WEATHER PROPSBOARDMATCHUPSMODELSSTATIONSREPLAYTHE LEAGUE
SCOUTING REPORT

Meet the models.

Different forecasts. The same weather. A record you can inspect.

THE SCOREBOARD

Forecast accuracy

ARCHIVE

D1 · same initialization, same station-days

  1. 01
    ICON5,433 shared cases
    0.0235Brier ↓
  2. 02
    IFS archive5,433 shared cases
    0.0286Brier ↓
  3. 03
    GFS5,433 shared cases
    0.0322Brier ↓
  4. 04
    AIFS Single5,433 shared cases
    0.0371Brier ↓

Reconstructed probabilities. Fixed temperature thresholds; lower Brier is better. Model versions and historical release times are not verified.

A FAIRER COMPARISON

Same cases.
Same clock.

Every ranked model is evaluated on exactly the same station-days at 00 UTC initialization, one, three, or five calendar days before the target date.

The five fixed high-temperature thresholds are 32, 50, 70, 90, and 100°F. Their Brier scores are averaged within each station-day, so extra props and extra runs do not earn extra weight.

This is a reconstructed research benchmark with a shared error model. It does not measure native ensemble calibration or establish a live league winner.

5,482 station-days · 49 stations
Updated Oct 19 · 00:44 UTC

THE ROSTER

Know who you’re comparing

Next on the roster

WeatherNext 3 and AIFS ENS will appear as separate products once their feeds, timestamps, and forecast fields have been verified. They do not inherit another model’s record.