Evaluating and Merging Satellite and Reanalysis Precipitation Products with Station Observations Using XGBoost in the Jinsha River Basin, China

The Jinsha River Basin constitutes the largest hydropower base in China. However, its complex terrain results in insufficient accurate data support for numerical forecasts, leading to low accuracy in precipitation predictions. To investigate the spatiotemporal distribution characteristics of precipitation in this basin with high precision, we evaluated the applicability of several mainstream precipitation products—GSMAP (Global Satellite Mapping of Precipitation), GPM-IMERG (Integrated Multi-satellite Retrievals for Global Precipitation Measurement), CMORPH (Climate Prediction Center Morphing technique), and ERA5 (European Center for Medium-Range Weather Forecasts Reanalysis 5)—in the Jinsha River Basin. Based on the XGBoost algorithm, we developed a merging model that integrates satellite and reanalysis data with station observations for daily-scale applications. The results indicate that the GSMAP-Gauge precipitation product exhibits strong performance in both quantitative accuracy and precipitation event detection, with a better correlation coefficient (CC = 0.66), the lowest root mean square error (RMSE = 4.45), and higher probability of detection (POD = 0.88) and critical success index (CSI = 0.59). The ERA5 and GSMAP-Gauge products performed well in detecting light rain events (daily precipitation < 10 mm), with hit rates of 0.92 and 0.90, respectively. Meanwhile, the GPM-IMERG and CMORPH-BLD products showed higher hit rates for heavy rain events (daily precipitation > 25 mm) compared to the other two products. Specifically, the POD indices for GPM-IMERG and CMORPH-BLD were 0.45 and 0.60, respectively, while those for ERA5 and GSMAP-Gauge were below 0.4. Following the precipitation merging experiment, the multi-source precipitation merged product (MSP) substantially enhanced the accuracy of precipitation estimates, and the spatiotemporal distribution characteristics of the merged data aligned more closely with the station observations. This study analyzes the strengths and limitations of various precipitation products in the Jinsha River Basin and provides a feasible multi-source precipitation data merging scheme, offering a novel approach to constructing high-precision daily precipitation datasets in complex terrain regions.