Prediction of the onset of the South China Sea summer monsoon based on machine learning

Accurately predicting the onset of the South China Sea (SCS) summer monsoon (SCSSM) is crucial for agricultural planning and disaster prevention for countries surrounding the SCS. However, due to the limited availability of data (only one data point per year), the prediction of the onset date of the SCSSM faces significant uncertainty. In this study, we propose a new forecasting method for the onset date of the SCSSM using the Random Forest regression model called MMPM (Multi-model Prediction Method). The MMPM generates a large number of randomly initialized predictive models (∼10 000) and selects the optimal one based on data from the past nine years to predict the date of SCSSM onset for the following year. By using this method, we effectively mitigate the impact of interdecadal variability and predict the date one month before the onset of the SCSSM using factors such as sea surface temperature, mean sea level pressure, and temperature tendency from the previous few months. Compared to an existing statistical forecasting method (SM-17), the MMPM improves the predictive performance by over 10%. This study demonstrates the effectiveness of an innovative strategy in overcoming data limitations and providing better tools for predicting important weather patterns. 准确预测南海 (SCS) 夏季季风 (SCSSM) 的爆发时间, 对于南海周边国家的农业规划和防灾减灾至关重要.然而, 由于可用数据有限 (每年仅有一个数据点) , SCSSM 爆发日期的预测存在较大不确定性.在本研究中, 我们提出了一种新的 SCSSM 爆发日期预测方法, 基于随机森林回归模型, 称为多模型预测方法 (MMPM, Multi-model Prediction Method) MMPM 随机初始化大量预测模型 (约 10 000 个) , 并基于过去九年的数据选择最优模型, 用于预测下一年的 SCSSM 爆发日期.通过该方法, 我们有效减弱了年代际变化的影响, 并利用前几个月的海表温度,平均海平面气压以及气温趋势等因子, 在 SCSSM 爆发前一个月进行预测.与现有的统计预测方法 (SM-17) 相比, MMPM 将预测性能提升了 10% 以上.本研究展示了一种创新策略在克服数据限制方面的有效性, 为重要天气模式的预测提供了更好的工具.