Abstract: Using the Spark GBT (Gradient Boosted Trees Regression) learning algorithm for quantitative precipitation estimation (QPE) in western China, a 3-D Barnes interpolation method was used to obtain the hourly precipitation grid field using the satellite cloud image data of Himawara-8 and ground observation data. The experimental results show that the introduction of satellite data to estimate precipitation can effectively improve the accuracy of the objective analysis of precipitation grid in the area with small station density.