The European model vs the American model: ECMWF vs GFS
Two global supercomputer models drive most of the world's forecasts. Here is how the European ECMWF and the American GFS actually differ, and why neither one is always right.
Two models, one planet
Almost every weather forecast you see starts with a global model: a giant physics simulation of the entire atmosphere, run on a supercomputer. Two of these models dominate the field. One is European, one is American, and professional forecasters watch both of them, every single day.
The European model is usually called the ECMWF, after the European Centre for Medium-Range Weather Forecasts that runs it. The American model is the GFS, the Global Forecast System, run by the United States agency NOAA. They solve the same physics over the same planet, yet they often disagree. Understanding why is the key to reading any forecast honestly.
ECMWF: the European heavyweight
ECMWF's flagship high-resolution forecast, produced by its Integrated Forecasting System (IFS), runs at roughly 9 km grid spacing. In other words, the model chops the globe into boxes about 9 km across. That is a finer grid than the American model uses, and it is one reason the European system has topped global accuracy rankings for years.
ECMWF is an independent intergovernmental organisation backed by more than 30 European member and cooperating states, based in Reading in the UK with a data centre in Bologna, Italy. Its headline high-resolution forecast has traditionally been anchored on two main runs a day, at 00 and 12 UTC, though ECMWF has expanded its higher-frequency products over time.
GFS: the American workhorse
The GFS is run by NOAA's National Centers for Environmental Prediction. It uses the FV3 (Finite Volume Cubed) dynamical core, runs at about 13 km resolution, and produces forecasts out to 16 days. Just as important, it updates four times a day, at 00, 06, 12 and 18 UTC, so it refreshes more often than the European headline run.
The GFS's other defining trait is openness. It has long been fully free and open: anyone, anywhere, can download the raw output at no cost. That is why the GFS quietly powers a huge share of the world's weather apps, websites and hobbyist forecasts.
Cost and openness
Historically this was the sharpest contrast between the two. The GFS was free; ECMWF's data sat largely behind a paywall and licensing terms. That has changed. ECMWF has moved through a phased open-data policy, starting an open subset of real-time forecasts in January 2022 and, from 1 October 2025, opening its entire real-time forecast catalogue for free at 25 km resolution in real time. The very highest-resolution 9 km products still carry licensing conditions for some professional uses, but the old picture of a completely closed European model no longer holds.
Track record: the Hurricane Sandy story
The most famous example of the two models diverging is Hurricane Sandy in October 2012. Days before landfall, the ECMWF consistently forecast Sandy taking a sharp northwest turn back into the US East Coast, while the GFS kept nudging the storm out to sea.
The ECMWF model was among the first to show the northwest turn at six and seven days out, correctly signalling a US landfall while the American guidance still favoured a track into the open Atlantic.
The European model was right, and that early warning helped emergency planners along the New Jersey and New York coasts. Later research in Geophysical Research Letters found the difference came down mainly to how each model handled a technical detail called cumulus parameterization, not resolution alone. It is a reminder that model skill is about the whole system, not just grid size.
Why forecasters watch both, and why neither always wins
Sandy made ECMWF famous, but it did not make it infallible. On any given day, either model can be closer to reality. The European model wins more verification scores on average over long records, yet the American model regularly beats it for specific storms, regions or timeframes, and its four-times-daily refresh can catch changes sooner.
That is exactly why experienced forecasters never rely on a single model. When ECMWF and GFS agree, confidence is high. When they split, that disagreement is itself useful information: it flags a genuinely uncertain situation where you should expect the forecast to shift.
- Resolution: ECMWF around 9 km, GFS around 13 km.
- Update frequency: GFS four times a day; ECMWF's headline run twice a day.
- Cost: GFS always free and open; ECMWF now free at 25 km in real time, with premium high-resolution tiers.
- Average accuracy: ECMWF usually leads global rankings, but not on every storm.
This is the whole reason a comparison site exists. Seeing the models side by side, rather than trusting one black-box app, tells you when the atmosphere is predictable and when it is not. You can compare them for your area on the compare US forecasts and compare UK forecasts pages.