Abstract: To improve the spatial resolution and expand the application scopes of GPM precipitation products, the downscaling study of GPM precipitation products was conducted based on the precipitation data of Guizhou Province by establishing multiple spatial downscaling models. Firstly, with the topographic factors including longitude, latitude, elevation, slope, and aspect as explanatory variables and the original GPM precipitation data as target variables, multiple downscaling models were established based on the methods of multivariate linear regression, geographically weighted regression, extreme learning machine, support vector machine, and random forest regression. Then multiyear average precipitation data were applied and assessed, and the optimal model was selected to conduct the spatial downscaling study of the annual and monthly precipitation amount in typical years in Guizhou Province. According to the results, the downscaling models except for the random forest regression model all performed well. Most especially, the multivariate linear regression model performed the most stably and effectively and yielded the highly improved downscaling results in terms of observation accuracy and spatial correlation. This study will provide a set of high-resolution gridded precipitation products for Guizhou Province and provide support for regional hydrometeorological research.