Correction of NWP Ocean Surface Wind Biases with Machine Learning

This work addresses the need for modelling and correcting the persistent Numerical Weather Prediction (NWP) local biases of the ocean surface wind forecasts. For such purpose, several NWP and ocean model output parameters are used as inputs to the machine learning and neural network models to generate the corrections of the NWP forecasts. The results show that such models are able to substantially reduce NWP local biases and therefore its overall error variance, opening the door for both its operational use as well the development of long-term data series of valuable ocean forcing datasets.