Deep Learning Emulator for the dynamics of sea ice within the MITgcm model

Still from video of model simulation output, Nguyen/Heimbach and Greg Foss, TACC, UT Austin, Arctic Atlantification, 2018.

Sea ice dynamics can be very computationally epxensive to simulate within general circulation models such as the MITgcm. Additionally, sea ice rheology is still not understood very well, making realistic simulations difficult.

We aim to leverage Machine Learning / Deep Learning models in order to tackle both of these problems. Essentially we ask the following questions, hoping to answer them in the future -

1. Can Deep Learning emulators reduce the computational cost of simulating sea ice dynamics realistically?

2. Can we infer sea ice rheology from real datasets?

3. Can we interpret our results by comparing to interpretable models, for example linear regression?

Further information forthcoming.