nextsim_surrogate

nextsim fully data-driven surrogate model

Installation

To install environment:

conda env create -f environment.yml

To activate the environment:

conda activate env

Dataset Build

Original Files are download from nextsim NANUK outputs [1]. Those files are available through the SASIP github. link to neXtSIM outputs. Forcings were dowloaded from ERA5 file link

Dataset is build and save under netCDF file with make_dataset script.

python make_dataset.py

To build the dataset with a TFRecord architecture (), use make_tfrecord script

python make_tfrecord.py

Note, the dataset takes a susbstantial amount of place on disk: 360Go for the .nc file, 270Go as .tfrecords

Run the code

The code is separated on two main section : data and src

• the data directory contains the code to build the dataset, once the original neXtSIM and ERA5 files are downloaded.
• the src file contains all the code to train the neurall network, and compute the test metrics.

To train the neural network:

python train.py

It takes around 18h on 1 GPU A100. All test metrics are computed with test.py and the Power Spectral Density is compute with

python compute_PSD.py

Code to plot the figure in the article is in the plot_article notebook.