Monthly extreme wave height prediction based on an LSTM-Stacking model

• A hybrid LSTM-Stacking framework is developed for monthly extreme wave height prediction. • Physics-informed constraints enhance long-term prediction stability and engineering reliability. • The proposed model demonstrates strong performance in long-term, small-sample, and cross-station predictions. Extreme wave height is a key parameter in ocean dynamics and a critical indicator for ocean engineering and marine risk assessment. However, existing prediction models often fail to adequately address the nonstationary characteristics of wave height time series. To overcome these limitations, this study proposes an extreme wave height prediction framework that integrates feature engineering with a Long Short-Term Memory (LSTM)-based Stacking ensemble learning approach. Time based and physics-informed features are constructed after missing-value imputation to form a multidimensional input dataset. LSTM networks are combined with eXtreme Gradient Boosting (XGB), Light Gradient Boosting Machine (LGB), Random Forest (RF), Support Vector Regression (SVR), and K-Nearest Neighbors (KNN) to capture nonlinear patterns and complex dependencies in wave height series, while a Stacking meta-learner is employed to optimize ensemble outputs. A physics-informed constraint calibrator is further introduced to enhance physical consistency and prediction accuracy. The model is validated using measured wave data from National Data Buoy Center (NDBC) station 42002 and evaluated using the coefficient of determination (R²), root mean square error (RMSE), mean absolute error (MAE), and mean square error (MSE). Comparative experiments demonstrate that the proposed LSTM-Stacking model outperforms standalone LSTM, XGB, and conventional Stacking approaches in monthly extreme wave height prediction, providing reliable forecasting support for offshore engineering and marine disaster prevention.