Real-time observations & ML-corrected forecast
No observations yet
Set up Netatmo API
| Time | Temp | Feels | Humidity | Wind | Gusts | Precip | Prob |
|---|
This forecast uses a V4 Hybrid ML approach to debias MeteoSwiss ICON-CH2 predictions: Conv1D+MLP for temperature, humidity, and rain (leveraging 12-hour temporal patterns), and XGBoost with lag features for wind variables. Trained on 8+ months of local data with day-out cross-validation.
Architecture: The hybrid model combines a current-timestep MLP with past-window Conv1D to capture temporal patterns. Wind uses XGBoost with 6-hour lag features and trend indicators. See GitHub for full methodology.