FourCastNet
AI / machine learningNVIDIA · 15 nm · Global
NVIDIA's machine-learning global model with experimental public runs.
When is FourCastNet updated?
Publication times for FourCastNet are not catalogued yet, so Ilma cannot say when it is updated. The provider's own pages are the place to check until they are added.
- Provider
- NVIDIA
- Region
- Global
- Resolution
- ~15 nm
- Licence
- See provider
Schedule
Publication times not yet catalogued.
What AI weather models are for
AI (machine-learning) models learn how the atmosphere evolves from decades of past weather, then produce a forecast in minutes instead of hours. They do well on large-scale patterns such as lows and fronts, but they are newer, often coarser, and less tested on local effects and extremes. Use them as an extra opinion next to a physics-based model.
FourCastNet has a grid spacing of about 15 nm, so each value stands for a box about that size. A model needs several grid boxes to show a feature, so details smaller than about 76 nm are smoothed out or missed.
Other global models
Comparing global models shows how certain the forecast is: when they agree, trust it more. Those marked “times known” have countdowns and alerts on Ilma.
- AuroraMicrosoft5.9 nm
- FengWuShanghai AI Laboratory13 nm
- ECMWF AIFSECMWF15 nmtimes known
- FuXiFudan University15 nm
- GraphCast / GenCastGoogle DeepMind15 nm
- NOAA AI-GFS (AIGFS)NOAA15 nmtimes known
- Pangu-WeatherHuawei15 nm
- AICONDWD—