Abstract: Limited by the coarse spatial resolution, Gravity Recovery and Climate Experiment（GRACE） satellite data is difficult to be applied in small or medium-sized areas. Therefore, based on the random forest algorithm, the monthly GRACE terrestrial water storage anomaly（TWSA） in central Yunnan from 2003 to 2020 had been improved from 1°×1° to 0.1°×0.1° from two scales, namely grid scale（Random Forest-Grid, RF-G） and regional scale（Random Forest-Zone, RF-Z）, respectively. The downscaling results are compared with the downscaling method based on PCRaster Global Water Balance（PCR-GLOBWB） hydrological model from the perspective of time and space to ensure accuracy. Furthermore, the Empirical Orthogonal Function（EOF） method is used to decompose the orthogonal pattern and analyze the characteristics of the TWSA downscaling results, enabling a deeper understanding of data characteristics and influencing factors. The results show that the RF-Z model outperforms in downscaling TWSA for the central Yunnan region, with the correlation coefficient of 0.99, the Nash-Sutcliffe efficiency coefficient of 0.97, the root mean square error is 6.68mm, and the mean absolute error of 5.22mm. Notably, the downscaled results from RF-Z successfully mitigate gridding artifacts. The variance contribution rate of the first four eigenvectors of EOF decomposition is 91.73%. The first mode is “high in the southwest and low in the northeast”, and its time coefficient has a significant seasonal rule. The second mode presents the distribution of “high in the northwest and low in the southeast” and the corresponding time coefficient shows an obvious decreasing trend. The third and fourth modes respectively present the distribution characteristics of “whole region type” and “high in northwest and low in southeast”. In addition, there is a strong correlation between TWSA and NDVI in the driving variables. This study can provide data support and technical guarantee for water resources management and ecological environment protection in central Yunnan.