Data Driven Methods for Submesoscale Modelling
This project contains files for the paper 'Data-driven methods for predicting oceanic upwelling at submesoscales' from the 2023 NFFDy Summer School in Cambridge.
Simulations of a small region of ocean were run using Oceananigans. Then a convolutional neural network (written using Pytorch) was used to predict the upwelling (depth-averaged vertical velocity) based on the surface density and pressure only.
See here for the summer school proceedings inclding our paper.