Against the background of global warming, summer extreme precipitation in Hunan has increased significantly.Therefore, improving the prediction accuracy of precipitation is of great practical significance for disaster prevention and mitigation in Hunan Province.Using the monthly precipitation data from meteorological stations in Hunan, the climate index sets from the National Climate Center (NCC) and the precipitation data from the hindcast experiments are performed using seasonal prediction models of NCC and NCEP (National Centers for Environmental Prediction).The recursive feature elimination (RFE) method is used to determine the key factors, and two statistical prediction schemes of summer precipitation in Hunan are established by three algorithms:multilayer feedforward neural network (FNN), support vector regression (SVR) and natural gradient boosting (NGBoost).The results show that the prediction model based on machine learning (ML) has superior ability to predict the distribution pattern of summer precipitation in Hunan.The respective average ACC skills of the two statistical schemes with lead times of 1 to 6 months are 0.15 and 0.19, which is a great improvement compared with the dynamic model.The respective average PS scores are 69.3 and 69.2, which are higher than the NCC model.The further analysis indicates that the preceding winter polar and mid-and high-latitude latitude circulation may be the main predictability sources of ML models with lead times of 1 to 3 months.Finally, the prediction skills of models with lead times of 4 to 6 months are likely derived from the precursory signal of sea surface temperature.