This study aims to improve the accuracy of flood simulation and forecasting based on improving the quality and spatiotemporal resolution of driven data (precipitation) in small and medium-sized watersheds. The utility of a multi-source precipitation merging method based on the mixed geographically weighted regression and MGWR-BI function was evaluated, and the effect of the merged precipitation to the calculation result of the WRF-Hydro model was studied. The merged precipitation data was used in the WRF-Hydro model for the flood forecasting in the Ziwuhe Basin, and the results were compared to the observation data of station. The comparison results show that the merged precipitation data have substantial higher quality than the downscaling CMORPH satellite data by comparing to the ground observations. Flood events are better captured by the WRF-Hydro model driven by the merged precipitation data than the downscaling CMORPH satellite data for small and medium-sized watersheds. Therefore, the WRF-Hydro model with merged precipitation data has a potential advantage of flood simulation and forecasting for small and medium-sized watersheds.