Qualidade das estimativas de precipitações derivadas de satélites na bacia do Alto Jacuí - RS

Abstract

The continuous increase in using satellite precipitation estimates as alternative sources for data have been increasing with the new technology of the devices. Therefore, the need for evaluating the quality and accuracy of these estimates is bigger. In this work, we assessed the TRMM satellite precipitation products 3B42 V6, 3B42 V7, and 3B42 Real Time, and the estimates from CMORPH method (RAW) by using the observed data from Alto do Jacuí basin region. To assess these products, we have used IPWG statistics to validate the estimate products such as PC (percent correct), H (hits), FAR (false alarm ratio) performance indexes, among others. We have also assessed products performance in detecting the occurrence and non-occurrence of different rainfall events. We have employed quantitative statistics to assess mean error (ME), root-mean-square error (RMSE), correlation coefficient (r), total errors and Nash-Sutcliffe (NS) efficiency coefficient (NSE). From the results, we have tried a methodology to improve them. The PC indexes showed an average of 81.3%, and they had a similar behavior among the products, while the H index showed an average of 60%. These numbers stress that the main difficulty is to detect rainfall events. The FAR index showed an average of 8% for V6 and 9% for CMORPH; 13% for V7, and 15% for Real Time, what is considered reasonable. In the qualitative assessment, we have emphasized the CMORPH product, which showed the best analysis results. This probably happens for being a method that uses more sources of information or for exploring this information more efficiently somehow. In the predictive potential evaluation, Real Time product had the worst results (NSE). V7 resulted in a small decrease of quality when compared to V6, although it was superior in other aspects. CMORPH overcame the other products, with an NSE average of 0.45. In the quantitative assessment, we have noted that V6 and CMORPH could estimate less than 50% of the total rainfall; V7 overestimated around 11% the total rainfall, while Real Time overestimated around 25% of the total. Despite the reasonable results, all the products showed good correlation (0.73). This made us try a method to improve the detection rate. Through modified double-mass equation, we had really significant improvements, except for V7. For instance, for the detection percentage of rainfall events > 60 millimeters, the V6 was almost null 1%; 41% for Real Time; and 1.1% for CMORPH. After the application of the improved method, the detection percentage increased to 53.6%, 50.8%, and 54.8% for those products respectively. These results indicate that the satellite rainfall estimates are an alternative source of data with a great spatial and temporal potential. Thus, the products can be improved to help the hydrological monitoring, mainly in areas with low quality of precipitation data. However, there are still many things to be improved about rainfall estimates, mainly in detecting rainfall, where we have found the biggest limitations.