OneScience-Group/FuXi-Weather

FuXi-Weather

# Model Introduction FuXi-Weather maps raw satellite observations to global forecasts through FuXi-DA and cascaded FuXi forecast models.

Paper: A data-to-forecast machine learning system for global weather
https://doi.org/10.1038/s41467-025-62024-1

Model Description

The system was proposed by teams from the Shanghai Academy of Artificial Intelligence for Science, Fudan University, CMA, and collaborators. It was trained with ERA5, microwave radiances from three polar-orbiting satellites, and GNSS radio occultation. Masked latent assimilation and Short/Medium forecasting support six-hour cycling and global forecasts to ten days.

Use Cases

Use Case	Description
Satellite assimilation	Fuse sparse observations and forecast backgrounds.
Global forecasting	Cascade short- and medium-range models.
Cycling analysis	Update global analyses and forecasts every six hours.
ModelScope/OneCode execution	Validate data, training, inference, metrics, and visualization.
Multi-GPU training	Start multi-process training through torchrun.

Usage Instructions

Use a GPU or DCU when available; CPU supports the default smoke configuration. DCU users should install a compatible DTK release.

hf download OneScience-Group/FuXi-Weather --local-dir ./FuXi-Weather
cd FuXi-Weather
python scripts/fake_data.py

For single-process training, use:

python scripts/train.py

For multi-process training, use:

torchrun --standalone --nproc_per_node=2 scripts/train.py

Run inference and evaluation with:

python scripts/inference.py
python scripts/result.py

Training jointly optimizes analysis and forecast objectives. Inference produces finite [2,12,20,16,16] cascaded forecasts, while evaluation reports lead-time RMSE under result/evaluation/.

Trained Weights

No weights are bundled under weight/. The FuXi model is available at https://zenodo.org/records/10401602, and the FuXi Weather model used by the paper is available at https://zenodo.org/records/15762985.

Citation and License

This repository is an independent engineering reproduction of the public FuXi-Weather specifications.

The original paper is licensed under CC BY-NC-ND 4.0; the original paper, official code, model weights, and related data remain subject to their respective licenses and terms.