<p>Numerical models have long been constrained by initial condition errors, systematic biases, and insufficient utilization of historical data, resulting in limited prediction skill for the spatial distribution of summer precipitation in North China. Therefore, taking North China as a case study, this paper fully leverages the advantages of Empirical Orthogonal Function (EOF) dimensionality reduction and Random Forest (RF) time-series prediction to transform the spatial distribution prediction problem into a time-series prediction task, and conducts a prediction study on the spatial distribution of summer precipitation in North China. The prediction results demonstrate that the multi-scheme RF ensemble model (MSEM_RF) not only ensures prediction accuracy but also significantly improves prediction stability. In the test set, the correlation coefficient between the prediction and the first principal component (PC1) reaches 0.84, with a range of only 0.12, while its spatial distribution prediction skill achieves 0.53. Analysis of the decision-tree structures indicates that the superior performance of MSEM_RF mainly arises from its ability to capture the southerly moisture transport toward North China and the associated regional upward motion, which are key dynamical processes controlling summer precipitation variability. In contrast, the numerical models failed to effectively capture key circulation features and demonstrated inferior prediction skill with spatial prediction skills all below 0.2. These results demonstrate the feasibility of integrating RF with EOF analysis for short-term climate prediction, providing a new perspective for improving the spatial prediction skill of summer precipitation in North China and potentially in other regions of China.</p>