About this product
AIFS-Single v2.0 is ECMWF's machine-learning forecast, run operationally alongside HRES. It is not a physical model: it has no equations of motion and no parametrised convection, and it produces a fifteen-day global forecast in minutes on a single accelerator where HRES needs a supercomputer for an hour. What it does have is forty years of reanalysis, and on the scores that matter most in the medium range it now beats the physical model it was trained on.
It is drawn in exactly the same colours as HRES, and that is the point. The interesting question about a machine-learning forecast is not whether it is good in the abstract — it is where it and the physical model disagree, and by how much. That comparison only means anything if nothing else on screen changes, so the ecCharts tables, the band breaks, the place picker, the pointer and the timeline here are the same instrument as the HRES explorer. And you no longer have to open the two in separate tabs to do it: Compare & verify in the map options draws HRES over this model at the same valid time, or subtracts one from the other on a diverging scale about zero.
The grid is 0.25° — about 28 km — which is what this model publishes, against 0.1° for HRES. That is an honest difference rather than a compromise: a data-driven model is not claiming valley-scale detail, and its value is in the pattern at day seven rather than the texture at hour six. So this page runs to fifteen days, six-hourly throughout, where the HRES explorer stops at six. The far end of the run is the part of a machine-learning forecast worth looking at, not an afterthought on it.
It is also why a difference map here is read with one caveat stated plainly: the two models are on different grids, so a difference is computed on the finer raster with the coarser one sampled into it. That is exactly right for the synoptic disagreement anyone opens this for, and it is not a claim about anything at 0.1° scale that only one of the two resolves.
The fields are what this feed actually publishes, which is not the same set as HRES. There is no wind gust, no CAPE and no snow depth, so those buttons are simply absent rather than filled with something approximate. In their place are three the physical feed does not carry and a fifteen-day horizon makes worth drawing: 100-metre wind at turbine height, and total precipitation and total snowfall accumulated since the run started — which, over a fortnight, is usually the number being asked for.
As on the HRES page, one raster covers the planet and every one of the 226 countries, twelve continental frames and eight ocean basins is a viewport on a field that is already drawn, so moving from Iceland to New Zealand is instant. The pointer reads a value back out of the picture, and reports the ecCharts band — “15 to 25 mm” — rather than a precision a banded chart never claimed.
Prefer a chart viewer to a free map? The legacy view bolts the camera down: a fixed set of framings, the whole run decoded up front, and loops that step instantly because nothing has to be re-fetched.
Data © ECMWF. Forecast Euro is not affiliated with ECMWF.