Model transferability for wave height prediction

Machine learning methods have been used to estimate missing data collected by wave buoys. The developed methods need a large amount of historical data related to wind conditions to estimate the missing wave heights. To deal with this problem, in this paper, the suitability of a deep neural network, composed of a cascade of convolutional and long short-term layers, to model transfer, is analyzed. Different methods are proposed and validated in a case study, showing the possibility of transferring knowledge acquired from a source buoy to a target buoy with a limited amount of data.