<p>The 7-day, 15-day, and monthly cumulative precipitation from CMIP6 climate models are important for geological hazard risk assessment and regional climate research. However, there are evident spatial and temporal heterogeneity issues, and it is difficult to accurately reflect the temporal changes and spatial distribution characteristics of precipitation. To improve the fitting accuracy, this study proposes a Bayesian Ridge Spatiotemporal Weighted Regression (BR-STWR) downscaling method, which constructs a spatial weighting matrix to enhance the model’s sensitivity to spatial variability in precipitation and derives a sequence of regression coefficients based on multi-year same-day mean precipitation and a sliding temporal window to account for the intra-annual temporal structure of precipitation. Using Yunnan Province as the main study area and Qinghai Province as the contrasting region, the BR-STWR model was developed based on regional observational data. Precipitation data from 12 CMIP6 climate models were corrected and compared with the Delta, CDF, GWR, and BRR methods. Results indicate that the BR-STWR method improves the overall accuracy of simulated precipitation at the 7-day, 15-day, and monthly scales. The correlation coefficients in Yunnan Province were 0.76, 0.85, and 0.90, respectively, while those in Qinghai Province were 0.78, 0.87, and 0.93, all of which surpass the comparison methods. Specifically, for the 7-day cumulative precipitation correction in Yunnan, the correlation coefficient of the BCC-CSM2-MR model increased from 0.51 to 0.76. These findings demonstrate that the BR-STWR method effectively mitigates the spatiotemporal biases of CMIP6 precipitation data and consistently enhances the fitting accuracy of 7-day, 15-day, and monthly cumulative precipitation across different regions and climate models, providing a reliable framework for precipitation bias correction.</p>