Abstract: Multi-source precipitation merging is a crucial way to estimate the spatiotemporal distribution of precipitation accurately. The commonly used merging methods mainly focus on bias correction of the total precipitation amount or precipitation intensity but often neglect to identify short-duration precipitation. In this study, we proposed a merging framework of multi-source precipitation by identifying rain and no rain and constructed a precipitation merging method considering both rain area identification and rainfall estimation by coupling the geographical weighted logistic regression (GWLR) and geographically weighted regression models (GWR). Then, the merging experiments of the Multi-Source Weighted-Ensemble Precipitation Version 2.1 (MSWEP V2.1) and the daily precipitation observed by the ground gauges network over the Han River basin were implemented. The results show that the proposed method successfully reproduces the spatial pattern of rain and no rain and catches the precipitation center. It overall strengthens the performance of MSWEP V2.1 to estimate ground precipitation, reduces the false alarm rate (R_FA) and false precipitation (P_F) by more than 60%, and improves the critical success index (I_CS) and Kling-Gupta efficiency coefficient (E_KG) by more than 40%. Moreover, the gains of correcting P_F and improving E_KG are higher than 10% against the spatially interpolated precipitation. Meanwhile, compared with reference data, precipitation fusion enhances the classification accuracy of heavy precipitation events (intensity ≥ 50 mm/d) by not less than 60%. The innovative method effectively improves the performance of precipitation estimation and provides a new idea for multi-source precipitation merging.