submeso_param_net

This is the code repository for Bodner, Balwada, and Zanna. A Data-Driven Approach for Parameterizing Ocean Submesoscale Buoyancy Fluxes (in prep).

• Plotting scripts are in notebooks

• Code for preprocessing the LLC4320 to generate training data is under scripts/preprocess_llc4320/

• Postprocessing diagnostics are under scripts/postprocess/

• Examples for training the CNN under the various test cases discussed in the paper can be found under scripts/nn_training_examples. These scripts correspond with the modules found in submeso_ml to perscribe the CNN architecture and hyperparameters, as well as the input features, specific regions/seasons, and filter scale.