Short-term forecasting of zonal and meridional wave energy flux in the Bay of Biscay using random forests

Three types of statistical models have been used to create up to 24h forecasts of the zonal and meridional components of wave energy flux levels at three directional buoys located in the Bay of Biscay. Hourly observations of the mean wave period and the significant height covering the 1999-2012 period have been used for this purpose. Additionally, data from the ocean (WAM model) and atmospheric components of the ERA-Interim reanalysis of the ECMWF have also been used. Those data have been splitted into a training database (1999-2005) used to build the models, and a test database (2006-2012) reserved to test them and assess their performance. The models used have been built using three different techniques: analogues, random forests (a machine learning algorithm) and a combination of both. For evaluation purposes, the performance of the models at each location has been compared at a 95% confidence level with the simplest prediction-persistence of levels- and also with the nearest gridpoint WAM forecasts. For forecasting horizons between 3 and roughly 16 hours at locations near the coast (where wave farms can be installed), among the statistical models, those built on random forests outperform the rest, including WAM and persistence.