A Novel Coupled Model for Monthly Rainfall Prediction Based on ESMD-EWT-SVD-LSTM

Precise predicting of rainfall is paramount for effective water resource management, ecological conservation, and the prevention of droughts and floods. Influenced by numerous variables, the process of rainfall is complex and the rainfall series exhibit high degrees of nonlinearity, making it challenging for traditional statistical prediction models to accurately capture the characteristics of rainfall series. Therefore, this paper proposes a new coupled model for predicting monthly rainfall based on Extreme-Point Symmetric Mode Decomposition (ESMD), Empirical Wavelet Transform (EWT), Singular Value Decomposition (SVD) and Long Short-Term Memory Neural Network (LSTM). By training and evaluating the ESMD-EWT-SVD-LSTM model on Kaifeng City’s monthly rainfall data from 2009 to 2020 and comparing its predictions with those of the ESMD-SVD-LSTM, SVD-LSTM, LSTM models, the analysis reveals that: the quadratic decomposition of ESMD-EWT and SVD denoising can further reduce the complexity of rainfall data, obtain more predictable feature IMFs, and enhance the precision in LSTM predicting; in comparison with alternative models, the ESMD-EWT-SVD-LSTM coupled model shows the highest accuracy in predicting results, with MAE of 4.96, RMSE of 6.13, and SI of 0.12, indicating that the ESMD-EWT-SVD-LSTM model has strong nonlinear process learning ability and accuracy in regional monthly rainfall prediction. This study can offer dependable scientific grounding and technical assistance for regional rainfall predicting, water resources planning, and disaster mitigation.