Presentation is loading. Please wait.

Presentation is loading. Please wait.

Empirical Analysis and Statistical Modeling of Errors in Satellite Precipitation Sensors Yudong Tian, Ling Tang, Robert Adler, and Xin Lin University of.

Similar presentations


Presentation on theme: "Empirical Analysis and Statistical Modeling of Errors in Satellite Precipitation Sensors Yudong Tian, Ling Tang, Robert Adler, and Xin Lin University of."— Presentation transcript:

1 Empirical Analysis and Statistical Modeling of Errors in Satellite Precipitation Sensors Yudong Tian, Ling Tang, Robert Adler, and Xin Lin University of Maryland & NASA/GSFC Sponsored by NASA ESDR-ERR Program

2 Motivation Two error sources in merged satellite data: -- the merging algorithm -- the upstream sensors Studying errors in the sensors is necessary in understanding errors in merged products 2

3 Outline To understand: empirical analysis of systematic errors: characterizing errors in passive microwave (PMW) sensors To quantify and to predict: statistical modeling of errors: with a measurement error model, to quantify both systematic and random errors Summary and Conclusions 3

4 Data and Study Period Time period: 3 years, 2009 ~ 2011 Ground reference: Q2 (NOAA NSSL Next Generation QPE), bias-corrected with NOAA NCEP Stage IV (hourly, 4-km) – Resolution: 5 minutes, 1 km, remapped to 5 mins,0.25 o Satellite sensor instantaneous rainfall measurements aggregated to 5 minutes time interval – Sensors: TMI, AMSR-E, and SSMIS – Imagers only for now – Resolution: 5 minutes, 0.25 o – Satellite data matched with Q2 over CONUS 4

5 5 Sensors covered by the study period

6 6 Q2 has biases and was corrected with Stage IV data Before After CPC Gauge Stage IV Radar

7 Sample sizes matched between sensors and Q2 7 AMSR-E TMI SSMIS F16 SSMIS F17

8 Mean Precipitation (Summer 2009~2011, units: mm/hr) 8 AMSR-E matched Q2 TMI matched Q2 SSMIS F16 matched Q2 SSMIS F17 matched Q2

9 Precipitation – Density Scatter Plots (Summer 2009~2011) 9 AMSR-E TMI SSMIS F16 SSMIS F17

10 More overestimates in SSMIS for summer 10 AMSR-E TMI SSMIS F16 SSMIS F17

11 11 AMSR-E TMI SSMIS F16 SSMIS F17 More underestimates in AMSR-E & TMI for winter

12 PDF Comparisons confirm season-dependent error characteristics 12 AMSR-E TMI AMSR-E TMI SSMIS F16 SSMIS F17 SSMIS F16 SSMIS F17 Summer Winter

13 13 A nonlinear multiplicative measurement error model: X i : truth, error free. Y i : measurements With a logarithm transformation, the model is now a linear, additive error model, with three parameters: A=log(α), B=β, and σ which can be easily estimated with ordinary least squares (OLS) method. Modeling the Measurement Errors: A-B-σ model

14 14 Clean separation of systematic and random errors More appropriate for measurements with several orders of magnitude variability Good predictive skills Tian et al., 2012: Error modeling for daily precipitation measurements: additive or multiplicative? to be submitted to Geophys. Rev. Lett. Justification for the nonlinear multiplicative error model

15 Spatial distribution of the model parameters 15 TMI AMSR-E F16 F17 A B σ(random error)

16 16 Probability distribution of the model parameters A B σ TMI AMSR-E F16 F17

17 Summary and Conclusions 1. what we did Created bias-corrected radar data for validation Evaluated biases in PMW imagers: AMSR-E, TMI and SSMIS Constructed an error model to quantify both systematic and random errors 17

18 Summary and Conclusions 2. what we found Sensor biases have seasonal and rain-rate dependency: summer – overestimates; winter: underestimates AMSR-E and TMI did better in summer; SSMI F16 and F17 in winter The multiplicative error model works consistently well Both systematic and random errors are quantified Model indicated AMSR-E had the lowest uncertainty Results useful for data assimilation, algorithm cal/val, etc. 18

19 Extra slides 19

20 20 What we did: 1.A nonlinear multiplicative error model 2.Constant variance in random errors 3.More appropriate for variables with several orders of variability 4.A parametric model is useful for data assimilation, cal/val What we found: 1.The model works well 2.Constant variance in random errors 3.More appropriate for variables with several orders of variability 4.A parametric model is useful for data assimilation, cal/val Summary and Conclusions

21 Summary and Conclusions what we did: AMSR-E and TMI underestimate rainfall in winter in Southeast US. AMSR-E, SSMIS F16 and F17 overestimate rainfall in Summer in Central and Southeast US. SSMIS F16 and F17 have high positive BIAS in Summer, over Central US; AMSR-E and TMI have high negative BIAS in Winter, over Southeast US. TMI performs the best compared with the other three sensors. 21

22 Biases become less pronounced with all-year data (2009~2011) 22 AMSR-E TMI SSMIS F16 SSMIS F17

23 23 Satellite Sensor Data Availability SSMI F No data SSMI F14 No data No data SSMI F15 No data , No data SSMIS F16 No data SSMIS F17 No data SSMIS F18 No data TMI No data AMSR-ENo data , , No dataMissing filesComplete

24 Precipitation – Density Scatter Plots (2009~2011) 24 AMSR-E TMI SSMIS F16 SSMIS F17

25 Precipitation – Density Scatter Plots (Winter 2009~2011) 25 AMSR-E TMI SSMIS F16 SSMIS F17

26 Sensors show mostly overestimates for summer 26 AMSR-E TMI AMSR-E TMI SSMIS F16 SSMIS F17 SSMIS F16 SSMIS F17 Summer

27 Spatial distribution of the model parameters (for winter) 27 A B σ TMI AMSR-E F16 F17

28 Spatial distribution of the model parameters for summer 28 A B σ TMI AMSR-E F16 F17


Download ppt "Empirical Analysis and Statistical Modeling of Errors in Satellite Precipitation Sensors Yudong Tian, Ling Tang, Robert Adler, and Xin Lin University of."

Similar presentations


Ads by Google