Annual Journal of Hydraulic Engineering, JSCE, Vol.

49, 2004, September

289


DEVELOPMENT OF 1-D CLOUD MICROPHYSICS
DATA ASSIMILATION SYSTEM (CMDAS) BY
USING AMSR-E DATA



Cyrus Raza MIRZA
1
, Toshio KOIKE
2
, Kun YANG
3
and Tobias GRAF
4


1
Student Member of JSCE, M. Sc., Dept. of Civil Eng., Uni. of Tokyo (Bunkyo-ku, Tokyo 113-8656, Japan)
2
Member of JSCE, Dr. Eng., Professor, Dept. of Civil Eng., Uni. of Tokyo (Bunkyo-ku, Tokyo 113-8656, Japan)
3
Dr. Eng., Associate Professor, Dept. of Civil Eng., Uni. of Tokyo (Bunkyo-ku, Tokyo 113-8656, Japan)
4
Student Member of JSCE, Dr. Eng., Dept. of Civil Eng., Uni. of Tokyo (Bunkyo-ku, Tokyo 113-8656, Japan)


In this paper, an 1-D Cloud Microphysics Data Assimilation System (CMDAS) is developed for retrieving
reasonable cloud distribution. The general framework of CMDAS includes the Kessler warm-rain cloud
microphysics scheme, a microwave radiative transfer model in the atmosphere and a global minimization
method of Shuffled Complex Evolution (SCE). The paper investigates potentials of the CMDAS to modify the
cloud properties by considering integrated cloud liquid water content as a assimilation parameter and to
introduce the heterogeneity into the initial state of the atmosphere, by applying the CMDAS to the satellite
microwave radiometer data set obtained by the international cooperative observation experiment, Wakasa
Bay Experiment 2003, in Japan. The simulated microwave brightness temperature by CMDAS is in good
qualitative and quantitative agreement with the observed one.

Key Words: Data assimilation, Cloud Liquid Water Content, Cloud Microphysics, Shuffled Evolution Method,
Kessler Warm Rain Macrophysics, Radiative Transfer Model, Microwave Remote Sensing


1. INTRODUCTION

For reliable numerical weather prediction, complete
and accurate description of the initial state of the
atmosphere is required. But the current operational
in-situ observation systems, which are used in
numerical weather prediction (NWP) to
intermittently update and improve our knowledge of
the current atmospheric state, can not observe the
cloud liquid water content (CLWC). Therefore, the
unavailability of CLWC in the observation data may
result into poor initialization of atmospheric model,
giving unreliable prediction of precipitation.
In order to address this problem, remotely sensed
data from space can be used for getting critically
important information for better understanding of the
cloud microphysics processes and for improving
predictability on its effects on weather. Various
studies have shown that the use of satellite derived
products in diabatic or physical initialization can
reduce the model spinup and improve the short-range
forecast which in turn can improve the first guess for
the analysis
1), 2), 3), 4)
. By viewing the significance and
sophistication of cloud microphysics processes and
its effects on the NWP model initialization, this
research is directed towards satellite data
assimilation.
Over the last few years, the variational form of
statistical estimation has been implemented at many
operational centres. Lorenc showed that the
statistical estimation problem could be cast in a
variational form (3D-Var) which is just a different
way of solving the problem that the so-called optimal
interpolation attempts to solve directly
5)
. Eyre
showed, in a 1D-Var context, that a variational
formulation leads to a more natural framework for
the direct assimilation of radiances instead of
retrieved temperature and humidity profiles. This is
also true for any indirect measurement of the state of
the atmosphere
6)
. Talagrand and Courtier showed
that the use of the adjoint of a numerical model
makes it possible to determine the initial conditions
leading to a forecast that would best fit data available
over a finite time interval
7)
. These two formulations
can be combined to yield what is now

290
















Fig.1 General framework of CMDAS

estimation problem or 4D-Var. Therefore, it has been
realized that the assimilation of satellite derived
rainfall and related data types can offer a way to
compensate for model deficiencies and tightness
estimates of atmospheric parameters given by
different assimilation systems.
The present study explores the variational
framework of an 1-D Cloud Microphysics Data
Assimilation System (CMDAS) for retrieving
reasonable cloud distribution. This paper investigates
potentials of the CMDAS to modify the cloud
properties by considering integrated cloud liquid
water content as a assimilation parameter and to
introduce the heterogeneity into the initial state of the
atmosphere, by applying the CMDAS to the satellite
microwave radiometer data set.

