Two-year national analysis
U.S. temperature forecast accuracy, 2024–2026
We compared archived raw National Blend of Models temperature forecasts with observed daily highs and lows for 146 station records represented by 150 U.S. city pages. The results show how error changes by location, season, and forecast day.
What the national pattern shows
Location matters. The station-record typical miss ranged from 1.8°F near Miami to 4.7°F near Jackson.
Lead time matters. The represented-station average grew from 2.4°F on forecast day 1 to 4.3°F on day 7.
These are model results. The rankings describe raw NBM temperature guidance across multiple model eras. They do not rank weather apps or every official NWS forecast.
Smaller typical errors
The ten distinct station records with the lowest combined error across days 1, 3, 5, and 7, labeled by a city they serve.
Larger typical errors
The ten distinct station records with the highest combined error in the same archive, labeled by a city they serve.
How to use these results
The report is best used to understand the normal scale of temperature forecast error and how quickly precision changes farther ahead. A smaller average error does not mean every individual forecast will be close, and a city represented by an airport station can have neighborhood-scale differences.
Read the full verification and calibration methodology or see the full seven-day accuracy distribution.
Valid dates 2024-08-02 through 2026-08-05; dataset generated 2026-08-12. Rankings use each verification station once so nearby city pages that share a station cannot alter the order. The CSV retains every city page and identifies its station.