“The Euro model” is what forecasters call ECMWF, and that is what this site is named after and what it draws — the model widely regarded as the most accurate in the world, rendered worldwide rather than over any one continent. It tries to do that without the things that usually come attached: no account, no tracking, no cookie banner, no advertising, and no proprietary “forecast engine” between you and the numbers the centre actually produced.

It is one of a small network. Forecast Eagle covers North America from NOAA sources; Forecast Canada covers Canada from MSC. Those two are organised by place; this one is organised by model, which is why it has no geographic limit. A separate site, Forecast European, will cover the continent itself.

The front page and /europe draw the centre's home continent on its own map: ETRS89-LAEA (EPSG:3035), the Lambert equal-area projection the European Union uses for its own maps, so every country keeps its true size. Every European country has a page there — its cities hour by hour, its capital from HRES, AIFS and the ensemble on one chart, and a fifteen-day ensemble outlook — built entirely from the same published frames and city values the explorers draw, with nothing computed on the side.

HRES at 0.1°. Most freely available ECMWF imagery is drawn from the 0.25° open-data feed. This site draws the 0.1° dissemination — a 3600 × 1801 global grid, about 11 km, roughly four times the detail. Hourly to 90 hours, three-hourly to six days, then six-hourly out to fifteen.

AIFS-Single v2.0, ECMWF's machine-learning forecast, is live too — at 0.25°, six-hourly to fifteen days. It is drawn in the same colours in the same instrument on purpose: the interesting question about a data-driven model is not whether it looks plausible on its own, but where it and HRES disagree.

The EPS ensemble is live as well, and it answers a different question from either: not what will happen, but how much of that to believe. Fifty-one members reduce to two fields — the mean, which past about day four beats any single run, and the spread between them, which says where the mean is a real forecast and where it is an average of genuinely different weather. Both are drawn, side by side, on the nine upper-air and pressure quantities ECMWF publishes them for.

Google DeepMind's WeatherNext 2 is the fourth model, and the only one that is not ECMWF's: a four-field subset of its ensemble mean at 0.25°, carried because it shares no centre, reanalysis or assumptions with the other three, so where it and ECMWF disagree the disagreement means something. It is an experimental feed, not an operational product, and every page that shows it says so.

Colours are the authentic ecCharts tables — the exact ECMWF Magics named colours in ecCharts palette order, on the published level breaks, shaded as discrete bands rather than a smooth ramp. Reproducing that banding is most of what makes a chart read as ECMWF's rather than as a generic rainbow. Every scale is pinned to absolute values and never re-fit per frame, because a loop whose colours rescale each step is actively misleading.

Not an official source, and not affiliated with ECMWF. Data is © ECMWF; the rendering, the wording and any judgement expressed here are this site's own and should not be read as the centre's.

Not a warning service. For watches, warnings and official guidance, go to your national meteorological service — DWD, Météo-France, the Met Office, AEMET and their counterparts — or to Meteoalarm, which collects them. A model chart is not a warning, however good the model.