mercator-ocean/GLONET

GLONET

Mercator's End-to-End Neural Forecasting System

GLONET is a neural network-based ocean forecasting system developed by Mercator Ocean International, which architecture is described in the scientific paper, outputting in seconds a 10-days daily forecasts of the temperature, sea surface height, salinity and currents at 1/4 degree resolution (1/12 degree effective resolution) over 21 depth levels.

Visit GLONET website for more information.

Forecast visualization examples

Current	Temperature	Salinity	Sea surface height

Architecture

Overview of GLONET's architecture containing different modules, particularly time-block designed to learn feature maps encapsulating initial conditions along with forcings. A spatial module architectured to learn multi-scale dynamics, and finally an encoder-decoder to fuse multi-scale circulations into a unified latent space.

Run inference

GLONET inference requires an NVIDIA GPU and Copernicus Marine credentials.

micromamba create -f inference/environment.yml
micromamba activate Glonet-Inference
export COPERNICUSMARINE_SERVICE_USERNAME="your-username"
export COPERNICUSMARINE_SERVICE_PASSWORD="your-password"
python inference/run_glonet_inference.py --forecast-date 2026-07-22

The command downloads and regrids the two initialization days, runs the three GLONET depth models, and writes one 10-day NetCDF forecast at 1/4° resolution.

Citation

GLONET's scientific paper is published in Journal of Geophysical Research: Machine Learning and Computation (AGU JDR).

@article{
  https://doi.org/10.1029/2025JH000686,
  author = {El Aouni, Anass and Gaudel, Quentin and Regnier, Charly and Van Gennip, Simon and Le Galloudec, Olivier and Drevillon, Marie and Drillet, Yann and Lellouche, Jean-Michel},
  title = {GLONET: Mercator's End-to-End Neural Global Ocean Forecasting System},
  journal = {Journal of Geophysical Research: Machine Learning and Computation},
  volume = {2},
  number = {3},
  pages = {e2025JH000686},
  doi = {https://doi.org/10.1029/2025JH000686},
  url = {https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2025JH000686},
  eprint = {https://agupubs.onlinelibrary.wiley.com/doi/pdf/10.1029/2025JH000686},
  note = {e2025JH000686 2025JH000686},
  year = {2025}
}

Licensing

This project is distributed under a dual licensing scheme to separate the code from the learned parameters (weights).

1. Source Code / Scripts

The source code and scripts provided in this repository are released under the European Union Public Licence (EUPL v1.2).

• See the LICENSE file for full details.

2. Model Weights

The model weights and associated statistics stored in the weights_and_stats/ directory are released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.

• You are free to: Share and adapt the weights.
• Under the following terms: You must give appropriate credit to Mercator Ocean International, provide a link to the license, and indicate if changes were made.

Disclaimer: This model is for research and demonstration purposes...

Contact

GLONET is implemented and developed by Mercator Ocean International. For any information about GLONET, you can contact glonet@mercator-ocean.eu.