MAE/SIO Recruitment Seminar - Physics-based and Data-driven Study of Air–Sea Processes

Physics-based models play an essential role in the simulations of the small-scale air–sea processes. Of equal importance are the emerging data-driven techniques because of their power in approximating complex dynamic systems. In the first part of my talk, I will present results obtained by large-eddy simulation of wind turbulence over a travelling wave subject to an impulsive mean wind speed increase or decrease. It is found that the turbulence motions have a delayed response to the change in the mean flow, while the wave-coherent motions have nearly real-time response. We observe an asymmetric hysteresis between the growing wind and the decaying wind in the evolution of the form drag and the viscous drag. Then, I will introduce a novel model based on a high-order spectral method for surface waves with the bottom boundary conditions computed from an internal wave solver. The model is validated against the perturbation theory in predicting the resonant wave energy growth in nonlinear triad interaction. We show that our model can effectively capture the surface manifestation of a large amplitude internal wave. In the final part, I will present a machine-learning based method for modeling the downwelling irradiance in the upper ocean to address the high computational cost issue in conventional Monte–Carlo (MC) simulations. Our method replaces the time-consuming photon tracking process with a pretrained deep neural network. The irradiance field predicted by the neural network agrees well with the MC simulation result but takes significantly less time to compute.