limitation in measuring small and large drops (it is rare            calculated as

                           Dmax
             < D n >= ∫           D n n( D)dD = N 0Λ −( µ + n+1) [γ ( µ + n + 1, ΛDmax ) − γ ( µ + n + 1, ΛDmin )]             (6)
                          Dmin

                                                                     An accurate way of estimating DSD parameters for a
                                                                     truncated spectrum should be based on the truncated
where γ(…) is an incomplete Gamma function. As                       moments as shown in Eq. (6) and to calculate the
expected, the truncated moments depend on the upper                  integral parameters used in DSD parameter retrieval
and lower limit of droplet size in the measured                      consistent with the truncated (or measured) raindrop
spectrum. If the moments obtained from (6) are used                  size range.
to fit Gamma distribution shown in Eq. (1) using the
above un-truncated moment method described in                        Using the expressions of truncated moments, Eqs. (3)
Eq.(2-5), the resultant DSD parameters may be in                     and (5) become
error.
[γ ( µ + 5, Λ Dmax ) − γ ( µ + 5, Λ Dmin )]
                                                                               2
       η=                                                                                                               (7)
             γ ( µ + 3, Λ D max )γ ( µ + 7, Λ Dmax ) − γ ( µ + 3, Λ Dmin )γ ( µ + 7, Λ Dmin )

and
                                                                                   1/ 2
                    <   D 2 > [γ ( µ + 5, ΛDmax ) − γ ( µ + 5, ΛDmin )] 
               Λ   =                                                                                                  (8)
                    <   D 4 > [γ ( µ + 3, ΛDmax ) − γ ( µ + 3, ΛDmin )] 

Eqs. (7) and (8) constitute joint equations for µ and Λ             µ = −0.036Λ2 + 1.2385Λ − 2.286                      (9)
for the truncated moments that are difficult to separate
from each other. Since the above equations are non-
                                                                   or
linear, an iterative approach is used for solving µ and
Λ.
                                                                    Λ = 0.037 µ 2 + 0.691µ + 1.926                     (10)
2.2 Analysis of µ - Λ relation
                                                                    without truncation, and
Video disdrometer measurements collected in
PRECIP’98 are used in this study. The data set is the
                                                                    µ = −0.040Λ2 + 1.405Λ − 2.461                      (11)
same as that reported in Zhang et al., 2001, except
minor revisions were made for splashing and wind
effects during DSD measurements. The video                         or
disdrometer was operated by Iowa State University in
east-central Florida during the summer. We use the                  Λ = 0.0365µ 2 + 0.735µ + 1.935                     (12)
above moment and truncated moment methods to fit
the measured DSDs with Gamma distribution. The                     for the estimate using the truncated moment method.
Gamma DSD parameters are calculated and analyzed.
Figure 1 shows the scatter plots of the fitted DSD                  It is interesting to note that the µ and Λ relations do
parameters (µ vs Λ). Figure 1a is obtained from the                not change much while the mean values of µ and Λ
un-truncated moment method and Figure 1b from the                  change from 4.09, 5.58 in Figure 1c to 3.25, 4.92 in
truncated moment method. There are a total of 1341                 Figure 1d for the truncated moment method.
data points covering 22 hrs and 21 minutes in 17 days
                                                                   Theoretically either µ(Λ) or Λ(µ) can be used. In
during PRECIP’98. Both Figures 1a and 1b show
                                                                   practice, however, there are some differences. We
correlation between µ and Λ. However, retrievals of µ
                                                                   notice that µ(Λ) fits large values better while Λ(µ) is a
and Λ obtained using the truncated moment method                   better fit for small values. Since heavy rain tends to
show better correlation than the corresponding set
                                                                   have small values of µ and Λ, we expect that Λ(µ) is a
retrieved using the untruncated moment method.
                                                                   better approach. To quantify the difference, the
Further analysis of raindrop spectra revealed the
                                                                   correlation coefficients between the fitted value and
correlation between µ and Λ is also reduced due to
                                                                   true value of µ and Λ were compared. They both have
incomplete sampling of DSD as a result of finite
                                                                   a correlation coefficient of 0.973. However, when µ
sampling volume of the video-disdrometer within a 1
minute sample time. To minimize the error due to                   and Λ are weighted by corresponding rain rate, the
sampling effects, data were filtered by allowing only              correlation coefficient increases to 0.956 for µ and
those with large rain rate >5 mm hr-1 and number of                0.982 for Λ. This supports the fact that the Λ(µ)
raindrops > 1000. The revised plot with the above-                 relation is better than µ(Λ) for high rain rate cases.
mentioned threshold are shown in Figures 1c and 1d.
The figures contain only 248 data points but captured              3. Relations among statistical moments
75% of the rainfall amount in Figures 1a,b. The
scatter plots shown in Figure 1c and 1d show less                  As mentioned in the Introduction, the three parameters
scatter and correlation between µ - Λ is higher. A                 of Gamma DSD are difficult to retrieve from limited
relation between Λ and µ is estimated using a                      radar measurements. The relation Λ(µ) or µ(Λ)
polynomial least-squares fit, and is given as                      derived in the previous section constitutes a
                                                                   constrained condition for Gamma distribution. The µ-
                                                                   Λ relation applied to Gamma DSD shown in Eq. (1)
Figure 1: Scatter plots of µ-Λ values obtained using moment method and truncated moment method. The plots in first
row include all the DSDs collected during 22 hrs and 21 minutes in 17 day of observations. The plots in second row
include measured DSDs only for R > 5 mm hr-1 and total raindrop counts > 1000.

