Three wind providers, one weather model, around 40 wind stations. Every ten minutes we measure how far each forecast sits from what the stations actually report. Lower is better.
Forecast vs. live wind, right now
All three providers run the same weather model: ICON-D2 from the German Weather Service. The raw data is identical. What differs is everything that comes after.
How often the provider pulls the freshest model run. How the model is downscaled to local terrain. Whether systematic over- or under-prediction is corrected. From which level the “10-metre wind” is interpolated. Sichtflug refreshes eight times a day, Burnair once a day, Open-Meteo regularly in the background. That cadence accounts for a large part of the deviation visible here.
We take each provider's most recent wind forecast and compare it to what the nearest weather station actually measured in the same hour. The difference is averaged across all stations and reported in km/h. Lower numbers mean the forecast was closer to the observation.
Around 40 wind stations in the Alps, picked by a simple rule: less than 200 metres line-of-sight from the nearest launch. At that distance the station and the launch sit inside the same model grid cell, so the forecast at the launch is also the forecast at the sensor. Elevation 400 to 3000 metres, location in or directly adjacent to the DACH region.
For privacy and competitive reasons we publish no station IDs, names, or coordinates. The selection is recomputed automatically every week and the snapshot is archived immutably.
Each cycle, every ten minutes, we pull the temporally closest live reading per station. Stations with a frozen sensor (no value change for two hours), with clear outliers (z-score above 5 in the trailing 24-hour window), or with a station outage are skipped for that cycle.
Live data comes from the same station network you see on the map. There is no pre-selection that favours any model.
Per station, hour, and provider: forecast minus observation. From those differences, the mean deviation across all stations. The live indicator at the top of this page is the current 10-minute cycle. The daily card in the chart is the running mean across the day.
We publish aggregates only, never per-sensor or per-second values. The full daily table is available as CSV further below.
Values under 5 km/h sit inside the shared model and measurement uncertainty. All three providers run the same weather model, and weather stations themselves vary by a few km/h depending on local exposure. Anything in that range is more noise than a clear quality signal.
Values above 5 km/h are a noticeable deviation. There it does start to matter which forecast you read.
A weather station measures wind at 2 to 10 m above the ground, the model computes it at the launch on ridge height. Both numbers are right, they just measure slightly different things. Out in the open they usually agree well, in sheltered sensor positions they can drift apart.
Live values every ten minutes. The daily value for the running day updates each cycle and stabilises towards evening. Station selection runs every Monday morning.
When current data from a provider is missing, we show no number. Better a gap than something stale.
Point forecasts cannot replace what's actually happening in the valley, behind the ridge, or at the venturi. When foehn hits, when a valley wind presses into the slope, when a thunderstorm builds, a station sees that earlier than any model. The forecast is your starting point, not your verdict.
We correct factual errors within 48 hours of notice. The email address sits at the bottom of the page. Please include the date and the observed value.
All sources are publicly accessible. Licences are honoured.