Advancing annual-scale significant wave height prediction: A comprehensive skill assessment of M5P decision tree approach

Accurate annual-scale prediction of significant wind-wave height is essential for maritime safety, coastal management, and renewable energy applications. While Machine Learning and numerical models offer powerful tools, their complexity and computational demands often limit practical implementation. This study evaluates the M5p Decision Tree, a computationally efficient and interpretable alternative, for significant height of wind waves prediction using ERA5 reanalysis data from the North Pacific Ocean. Feature selection was performed using Decision Tree analysis and compared with the Backward Elimination method. Results indicate that wind speed is the most influential predictor, and segmentation of data by month improves model performance, reducing RMSE by up to 30% and enhancing accuracy. Limiting tree depth to 5 levels improves interpretability while maintaining similar predictive accuracy. The M5P DT model with monthly data segmentation showed the best performance, with RMSE values as low as 0.059 and a correlation coefficient of 0.998, compared to 0.130 and 0.989 for the model excluding wind speed. The study demonstrates that M5P DT offers a practical, interpretable, and efficient approach for significant height of wind waves forecasting, outperforming traditional models in both accuracy and computational efficiency.