2. DEVELOPMENT OF CMDAS

Fig. 1 shows the general framework of the 1-D
CMDAS in this study. It includes the Kessler
warm-rain cloud microphysics scheme as a model
operator, a microwave radiative transfer model
(RTM) for the atmosphere as an observation operator
and a global optimization method of SCE.
In order to provide the first guess to run CMDAS, a
non-hydrostatic model named Advanced Regional
Prediction System (ARPS) developed by the Center
for Analysis and Prediction of Storms (CAPS) at the
University of Oklahoma is used by coupling with the
global analysis data set. ARPS includes four
packages: an atmospheric model, a land surface
scheme, a radiation package and parameterization of
cloud microphysics. Therefore from simulation of
ARPS, control variables of relative humidity, mixing
ratios of cloud water and rainwater, accumulated grid
scale rainfall, rain rate profile, pressure and base state
air density are obtained in order to run the model
operator.
Considering that the satellite microwave radiometer
can provide an integrated value of CLWC (ICLWC)
but not a vertical profile, we chose the ICLWC as an
assimilation target parameter. In addition, we
introduce a parabolic CLWC distribution with zero
values at cloud top and bottom, for combining the
ARPS outputs and the assimilation target, as it looks
feasible regarding to the cloud system development.
The Kessler warm-rain cloud microphysics scheme
is run by using the available first guess of the control
variables provided from ARPS and modifies the
variables of water vapor specific humidity, pressure,
air temperature, perturbation potential temperature.
All these variables along with other supplementary
parameters, salinity, satellite frequency, drop size
distribution, are input to the observation operator i.e.,
radiative transfer model (RTM) to estimate the
microwave brightness temperature observed by
satellites.
During the assimilation process, the ICLWC is
updated, until the simulated microwave brightness
temperature (TB
sim
) is in good agreement with the
observed one (TB
obs
). Then SCE minimization
method tries to find the best assimilated parameter to
minimized the discrepancy between TB
sim
and TB
obs
..
In this way, we found better initial condition of the
targeted parameter.
Each component of the algorithm is described in
detail as follows.

(1) Cost Function
The assimilation scheme is used to minimize the
cost function J by adjusting state vector x
9)
. In
general J can be separated into two different costs
i.e., the background error J
B
and the observation error
J
0
:

0
J J J
B
+ = (1)

For this application the background error J
B
is
neglected. Therefore J reduces to the observation
error J
0
, which is expressed as:
) (
0 0
x J J = (2)
| | ( ) | | ( )

=
N
i
i i
T
i i
y x H R y x H J
1
0 1 0
0
2
1
(3)
y
i
0
is a vector representing the satellite observation at
89 GHz (vertical and horizontal polarization), which
will be assimilated. H is the radiative transfer model
(observation operator) and R is the error covariance
matrix of the observation. x
i
represents the state
vector at time t
i
calculated by the cloud microphysics
model M (model operator) using the state vector x
0
at
time t
0
as initial conditions.
Tb
obs

IF Jmin
No
Yes
ARPS Model Output
(Initial Guess)
Observation Operator
(RTM) (Tb
sim
)

Model Operator
(Cloud Microphysics)
Cost (J)= (Tb
sim
- Tb
obs
)
2

Global Optimization
Scheme (Shuffled
Complex Evolution)
Optimized
Initial
Condition
Cloud Parameters Update

291
R is assumed as unit matrix because of no
information to estimate it while environmental
forcing data are neglected at the moment. The state
vector x
i
comprises of
) , (
0 i i
t x M x = (4)
where x
o,
x
i
represents the intial state and state at time
t
i
of the ICLWC respectively. Combining Eq.(3) and
Eq.(4) yields the cost function for the assimilation
scheme:
| | ( )
| | ( )
0
0
1
0
0 0
) , (
) , (
2
1
) (
i i
N
i
T
i i
y t x M H
y t x M H x J

=

=
(5)
The above equation shows that the TB
obs
is directly
included into the minimization process of the cost
function J as
0
i
y .