reduces to a two parameter DSD and is dubbed as a            constrained Gamma DSD can be found by taking a
constrained Gamma DSD.                                       ratio as

Statistical moments are integral parameters of a DSD.        < Dk >                            Γ( µ + k + 1)
They are directly related to polarimetric radar                         = Λ ( µ ) − ( k −l )                 ,      (13)
measurements and rainfall rate. For a radar wavelength          l                          Γ( µ + l + 1)
large compared to hydrometeors as in the case of S-
band radar measurement of rain: the reflectivity factor      i.e.
is the 6th moment (Z = ); specific differential
phase is proportional to the 4.6th moment (KDP ∝                                           Γ( µ + k + 1)
) and rain rate is related to the 3.67th moment (R     < D k >= Λ ( µ ) − ( k −l )                 < Dl > .   (14)
                                                                                           Γ( µ + l + 1)
~ ). It is important to know the relation among
the moments and hence the relations between radar
measurements and rain parameters. Instead of                 It should be pointed out that Eq. (14) does not specify
eliminating the median volume diameter as in Ulbrich,        a linear relation between the moments. Actually it
1983; Testud, et al.; 2001, we find the relation             does not even guarantee a functional relation since all
between two moments by canceling N0. Therefore, a            of the moments depend on µ which is a variable. A
relation between the kth and lth moments of a                linear relation exists only when the shape of DSD is
                                                             fixed such as the equilibrium shape of DSD [List, et
                                                             al., 1988] or constrained Gamma DSD with a constant
µ in (14). The shape of DSD (or µ) however, usually          ranges of rain rate than typical scatter in Z-R relations
depends on rain type and rain rate, and should not be        [Doviak and Zrnic, 1993; Ulbrich and Atlas, 1998].
treated as a constant.                                       For rain estimation from specific differential phase, a
                                                             relation of R = 40.56 K DP is currently used. It is
                                                                                        0.866

The relation between moments can be used for                 noted that the relation tends to underestimate rain,
specifying rain rate as a function of various radar          especially for low rainfall [Brandes et al., 2001].
measurements. Since the reflectivity factor is the 6th       Recently, ZDR was combined with KDP to reduce
moment, rain rate (R in mm hr-1 ) is proportional to the     effects due to an apparent change in drop shape as a
3.67th moment given as                                       result of canting and oscillation [Ryzhkov and Zrnic,
                                                             1995]. Similar to the procedure for deriving the R(Z,
R = 7.125 × 10 −3 < D 3.67 > .                        (15)   ZDR) relation, a physically-based R(KDP,ZDR) relation
                                                             can be obtained as follows:
Letting k = 3.67 and l = 6 in Eq. (14) and substituting
into (15), we have                                                   72.28                    −(1.118−0.240σ φ +3.495σ φ2 )
                                                             R=                    K DP Z DR                                  .   (20)
                                                                  (1 − 2σ φ2 )
                                    Γ ( µ + 4.67)
R = 7.125 × 10 −3 [ Λ ( µ )] 2.33                 Z   (16)
                                      Γ ( µ + 7)
                                                             The coefficient of 72.28 is almost twice that currently
                                                             being used 40.56. Therefore, it is expected Eq. (20)
         where                                               will provide a better rain rate estimation, especially for
                                                             low rain rate.
                                    Γ( µ + 4.67)
          F ( µ ) = [Λ ( µ )]2.33                . (17)
                                     Γ( µ + 7)               Specific differential phase KDP is an important
                                                             parameter which can be used for a self-consistent
                                                             check among Z, ZDR and KDP and attenuation
This shows that the R-Z relation is governed by the          correction of reflectivity due to rain. KDP in deg km-1
shape parameter µ. For constrained Gamma DSDs, µ             is related to Z and ZDR in linear units as
uniquely determines ZDR [Zhang et al. 2001] for
assumed axis ratio and canting angle of raindrops.
Thus F(µ) as a function of ZDR can be obtained as            K DP = 5.304 × 10−5 (1 − 2σ φ2 )
                                                                                                          .                       (21)
                                                                  − (2.011− 0.423σ φ + 6.27σ φ2 )
                               −bR                           ZZ
          R = 3.73 × 10 −3 Z Z DR                     (18)
                                                                  DR

