Statistical forecast of the marine surge

Gonzalo Iñaki Quintana, Pierre Tandeo, Lucas Drumetz, Laurent Leballeur, Marc Pavec

Abstract: This paper studies different machine learning methods for solving the regression problem of estimating the marine surge value given meteorological data. The marine surge is defined as the difference between the sea level predicted with the tides equations, and the real measured sea level. Different approaches are explored, from linear regression to multilayer perceptrons and recurrent neural networks. Stochastic networks are also considered, as they enable us to calculate a prediction error. These models are compared with a baseline method, which uses physical equations to calculate the surge. We show that all the statistical models outperform the baseline, being the multilayer perceptron the one that performs the best. (It reaches an \(R^2\) score of 0.68 and an RMSE of 7.3 cm.)