Abstract: Based on the National Climate Center Climate System Model BCC_CSM1.1m (Beijing Climate Center Climate System Model) and the NCEP/NCAR climate prediction model CFSv2 (The NCEP Climate Forecast System Version 2) of the United States, two dynamic statistical downscaling prediction models of the precipitation during flood season in the Yangtze River Basin are established correspondingly. The forecasting skills and sources of differences between the two models are compared. The global geopotential height fields at 500 hPa and 200 hPa produced by the two models from February are selected as the predictors, and the models are established by combining the interannual increments and the empirical orthogonal decomposition (EOF) iteration method (the test scheme named DY_CSM1.1 and DY_CFSv2). This study found that:(1) The increase of the truncated explained variance in the EOF iteration method enhances the synergy and stability of the predictors, thereby significantly improving the forecasting skills, which indicates that 98% of the truncated explained variance is the optimal parameter of the model. (2) The prediction effect of the optimal parameters of the two models is better than the original precipitation prediction of the model, and the DY_CSM1.1m prediction skill is higher, especially in the main stream of the Yangtze River. The 29-year average of the spatial anomaly correlation coefficient ACC score can reach 0.43 and 0.39, respectively. When the predicted interannual increment percentage of precipitation is converted to the flood season precipitation anomaly percentage, the ACC scores dropped to 0.27 and 0.22, but are still higher than the model's original predictions. (3) The ACC score of DY_CSM1.1m has a high correlation with the interannual increment of the West Pacific Subtropical High Ridge Position Index (WPSHRP) (but DY_CFSv2 has no such relationship). The inter-annual increment of precipitation in the flood season in the Yangtze River Basin also has a high correlation with the inter-annual increment of WPSHRP. Therefore, the simulation performance of BCC_CSM1.1m in terms of the mid-low latitude geopotential height field is better than that of CFSv2, which is an important reason for the higher forecasting skills of the model after downscaling. It is evidenced in the typical flood years of 1998 and 2020.