                                                             In the following section the above-derived closed
where the exponent bR depends on the effective               expressions are used for estimating rain rate and also
canting angle of raindrops. Assuming the mean                verifying self-consistency among Z, ZDR and KDP.
canting angle is zero, the coefficient bR is expressed as
a function of the standard derivation of effective           4. Rain Rate Estimation
canting angle (σφ ) as
                                                             During PRECIP’98, NCAR's S-Pol radar was
         b R = 3.130 − 0.667σ φ + 9.77σ φ .     2
                                                      (19)   deployed for evaluating the potential of polarimetric
                                                             radar for estimating rain in a tropical environment.
                                                             The rain gauge measurements were also collected
where σφ is in radian. Compared with the widely used         during the project. A detailed description of the project
                        −3      −4.86
relation R = 4.84 × 10 Z Z DR [Sachidananda                  and instrument deployment is illustrated in Brandes, et
and Zrnic, 1987], Eq. (18) has a smaller coefficient         al., 2001.
and lower value of exponent for ZDR. In general, it
gives a lower rain rate estimation, which agrees better      4.1 Comparisons of rain estimation
with gauge measurements [Brandes et al., 2001] for
specified Z and ZDR. Figure 2 shows the relation             Figure 3 shows comparisons between radar estimated
between Z and R for various ZDR. As we expect, a             rain rate and the corresponding rain gauge
fixed Z gives a higher rain rate for a smaller ZDR than      measurements. The rain rate is plotted as a function of
with a larger ZDR because small ZDR is related to a          time for the rain gauge location.
larger amount of smaller raindrops. Specification of         The radar estimated rain is calculated using both the
ZDR confines the Z-R relation constrained to smaller         classic relations and the newly derived relations in this
                                                             paper for comparison. For the result using classic
                                                             relations (Figure 3a), both R(Z) and R(KDP)
Figure 2: Relationship between reflectivity and rain rate for various differential reflectivity factors. Larger ZDR
corresponds small µ. The thickness of each line represents the margin due to effective canting angle of raindrops.

underestimate rain by 40% while R(Z, ZDR)                      4.3 Indirect verification of µ-Λ relation and self-
overestimate rain by more than 20%. This shows the             consistency in polarization radar observation
inconsistency among the empirical relations. The
inconsistency has also been found in large data sets           To verify the validity of constrained Gamma rain DSD
[Brandes et al., 2001]                                         and self-consistency in polarization observation, we
                                                               compare the estimated differential phase and specific
Using the relations derived from constrained Gamma             differential phase with those measured. The estimated
DSD, we plot the estimated rain rate in Figure 3b. The         KDP is obtained from power measurements; i.e., Z and
physically-based relations are shown to give consistent        ZDR using Eq. (21) with a specified σφ = 12o
results and agree with the gauge measurement much              [Ryzhkov et al., 1999] and the estimated φDP is
better than the results obtained using classic relations       calculated as φ DP
                                                                                e
                                                                                   = 2∫ K DP
                                                                                           e
                                                                                              (l ) dl where l is the
based on a fixed power law. This is because the
                                                               distance along the radial. Figure 4 shows comparisons
constrained Gamma DSD based relations use the DSD
                                                               of differential phase between measurements and
information directly. The classic relations, however,
                                                               estimations. Figure 4a is an example of differential
were obtained using least-squares fitting, which
depends on the selection of DSD data sets that were            phase plotted as a function of range. As expected,
chosen for the fitting; for example, DSDs with a               estimated φDP monotonically increases with range and
certain range of DSD parameters and weight might               agrees well with the mean of the measurement. The
not be a true representation of natural rain drop size         statistical comparison is shown in the scatter plot in
distribution.                                                  Figure 4b. There are 20 rays and 1966 data points. The
                                                               mean of the measured φDP is 11.65 degrees and the
                                                               mean of the estimated value is 11.77 degrees. The bias
                                                               is negligible.
(a)

