CMCC Seminar
24 May 2024 – h. 10:00 a.m. CEST at Viale B.Pichat 6/2 2nd floor, room BP-2A and via Zoom  (access code 525904)

Most machine learning results are limited in interpretability and generalization over different computational grid resolutions, initial and boundary conditions, domain geometries, and physical or problem-specific parameters. To address these challenges, we augment existing/low-fidelity dynamical models directly in their partial differential equation forms with both Markovian and non-Markovian neural network closure parameterizations. We apply our results to advecting nonlinear waves, shocks, chaotic balanced equations, and ocean acidification models. Our learned generalized neural closure models (gnCMs) find leading numerical error terms, discriminate among candidate functional forms in an interpretable fashion, achieve generalization, discover missing chaotic physics, and compensate for the lack of complexity in simpler models. We then investigate the effectiveness of deep neural operator models for reproducing and predicting fluid flows and realistic ocean simulations. We showcase applications to ocean forecasting in the Gulf of Mexico, Middle Atlantic Bight, and Massachusetts Bay, learning from high-resolution data-assimilative simulations employed for real sea experiments. We confirm that trained deep neural operator models are capable of predicting idealized periodic eddy shedding. For realistic ocean surface flows, they can predict several of the features and show some skill, providing potential for applications.