Correcting Predictions from Simulating Wave Nearshore Model via Gaussian Process Regression

Accurately predicting wave height has a significant impact on offshore production and marine transportation. Numerical model predictions are the most commonly used wave height prediction method. But all numerical models are hard to describe the rules of ocean physics numerical model completely. Machine learning methods learn the inherent rules of data and are widely used in various predictions. In this paper, we try to use machine learning to correct errors in numerical model predictions. We count wave height numerical model predictions and observations for one year and find that the residuals of them are subject to a Gaussian distribution. Therefore, we propose a method for correcting wave height predictions from the Simulating Wave Nearshore (SWAN) model based on Gaussian process regression (GPR). To this end, we use residuals of wave height observations and the predictions of the SWAN model as input. We then train a GPR model for correcting the next time step wave height. Experimental results reveal that the proposed method predicts the wave height more accurately compared with the maritime numerical model.