An ensemble model with feature selection for nearshore wave forecasting

Abstract

The study proposes an ensemble one-week ahead Wave Forecast of Nearshore Waves (OWFNW) framework for managing shipping and construction in marine work sites. The framework uses XGBoost with feature selection (FS_XGBoost) for forecasting at 5 stations on the Japanese coast. XGBoost-based wave models are developed for each station, transforming global wave data into nearshore wave predictions. Models are trained using four different training sets from the Japan Meteorological Agency (JMA), National Oceanic and Atmospheric Administration (NOAA), European Centre for Medium-Range Weather Forecasts (ECMWF) and Nationwide Ocean Wave information network for Ports and HarbourS (NOWPHAS). The results indicate that selecting features enhances the model's prediction accuracy and refining prediction errors. The methodology can be applied to other nearshore seas where global wave forecast data is available.