Short-term Storm Surge Prediction based on Variational Mode Decomposition and Temporal Convolutional Network

Storm surges have caused severe casualties and economic losses. Accurate prediction of storm surge levels is crucial for disaster assessment, early warning, and effective disaster management. Compared to numerical simulation methods, machine learning approaches are relatively more efficient and direct. However, traditional machine learning models often fail to significantly remove noise and effectively capture long-term sequential dependencies. Therefore, this paper proposes a hybrid storm surge prediction model that combines Variational Mode Decomposition (VMD) with Temporal Convolutional Networks (TCN). By utilizing the VMD method, the model effectively decomposes storm surge signals into intrinsic mode functions (IMF) to enhance prediction accuracy by removing noise and extracting key features. The TCN component captures long-term dependencies in time series data, making the prediction of storm surge levels more precise. Upon comparison with conventional models such as Convolutional Neural Networks (CNN), Long Short-Term Memory networks (LSTM), Gated Recurrent Units (GRU), and standard TCN, the proposed VMD-TCN model exhibits significant advantages. The forecasting performance metrics Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Correlation Coefficient (CC) are 7.2254 cm, 9.4548 cm, and 0.9251 respectively, indicating that the model proposed in this paper performs well in predicting storm surges.