Skillful prediction of boreal winter-spring seasonal precipitation in Southern China based on machine learning approach and dynamical ENSO prediction

El Niño-Southern Oscillation (ENSO) and the antisymmetric combination mode (C-mode) have a significant impact on the seasonal precipitation in southern China (SC) during the boreal winter and spring seasons. Accurate seasonal precipitation prediction in SC is closely related to the effective foresight during the development of ENSO and C-mode events. While the prediction of ENSO events has been successful, accurately predicting C-mode events remains challenging. In this study, a machine-learning forecast model by employing the Long Short-Term Memory (LSTM) neural network is proposed to produce skillful predictions of C-mode events for lead time up to six months. During the test period from 2007 to 2021, the model, incorporating additional information about subsequent ENSO forecast obtained from the dynamical ENSO prediction model, achieves significantly better performance compared to models that only consider historical data. Based on the reliable prediction results of the C-mode event, the multiple regression used ENSO mode and C-mode index are constructed to effectively predict winter (spring) seasonal precipitation at a lead of five (four) months, respectively in SC. This statistical prediction model also demonstrates the ability to capture the extreme precipitation patterns associated with strong ENSO events.