(2) Model Operator
To address the condition of Japan Sea in winter,
the effect of the ice particles physics have to be
considered but however, the ice schemes consume
long computational time and physically more
complex. In order to understand the effectiveness of
CMDAS framework along with computationally less
cost, we select the Kessler warm rain microphysics
scheme as a model operator in this study.
As at this stage of the paper, we focus only on the
assimilation of cloud rather than assimilation of
precipitation. Due to that reason, the snowfall
particles have not been introduced into this data
assimilation system. Therefore to remove
complicated processes including the evolution of
cloud condensate from very small water or ice
particles up to precipitation-size particles, we chose
simple microphysical parameterizations, employing
as few field variables as possible to represent cloud
condensate
9)
seems more reasonable at a first step of
system development rather than using complex ice
microphysics.
Fig. 2 explains the processes involved in Kessler
scheme
10),11)
, which includes three categories of
water i.e., water vapor, cloud liquid water and rain
water. Each of the liquid water forms is implicitly
characterized by a droplet distribution. Small cloud
droplets are first formed when the air becomes
saturated and nucleation occurs. If the cloud
water-mixing ratio exceeds a threshold value,
raindrops are formed by auto-conversion from the
cloud droplets.
The raindrops then collect smaller cloud
droplets by accretion as they fall at their terminal
speed. The evaporation rate is used only when the air
is unsaturated. The saturation adjustment scheme


Fig.2 Processes involved in Kessler scheme


computes the amount of water vapor converted to
cloud water if super-saturation exists, or the amount
of cloud water evaporated if sub-saturation exists.
The adjusted values for temperature, water vapor
specific humidity, cloud water mixing ratio and
rainwater mixing ratio are obtained at each
integration time step and are finally passed to the
observation operator.

(3) Observation Operator
RTM as the observation operator is used to
calculate the microwave brightness temperature
corresponding to the outputs from the model
operator. Considering the scattering effect of snow
particle as well, the 4-stream fast model developed
by Liu
12)
is used as an observation operator.
The radiative transfer in a plane-parallel and
azimuthally symmetric atmosphere with spherical
particles can be expressed by Tsang and
Kong
13)
.
( ) 6
1
1
) ( ) 1 (
'
) , (
) , (
2
) , (
) , (
) , (
) , (
0
1
1
0

u
u
u

u
u
u
u

u
B
d
I
I
P P
P P
I
I
I
I
d
d
H
V
HH HV
VH VV
H
V
H
V
where ( ) u ,
P
I is the radiance at optical depth in
direction u for vertical or horizontal polarization,
o
is the single-scattering albedo, ( ) B is the Plank
function and
ij
P (i, j = H or V) is the scattering phase
functions. The 4stream model solves Eq.(6) by using
the Discrete Ordinate Method (DOM) and
introducing the approximations that no
cross-polarization exist. The scattering phase
function is expressed by the Henyey-Greenstein
formulation
14)
. These considerations and the
limitation of the number of streams to four make it
possible to solve Eq. (6) efficiently from the

292
computational point of view. In order to solve Eq. (6)
boundary conditions for the top and the bottom of the
atmosphere need to be applied. The upper boundary
layer is assumed to be a constant stellar background
with a brightness temperature of 3
o
K. The bottom of
the atmosphere is bounded by the ocean surface.
As we are not using ice microphysics as a model
operator, therefore scattering effect of snow particles
are not taken into consideration.

(4) Shuffled Complex Evolution (SCE) Method
In order to reduce the cost function, the algorithm
uses heuristic global optimization method of SCE
15)
,
which has become one of the most popular among
water resources engineers. It involves the evaluation
of the function usually at a random sample of points
in the feasible parameter space, followed by
subsequent manipulations of the sample using a
combination of deterministic and probabilistic rules.
It guarantees asymptotic convergence to the global
optimum. The basis of its algorithm is based on an
iterative method, where the difference between
modeled and observed brightness temperature is
minimized by adjusting the assimilated parameter.
Mathematically, this can be expressed as
( ) ) 7 ( min
,
1
|
.
|

\
|

=
P M T
i B
L
i

where
i B
T
,
is the brightness temperature observation
at a specific frequency and polarization i and M is
the radiative model using the atmospheric parameters
of interest P which in our case is the ICLWC.

3. CASE STUDY

A case study is conducted over the Sea of Japan on
25
th
Jan 2003 under the strong winter monsoon. The
domain area includes the field of the Wakasa Bay
Experiment 2003, which is close to the city of Fukui.
Its size is 240 km x 240 km x 11 km with vertical and
horizontal resolutions of 0.375 km and 4 km
respectively. The GANAL data 136
o
E longitude and
36.2
o
N latitude is used as the initial condition of the
simulation.
Rough estimate of cloud top (Fig. 3) of 3500 m is
derived from available data from
Moderate-resolution Imaging Spectroradiometer
(MODIS) onboard of Aqua. This ensures that the
time of the cloud top and the microwave brightness
temperature observations should be matched, while
no observation data of the cloud bottom is available;
therefore it is assumed to be 2500 m below the cloud
top.
To have initial condition (x
o
) of CMDAS, firstly
















Fig.3 Estimated Cloud Top from MODIS for 25
th
Jan 2003 at
03:55z.

