Graupel Weather

How do ECMWF, GFS and HRRR weather models compare?

ECMWF IFS and NOAA’s GFS are global weather models that use physics to simulate the atmosphere. ECMWF AIFS uses machine learning for its global forecast. NOAA’s HRRR focuses on short-range weather, and Graupel Weather uses its feed for the contiguous U.S.

Who runs ECMWF IFS, ECMWF AIFS, GFS and HRRR?

ECMWF IFS
The Integrated Forecasting System is the physics-based forecast system from the European Centre for Medium-Range Weather Forecasts.
ECMWF AIFS
The Artificial Intelligence Forecasting System is ECMWF’s machine-learning forecast. Graupel Weather uses the single-run feed.
GFS
The Global Forecast System is a worldwide forecast model from NOAA.
HRRR
The High-Resolution Rapid Refresh is NOAA’s short-range model. The contiguous U.S. version uses a 3 km grid and runs hourly.

What else can I compare?

NBM
NOAA’s National Blend of Models combines guidance from several models. The row uses the feed for the contiguous U.S.
ICON
Germany’s national weather service, DWD, runs ICON for global and regional forecasts.
GEM
Environment and Climate Change Canada runs this global weather model.
UKMO
Forecasts from the UK Met Office’s global and regional models.
ARPEGE
Météo-France’s global weather model.
JMA
Forecasts from the Japan Meteorological Agency. Its global GSM feed has a 0.5° grid and runs every six hours.
NWS
The official U.S. National Weather Service forecast, edited by forecasters. It covers U.S. places.
WeatherNext 2
Google DeepMind’s AI ensemble, a group of possible forecasts. It uses a 0.25° grid, runs at 00 and 12 UTC, and provides output in six-hour steps. This row is web-only.

The main forecast uses Open-Meteo’s Best match, which combines model data. WeatherNext 3 is a Google DeepMind model and has no live forecast data in Graupel Weather yet.

What does it mean when weather models disagree?

Models can start from different estimates of current conditions and use different ways to predict what happens next. Compare the same place and time, and look at what changes across the forecasts. If some show rain in the morning and others show it in the afternoon, treat the timing as uncertain.

When forecasts split, the tally counts the wet and dry forecasts and notes any rows without enough data. The chance of rain, where shown, is a separate value from the forecast source or the WeatherNext 2 ensemble.

Where can I compare weather models side by side?

Pick a place in Graupel Weather and open Models to compare temperature and precipitation forecasts side by side. Switch between Hourly and Daily views, and check which rows have data for the time you care about. A blank means data is missing for that row and time.

Open the app · View on Google Play

Sources