Ocean General Circulation Models

Can we learn to emulate physical models from simulations?

CNRS
MEOM

There are many sophisticated Ocean GCM's that exist within the community like NEMO and MOM6. However, these are very complex systems with many internal parts and parameterizations. There have been some attempts to convert some of these mega systems into differentiable models [Häfner et al., 2018] or rebuild them from scratch [Ramadhan et al., 2020]. However, it would undeniably be a huge effort to convert all of these systems into differentiable models. In addition, it might not be useful to have such a large differentiable system because the back-propagation process throughout the entire system in a learning-based setting might not be feasible.

A different approach is to train surrogate models to emulate individual components of the system.

References
  1. Häfner, D., Jacobsen, R. L., Eden, C., Kristensen, M. R. B., Jochum, M., Nuterman, R., & Vinter, B. (2018). Veros v0.1 – a fast and versatile ocean simulator in pure Python. Geoscientific Model Development, 11(8), 3299–3312. 10.5194/gmd-11-3299-2018
  2. Ramadhan, A., Wagner, G., Hill, C., Campin, J.-M., Churavy, V., Besard, T., Souza, A., Edelman, A., Ferrari, R., & Marshall, J. (2020). Oceananigans.jl: Fast and friendly geophysical fluid dynamics on GPUs. Journal of Open Source Software, 5(53), 2018. 10.21105/joss.02018