Machine Learning-based technique for the correction of numerical wave modelling over the Northern Tyrrhenian Sea

Human activities along shorelines and open water commonly rely on public forecasts to perform such activities safely. Metoceanic information, such as tides, currents, wind, and ocean waves, is estimated using numerical models and validated, adjusted, or corrected from in situ real-condition measurements. However, the models for calculating these parameters, the adjustment methodologies, or data assimilation may not be accurately adjusted to specific local conditions. Regarding ocean wave forecasting, although the development of new technologies and artificial intelligence has contributed to reducing the error between the modeled variables from various numerical models validated worldwide and those recorded insitu, further work is still required. The present study performs corrections with Generalized Additive and XGBoost models between the corrected significant wave height obtained from different altimeter and in situ buoy missions in the Tyrrhenian Sea. In particular, adjustment methods are applied to output variables from numerical wave models, such as significant wave height and peak period. Results have shown improvements in adjustment through Machine Learning methods. Coefficient of determination has increased from approximately 0.30 to 0.65, and the root mean square error fell from 0.55 to less than 0.3 over the study region. Trained Machine Learning models are generated from the correction methods over the area of study in the Tyrrhenian Sea, which can be reused to correct data from alternative models or predictive wave models as an application of the present research.