About this product
ARPEGE is Météo-France's operational forecast model, and it is not a regional model with a global parent. ARPEGE is one global run on a stretched, tilted grid: its resolution is highest over France — about 5 km — and falls away smoothly to roughly 24 km at the antipode. What you are looking at is the 0.1° European domain Météo-France publishes as open data, which is the interpolation of the part of that grid which is actually fine. This is the model a French forecast is made on.
The frame is the whole domain, edge to edge — 32°W to 42°E, 20°N to 72°N — and not a box around France. French weather arrives from the Atlantic: the depression that decides a Breton forecast is eight hundred miles out to sea the day before, and the ridge that decides a July one sits over the Azores. A map cropped at the Channel cannot show a reader the thing that is about to happen to them. Every place in the picker is a viewport on that one raster, so cropping it would only remove somewhere the same picture already serves for free.
The surface fields are hourly the whole way to +102. Temperature, dewpoint, humidity, pressure, the 10-metre wind, all four cloud decks, gust, precipitation and snowfall: 103 real forecast hours, four times a day. That is the thing this model does that ICON-EU does not — the nest goes three-hourly after +48, so from two days out its timeline steps in threes while this one keeps stepping in ones, all the way to four and a quarter days.
The fields aloft do not, and the scrubber says so. Météo-France thins after +48: the 500 hPa height, both upper temperatures, 700 hPa humidity, the 250 and 850 hPa winds, CAPE and total column water all go three-hourly from +49, and visibility stops at +48 altogether. Nothing here fills those gaps in. Each field's timeline is built from the frames that field actually has, so a shorter scrubber on a field aloft is the feed being reported honestly rather than a hole in this site.
Colours and band breaks are identical to ICON-EU's and to ARPEGE 0.25°'s, on purpose. The question worth asking of two regional models over the same ground is where they disagree — over the Alps, the Pyrenees, the Massif Central, the Rhône valley in a mistral — and that comparison means nothing if anything other than the forecast changes on screen. Use Compare & verify to swipe one over the other, or to subtract them.
Precipitation and snowfall are published as totals since the start of the run and are differenced back to per-step here, so every frame states the interval it covers. Snowfall is the convective and grid-scale parts summed, because neither alone is the forecast. Gust and visibility are per-step extrema rather than instants — the strongest gust and the worst visibility of that hour, which is the reduction that actually closes a runway.
Visibility is here and is on no other model on this site, because no DWD feed publishes it. It is drawn from the hourly minimum and the feed clips at 20 km, so the top of the scale is never reached: this is a field about fog in the Rhône and Garonne valleys and haar in the Channel, not about how clear a clear day is. There is no 2-metre maximum or minimum temperature on this domain, because Météo-France publishes no period extremum of it in the 0.1° European download — but the same run's 0.25° global output does carry both, over a three-hour window.
The domain stops at 32°W, and ARPEGE 0.25° is where to look past it. That page is the same forecast on the global grid — measured, the two agree over France to 0.02 K in mean 2-metre temperature — so it is not a second opinion, it is this one with the boundary removed. Use it for the Atlantic beyond the frame, for anything the far side of the Urals, and for the two poles, which no map on this page can draw at any size.
Data © Météo-France, from their open-data service under the Licence Ouverte / Etalab. Forecast France is not affiliated with Météo-France, and is not a warning service — for warnings see Vigilance Météo-France.