1. New GSMaP over-land MWI algorithm
5th IPWG Workshop
New GSMaP over-land MWI algorithm
Kazumasa Aonashi
Meteorological Research Institute
Japan Meteorological Agency
My name is Kazumasa Aonashi, MRI/JMA.
2010/10/11
5th IPWG Workshop

2. Outline Introduction New over-land MWI algorithm Results
GSMaP MWI Algorithm
Problems in over-land retrieval
New over-land MWI algorithm
Basic idea
Estimation of Index of Dtop(r8537) and SRR(sigma85)
Statistical correction using r8537 and sigma85
Results
Over-land retrieval for 2004
Pakistan Flood case (July, 2010)
Summary & Future directions
The outline of my today’s talk is as follows:
2010/10/11
5th IPWG Workshop

3. MWR precip retrieval algotrithm
GSMaP Project
Aonashi (MRI)
Kubota (JAXA)
Shige (U. Kyoto)
TRMM
TMI
Aqua
AMSR-E
DMSP
SSM/I
NOAA
AMSU
MWR precip retrieval algotrithm
L2 Product from each sensor
Geo. Satellite
Mix
Infrared radiomter
Cloud motion vector
As Aonashi-san pointed out, microwave radiometer is an excellent tool for the accurate rainfall estimation. Up untill now, we have four sensors. The TRMm/tmi, aqua/amser-e, adeos/amsr, and dmsp/ssmi .by combininb these sensosrs, we have the composite product.
As is showed earlier, currently we have several satellites aboard microwave radiometere available now
Near Real Time
System
At JAXA/EORC
Kachi & Kubota
(JAXA)
L3 Composite
product
0.1degree
1 hour.
6 hour
1 day
1 month
Ushio (U. Osaka)
Composite algorithm of
IR and MWR
0.1degree・30min
2010/10/11
5th IPWG Workshop 4 4

4. Basic Idea of the Retrieval Algorithm
PCT37, PCT85　over land
Observed
TBs
Retrieval Calculation
Forward calculation
Statistical
Precip-related
Variable Models
RTM
Screening
Inhomogeneity
estimation
Scattering part
Emission part
Look-up
Table
Precip Profiles
DSD
Mixed phase
inhomogeneity
The basic idea of this algorithm is to find the optimal precipitation that gives RTM-calculated TBs fitting best with the observed TBs:
PCT37, PCT85 over land;
TB10v,TB19v, TB37v, PCT37, PCT85 over sea.
This algorithm consists of two parts:
Forward calculation part that computes LUTs between precip and TBs by incorporating the precip cloud models into RTM;
Retrieval part that derives precip from the observed TBs using the LUTs.
Precip.
Find the optimal precipitation that gives RTM-calculated TBs
fitting best with the observed TBs:
2010/10/11
5th IPWG Workshop

5. Statistical precip-related var. models from TRMM observation
Precip type classification
Data base
10 types (land 6, sea 4) are classified from TRMM PR data
(2.5 deg, 3 monthly)
Precip Profile
(land) 0: thunderstorm, 1: shower, 2: shallow, 3: frontal rain, 4: organized rain 5: highland
(sea) 6: shallow 7:frontal rain,
8:transit, 9:organized rain
Precip profile data base
As for the precipitation profile, we used statistical profile data produced from TRMM PR 2A25 datasets.
Height from 1 deg level [km]
Example：
TRMM PR averaged preciptation profiles for each type, surface precip, conv/stra
1℃ level
2010/10/11
Rainfall rate [mm/h]
5th IPWG Workshop 9

6. GPROF,GSMaP vs PR for July ‘98
(b)
Over Land
GPROF retrieval (mm/hr)
GSMaP retrieval (mm/hr)
PR precip rate (mm hr-1)
PR precip rate (mm hr-1)
(c )
(d)
Over Ocean
GPROF retrieval (mm/hr)
GSMaP retrieval (mm/hr)
PR precip rate (mm hr-1)
PR precip rate (mm hr-1)
2010/10/11
5th IPWG Workshop

7. Rainsurf vs. Rainspc (Land ‘98) in terms of Dtop (toplev-ZFLH) and SRR(stratiform rain ratio)
Dtop > 4km
Dtop <2km
SRR > 0.7
SRR < 0.3

8. New over-land algorithm
Estimation of Index of Dtop and SRR
Statistical correction using these indexes
2010/10/11
5th IPWG Workshop

9. Index of Dtop, R8537 Index of Dtop:
the ratio of TB85 depressions to TB 37 depressions (R8537).
R8537 in terms of ratio of precipitation retrieved from TB85 to TB37 using the conventional GSMaP algorithm.
Retrain.v
match-up data　(Land ‘98)

10. Ratio of TMI scattering signals to PR in terms of precip top level (July, 1998)
(Rain37/rainsurf)
(Rain85/rainsurf)
In addition, the over-land algorithm tends to overestimate precip for disturbances with high top levels.
As these scatter diagrams show, ratio of rain85 to surface precip has significant sensitivity to PR top levels, compared to rain37 to precip ratio.
This frequency dependency suggests problems in our PSD model.
PR top level –FLH (m)
PR top level –FLH (m)
TB85 depr becomes larger than TB37 depr.
for higher Dtop.
2010/10/11
5th IPWG Workshop

11. Forward calculation with different Dtoplev
TB85 depression became larger than the TB37 depression
for precipitation with higher top levels.

