PEHRPP Workshop Dec. 3, 2007 Geneva GSMaP Passive Microwave Precipitation Retrieval Algorithm Kazumasa Aonashi (MRI/JMA)
Passive Microwave Precipitation Retrieval GSMaP Retrieval Algorithm Global Satellite Mapping of Precipitation Project started in 2003. Leader: Prof. Ken ich Okamoto (Osaka Pref. Univ.) Funded by JST/CREST The goal is to produce accurate precip map using mainly satellite microwave radiometer. Passive microwave precip retrieval (Aonashi) Passive microwave precip+ IR wind (Dr. Ushio)
Outline Introduction Algorithm Description Forward Calculation (precip cloud models) Retrieval Part Validation (TRMM PR, Ground-based Obs.) Future Directions Improvement of Scattering part Summary
Algorithm Description Forward Calculation Retrieval Part
Basic Idea of the Retrieval Algorithm Forward calculation Precip Cloud Models FLH Precip Profiles DSD Mixed phase inhomogeneity RTM Look-up Table Observed TBs Precip. Retrieval Calculation Screening Inhomogeneity estimation Scattering part Radiation part Find the optimal precipitation that gives RTM-calculated TBs fitting best with the observed TBs: PCT37, PCT85 (land) TB10v,TB19v, PCT37, PCT85 (sea)
Precipitating Cloud Model for forward calculation Stratiform/ Convective Freezing Level Height Frozen Precip Rain Precipitation typ Precipitation Profile model Freezing Level Mixed-phase model Particle Size Distribution Atmosphere & Surfac (GANAL)
Parameters used in the Algorithm: Atmospheric & surface variables Atmospheric variables (Temp,FLH), surface variables(ts, SSW, SST) are derived from the Global Analysis data of JMA Temperature bias of GANAL against sonde Freezing Level Height for Jan.1, 2003
Precip type classification Data base 10 types (land 6, sea 4) are classified from TRMM PR data (2.5 deg, 3 monthly) Precipitation Profile Model Height from 1 deg level [km] Precip Profile 1 level (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 Example: TRMM PR averaged preciptation profiles for each type, surface precip, conv/stra Rainfall rate [mm/h]
Particle Size Distribution DSD for rain: Kozu model (2A25 average distribution calibrated with averaged epsillon) epsillon =1 for stratiform rain μ N( D) = N0D exp( ΛD) PSD for frozen particles: Marshall-Palmer distribution Data base of conv. Epsillon Averaged for each precip type
Nishitsuji (Mixed-Phase) Model for Stratiform Rain On the basis of the filed experiment, the following parameters are modeled Volume liquid water fraction (Pw) ε s 1 ε w 1 ε i 1 ε a 1 = Pw + Pi + Pa shape parameter of the dielectric constant (U) ε + U ε + U ε + U ε + U DSD parameter (B) is a function of Pw Ba 3 1 N( D) = N 010 ( m mm ) (a Density ρ= Pw Fall velocity Magono-Nakamura(1965) for snow and Foot and Du Toit for rain s w i a : radius in cm) Pw and U profile U 10 0 10 2 10 4 10 6 0 Relationship between B and Pw Distance below BB (km) 0.5 1 U PW B 40 20 0 0.5 1 PW 10 3 10 2 10 1 10 0
where μ = P is LUT calculation (1) TBs for homogeneous precip dtb( τ, μ, ϕ) μ = TB (1 ω0 ) T ( τ ) dτ ω0 ' ' ' ' ' ' P( τ, μ, ϕ, μ, ϕ ) TB( τ, μ, ϕ ) dμ dϕ 4π phase cosθ, τ = function K K dz, ω /( K Radiative Transfer Code (Liu,1998) One-dimensional model (Plane-parallel) Mie Scattering (Sphere) 4 stream approximation Calculate TBs for homogeneous, convective & stratiform precip with each precip types. K ab + sc 0 = sc ab + sc ), K
LUT calculation (2): LUTs for inhomo precip The calculated TBs are converted into TBs for inhomogeneous precip with Aonashi and Liu s s method (2000). LUT used for retrieval is weighted average of convective & stratiform TBs. 2.0 LOOK-UP TABLE (LUT) 1.5 STD of Log(Pr) ) is estimated from STD of Log(rain85) statistically. SIGMA85 1.0.5 0.414+ 0.678*sigma85o SIGMA85O 0.0 0.0.5 1.0 1.5 2.0 STDLGPRH STD LN(PR) (PCT85ra < 260K)
Flow of the Retrieval Part Observed TBs Screening of Precip Areas rain flag Inhomo. Estimation / LUT selection LOOK-UP TABLE (LUT) Inhomogeneity (STD of Log (Pr)) First-guess of Precipitation (scattering) rain37 PCT37+LUT rain85 PCT85+LUT TB10v,TB19v Minimization of Σ(TBc-TBo)**2 Over Ocean Retrivals Over Land Retrievals
Validation Comparison with TRMM PR & Ground-based observations
Multi-Satellite Precip Composite (GSMaP_MWR, daily precip) TMI+AMSR+AMSR-E+SSM/I (F13, F14, F15), 0.25 0.25
Zonal Mean TRMM Precip over Ocean GSMaP, GPROF, PR (1998~2006) PR 2A25V6 TMI 2A12V6 GSMaP V4.8.4 1998-2006 PR swath only
Comparison with ground-based radar (0.25 x 0.25 deg in lat-lon lon grid) COBRA(Okinawa) 4 cases in June 2004 Kwajalein Radar 10 cases in May 2003 Corretation:0.82(No:253) RMSE:1.37 mm/hr Correlation:0.65(No:1139) RMSE:1.78 mm/hr
PR 2A25V6 TMI 2A12V6 GSMaP V4.8.4 1998-2006 PR swath only Zonal Mean TRMM Precip over Land GSMaP, GPROF, PR (1998~2006)
Comparison with GPCC data Comparison with GPCC data GSMaP_MWR:monthly mean preciptation (1x1 deg) GPCC Monthly Precipitation (Monitoring) Product (Rudolf et al. 2006): Correlation: 0.80 for 45S~45N (sample: 69440) 0.85 for 15S~15N (sample: 2177) Fitting: y=1.14 x + 10.1 [40S~40N] y=1.21 x + 19.2 [15S~15N]
Ratio of TMI scattering signals to PR in terms of precip top level (July, 1998) (Rain37/rainsurf) (Rain85/rainsurf) rain37/rainsurf 50 40 30 20 10 5 4 3 2 1.5.4.3.2.1.05.04.03.02.01 R 2 乗 = 0.0843-2000 2000 6000 10000 0 4000 8000 PR DTOP top level FLH (m) rain85/rainsurf 50 40 30 20 10 5 4 3 2 1.5.4.3.2.1.05.04.03.02.01 R 2 乗 = 0.2697-2000 2000 6000 10000 0 4000 8000 PR DTOPtop level FLH (m) Precip is over-(under-)estimated for PR high (low) top level. (Rain85/rainsurf) is sensitive to PR top level.
