Bias correction of satellite precipitation estimates over Thailand

Abstract

Over the years, meteorological satellite instruments have produced Satellite Precipitation Estimates (SPEs) that can supply rainfall intensity rates globally. However, these datasets do not directly reflect the actual values of ground measurements so it is imperative to correct the systematic biases of SPEs to produce reliable hydrologic models. Thus, the aim of this study is to assess the effectiveness of bias correction of SPE products over Thailand. The Precipitation Estimates from Remotely Sensed Information using Artificial Neural Networks – Cloud Classification System (PERSIANN-CCS), Global Satellite Mapping of Precipitation - Near Real Time (GSMaP_NRT), and Integrated Multi-satellitE Retrievals for GPM (IMERG) Early version were evaluated in comparison to the Thai Meteorological Department (TMD) gauge measurements from 2003 to 2018. Subsequently, the SPEs were corrected by using Scaling, Quantile Mapping (QM), and an Artificial Neural Network (ANN) correction. Both the original PERSIANN-CCS and IMERG Early generally exhibit overestimation over Thailand while the GSMaP_NRT slightly underestimate rainfall. The original IMERG Early also shows the least RMSE overall, followed by GSMaP_NRT, then by PERSIANN-CCS. GSMaP_NRT shows the highest Equitable Threat Score (ETS) while IMERG Early has the lowest ETS because it has large amounts of false alarms. All products exhibit higher errors during the wet season, high underestimation during heavy and extreme rainfall, and higher errors near the coastal areas where high rainfall occurs. IMERG Early also shows the least RMSE in all river basins. After bias correction, the adjusted IMERG Early dataset still provides the least RMSE for all basins regardless of which correction method was applied. The ANN bias correction method performs the best among the three methods in terms of RMSE. However, it increases the underestimation and RMSE of extreme rainfall events and worsens ETS of PERSIANN-CCS and GSMaP_NRT. Only the QM bias correction is able to consistently reduce errors of extreme rainfall and improve ETS. Overall, the ANN adjusted IMERG Early dataset has the least RMSE in all river basins.