On the inference of the Peak Wave Period using Satellite Altimetry measurements

Satellite altimeters provide valuable measurements of ocean surface parameters such as significant wave height and wind speed but do not directly observe wave period, a key variable for sea state characterization and offshore applications. This work presents a machine learning-based methodology to estimate the peak wave period using altimeter-derived data. Leveraging a large dataset of numerical wave simulations and in-situ observations, we develop a two-step approach: first, a classification model discriminates between locally generated windsea and remotely generated swell conditions; second, separate regression models are trained for each wave type to predict the peak wave period. The models are implemented using Random Forests and incorporate spatial and temporal features to enhance accuracy. Results show high predictive performance, with mean absolute errors around 0.3 seconds and coefficients of determination exceeding 0.93 in test sets. The methodology enables enhanced sea state reconstruction from satellite observations, supporting improved ocean monitoring and offshore planning.