Meteo·Models

Weather models

Weather forecast models, explained

Every forecast you read comes from a numerical weather prediction (NWP) model — a physics simulation of the atmosphere run on a supercomputer. No single model is right all the time, which is why comparing them is useful. Here are the models you can overlay in Meteo·Models, and what each is good for.

Global vs regional models

Global models cover the whole planet and run out one to two weeks, so they trade resolution for reach. Regional models such as AROME (Météo-France, ~1.3 km) cover a limited area a couple of days out at much higher resolution, which lets them resolve thunderstorms, fog and terrain effects that global models smooth over. Meteo·Models focuses on the global models below, where multi-model comparison is most valuable.

The global models you can overlay

ModelRun byCoverageResolutionUpdates (UTC)Good for
ECMWF IFSECMWF (Europe)Global~9 km (0.25° open data)00 / 12 UTCThe reference global model; most accurate in the medium range.
ECMWF AIFSECMWF (Europe)Global0.25°00 / 12 UTCECMWF’s machine-learning model — fast, and competitive with the physics-based IFS.
GFSNOAA / NCEP (USA)Global~13 km (0.11° / 0.25°)00 / 06 / 12 / 18 UTCRuns four times a day — the freshest global data; free and open.
NOAA AIGFSNOAA (USA)Global0.25°00 / 06 / 12 / 18 UTCNOAA’s AI-based global forecast, run alongside the GFS.
ICONDWD (Germany)Global~13 km00 / 06 / 12 / 18 UTCIcosahedral, non-hydrostatic grid; strong over complex terrain.
ARPEGEMétéo-FranceGlobal (stretched)~5 km over Europe00 / 06 / 12 / 18 UTCHigher resolution over Europe, coarser elsewhere.
UKMO GlobalUK Met OfficeGlobal~10 km00 / 12 UTCThe Unified Model; consistently among the top global performers.
GEMCMC (Canada)Global~15 km00 / 12 UTCEnvironment Canada’s Global Environmental Multiscale model.
JMA GSMJMA (Japan)Global~13–20 km00 / 06 / 12 / 18 UTCJapan’s Global Spectral Model; strong over the western Pacific.
CMA GRAPESCMA (China)Global~0.25°00 / 12 UTCChina Meteorological Administration’s global model.

Because each model has different physics, resolution and data assimilation, they disagree — especially for precipitation and for anything beyond about five days. That disagreement is exactly what the certainty mask visualises. See how the consensus is computed on the methodology page. Read the methodology →

Open the live model map GFS vs ECMWF