MeteoModels

Model profile

The ECMWF model, explained

Last updated: 14 November 2026

When a forecaster mentions “the European model” or “the Euro”, they mean the Integrated Forecasting System (IFS) run by ECMWF. It has been the most accurate global weather model on average for decades. This guide explains what it is, why it performs so well and where its limits are.

ECMWF at a glance

ECMWF IFS
Stands forEuropean Centre for Medium-Range Weather Forecasts
Founded1975, as an intergovernmental organisation
SitesReading (UK) · Bologna (Italy) · Bonn (Germany)
ModelIntegrated Forecasting System (IFS)
Resolution~9 km, 137 vertical levels
Main runs00 / 12 UTC, to 15 days
Extra runs06 / 18 UTC, to 6 days
Ensemble51 members at ~9 km
Open data0.25° and 0.1° grids, free to use

What ECMWF is

ECMWF is the European Centre for Medium-Range Weather Forecasts. It is not a national weather service and it does not issue public forecasts or warnings. It is an intergovernmental organisation, founded in 1975 and funded by more than 30 member and co-operating states, with one core job: produce the best possible global forecasts for the medium range — roughly three to fifteen days ahead — and hand them to its members’ national weather services.

Pooling the resources of many countries into one centre is a large part of why it leads. ECMWF operates one of the biggest supercomputing facilities in meteorology, in Bologna, and employs several hundred scientists who work on a single forecasting system.

How the IFS works

The Integrated Forecasting System is a physics-based model. It represents the atmosphere on a grid with cells about 9 km across and 137 levels from the surface to the edge of space, and it is coupled to models of the ocean, sea ice, ocean waves and the land surface, because all of them exchange heat and moisture with the air.

The part that matters most is the starting point. Before each forecast, the IFS blends tens of millions of observations — from satellites, aircraft, weather balloons, ships, buoys and surface stations — with its own previous short-range forecast to build an estimate of the current state of the atmosphere. ECMWF’s method for this, called 4D-Var, is widely regarded as the best in the field, and a better starting state leads to a better forecast at every lead time.

The control run and the ensemble

Every ECMWF forecast is an ensemble of 51 runs. One, the control, starts from the best estimate of current conditions; this is the forecast most maps and apps show, and the one long known as the high-resolution run or HRES. The other fifty start from slightly perturbed initial conditions. Since 2023 all 51 have run at the same ~9 km resolution, and since a 2026 upgrade the former high-resolution run itself serves as the ensemble’s control.

The ensemble exists because the atmosphere is chaotic: tiny errors in the starting state grow over time. If all 51 members tell the same story, the forecast is robust. If they scatter, it is uncertain. The main runs go out to 15 days, with longer extended-range and seasonal forecasts produced on separate schedules.

When it updates

The main runs start at 00 and 12 UTC and forecast 15 days ahead. Shorter runs at 06 and 18 UTC cover the first six days. Because the IFS waits for more observations to arrive before it starts and is expensive to compute, its output reaches users later than the GFS: data from the 00 UTC run is typically available around 06 to 08 UTC.

That means the ECMWF forecast you are looking at can be several hours older than the newest GFS or ICON run. In fast-changing weather, freshness can matter as much as average skill.

How accurate ECMWF is

On the standard measure used to compare global models — how well the forecast pattern of pressure in the middle atmosphere matches what actually happened — ECMWF has led for decades. A common rule of thumb is that its forecasts stay useful about a day longer than its nearest competitors: its six-day forecast is roughly as good as another model’s five-day forecast.

The best-known illustration is Hurricane Sandy in 2012. Around a week ahead, ECMWF’s forecasts consistently showed the storm making an unusual left turn into the north-eastern United States, while the GFS at first took it harmlessly out to sea. ECMWF was right, and the episode triggered major investment in the American model.

The gap has narrowed since then, and ECMWF does not win every forecast. Its advantage is an average over many days and the whole globe. On any given day, for any given town, another model can be closer to the truth — particularly for rainfall amounts, which every model struggles with.

ECMWF’s AI model: the AIFS

In 2025 ECMWF made a second, very different model operational: the Artificial Intelligence Forecasting System, or AIFS. Instead of solving physical equations, it is a machine-learning model trained on decades of past weather. It runs in minutes, not hours, and on several headline scores it matches or beats the physics-based IFS, especially for the large-scale pattern and for tropical cyclone tracks.

ECMWF runs the two side by side. The IFS still provides the starting conditions that the AIFS needs, and it remains the reference for fine detail and for intense, small-scale extremes.

Is ECMWF data free?

For most of its history ECMWF data was expensive, which is why consumer apps were built on the free GFS. That has changed. ECMWF now publishes a large open-data set of its forecasts under an open licence that allows commercial use. It started at coarser resolution, moved to a 0.25° grid in 2024, and since October 2026 has also been available at 0.1°, close to the model’s native 9 km.

MeteoModels reads that open data, so the ECMWF layer on the map comes from the same IFS forecast that national weather services use, at the open-data grid spacing.

How to read ECMWF alongside other models

ECMWF is the best single default for a medium-range forecast, but the most reliable signal is agreement. When ECMWF, GFS, ICON and the UK Met Office model all show the same thing, you can plan around it. When ECMWF is on its own, even the best model on average may be the outlier that day. Overlay them on the MeteoModels map and the certainty mask shows which situation you are in.

Frequently asked questions

What does ECMWF stand for?

ECMWF stands for the European Centre for Medium-Range Weather Forecasts. It is an intergovernmental organisation founded in 1975 and supported by more than 30 European states.

Is ECMWF the same as the “Euro” or European model?

Yes. “The Euro” and “the European model” are informal names, used mostly in North America, for ECMWF’s Integrated Forecasting System (IFS).

How often does the ECMWF model update?

The main forecasts run twice a day, at 00 and 12 UTC, out to 15 days. Shorter runs at 06 and 18 UTC cover the first six days.

How far ahead does ECMWF forecast?

The main 00 and 12 UTC forecasts and the 51-member ensemble run 15 days ahead. Separate extended-range and seasonal forecasts look weeks to months ahead at lower detail.

Is ECMWF more accurate than GFS?

On average, yes. ECMWF has held roughly a one-day lead in medium-range skill for many years. It does not win every individual forecast, which is why comparing models is more reliable than trusting one.

Keep reading

Overlay ECMWF on the live map All guides