Modeling Significant Wave Heights for Multiple Time Horizons Using Metaheuristic Regression Methods

The study examines the applicability of six metaheuristic regression techniques—M5 model tree (M5RT), multivariate adaptive spline (MARS), principal component (PCR), random forest (RF), partial least square (PLSR) and Gaussian process (GPR)—for predicting short-term significant wave heights from one hour to day ahead. Hourly data two stations, Townsville Brisbane Buoys, Queensland, Australia, historical values were used as inputs for predictions. methods assessed based on root mean error, absolute determination coefficient new graphical inspection (e.g., Taylor violin charts). On basis error (RMSE), (MAE), (R2) statistics, it was observed that GPR provided best accuracy in single-time-step multi-time-step heights. RMSE, improved M5RT, MARS, PCR, RF PLSR by 16.63, 8.03, 10.34, 3.25 7.78% (first station) 14.04, 8.35, 13.34, 3.87 8.30% (second test stage.