                                                                                     (b)

Figure 3: Comparison between various rain rate estimators and gauge measurements. (a) Fixed power-law rain
estimators are used for rain retrieval. Polarization radar-based rain accumulation show both under and over
estimation compared to the gauge observations and (b) Analytically derived rain estimators produce almost same
rain accumulation.

5. Summary and Discussions                                  DSD allows the DSD parameters retrieved from
                                                            polarization radar measurements: reflectivity and
Analytical relations for rain rate estimation and self-     differential reflectivity. General relations among the
consistency among Z, ZDR and KDP were investigated          moments and physical quantities such as reflectivity,
using constrained Gamma rain DSD in this paper. The         propagation phase and rain rate were derived. The
constrained condition (µ - Λ relation) is derived from      physically-based relations are used for a rain
video-disdrometer measurements. The two parameter           estimation and self-consistency check. It has been
(a)

                                                                                  (b)

Figure 4: Indirect verification of Λ-µ relation and self-consistency among polarization radar observations. (a) An
example of measured and estimated φDP along a radar beam, and (b) scatter plot of measured and estimated φDP a
number of segments.

shown that: (1) the µ - Λ relation holds at least for the     measurement with results being consistent with in situ
data collected in sub-tropical rain events such as in         observations.
Florida, and (2) physically-based relations show
smaller bias in rain estimation from polarimetric radar       Polarimetric measurements are sensitive to DSD,
                                                              shape and canting angle. Even though the µ - Λ
relation simplifies DSD representation, any difference    Haddad, Z.S., S.L. Durden and E. Im, 1996:
between assumed and actual microphysical parameters                Parameterizing the raindrop size distribution,
such as shape and canting angle might introduce                    J. Appl. Meteor, 35, 3-13.
significant uncertainties in polarization radar-based     Illingworth, A.J. and T.M. Blackman, 1999:The need
retrieval. The equilibrium shape of raindrops is                   to normalize RSD based on the Gamma RSD
assumed in this study while some observations suggest              formulation and implication for interpreting
that a more spherical shape should be adapted. The                 polarimetric radar data. Preprints, 29th Int.
equilibrium shape of raindrops is assumed for                      Conf. On Radar Meteor., Montreal. Amer.
maintaining continuity with earlier studies. A more                Meteor. Soc., Boston, 629-631.
accurate model may be used in future studies when the     Jameson, A.R., 1983a: Microphysical interpretation
axis ratio and canting angle are well understood.                  of multiparameter radar measurements in rain
                                                                   – Part I: Interpretation of polarization
Acknowledgements                                                   measurements and estimation of raindrop
                                                                   shapes, J. Atmos. Sci., 40,1792-1802.
The authors wish to thank Witold F. Krajewski and         Jameson, A.R., 1983b: Microphysical interpretation
Anton Kruger of Iowa State University for making the               of multiparameter radar measurements in rain
video disdrometer data available. They would also like             – Part I: Estimation of raindrop distribution
to thank J. Lutz and M. Randall of NCAR/ATD for                    parameters by combined dual-wavelength and
smooth operation of S-Pol radar. The study was partly              polarization measurements, J. Atmos. Sci.,
supported by funds from the National Science                       40,1803-1813.
Foundation that have been designated for the U.S.         Kozu, T., and K. Nakamura, 1991: Rain parameter
Weather Research Program at National Center for                    estimation from dual-radar measurements
Atmospheric Research (NCAR).                                       combining reflectivity profile and path-
                                                                   integrated attenuation, J. Atmos. and Ocean.
References                                                         Tech., 8, 259-270.
                                                          Oguchi, T., 1983: Electromagnetic wave propagation
Brandes, E.A., and A.V. Ryzhkov, and D.S. Zrnic,                   and scattering in rain and other hydrometeors.
         2001: An evaluation of radar rainfall                     Proc. IEEE, 71, 1029-1078.
