Verified blend accuracy
Does blending weather models beat the best one?
Measured against observed snowfall, SnowSure's blended forecast has 8.9% lower error than JMA — the single most accurate weather model — across 43,647 verified forecasts.
Ground truth: measured (SNOTEL + resort-reported), snow days only. Machine-readable: /api/v1/blend-accuracy.
Forecast Trust — Verified, not guessed — corpus scale, blend vs best model, and the full record.
Blend vs best single model, by horizon
| Horizon | Blend MAE | Best model | Best-model MAE | Improvement | Samples |
|---|---|---|---|---|---|
| Overall | 5.38 cm | JMA | 5.69 cm | +5.5% | n = 113,574 · as of 2026-07-28 |
| 1–3 days out | 4.89 cm | JMA | 5.37 cm | +8.9% | n = 43,647 · as of 2026-07-28 |
| 4–7 days out | 5.51 cm | JMA | 5.78 cm | +4.7% | n = 58,156 · as of 2026-07-28 |
| 8–14 days out | 6.51 cm | ECMWF | 6.76 cm | +3.8% | n = 11,771 · as of 2026-07-28 |
Lower MAE = closer to observed snowfall. Improvement = how much lower the blend's error is than the best single model. Updated 2026-09-04.
Every model, ranked by error
| Model | MAE (cm) | Samples |
|---|---|---|
| JMA | 5.37 | 43,645 |
| ECMWF | 5.46 | 5,903 |
| ICON | 5.5 | 43,646 |
| Météo-France | 5.63 | 43,644 |
| GEM | 5.73 | 43,645 |
| GFS | 5.91 | 43,647 |
See also: most accurate model by region · Forecast Trust.
