Crepes_complete repository: Learning-based calibration of ocean carbon models to tackle physical forcing uncertainties and observation sparsity

Repository for the paper "Learning-based calibration of ocean carbon models to tackle physical forcing uncertainties and observation sparsity.

Description: This study is part of the PhD project "Carbon REconstructed Per an Emulator that is Supervised" (Carbone REconstruit Par Emulateur Supervisé). It contains 6 different files:

• spec-file.txt contains all the packages installed thanks to conda with the effective versions
• Dataset_Generator.py To generate all the necessary data sets.
• DA_method.py To apply the DA-based method on a data set.
• NN_method.py To train and validate a NN upon the generated data set.
• Functions.py contains all the functions used to plot/analyse the data.
• Article_plots.py plots the figures that mix both DA and NN results.

For a use without errors:

1. Install the correct packages with their associated version with the spec-file.txt -> In the command prompt: conda create --name MyEnv --file spec-file.txt -> Add the Lightning package that cannot be installed with conda: pip install https://github.com/Lightning-AI/lightning/archive/refs/heads/release/stable.zip -U
2. Generate the different data sets: run Dataset_Generator.py
3. Use freely the different methods (run DA_method.py or NN_method.py)

/!\ The Article_plots.py script will work only if results have been generated for each 9 scenarii with both DA and NN methods.