12. Index of Dtop, R8537 Index of Dtop:
the ratio of TB85 depressions to TB 37 depressions (R8537).
R8537 in terms of ratio of precipitation retrieved from TB85 to TB37 using the conventional GSMaP algorithm.
Retrain.v
match-up data　(Land ‘98)

13. Index of SRR, sigma85
First guess of Sigma85 from Rain85 within the TB10v FOVs.
Adjustment using the statistical relationship between Sigma85 and PR (Kubota et al. 2009).
Retrain.v
match-up data　(Land ‘98)

14. Statistical correction using r8537 and sigma85

15. Rainsurf vs. rain37 Land ’98 Class by (r8537, sigma85)

16. Rainsurf vs. rain85 Land ’98 class by (r8537, sigma85)

17. Rainsurf vs. rain37 Land ’98 (ktype
Rainsurf vs. rain37 Land ’98 (ktype !=6) class (r8537,sigma85): Seasonal change
Winter
Spring
Summer
Autumn

18. Rainsurf vs. rain37 Land ’98 class (r8537,sigma85): Precip type change
ZKTYPE=5
ZKTYPE=6
TRMM data sets for 1998 are classified
by (R8537,sigma85) for each precip type.

19. Statistical correction using (R8537,sigma85)
TRMM data sets for 1998 are classified by (R8537,sigma85) for each precip type.
Linear fitting coefficients between rain37, rain85 and PR surface precipitation rates.
Weighted sum of corrected rain37 and rain85.

20. Over-land retrieval for 2004 Pakistan Flood case (July, 2010)
Results
Over-land retrieval for 2004 Pakistan Flood case (July, 2010)

21. Correction of over-land retrievals using (R8537, sigma85)
Rainspc vs. rainsurf for Land Jan. ‘04
Before Correction
After Correction
( )
( )

22. Rainsurf vs. Rainspc (Land Jan. ‘04) Class by Dtop and SRR
Sigma85>1
R8537>1
Sigma85>1
R8537<1
Sigma85<1
R8537>1
Sigma85<1

23. Rainsurf vs. Rainspc (Land Jan. ‘04) Class by Dtop and SRR
Sigma85>1
R8537>1
Sigma85>1
R8537<1
Sigma85<1
R8537>1
Sigma85<1

24. Pakistan Flood case (July 21, 2010)

25. Future Directions Improvement of the forward calculation
Apri-ori information for precip-related variables

26. Improvement of the forward calculation:
Non-spherical particle effects
Introducing the effect into RTM:
1) Realistic non-spherical model
2) FDTD calculation
3) Parameterization for fast RTM
2010/10/11
5th IPWG Workshop

27. Realistic non-spherical frozen particle model
(Ishimoto, 2008)
Fractal dimension Ｄｆ
1.8
1.9
2.0
2.1
2.2
2.3
2.4
Monte Carlo
calculation of
particles for given
fractal dimension
2010/10/11
5th IPWG Workshop

28. D0: diameter of solid sphere
Parameterization introduced to the Fast RTM code
1) Approximation of Roeff
in terms of Dmax
2) Softness parameter
in terms of df, f,
and the frequency
Non-
Sphere
SP=(D-D0)/(Dmax-D0)
Keeping the single scattering properties
Sphere
We are also planning to introduce scattering properties of non-spherical frozen particles. Though some studies used DDA for this purpose, we think this method is too expensive. Hence we are trying to use SP proposed by Liu (2004) that seeks spherical particles with the same single scattering properties of the particles. D
Dmax
D0: diameter of solid sphere
2010/10/11
5th IPWG Workshop 45 45

29. Orographic Rainfall (Shige & Taniguchi, 2010)
Warm rain with less ice precipitation
Freezing Level
Orographic
A large amount of condensates
Low-level orographic lifting
Ice
Water
Maritime Air
Mountain
Ocean 47

30. Topographically induced upward motion
Classification of Orographic/No-Orographic Rain (Shige & Taniguchi, 2010)
Topographically induced upward motion
Low-level moisture convergence
Slope from GTOP
Winds and Water Vapor from GANAL
Winds from GANAL

31. Summary New over-land algorithm: Estimation of Index of Dtop and SRR
Statistical correction of LUTs
Results
Future directions
2010/10/11
5th IPWG Workshop