Interpretable Machine Learning-based Radiation Emulation for ICON

This repository contains the code for the interpretable ML-based radiation emulator for ICON. The corresponding paper is published:

Hafner, K., Iglesias-Suarez, F., Shamekh, S., Gentine, P., Giorgetta, M. A., Pincus, R., & Eyring, V. (2025). Interpretable machine learning-based radiation emulation for ICON. Journal of Geophysical Research: Machine Learning and Computation, 2, e2024JH000501. https://doi.org/10.1029/2024JH000501

If you want to use this repository, start by executing

conda env create -f environment.yml
conda activate hafner_ml_rad

Repository content

• evaluation contains some functions for prediction and evaluation
• models contains the NN architecture including preprocessing layer
• nn_config contains the configuration of all NNs
• plotter contains plotting functions
• preprocessing contains the normalization file and data loader
• utils contains some helper functions
• train_coarse_levante.py contains the training script
• eval_coarse_levante.py contains the evaluation script
• eval_coarse_levante.ipynb contains some further evaluation