         estimates from specific differential phase, J.   Richter, C., and M. Hagen, 1997: Drop-size
         Atmos. Ocean. Tech., 18, 363-375.                         distributions of raindrops by polarization
Brandes, E.A., and G. Zhang, and J. Vivekanandan,                  radar and simultaneous measurements with
         2001: Experiments in rainfall estimation with             disdrometer, windprofiler and PMS probes,
         a polarimetric radar in a subtropical                     Q. J. R. Meteorol. Soc., 122, 2277-2296.
         environmemnt, Submitted to J. Atmos. Ocean.      Ryzhkov, A. V., and D. S. Zrnic, 1995: Comparison of
         Tech.                                                     dual-polarization radar estimators of rain. J.
Bringi, V. N., and A. Hendry, 1990: Technology of                  Atmos. Oceanic Technol., 12, 249-256.
         polarization diversity radars of meteorology,    Ryzhkov, A. V., D. S. Zrnic, G. Huang, E. A. Brandes,
         Radar in Meteorology, American Meteorl.                   and J. Vivekanandan, 1999: Characteristics of
         Society, Boston, 153-190.                                 hydrometer orientation obtained from
Bringi, V.N, and V. Chandrasekar, 2000:                            polarimetric radar polarimetric measurements
         Polarimetric Doppler Weather Radar-                       in a linear polarization basis, Proc. of
         Principle and Application, Oxford Press.                  IGARSS’99, Hamburg, Germany, 702-704.
Chandrasekar, V. and V.N. Bringi, 1987: Simulation        Sachidananda, M., and D.S. Zrnic, 1987: Rain rate
         of radar reflectivity and surface                         estimates from differential polarization
         measurements of rainfall. J. Atmos. Ocean                 measurements, J. Atmos. Oceanic Technol.,
         Tech., 4, 464-478.                                        4, 588-598.
Dou, X., W. Liu and P. Amayene, and J. Liu, 1999:         Seliga, T. A., and V.N. Bringi, 1976: Potential use of
         Optimization of the parameter of the raindrop             radar differential reflectivity measurements at
         size distribution in rain rate measurement by             orthogonal polarizations for measuring
         airborne radar. Quart J. of Applied Meteor.,              precipitation, J. Appl. Meteor., 15, 69-76.
         Beijing, 10, 293-298.                            Seliga, T. A., and V.N. Bringi, 1978: Differential
Doviak, J.D, and D.S. Zrnic, 1993: Doppler Radar                   reflectivity and differential phase shift:
         and Weather Observations, 2nd ed. San                     Application in radar meteorology, Radio Sci.,
         Diego: Academic Press.                                    13, 271-275.
                                                          Testud, J., S. Oury, R.A. Black, P. Amayenc and X.
                                                                   Dou, 2001: The concept of "normalized"
distributions to describe raindrop spectra: a   Williams, C.R., A. Kruger, K.S. Gage, A. Tokay, R.
         tool for cloud physics and cloud remote                  Cifelli, W.F. Krajewski, and C. Kummerow,
         sensing, Accepted by J. Appl. Meteor..                   2000: Comparison of simultaneous rain drop
Tokay, A. and D.A. Short, 1996: Evidence for tropical             size distribution estimated from two surface
         raindrop spectra of the origin of rain from              disdrometers and a UHF profiler, Geophys.
         stratiform versus convective clouds, J. Appl.            Res. Letters, 27, 1763-1766.
         Meteor., 35, 355-371.                           Willis, P.T., 1984: Functional fits to some observed
Ulbrich, C.W., 1983: Natural variations in the                    drop size distributions and parameterization
         analytical form of the raindrop size                     of rain, J. Atmos. Sci., 41, 1648-1661.
         distribution,'' J. Appl. Meteor., 22, 1764-     Yuter, S., and R. A. Houze, 1997: Measurements of
         1775.                                                    raindrop size distributions over the Pacific
Ulbrich, C.W., and D. Atlas, 1998: Rain microphysics              warm pool and implications for Z – R
         and radar properties: analysis methods for               relations, J. Appl. Meteor., 36, 847-867.
         drop size spectra, J. Appl. Meteor., 37, 912-   Zhang, G., J. Vivekanandan, and E. Brandes, 2001: A
         923.                                                     method for estimating rain rate and drop size
Vivekanandan, G. Zhang, and A.V. Ryzhkov 1999b:                   distribution from polarimetric radar
         Estimation of canting angle distribution of              measurements, IEEE Trans. on Geoscience
         raindrop spectra using radar measurements.               and remote Sensing, 39, 830-841.
         Bull. Int. Radar Symp., Calcutta, India.
You can also read