Family of ML weather/ocean forecasting foundation models from the Fudan AI Innovation & Incubation Institute (Hao Li group). FuXi-2.0 couples atmosphere and ocean surface for practical medium-range forecasting, beating ECMWF HRES on key variables including wind power. FuXi-Ocean (2025) is the first data-driven global ocean forecast at eddy-resolving 1/12° resolution, sub-daily, down to 1500m depth, with a Mixture-of-Time module. FuXi-ENS (Science Advances 2025) is a VAE-based ML ensemble that beats the ECMWF ensemble's CRPS on 98.1% of variable/lead-time combinations — with strong adoption in the AI-weather community.

Model Details

Variants

Name Parameters Notes
FuXi-2.0 Atmosphere + ocean-surface coupled medium-range model (arXiv:2409.07188)
FuXi-Ocean First eddy-resolving 1/12° global ocean forecast model (arXiv:2506.03210)
FuXi-ENS VAE-based ensemble weather model, Science Advances 2025 (arXiv:2405.05925)

Paper

weatherclimatescientific-fmresearch