The GOES-R Rainfall Rate, Rainfall Potential, and Probability of Rainfall Algorithms Bob Kuligowski, NOAA/NESDIS/STAR Yaping Li, Zhihua Zhang, Richard Barnhill, I. M. Systems Group 5 th International Precipitation Working Group (IPWG) Workshop Hamburg, Germany, 12 October 2010 1
Outline · Review of GOES-R Status and Capabilities · GOES-R Algorithm Working Group · Algorithm Descriptions and Examples » Rainfall Rate Algorithm » Rainfall Potential Algorithm » Probability of Rainfall Algorithm · Algorithm Validation · Status and Future Work · Summary 2
Review of GOES-R Status and Capabilities · Anticipated launch in late 2015 · Advanced Baseline Imager (ABI) § Increase from 5 to 16 spectral bands § Improved spatial resolution (4 2 km IR; 1 0. 5 km VIS) § Faster scanning (5 -min full disk vs. 30 min) · GOES Lightning Mapper (GLM) § Detects total lightning, not just cloud-to-ground § Single-channel, near-IR optical detector § Spatial resolution of ~10 km 3
GOES-R Algorithm Working Group (AWG) · Algorithm Teams (AT’s) working together to develop a prototype GOES-R ground processing system · Hydrology AT Products: » Rainfall Rate / QPE (current) » Probability of Rainfall (next 0 -3 h) » Rainfall Potential (next 0 -3 h) · Hydrology AT Members: § § § Bob Kuligowski (STAR/SMCD), Chair Ralph Ferraro (STAR/CORP) Kuo-Lin Hsu (UC-Irvine) George Huffman (NASA-GSFC/SSAI) Sheldon Kusselson (OSDPD/SSD/SAB) Matthew Sapiano (UMCP/ESSIC) 4
Rainfall Rate Algorithm Description · IR algorithm calibrated in real time using MW rain rates » IR continuously available, but weaker relationship to rain rate » MW more strongly related to rain rate, but available ~every 3 h · Calibration by type and region » Three cloud types: – “Water cloud”: T 7. 34<T 11. 2 and T 8. 5 -T 11. 2<-0. 3 – "Ice cloud": T 7. 34<T 11. 2 and T 8. 5 -T 11. 2≥-0. 3 – "Cold-top convective cloud": T 7. 34≥T 11. 2 » Four geographic regions: 60 -30ºS, 30ºS-EQ, EQ-30ºN, 30 -60ºN · Two retrieval steps: » Rain / no rain separation via discriminant analysis » Rain rate via multiple linear regression 5
Rainfall Rate Algorithm Description · 8 predictors derived from 5 ABI bands T 6. 19 T 8. 5 - T 7. 34 S = 0. 568 -(Tmin, 11. 2 -217 K) T 11. 2 - T 7. 34 Tavg, 11. 2 - Tmin, 11. 2 - S T 8. 5 - T 11. 2 T 7. 34 - T 6. 19 T 11. 2 - T 12. 3 · 8 additional nonlinear predictors » Regressed against the MW rain rates in log-log space 6
Rainfall Rate Algorithm Description · Initial SCa. MPR rain rates strongly underestimate heavy rain · Adjust distribution » For each class and region, match the CDF of the SCa. MPR rain rates against the CDF of the target MW rain rates » Create an interpolated LUT to modify the SCa. MPR rain rate distribution 7
Rainfall Rate Algorithm Description Apply most recent calibration in between new MW overpasses Retrieve rain rates from ABI data Update calibration when new MW rain rates available 8
Rainfall Rate Examples Radar Rainfall Rate 9
Rainfall Potential Algorithm Description · Identify features (clusters) in Rainfall Rate imagery » Filter rain rate image to reduce noise » Use cost minimization to organize pixels into clusters » Combine smaller clusters into larger ones · Determine motion vectors between features in consecutive images » For each cluster in current image, determine spatial offset that maximizes match with corresponding cluster in previous image » Objectively analyze the resulting spatial offsets for all clusters to create a spatially distributed motion field ● Apply motion vectors to create rainfall nowcasts » In 15 -minute increments… – Project each pixel forward in time based on motion vectors – Project motion vectors forward in time » Sum 15 -min rate fields to get a 3 -hour total 10
Rainfall Potential Examples Radar Rainfall Potential 11
