Abstract: Precipitation is the basic component of the earth water cycle. As a water flux, it connects the atmospheric process with the surface process, and has important significance in meteorology, climatology and hydrology. Due to the strong temporal and spatial variability of precipitation, it is one of the most difficult hydrological variables to measure accurately at present. Accurate precipitation data with high temporal and spatial resolution is very important for many applications such as hydrological and meteorological analysis. This paper takes Hanjiang River Basin as the research area and puts forward a two-step downscaling-merging method. By using the complementary characteristics of data availability and accuracy of precipitation observed by rainfall gauges and retrieved by satellites, a high-quality daily precipitation product with a spatial resolution of 0.01 can be generated by fusing gauge observations and GPM satellite precipitation products. The obtained fused precipitation product is driven by semi-distributed hydrological model WASMOD-D to simulate the rainfall-runoff process, and its runoff simulation effect is verified. The results show that:（1） The downscaling algorithm based on random forest model not only significantly improves the spatial resolution of GPM precipitation, but also maintains good accuracy.（2）The linear fusion model of precipitation data based on the co-Kriging method, the fusion scheme greatly improves the estimation accuracy of GPM precipitation, with the average absolute error and root mean square error reduced by 32.38% and 21.38% respectively, and the bias dropped below 1%；（3） Considering the simulation results of daily runoff under two different scenarios, the overall simulation effect of integrating precipitation data is the best because of combining the advantages of satellite precipitation data and gauge observations, and the overall improvement effect is obvious. This paper provides a new idea for the data fusion method based on Satellite-Gauge（S-G）,and the research results can be used as a reference for obtaining highresolution and high-precision precipitation data.