Predicting turbulence statistics of wind flow over oceanic waves using data-driven convolutional neural networks

To efficiently assess the wave effect on the wind field, we developed an autoencoder convolutional neural network (CNN) based on data obtained from large-eddy simulation (LES) of wind turbulence over complex surface waves. We performed simulations for wind fields of four characteristic wind velocities and two wave conditions to produce the training and validation data. is then The CNN was then trained using the LES results of two wind velocities with various wave conditions and employed to reconstruct the time-averaged velocity field, wave-induced fluctuation, and turbulent kinetic energy of the wind field over oceanic waves. The validation study results show good agreement between the CNN predictions and LES for the turbulence statistics of the wind flow in the selected oceanic environment.