ARPS model simulation is setup to run at 25
th
Jan
2003 from 00:00z to 03:55z by initializing it with the
use of external global re-analysis (GANAL) data set.
The simulated output of ARPS at 03:30z is then
used as a initial condition (i.e first guess) in order to
run the assimilation system of CMDAS. The
CMDAS is run for assimilation window of 25
minutes in order to compare the results of TB
sim
and
TB
obs
at the observation time of 03:55z.

(1) Data Sets
a) AMSR-E Brightness Temperature Data
The AMSR-E brightness temperature data is
provided by Japan Aerospace Exploration Agency
(JAXA) for the area of 131
o
E to 141
o
E and 31
o
N to
41
o
N. The data is grided at a resolution of 0.05
o
. It
should be noted that the actual average footprint size
for the 89.0 GHz frequency channel is 5.4 km
16)

which represents around 0.04
o
at this latitude.
b) GANAL Atmospheric Data
The GANAL data set provided from the Japanese
Meteorological Agency (JMA) contains sea level
pressure, geo-potential height, air temperature and
dew point temperature with a 1.25
o
for horizontal
resolutions, and 13 vertical levels from surface to 100
hPa, which correspond to a height of 16 km and is
sufficient for the radiative transfer in the microwave
region. It provides temperature, pressure and relative
humidity every 12 hours (00z and 12z).

(2) Results and Discussions
In this paper a basic study is presented to develop an
algorithm in order to get a reliable cloud distribution
by applying the CMDAS to the AMSR-E data and
the GANAL outputs.
The CMDAS during the assimilation of ICLWC



293














Fig.4 TB
sim
at 89.0H GHz without CMDAS retrieved for 25
th
Jan
2003 at 03:55z.

modify the water vapor field, temperature field,
pressure field, mixing ratio of cloud water and
rainwater.
In order to identify effects of the CMDAS on the
cloud distribution mapping, the results of the
simulation in two cases, i.e. without the CMDAS
(Case 1) and with (Case 2) the CMDAS, are
compared.
For Case 1, the simulated microwave brightness
temperature at 89.0H produces a completely
homogeneous structure of brightness temperature as
shown in Fig.4. It cant be successful to generate
cloud formation effectively mainly due to
homogeneity of the external GANAL data set i.e. the
poor initial conditions. It is also suspected that water
vapor is not sufficient for reliable prediction of cloud
distribution and precipitation. Therefore, there is
need to incorporate the information of CLWC in the
initialization of model for improved forecast of
precipitation.
In Case 2, the short assimilation time period is
selected due to neglecting impacts of environmental
forcing, which cause changes very dramatically, for
checking the impacts of the CMDAS at the cloud
distribution mapping.
Fig. 5 shows that the assimilation system improves
the performance of cloud microphysics scheme
significantly. Thats mainly due to intrusion
of heterogeneity into the external GANAL data,
which resultantly improved atmospheric initial
conditions by considering the ICLWC as the
assimilation parameter.
Fig. 6 shows the difference between observed and
simulated brightness temperature after the iteration
stopped in Case 2. Almost all parts of the simulation
result show very good agreement with the observed
one with around K 2 discrepancy. Fig. 7 and
























Fig.6 Difference between TB
sim
(Fig. 5) and TB
obs
by AMSR-E at
89.0H GHz for 25
th
Jan 2003 at 03:55z.












Fig. 5 TB
sim
at 89.0H GHz with CMDAS for 25
th
Jan 2003 at 0





Fig.7 Initial ICLWC [g/m
2
] retrieved for 25
th
Jan 2003 at 03:30z.



Fig. 5 TB
sim
at 89.0H GHz with CMDAS for 25
th
Jan 2003 at
03:55z.

294














Fig.8 Final ICLWC [g/m
2
] retrieved for 25
th
Jan 2003 at 03:55z.

Fig. 8 reveal that the CMDAS has significantly
improved the amount of assimilated ICLWC. Final
ICLWC (Fig. 8) result shows the comparable
structure of cloud system with MODIS image for
cloud top (Fig. 3). ICLWC horizontal spatial
distribution output seems better than simulation
results without the CMDAS (Case 1).

