Model comparison
ICON vs ECMWF: how the two models compare
Last updated: 14 November 2026
ICON and ECMWF are both European, both global and both free to view, but they are built for different jobs. ECMWF aims at the best possible forecast three to ten days ahead. ICON is designed to scale from a global forecast down to kilometre-scale detail over Germany. Here is how they compare.
Side by side
| ICON | ECMWF IFS | |
|---|---|---|
| Run by | DWD (Germany) | ECMWF (intergovernmental) |
| Global resolution | ~13 km | ~9 km |
| Full-length runs | 00 / 12 UTC | 00 / 12 UTC |
| Extra runs | 06 / 18 UTC, to 5 days | 06 / 18 UTC, to 6 days |
| Forecast range | 7.5 days | 15 days |
| High-res nests | ICON-EU 6.5 km · ICON-D2 2.2 km | None (global only) |
| Ensemble | 40 members | 51 members at ~9 km |
| Availability | A few hours after run time | Later than ICON |
Resolution
Globally, ECMWF is the finer of the two: about 9 km against ICON’s 13 km. Finer grid spacing helps with coastlines, mountain ranges and the structure of intense lows and tropical cyclones.
Over Europe the picture reverses. ICON-EU runs at 6.5 km and ICON-D2 at 2.2 km over Germany, the Alps and neighbouring countries. At 2.2 km the model simulates individual thunderstorms explicitly, which no global model can do. For a summer storm day in Bavaria or a foehn event in Switzerland, the ICON nests carry detail that ECMWF’s grid cannot.
Forecast range and timing
ECMWF forecasts twice as far: 15 days against 7.5 for ICON. If you are looking at the weekend after next, ECMWF covers it and ICON does not.
ICON is quicker to arrive. ECMWF waits longer to collect observations before it starts and takes longer to compute, so its output is published later. Early in the morning, European time, the newest ICON run is often available before the newest ECMWF run.
Accuracy
On global verification scores ECMWF leads every other model, ICON included, and the lead grows with forecast length. By day five or six, ECMWF’s advantage in the large-scale pattern is clear. Much of that comes from its data assimilation, the process that builds the starting state from observations, and from its finer global grid.
In the first two to three days over Europe, the difference is smaller and ICON’s high-resolution nests can make it the more useful model for local weather: valley fog, convective showers, wind in complex terrain.
ECMWF’s larger, higher-resolution ensemble is also the stronger tool for uncertainty. It gives a more reliable picture of the range of possible outcomes than any other global ensemble.
Which to use when
For days four to ten, default to ECMWF and use ICON, the GFS and others as a cross-check. For the next 48 hours in central Europe, ICON-D2 and ICON-EU carry more local detail. For days two to four, look at both: this is where they are closest in skill and where agreement between them is most informative.
One practical note: MeteoModels shows the global configuration of each model, so the comparison on the map is ICON global against ECMWF’s open data, like for like.
When they disagree
Because the two models are developed independently, agreement between them is real evidence. If both place a front over the same region at the same time, it is a robust signal. If ECMWF has heavy rain where ICON is dry, neither should be taken at face value until the next runs come in.
Overlay the two on the MeteoModels map and switch on the certainty mask: bright where they agree, faded where they do not.
The verdict
ECMWF is the more accurate global model and the better guide beyond day four. ICON is fresher and, through its nests, sharper for short-range weather in Europe. They complement each other — and when they agree, you can be confident.
Frequently asked questions
Is ECMWF more accurate than ICON?
On global average scores, yes, and the difference grows with lead time. In the first two to three days over Europe the two are much closer, and ICON’s high-resolution regional versions can be more useful for local detail.
Which has higher resolution, ICON or ECMWF?
Globally, ECMWF at about 9 km against ICON’s 13 km. Over Europe, ICON’s nested models are finer: 6.5 km for ICON-EU and 2.2 km for ICON-D2.
Which updates more often?
Both run at 00, 06, 12 and 18 UTC, but ICON’s output is published sooner after each run starts, and ECMWF’s 06 and 18 UTC runs stop at six days.