About this product
ICON is the Deutscher Wetterdienst's global forecast model, and the run every German forecast is ultimately built on. It solves the non-hydrostatic equations on an icosahedral-triangular mesh — nearly equilateral triangles wrapped round the sphere, which is why it has no pole problem and no converging meridians to special-case. The operational mesh is R03B07, about 13 km, and DWD publishes it interpolated to a regular 0.125° lat-lon grid. That is what is drawn here: roughly twice the detail of the 0.25° open data most public sites draw a global model from.
Colours are the authentic ecCharts tables — ECMWF Magics named colours in ecCharts palette order — and that is deliberate rather than an accident of where this framework came from. Reading a DWD forecast against an ECMWF one is a comparison forecasters actually make, and it means nothing unless the two are drawn on identical band breaks. Where ICON publishes a field ECMWF's tables do not cover, the table is built here and the range it was measured over is recorded beside it.
One raster covers the planet, so every country, continental frame and ocean basin in the picker is a viewport on a field that is already drawn — which is why moving from Bavaria to the Bering Sea is instant. That is affordable because each field is quantised to its own band count rather than to a flat 128 levels: a banded chart only ever claims which band a pixel is in, so the extra levels encoded noise nobody could read.
Steps are hourly to +78 hours, then three-hourly to +180. The hourly section reaches further than any single feed on the ingest host provides, because it is stitched from two of them — DWD's long run out to seven and a half days, and a second hourly fetch that reaches +78. Both are the same model on the same grid, so a frame is identical whichever one served it. The 06Z and 18Z runs stop at +120, which is DWD's own schedule rather than a gap here: the catalog carries what each run actually produced, so the scrubber is shorter on those cycles and says so.
Precipitation and snowfall are published by DWD as totals since the start of the run and are differenced back to per-step here, so every frame states the interval it actually covers — 12 mm in one hour and 12 mm in three are very different forecasts, and the readout says which. Snowfall is the sum of the convective and grid-scale parts, because neither on its own is the forecast.
Compare & verify draws a second model over this one at the same valid time, or subtracts it. ICON against ICON-EU is the comparison worth having: the nest is this run at twice the resolution, in identical colours on identical band breaks, so a swipe or a difference map shows where the extra detail actually changes the forecast rather than where two rendering pipelines do.
The poles need a projection this map cannot give them — Web Mercator is infinite at 85° — so they get their own. The polar view draws the same fields on a stereographic raster, which is where a 13 km global model has the most to say and the fewest observations behind it.
Data © Deutscher Wetterdienst, from opendata.dwd.de under the DWD open-data terms. Forecast Germany is not affiliated with DWD.