Presentation is loading. Please wait.

Presentation is loading. Please wait.

VALIDATION AND IMPROVEMENT OF THE GOES-R RAINFALL RATE ALGORITHM Background Robert J. Kuligowski, Center for Satellite Applications and Research, NOAA/NESDIS,

Similar presentations


Presentation on theme: "VALIDATION AND IMPROVEMENT OF THE GOES-R RAINFALL RATE ALGORITHM Background Robert J. Kuligowski, Center for Satellite Applications and Research, NOAA/NESDIS,"— Presentation transcript:

1 VALIDATION AND IMPROVEMENT OF THE GOES-R RAINFALL RATE ALGORITHM Background Robert J. Kuligowski, Center for Satellite Applications and Research, NOAA/NESDIS, College Park, MD Bob.Kuligowski@noaa.govBob.Kuligowski@noaa.gov Yaping Li, I. M. Systems Group, College Park, MD Current-GOES Version and AnalysisNext Steps Brief Algorithm Description The GOES-R Rainfall Rate algorithm will estimate instantaneous rain rate every 15 min on the ABI full disk at the IR pixel resolution. The rain rates are derived from the ABI IR bands, calibrated against rain rates from microwave (MW) instruments. This allows the rapid refresh and high spatial resolution of GEO IR data while trying to capture the accuracy of LEO MW rain rates. A rolling-value matched MW-IR calibration dataset is updated when new MW rain rates become available (Fig. 1) and the updated calibration is applied to independent ABI data (Fig. 2): Discriminant analysis is used to select the best two rain / no rain predictors and coefficients based on matches with the MW rain rates; Linear regression is used to select the best two rain rate predictors and coefficients (including nonlinear transformations of the predictors) based on matches with the MW rain rates. To correct regression-induced distortions in the distribution, the derived rainfall rates are matched against the training MW rain rates to create a lookup table (LUT) for adjusting the resulting rain rates. To account for differences among precipitation regimes, separate calibrations are performed for each 30-degree latitude band and for three cloud types based on brightness temperature differences (BTDs; Fig. 2). DISCLAIMER: The contents of this poster are solely the opinions of the author and do not constitute a statement of policy, decision, or position on behalf of the GOES-R Program Office, NOAA, or the U.S. Government. Acknowledgment: This work was supported by the GOES-R Program Office. Current-GOES Version A simplified version of the algorithm (no 6.2, 8.5, or 12.0 µm bands) has run in real time on current GOES since 2011 for evaluation and to support GOES-R Proving Ground activities. Impacts of fewer bands: 1 less algorithm class (“water cloud” and “ice cloud” combined since no 8.5 and 12.0 µm bands) Half as many available predictors This version is being evaluated against Multisensor Precipitation Estimator (MPE) hourly 4-km radar / gauge fields over the CONUS. Current-GOES Statistical Comparison The simplified GOES-R rain rates are slightly less correlated with MPE than the operational Hydro- Estimator (H-E), (Fig. 3a). The GOES-R algorithm has a very strong wet bias compared to both algorithms. (Fig. 3b). However, the GOES-R algorithm shows a conditional dry bias for “hit” pixels (Fig. 3c), and misses more rain than the H-E (Fig. 3d). The GOES-R wet bias is from false alarms (Fig. 3e). “Full” Algorithm Statistical Comparison To determine the effect of fewer bands on performance, rain rates from the “full” algorithm on METEOSAT Spinning Enhanced Visible InfraRed Imager (SEVIRI) data during 5-9 January, April, July, and October 2005 were compared to the H-E using Tropical Rainfall Measuring Mission (TRMM) Precipitation Radar (PR) rainfall rates as “ground truth”. The full version of the algorithm has a much higher correlation than the H-E (Fig. 4b), similar hit bias (Fig. 4c), and much less missed rainfall (Fig. 4d). Thus, the removal of these 3 bands significantly degrades algorithm performance and this needs to be considered during pre-launch evaluation. However, false alarm rainfall remains a significant problem (Fig. 4e)—it is not just from simplifying the algorithm. Address False Alarms Reducing permissible bias during rain / no rain calibration reduced false alarms / wet bias somewhat but not completely (Fig. 5). Other possible causes of bias are still being investigated, including uncorrected GOES limb effects. Figure 2. GOES-R Rainfall Rate Algorithm data processing diagram. Figure 1. Illustration of the rolling-value matched MW-IR data file. Other Planned Work Apply GOES-R Risk Reduction work by Li et al. to real-time GOES cloud property information and evaluate impact on real-time GOES cloud property information on warm-cloud light rainfall which typically IR and MW have difficulty detecting. Experiment with a model PW / RH adjustment to rain rates to account for moisture availability and subcloud evaporation of hydrometeors. Continue experiments with orographic rainfall modulation. Incorporate findings from other GOES-R Risk Reduction partners (Adler et al., Rabin, Dong, etc.) Figure 4. Same as Fig. 3 but vs. TMI rain rates and using the full version of the GOES-R Rainfall Rate algorithm for 5-9 January, April, July, and October 2005. a)b) c)d) e) Figure 5. Same as Figs. 3b and 3e, except for the current (“control”) and bias-constrained “modified” version of the algorithm for 26 June – 9 July 2012. a) b) Figure 3. Tukey box plots of (a) correlation coefficient ; (b) volume bias ratio; (c) additive hit bias (volume of excessive hit rain / volume of observed rain); (d) missed rainfall (normalized by volume of observed rainfall); and (e) false alarm rainfall (normalized by volume of observed rain vs. 1-h MPE for the current- GOES version of the GOES-R Rainfall Rate algorithm and the H-E for 1 September 2011 – 31 August 2012. b)a) c)d) e)


Download ppt "VALIDATION AND IMPROVEMENT OF THE GOES-R RAINFALL RATE ALGORITHM Background Robert J. Kuligowski, Center for Satellite Applications and Research, NOAA/NESDIS,"

Similar presentations


Ads by Google