Improving the accuracy of global ECMWF wave height forecasts with machine learning

Abstract

Significant wave height (SWH) stands as one of the most crucial parameters for maritime activities. However, even the SWH data from the widely utilized European Centre for Medium-Range Weather Forecast Integrated Forecasting System (ECMWF-IFS) carries errors and uncertainties. In this study, the Light Gradient Boosting Machine (LightGBM) is used to inference the global ECMWF-IFS SWH forecast biases. The results demonstrate that globally, the LightGBM reduces the root mean square error by 10–20 %. Particularly noteworthy is the enhanced forecast accuracy observed in the western Pacific during late summers. Furthermore, the corrected forecast results during Super Typhoon Lekima in 2019 showcase the capability of model to effectively enhance the forecast accuracy of typhoon-induced wind waves, even when four typhoons occur concurrently. This study establishes the feasibility of LightGBM in inferencing single-step SWH forecast bias and presents a cost-effective model for enhancing global wave forecasts.