Bias Correction of Satellite Data in GRAPES-VAR

Transcription

1 Bias Correction of Satellite Data in GRAPES-VAR Wei Han Chinese Academy of Meteorological Sciences, CMA ITSC15, , Italy

2 Outline Status of GRAPES-3DVAR Main components of GRAPES Usage of satellite data in GRAPES-Var Improvement of satellite data assimilation Bias Correction Error Tuning Impact on Analysis and Forecast Global Model Typhoon Forecast Discussions and Ongoing Work Developed Different Observation Operators Monitoring the Innovations And Tuning Parameters Impact

3 Main components of Chinese new generation NWP system Global/ Regional Assimilation and Prediction System GRAPES- Var GAPRES- Global GRAPES- Meso

4 Data usage of GRAPES-Var TEMP Conventional data SYNOP SHIPS AIREP ATOVS Assimilated directly Remote sensing data AMV Weather Radars

5 ATOVSFrom Regional to Global Since Oct Global l1b data from NOAA/NESDIS Local received HRPT data from NSMC PREPROC ATOVS Level 1C data ATOVS Level 1D data Typhoons Origin NOAA16/17 AMSUB CH (-3h~3h)Local received NOAA16/17 AMSUB CH (-3h~3h)From NESDIS

6 Satellite Data Assimilated In GRAPES in Nov xAMSU-B (NOAA-16/17) Winds from 5 GEOS (Met-5/7 GOES-9/10/12) OAA16/17 AMSUB GTS AMVs OAA15/16 AMSUA 2xAMSU-A (NOAA-15/16) FY2C AMVs Winds from FY2C

7 Preprocessing of Satellite Data in GRAPES-Var Quality control ( by NSMC ) Thinning Error Correlation (Not represented in the obs. covariance ) To reduce data volume Observational error Assignment Statistics of the Innovations Tuning of the Error Setting Bias correction Global Model Regional Model

8 Bias Correction Scheme in Practice: Harris and Kelly(2001) s scheme Scan Bias s=<d j (θ)-d j (θ=0)> Air Mass Bias : b=h(x b )-y-s Least Square p (predictors)b : Air Mass Bias Solution b = Ap + T T 1 c A= bp (pp ) c = b Ap Predictors: Scan Bias Latitude Band Average Air Mass Bias Thickness between hPa Thickness between hPa Surface temperatures Integrated water vapor

9 AMSUA Bias : H(xb)-Yo Ch5-Ch10:Cold Bias of Obs. Ch1-Ch4 and Ch11-Ch15: Warm Bias of Obs.

10 Estimation of std. of AMSUA Error

11 AMSUB Bias of H(xb)-Yo Bias Std. NOAA16 NOAA17

12 Tuning of Obs. Error Step1: Tuning of Obs. Error based on Innovation Statistics ε, ε, K, ε, K, ε o o o o sound synop amsu type N Step2: Tuning of Different Observations o o J( α) = J( ε, αε ) J = sound amsu p 2

13 Noaa16 AMSUA

14 AMSUB NOAA16 NOAA

15 Impact of Bias Correction On Analysis Increments On Forecast Global Model Verification against its own analysis(15 days) ACC: Anomaly Correlation Coefficient Typhoon Track Forecast Regional Model

16 dxa: T, q Average of days Ananlysis Increments

17 dxa: T, q Average of days 500hpa

18 500mb ACC ( ,144h Forecast) N-90N,0-360,500hPa BC Ctrl NoBC S-90N,0-360,500hPa BC Ctrl NoBC Anomaly Corr Anomaly Corr Forecast Day Forecast Day ctrl NoBC BC ctrl NoBC BC

19 dxa: T, q Average of days Ananlysis Increments

20 200hpa Analysis Increments

21 Bias Correction For Assimilation of ATOVS in GRAPES: Impact on Typhoon Forecast Global Model (One Case: MATSA) Regional Model (One Case: RANANIM)

22 GRAPES Global Model: (UTC),120h Forecast Different Initial Value only N: NCEP Analysis G: GTS S: GTS+AMSU+BC O: Obs.

23 Animation of 700hpa humidity 144h Forecast GRAPES Global Model with Assimlation of AMSU+BC

24 NOAA16:AMSUA-ch5, warm core

25 Ch5,6,7,8 Innovation of AMSUA(NOAA16) Ch5-Ch8 Ch5 Ch6 Ch7 Ch8

26 Ch5,6,7,8 After Q.C. Without B.C Ch5 Ch6 Ch7 Ch8

27 Ch5,6,7,8 After Q.C. With B.C Ch5 Ch6 Ch7 Ch8

28 Distribution of innovation of effective obs. Dashed line: without B.C Solid Line : with B.C. After Q.C.

29 H(xa)-Yo Residual Ch5 Ch6 Ch Ch8 Ch9 Ch10

30 RANANIM:0414 AMSU+BC No AMSU ,UTC AMSU

31 Precip. Forecast Durinig Rananim Landfall,24h Obs. 14Z13AUG2004 Forecast Arrows: 10m Wind Vectors Shaded6-hour Accumulated Rainfall

32 Precip. Forecast Durinig Rananim Landfall, 30h Obs. 20Z13AUG2004 Forecast Arrows: 10m Wind Vectors Shaded6-hour Accumulated Rainfall

33 Discussion GRAPES-3DVAR is Moving to a New Stage Operational implementation Refinement Bias Correction and Obs. Error Tuning VERY Important Preliminary results is promising Need More Work Positive Impact of Satellite Data Assimilation Global Medium Forecast Typhoon Forecasts Other Verifications is Ongoing(Precip.,etc.) Developed Different Observation Operators Monitoring the Innovations And Tuning Parameters

34 Ongoing Work Thinning( Resolution of Obs., Analysis and Model) Global Model Regional and Mesoscale Model Constant Distance Thinning Bias Correction Predictors Parameter Estimation Method Bias Model Treatment of Coast (No Options in RTTOV) Observation Error Setting and Online Tuning Interaction with Q.C. Diagnosis of E(Jmin)=p/2 (Talagrand,1999; Chapnik,2006)

35 Thank you for Attention