Future Directions PSD, densities of frozen particles Scattering properties of Non-spherical particles
Estimation of realistic PSD, density etc. PSD (field campaign) RTM simulation Observed Radar & MWR data
Scattering properties of non-spherical frozen particles (Liu, 2004) DDA :Each An of the actual dipoles target is subject is approximated to an electric by an field array which of dipoles. is the sum of the incident wave and the electric fields due to all of the other dipoles. Non- Sphere From Mishchenko et al (2000) Dmax D0: diameter of solid sphere SP=(D-D0)/(Dmax-D0) D Keeping the single scattering properties Sphere
Summary GSMaP passive microwave precipitation retrieval algorithm: Atmopheric & surface variables from GANAL Precip profile, DSD & inhomo. from PR statistics Mixed-phase model The retrieved precipitation agreed well with PR, radar data over ocean. The over-land algorithm underestimated the GPCC precipitation, and showed bias in terms of precipitation top level. Introduction of the scattering of non-spherical particles, realistic PSD etc.
Thank you. END
放射伝達方程GSMaP MWR algorithm アルゴリズム本体 ( リトリーバル ) 観測データ 降水物理モデル ( フォワード計算 ) 放射アルゴリズム (10, 19, 37GHz) 陸上降雨有無判定 海上降雨有無判定 海岸降雨有無判定 降水の非一様性補正 散乱アルゴリズム (85, 37GHz) 各種判定補正降水強度推定降水強度式ルックアップテーブル (LUT) GANAL( 大気 地表面物理量 ) 降水タイプ分類 ( 陸上 6 種 海上 4 種 ) 降水 ( 鉛直 ) プロファイルモデル 雨滴粒径分布モデル 融解層モデル 層状性降雨 対流性降雨分類 衛星が観測するのは 放射 散乱強度の積分値を表す輝度温度である 降水物理モデルを仮定して放射伝達方程式を計算し 輝度温度と降水強度の関係をテーブル化し 観測値に近い輝度温度を与える降水強度を解としている
Particle Size Distribution Model 全球で取得可能なDSD パラメータ (TRMM PRの ε) の分布と 降水タイプ分類のパターンの類似性から 降水タイプ分類と関係づけたデータベース 450 400 350 300 kt-5deg98-04 kt-disd01-03 kt-1deg98-05 地上ディスドロメータデータによる検証 a 250 200 150 100 50 Jan Feb Mar Apr May Jun Jul Kototabang Aug Sep Oct Nov Dec Jan TRMM PR から推定した Z-R 関係の係数 a とディスドロメータから推定した a の比較. コトタバン ( 西スマトラ 山岳地帯 ) Month
リーバル値と PR( 降水レーダ ) の地上降水 強度の比較 (98 年 7 月 ) 30 30 25 25 20 20 RAIN3785 15 10 5 0 R 2 乗 = 0.4161 0 5 10 15 20 25 30 RAINSURF 15 10 5 0 0 5 10 15 20 25 30 RAIN37 RAINSURF R 2 乗 = 0.3967 RAIN85 RAINSURF R 2 乗 = 0.3074 rain85/rainsurf 50 40 30 20 10 5 4 3 2 1.5.4.3.2.1.05.04.03.02.01 R 2 乗 = 0.2697-2000 2000 6000 10000 0 4000 8000 rain37/rainsurf 50 40 30 20 10 5 4 3 2 1.5.4.3.2.1.05.04.03.02.01 R 2 乗 = 0.0843-2000 2000 6000 10000 0 4000 8000 DTOP 降水トップの高い ( 低い ) 降水を過大 ( 過小 ) 評価する特に 85GHz の散乱シグナルは降水トップへの感度が大きい DTOP
Evaluation using PR Match-up data We were able to generate 141 AMSR-E vs. PR match-up data within observation time difference 5 minutes for Jan,Feb,Mar,Jun,Jul,Aug 2003. This figure shows distribution of the match-up locations. Total:141
Evaluation using PR Match-up data Distribution of the bias -2-1 -0.5 0 0.5 1 2
Evaluation using PR Match-up data This figure shows the histogram of BIAS which are calculated from 141 match-up data. Liu Petty Aonashi Average=0.08 Average=0.12 Average=-0.04
Evaluation using PR Match-up data Distribution of the RMSE 0 1 2 3 <
Evaluation using PR Match-up data This figure shows the frequency of RMSE which are calculated from 141 match-up data. Liu Petty Aonashi Average=1.40 Average=1.40 Average=1.33