Abstract: To accurately acquire the spatial distribution and resources of precipitation in small mountainous watersheds, this study employed the Kriging interpolation method for spatial downscaling of low-resolution satellite data. It integrated local satellite and observational data using the long short-term memory (LSTM) network, enhancing the temporal correlation between satellite and observed precipitation by incorporating antecedent precipitation information. This model was further utilized to estimate the spatial distribution of watershed precipitation. The results indicate that, spatially, the fusion model captures the location of rainstorm centers with greater precision. In terms of precipitation amount, the proposed model shows a probability of detection and a critical success index of 0.60 and 0.50, respectively, under short-duration intense rainfall, improving the original low-resolution satellite rainfall data to better approximate actual conditions. As for the number of precipitation stations, the accuracy of the merged precipitation data increases with the number of stations, reaching stability when a critical value of station density is achieved. The Kriging-LSTM model offers a novel approach for precisely acquiring precipitation resources in small mountainous watersheds.