MeteoStack
Model guide

Every weather model,
explained.

A weather model is a computer simulation of the atmosphere run by a national weather service. Each one starts from slightly different observations and physics, so they disagree, and that disagreement tells you how certain a forecast really is. Here are the 12 models MeteoStack stacks.

ModelRun byResolutionRangeType
ECMWF IFS European Centre for Medium-Range Weather Forecasts 9 km (IFS HRES) 15 days Physics (NWP)
ECMWF AIFS European Centre for Medium-Range Weather Forecasts 0.25° (~28 km), 6-hourly steps 15 days Machine learning
NOAA GFS NOAA / National Centers for Environmental Prediction ~13 km (blended with HRRR 3 km over the US) 16 days Physics (NWP)
DWD ICON Deutscher Wetterdienst (German Weather Service) 13 km global, 7 km Europe (ICON-EU), 2 km Central Europe (ICON-D2) 7 days Physics (NWP)
Canadian GEM Environment and Climate Change Canada ~15 km global; 2.5 km HRDPS over Canada 10 days Physics (NWP)
JMA GSM Japan Meteorological Agency ~55 km open data global; 5 km MSM over Japan 11 days Physics (NWP)
Météo-France ARPEGE / AROME Météo-France ~25 km world, 10 km Europe, 1.3 km AROME over France 4 days Physics (NWP)
UK Met Office Met Office (United Kingdom) ~10 km global; 2 km UKV over the UK 7 days Physics (NWP)
CMA GRAPES China Meteorological Administration ~15 km 5 days Physics (NWP)
BOM ACCESS-G Australian Bureau of Meteorology ~12 km 10 days Physics (NWP)
KMA KIM Korea Meteorological Administration ~12 km global; 1.5 km over Korea 12 days Physics (NWP)
MET Norway Nordic Norwegian Meteorological Institute 1 km over Scandinavia, global fallback elsewhere 3 days Physics (NWP)

ECMWF IFS

European Centre for Medium-Range Weather Forecasts · European model, Euro, IFS, ECMWF HRES

The "European model". For two decades the reference global model in independent verification scores, especially beyond day 3.

ECMWF AIFS

European Centre for Medium-Range Weather Forecasts · ECMWF AI model, Artificial Intelligence Forecasting System

ECMWF's machine-learning model, operational since 2025. Trained on decades of reanalysis, it runs in minutes instead of hours.

NOAA GFS

NOAA / National Centers for Environmental Prediction · American model, Global Forecast System

The "American model". Free, global, four runs a day out to 16 days, the most widely redistributed forecast on Earth.

DWD ICON

Deutscher Wetterdienst (German Weather Service) · German model, ICON-EU, ICON-D2

The German model. Its seamless blend nests a 2 km model inside a 7 km model inside the global one, which makes it superb over Europe.

Canadian GEM

Environment and Climate Change Canada · Canadian model, CMC, GDPS, HRDPS

The Canadian model. A solid third opinion in North America, with a 2.5 km high-resolution nest.

JMA GSM

Japan Meteorological Agency · Japanese model, GSM, MSM

The Japanese model. Coarse globally in open data but a valuable independent member, especially for the West Pacific.

Météo-France ARPEGE / AROME

Météo-France · French model, ARPEGE, AROME

The French models. AROME at 1.3 km is one of the sharpest operational models for storms and Mediterranean downpours.

UK Met Office

Met Office (United Kingdom) · UK model, Met Office Unified Model, UKV

The UK Met Office Unified Model, consistently among the top-scoring global models, with a 2 km UK nest.

CMA GRAPES

China Meteorological Administration · Chinese model, GRAPES-GFS

China's global model. An independent opinion that widens the spread of the stack.

BOM ACCESS-G

Australian Bureau of Meteorology · Australian model, ACCESS

The Australian model, built on the Met Office Unified Model. The Southern Hemisphere specialist.

KMA KIM

Korea Meteorological Administration · Korean model, KIM, GDPS, LDPS

The Korean Integrated Model, with a 1.5 km nest over the Korean peninsula.

MET Norway Nordic

Norwegian Meteorological Institute · yr.no model, MET Nordic, MEPS

The model behind yr.no. A 1 km nowcast-grade forecast for the Nordics.

Which weather model is the most accurate?

In global verification published by the World Meteorological Organization's lead centres, ECMWF has held the top spot for medium-range forecasts for most of the past twenty years, with the UK Met Office usually second and GFS, GEM and JMA behind. Machine-learning models such as ECMWF AIFS now match or beat physics models on many large-scale scores.

But "most accurate on average" is not "right for your town tomorrow". High-resolution regional models: ICON-D2, AROME, HRRR, UKV: often win the first 48 hours locally. That is why MeteoStack shows them all and computes a consensus: the median of many independent models beats any single model over time.

Popular comparisons