History
30 days of forecast vs. what actually happened at the lake. Click a single day to see its details.
Last 30 days
Rule-based forecast (actually used)
ML-based forecast — logistic (experimental)
ML-based forecast — HistGradientBoosting (black-box model, experimental)
Actual measurements (session ≥ 1 h)
Thu Jun 25
Fri Jul 24
session (≥ 1 h ≥ 11 kt)
marginal (≥ 1 h ≥ 8 kt)
no wind
no data
thunderstorm (forecast rows: predicted · actual row: observed)
Jul 19, 2026
avg wind
gust
ignition (8 kt)
session (11 kt)
For Jul 19, 2026
NO GO
700 hPa crossflow 34 kt — valley decoupled
⚡ Storm risk: the thermal often runs well until the front — watch the sky, come in early.
🤖 ML — logistic
experimental
NO GO
GO 17% ·
MAYBE 19% ·
NO GO 65%
Strongest inputs —
(−) yesterday rain, pressure Δ (thermik), daytime cloud.
🤖 ML — HGB · black box
experimental
NO GO
GO 2% ·
MAYBE 7% ·
NO GO 91%
Most distinctive inputs: daytime cloud, Innsbruck pressure, Bolzano pressure.
Experimental shadow models — they never drive the verdict.
How good are they? →
How the day was forecast at the time, before the rules were recalibrated — shown for comparison only.