A two-step hybrid system towards optimized wave height forecasts

The accurate prediction of wave height is of paramount importance in many oceanic, coastal and offshore activities. To this end, the traditional approach is the application of numerical wave prediction models with great success but also drawbacks under particular conditions. In this work, a two-step hybrid system is proposed to optimize wave height forecasts. It is based on the well-known wave model WAM, which is primarily used to obtain the simulated parameters, and the gradient boosting regression tree algorithm, which is subsequently applied to improve the wave height forecasts of the numerical model. The developed system and the underlying algorithm, use as features the information available only from the WAM model, without the need of meteorological data records from buoys or other sources. These are replaced with the WAM model simulated parameters. The predictive performance of the proposed method is evaluated in three different study areas, in order to avoid any area dependencies, and to justify the method’s applicability despite the local wave climatology. The results demonstrate that, the developed hybrid system can output more accurate wave height predictions than the original WAM forecasts, in terms of different statistical divergence metrics. In addition, the system’s predictions follow a realistic trend, in accordance with the general pattern of the real values of wave height, while the reduction of the systematic error of the WAM model is significant.