4. CONCLUDING REMARKS

The CMDAS shows the ability to introduce the
spatial heterogeneity of ICLWC into the
downscaling for the global model to the regional one.
Regarding the future step, there is need to consider
the environmental forcing data for the assimilation
system in order to dynamically modify the
atmospheric fields.
Furthermore, there is need to validate the products
of CMDAS i.e., whether the new initial conditions
provided by CMDAS can i mprove weat her
prediction of forecast or not. Therefore by using
CMDAS output as new initial conditions of the
mesoscale model especially for remote areas where
no gauge network can be found, we can make initial
conditions for finer grid scale model in combination
with observed satellite brightness temperature data.
By this downscaling method, we may get more
reliable forecast of precipitation, which will
ul t i mat el y cont r i but e t o ungauged basi ns.

ACKNOWLEDGMENT: This study was carried
out as a part of Coordinated Enhanced Observing
Period (CEOP) funded by the Japan Science and
Technology Agency and the Special Coordination
Funds for Promoting Science and Technology. The
authors express their so deep gratitude to them.
REFERENCES
1) Krishnamurti, T. N., H. S. Bedi, and K. Ingles: Physical
Initialization using SSM/I rin rates. Tellus, 45A, pp.
247-269, 1993.
2) Puri, K., and M. J. Miller: The use of satellite data in the
specification of convective heating for diabatic initialization
and moisture adjustment in numerical weather prediction
models. Mon. Wea. Rev., 118, pp. 67-93, 1990.
3) Heckley, W. A, G. Kelly, and M. Tiedtke: On the use of
satellite derived heating rates for data assimilation within
tropics. Mon. Wea. Rev., 118, pp. 1743-1757, 1990.
4) Mathur, M. B., H. S. Bedi, T. N. Krishnamurti, M.
Kanamitsu, and J. S. Woolen: Use of satellite derived
rainfall for improving tropical forecasts. Mon. Wea. Rev.,
1202, pp. 2540-2560, 1992.
5) Lorenc, A.C., S. P. Ballard, R. S. Bell, N. B. Ingleby, P. L.
F. Andrews, D. M. Barker, J. R. Bray, A. M. Clayton, T.
Dalby, D. Li, T. J. Payne and F. W. Saunders: The Met.
Office Global threedimensional variational data
assimilation scheme. Quart. J.R. Meteor. Soc., 126, pp.
2991-3012, 2000.
6) Eyre, J.R.: Inversion of cloudy satellite sounding radiances
by nonlinear optimal estimation. I: Theory and simulation
of TOVS. Quart. J.R. Meteor. Soc., 115, pp. 1001-1026,
1989.
7) Talagrand O. and P. Courtier: Variational assimilation of
meteorological observations with the adjoint vorticity
equation. I. Theory. Quart. J. R. Met. Soc. , 113, pp.
1311-1328, 1987.
8) Philippe Courtier : Variational methods. J. Meteor. Soc.
Japan, 75(1B), pp.211218, 1997.
9) Grabowski, W. W.: Toward cloud modeling of large scale
tropical circulations: a simple cloud microphysics
parameterization. J. Atmos. Sci., 55, pp. 3283-3298.
10) Klemp, J. B., and Wilhelmson: The simulation of
three-dimentional convective storm dynamics. J. Atmos.
Sci., 35, pp.1070-1096, 1978.
11) Soong, S-T., and Y. Ogura: A comparison between
axi-symmetric and slab-symmetric cumulus cloud models.
J. Atmos. Sci., 30, pp.879-893, 1973.
12) Liu, G.: A Fast and Accurate Model for Microwave
Radiance Calculations. J. Meteor. Soc. Japan, 76(2),
pp335-343, 1998.
13) Tsang, L., and J. A. Kong: Theory of thermal microwave
emission from a bounded medium containing spherical
scatteres. J. Applied Physics, 48, pp.3593-3599, 1977.
14) Heney, L. C., and J. L. Greenstein: Diffuse radiation in the
galaxy. Astrophysical Journal, 93, pp.70-83, 1941.
15) Duan , Q., S. Sorooshian, and V. Gupta: Effective and
efficient global optimization for conceptual rainfall-runoff
models. Water Res. Research, 28(4), pp.1015-1031, 1992.
16) JAXA: AMSR/AMSR-E data for validation. 2004.

(Received September 30, 2004)