E_GSMaP precipitation dataset reforecasted by RF-WMRA: Description and validation.

Accurate precipitation data is an important state variable in various application fields such as geohazard early warning, flood, and drought hazard monitoring and evaluation. This study proposed a precipitation reforecast inversion model based on Random Forest and Wavelet Multi-Resolution Analysis (RF-WMRA) method. In the basin along the Sichuan-Tibet Railway, we constructed the reforecast inversion model by using the fifth generation of atmospheric reanalysis of the European Centre for Medium-Range Weather Forecasts (ERA5) and Global Satellite Mapping of Precipitation (GSMaP) precipitation data and considering several geo-environmental variables such as elevation, humidity, temperature, and wind speed. In turn, we could obtain a set of E_GSMaP precipitation data with high spatial resolution (0.1°) from 1961 to 2020 in the study area. In addition, we also compared and analyzed the accuracy of the E_GSMaP data with both the ERA5 and GSMaP data from the perspective of extreme precipitation event monitoring. The results showed that: 1) The Correlation Coefficient (CC) of the RF reforecast model was 0.89, Root Mean Squared Error (RMSE) was 1.65 mm, Relative Bias (BIAS) was 0.01 %, Mean Absolute Error (MAE) was 0.42 mm, and Kling-Gupta efficiency (KGE) was 0.80. CC and KGE increased to various degrees before and after WMRA model correction, RMSE, BIAS, and MAE decreased to various degrees. It indicated that this study's RF-WMRA reforecast inversion model performed better. 2) If extreme precipitation events in the study area were analyzed, we recommend the E_GSMaP data given the performance of three datasets, E_GSMaP, ERA5, and GSMaP, on seasonal and monthly time scales as well as in total extreme precipitation. The results of this study can provide basic data for regional geological disasters and hydrometeorological research and help to respond to the risk of geological disasters and climate change.