How accurate we are
- Forecast × observation pairs
- 48,732
- Stations
- 22
- Resorts
- 45
Weak spots
- The model overestimates sun (+1.8 h).
- Alps not yet verified against stations.
- Fog during inversions – lower accuracy.
Sunny day
- Hit rate
- Events detected (POD)
- False alarms (FAR)
Data
| Lead time | n | Hit rate | Events detected (POD) | False alarms (FAR) | Mean error (MAE) | Bias forecast − observation | Brier score |
|---|---|---|---|---|---|---|---|
| +1 d | 5,445 | 87% | 80% | 23% | – | – | 0.1 |
| +2 d | 5,445 | 85% | 76% | 27% | – | – | 0.1 |
| +3 d | 5,445 | 83% | 63% | 26% | – | – | 0.1 |
| +5 d | 5,445 | 78% | 42% | 30% | – | – | 0.2 |
| +7 d | 5,445 | 71% | 0% | – | – | – | 0.2 |
Fresh snow
- Hit rate
- Events detected (POD)
- False alarms (FAR)
Data
| Lead time | n | Hit rate | Events detected (POD) | False alarms (FAR) | Mean error (MAE) | Bias forecast − observation | Brier score |
|---|---|---|---|---|---|---|---|
| +1 d | 7,169 | 94% | 64% | 40% | – | – | – |
| +2 d | 7,169 | 94% | 56% | 37% | – | – | – |
| +3 d | 7,169 | 93% | 44% | 38% | – | – | – |
Method
Forecast issued N days ahead vs. station measurements. Hit rate = share of days when forecast and observation agree. Positive bias = the forecast overestimates.