How our wind forecast works
Our forecast is new and works differently from the usual weather apps: it learns from real measurements at the spot, and instead of a single number it tells you how likely wind is. Here is a short guide to reading it.
What is different?
Weather models compute on a grid several kilometres wide. They only roughly capture mountains, valleys and shores – which are exactly what shape the wind on a lake. For years we have compared the weather models with our measuring stations at the spot. From this a machine-learning model learns how the wind actually arrives at this very spot and corrects the weather models accordingly. Every spot has its own model.
Why probabilities?
Weather is never certain. A single number like “14 knots” pretends a precision that does not exist. So we tell you how often each kind of wind came in such a weather situation. “Windy 70 %” means: on 7 out of 10 comparable days it was windy – on 3 it was not. 70 % is a good chance, but no guarantee. The closer the day, the more reliable the forecast.
The levels
Every day gets a level. We rate the mean wind between 7 am and 7 pm, the same at every spot. One hour is enough.
Calm – below 6 knots.
Light wind – from 6 knots.
Windy – from 11 knots, enough for a session with big gear.
Big Day – from 14 knots.
The labels: which wind
The label tells where the wind comes from on that day. Which labels a spot has and how we recognise them there is explained on the spot's details page.
Thermal – on sunny days mountains and valleys heat up and air flows in. Usually from midday, steady.
Foehn – warm, dry south wind across the Alps, often already in the morning, strong and gusty.
Foehn, then front – the foehn collapses during the day, a front approaches and the wind turns to north. Where we can calculate it, the label shows roughly from when to expect the north wind.
Inversion – cold air, often with fog or low stratus, lies in the foreland while the sun shines on the lake. At Lake Walchensee this often pushes a north wind over the Kesselberg.
Ora and Pelér – at Lake Garda: the Ora is the thermal south wind in the afternoon, the Pelér the north wind in the morning.
West wind and east wind – wind that comes with the large-scale weather pattern (e.g. at Lake Chiemsee).
Cold air – downdraft from thunderstorms in the mountains, short and strong (Lake Kochel).
Front – a front passes through: short, strong wind, then it clouds over.
How to read the forecast
Day tile: the most likely level and its probability. The coloured bar below shows how the chances split across all levels – a day with “Light wind 50 %” can still have a 30 % chance of being windy.
Label (e.g. thermal, foehn, inversion): the wind system we expect for the day (see below). It only appears on days with at least “Light wind”.
Hourly wind: the line is the most likely value, the light band the range in which the wind lies in 8 out of 10 cases. A wide band means the situation is uncertain.
Details: why we forecast this way
The Details tab shows the drivers behind the forecast, for example pressure differences across the Alps, the wind aloft or the sun. The dashed lines are limits we derived from our measurements at the spot: when the forecast reaches the line, there was usually wind in the past. Under “Wind systems at this spot” you can read which winds work well there – and which do not.
Review: how well we did
In the Review tab we compare our forecast with the measurement every day – including the days we got it wrong. So you can see for yourself how reliable the forecast is at your spot. The green model label at the top names the model with its date, station and number of inputs.
How our model is built
The idea is not new. Back in 2018 we built a model that learned from the station's measurements alone how the wind would develop over the next hours. For the next three hours it was better than the well-known weather services. But it had a fundamental limit: it could not see beyond the station, so it missed any weather approaching on a larger scale.
Since 2022 Open-Meteo has published historical forecasts, meaning what the weather models actually predicted in the past. After four years this data set is large enough to train on, and it keeps growing every year.
Today we take the forecast of the weather models as a starting point and correct it statistically towards the measurement at the spot (technical term: MOS). The model also gets the signals that drive the wind: pressure differences, wind aloft, sunshine and the layering of the air. Time is especially important. Thermal wind follows a rhythm: it builds up with the sun, peaks in the afternoon and dies down in the evening. Over the year this changes with the height of the sun and the length of the day.
We learned a lot from Simon Stieber. For Lake Walchensee he worked out under which conditions the thermal wind kicks in. We adopted his drivers: the pressure difference Munich–Innsbruck, the foehn pressure difference Bolzano–Innsbruck, the sunshine and the wind aloft. Unlike a set of rules that only answers whether the thermal kicks in, our model forecasts the wind itself: mean wind and gusts hour by hour, with a range. From this follows how likely each level is, whether the wind comes from a thermal, wind aloft or a front.
Limits today and why we keep improving
At many spots we have measured wind for more than ten years, but historical forecasts only since 2022. For everyday situations like the Ora at Lake Garda or the thermal wind at Lake Walchensee that is enough. Rare weather situations, however, the model has seen too rarely to predict them reliably. At Lake Garda these are, for example, the north foehn (Balin) with strong north wind, often in winter, and the transition periods in spring and autumn.
The data set grows every year, and with it the model also learns the rare situations better. You can help: in the reports you will be able to rate for every spot with a forecast whether it was right (real error, borderline or correct, plus a comment). We use these ratings specifically to improve the difficult days.
Our forecast is a statistical forecast, not an official weather forecast and no substitute for severe weather warnings. Short events such as thunderstorm gusts or single fronts are hard to predict to the hour. Decide on site according to the conditions. The webcam and the live measurement help with that.
Statistical forecast, no warranty – not an official weather forecast and no substitute for severe-weather warnings. Water sports at your own risk; judge the conditions on site. Terms of use