This repository is the official implementation for the paper: Armand de Villeroché et al. (2025), Physics-informed neural surrogates for atmospheric dispersion in industrial sites, submitted to "Air Quality, Atmosphere & Health".
This repository contains the code for the model presented in the article, as well as the configuration files and analysis scripts for the studies presented in the article.
The data is available on demand.
Requiered packages are numpy, tensorflow 2.14 and vtk. matplotlib and cmocean are also used for plots.
To run a study, create an environement variable PATH_CASES containing the path to the raw dataset. Then launch the study with python -m pinnatmo.train path/to/study/config/file/.
If you have further questions, please feel free to create a GitHub issue.