Abstract: High-resolution precipitation data can be directly used to characterize the spatial-temporal differentiation features of precipitation after the accurate simulation of hydrological, meteorological, and biological systems. Therefore, it is crucial to implement the spatial downscaling for the precipitation products with the low spatial resolution. However, it is still lacking on the precision and detail features in the downscaling precipitation data, due to fail to consider the scale variations in the spatial distribution of precipitation. In this study, a spatial downscaling approach was proposed to improve the TRMM precipitation data in the Wei River basin (WRB). A Multi-scale Geographically Weighted Regression (MGWR) model was also integrated to enable the conditional relationships between the response and predictor variables that changed at the spatial scales. Therein, the goodness of fit (R2), relative deviation (BIAS), and Root Mean Square Error (RMSE) were employed to verify the TRMM satellite precipitation product, compared with the actual precipitation data from meteorological stations. Normalized Difference Vegetation Index (NDVI), digital elevation model (DEM), slope, aspect, latitude and longitude were induced as the Geographic Environmental Factors (GEFs). The MGWR models with the monthly TRMM precipitation data were constructed to further investigate the scale effects of factors on precipitation distribution. The spatial downscaling of TRMM production data was then implemented using a scale conversion process. Finally, the reliable spatial downscaling was achieved in the TRMM precipitation products. The results illustrated that: 1) The TRMM precipitation data was better suited for use at different scales in the WRB. An acceptable fitness was found in the statistics of R2 (0.807), BIAS (2.909%), and RMSE (83.477 mm) at the annual scale. Specifically, the maximum R2 was 0.847 at the seasonal scale, the largest RMSE was 62.393 mm in the summer, and the BIAS values were lower in all four seasons. More importantly, the R2 varied between 0.456 and 0.815 on the monthly scale, with the smallest value in June and the largest value in September. The BIAS was positive in the most months, indicating that the TRMM product data generally overestimated the precipitation. The RMSE index was fallen in the range of 3.507-39.342 mm, which was lower than those on the annual and seasonal scales. 2) Different scale characteristics were found in the influence of various GEFs on the spatial pattern of precipitation divergence in wet and dry years. Slope was set as the global scale, whereas the DEM, NDVI, aspect, latitude, and longitude were the local effects on the precipitation in wet years, and all GEFs were used the local impacts in dry years. 3) The more precise data was obtained in the downscaled TRMM on the watershed and station scales, compared with the product data, indicating an increase in the R2 of the entire watershed of 3%, while a decrease in the RMSE of 1 mm. However, the accuracy of station downscaling precipitation data at the annual scale was worse than that at the monthly scale, due to the accumulation of errors on the time scale, as shown by the R2 range decreasing, while the RMSE range increasing. 4) The downscaled TRMM data presented the better detailed characteristics, the greater precision at annual and monthly scales, and a much more delicate geographical distribution than the product data. The finding can provide a strong data support for the hydrological design in the areas with less precipitation data.