Probability of Rainfall Algorithm Description · Inputs » Rainfall Potential algorithm output (3 -h total) » Intermediate (every 15 min) rainfall nowcasts from the Rainfall Potential algorithm. · Calibrated using conditional probability tables » Rainfall Potential ≥ 1 mm: total number of raining 15 -min periods » Rainfall Potential <1 mm: distance to nearest raining pixel · Calibrated against the Rainfall Rate product » Eliminate uncertainties associated with Rainfall Rate errors; » Allow much more spatially widespread calibration (ground truth is generally available over Western Europe only) 12
Probability of Rainfall Examples Radar Probability of Rainfall 13
Validation: Truth Data · Time scales ≤ 3 h, so must validate against radar · Validation datasets in SEVIRI region: · Tropical Rainfall Measuring Mission (TRMM) Precipitation Radar for Rainfall Rate · Nimrod radar data from the British Atmospheric Data Centre (BADC) for all 3 algorithms 14
Rainfall Rate “Fuzzy” Validation · Pixel-by-pixel comparisons difficult » Instantaneous rain rate varies too much at small scales · Neighborhood comparison » Compare to most similar nearby value (Ebert 2008) » Better indication of usefulness » Not needed for 3 -h Rainfall Potential / Probability 15 15
Rainfall Rate Validation Comparison with collocated TRMM PR for 6 -9 January, April, July, and October 2005 and all of January 2008. 16
Rainfall Rate Validation CDF of (absolute) errors of Rainfall Rate pixels with rates of 9. 5 -10. 5 mm/h vs. TRMM PR for 51 days: 6 -9 January, April, July, and October 2005. CDF of (absolute) errors of Rainfall Rate pixels with rates of 9. 5 -10. 5 mm/h vs. NIMROD radar data for 34 days: 69 April, July, and October 2005. 17
Rainfall Potential Validation Comparison with collocated Nimrod radar for 6 -9 April, July, and October 2005. 18
Probability of Rainfall Validation Reliability diagram of Probability of Rainfall vs. Nimrod radar data for 5 -9 April, July, and October 2005. 19
Validation Summary vs. Spec Validation versus TRMM PR for 51 days of data: 6 -9 January, April, July, and October 2005 and all of January 2008: Rainfall Rate (mm/h) Requirement vs. TRMM radar Accuracy Precision 6. 0 9. 0 4. 9 8. 9 Rainfall Rate (mm/h) Validation against Nimrod for 6 -9 April, July, and October 2005: Requirement vs. NIMROD Accuracy Precision 6. 0 9. 0 8. 6 9. 7 Rainfall Potential (mm/3 h) Requirement Evaluation Accuracy Precision 5. 0 2. 4 3. 1 Requirement Evaluation Probability of Accuracy Precision Rainfall (%) 25 40 25 71 20
Status and Future Work · Rainfall Rate: » » » Delivered “final” algorithm to System Prime 30 Sep 2011 Validation against an additional 4 months of data ongoing Developing real-time and “deep-dive” validation tools for further evaluation and potential improvement » “Maintenance” delivery 30 September 2012 that incorporates feedback from “deep-dive” validation · Rainfall Potential: » Optimizing parameters; “final” internal delivery May 2011 » Final algorithm delivery to System Prime by 30 Sep 2011 · Probability of Rainfall: » Continuing to recalibrate; “final” internal delivery May 2011 » Final algorithm delivery to System Prime by 30 Sep 2011 21
Summary · Three rainfall-related algorithms for GOES-R: » » » Rainfall Rate Rainfall Potential (0 -3 h) Probability of Rainfall (0 -3 h) · Performance: » Rainfall Rate and Rainfall Potential meet GOES-R spec » Probability of Rainfall partially meets spec and is being recalibrated · Future Work: » Rainfall Rate has been finalized and is in the validation stage » Rainfall Potential and Probability of Rainfall are still being modified; final delivery September 2011 22
Questions? Bob. Kuligowski@noaa. gov 23
Скачать презентацию The GOES-R Rainfall Rate Rainfall Potential and Probability
82a960cd9a6acfc30fdd26a4768c1f40.ppt