Abstract: As a crucial transboundary river system in Asia, the accurate prediction of summer precipitation in the Lancang-Mekong River Basin (LMRB) is of great significance for regional water resource management and flood disaster prevention. Based on historical hindcast experiments from 1991 to 2020 using the dynamical climate prediction system of the Institute of Atmospheric Physics, Chinese Academy of Sciences (IAP-DCPv3.5), this study first evaluates the system"s prediction performance for summer precipitation anomalies in the LMRB, as well as the impact of the traditional empirical orthogonal function (EOF) correction method on prediction skill. Subsequently, by incorporating key atmospheric circulation features that influence basin precipitation, a precipitation bias correction scheme based on the U-Net model is proposed. The circulation feature extraction method within the model is effectively optimized, significantly enhancing the prediction skill of IAP-DCPv3.5 for summer precipitation in the LMRB. Verification results from an independent testing period spanning 2021–2024 demonstrate that the U-Net correction scheme substantially improves the model"s prediction performance. For summer mean precipitation initialized in March, May, and June, the mean absolute error (MAE) is reduced by over 50% compared to the original predictions. Concurrently, it improves the spatial distribution characteristics of the summer precipitation predicted by IAP-DCPv3.5 and overcomes the instability inherent in the traditional EOF method. For summer precipitation anomalies initialized in June, May, and March, the average spatial correlation coefficients (PCC) of the original predictions are merely 0.09, -0.01, and 0.09, respectively. After applying the U-Net correction, the corresponding PCCs increase to 0.31, 0.21, and 0.19. In contrast, when using the EOF method, the PCCs for predictions initialized in May and June actually decrease to -0.02 and 0.02, respectively. Furthermore, an analysis of the typical basin-wide drought year in 2021 reveals that the U-Net method successfully predicts precipitation magnitudes and spatial distribution patterns that are highly consistent with observations; notably, the prediction initialized in March achieves a PCC as high as 0.50. This study demonstrates that the deep integration of deep learning with numerical model predictions is an effective approach to improving subseasonal-to-seasonal (S2S) climate prediction skills and extending prediction lead times.