Training-Free Data Assimilation with GenCast 🌍

This repository contains the code for the paper Training-Free Data Assimilation with GenCast by Thomas Savary, François Rozet and Gilles Louppe. The paper was published in 2025 in the Tackling Climate Change with Machine Learning Workshop at NeurIPS.

Code

The methods described in the paper are implemented in the filtering folder. In particular, it contains the following files:

• filtering/wrapper/denoisers.py that implements an MMPS denoiser using the "basic" GenCast denoiser to draw samples from the optimal proposal distribution $q(x_{k+1} \mid x^{k}, y^{k+1}) = p(x_{k+1} \mid x^{k}, y^{k+1})$.
• filtering/fa_apf.py that implements the Fully Adapated Auxiliary Particle Filter (FA-APF) with covariance inflation to control the degeneracy of the weights.

To do other experiments, users can modify the configuration files in the config folder, as well as observations parameters (mask, covariance, ...) in the data/observations folder.

Model and data

Our work build on GenCast, a diffusion-based emulator of the atmosphere developed by Google. GenCast's denoisers were trained on ERA5, a global atmospheric reanalysis dataset covering the period from 1940 to present and produced by the ECMWF (the European Centre for Medium-Range Weather Forecasts).

Citation

If you find this work useful in your research, please consider citing:

@misc{savary2025trainingfreedataassimilationgencast,
      title={Training-Free Data Assimilation with GenCast}, 
      author={Thomas Savary and François Rozet and Gilles Louppe},
      year={2025},
      eprint={2509.18811},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2509.18811}, 
}