Toward a New Parameterization of Fine-Scale Ocean-Atmosphere Interactions Based on a Machine Learning Approach

In the last decades, mesoscale air-sea interactions have received increasing interest from the scientific community. Mesoscale thermal (sea surface temperature influence, TFB) and mechanical (oceanic surface current influence, CFB) air-sea interactions have been shown to have a strong influence on the wind up to the troposphere and on ocean dynamics. However, from an oceanic perspective, running an atmospheric model is very expensive. To overcome this issue, we have developed a convolutional neural network (CNN) that aims to reproduce the mesoscale ocean-atmosphere interactions. Training was performed with simulated data from a realistic coupled ocean-atmosphere tropical channel simulation (45°S- 45°N) using NEMO for the ocean model, WRF for the atmosphere model, and the OASIS3-MCT coupler. As a first step, the CNN was trained over two energetic regions (the Agulhas Current and the Kuroshio) to predict mesoscale surface stress anomalies from large-scale atmospheric and mesoscale oceanic inputs. Validation over the Gulf Stream and other regions shows that the CNN successfully reproduces the surface stress anomalies associated with both TFB and CFB.  In a second step, to parameterize the mesoscale ocean-atmosphere interactions, we coupled the CNN to NEMO via an Eophis library (pyOASIS) and ran a simulation over the tropical channel configuration. In this talk, we will present our main results in terms of oceanic energetics and ocean-atmosphere energy transfer.