Next-day weather model rankings
HRRR is the best-ranked model on the next-day board. HRRR has a published rank in 206 markets. Those ranks average about 209 scored days. That board ranks tomorrow's high, low, and rain chance together against the National Weather Service. The National Weather Service baseline still did better on those same days. HRRR's average miss on highs is 2.6°F. 5 models have cleared that line. Compared with the National Weather Service.
Tomorrow only. Every model that publishes a next-day forecast can rank here, including the short-range models. A model is listed after 90 days scored in at least 30 markets. The National Weather Service is the baseline, so it is not ranked against itself.
How to read this. A shorter miss is closer to the weather that happened. Beats the NWS means a smaller miss than the National Weather Service on the same days. Behind the NWS means a larger miss. Early read means at least 90 days scored. Proven means 180.
The technical details
Skill is 1 minus this forecast’s mean loss divided by the National Weather Service mean loss on the same days. Loss is the mean absolute error for temperature and the Brier score for rain chance. Zero is a tie. Positive beats the Weather Service. Negative falls behind it.
Days scored are matched days: issue dates with a forecast from this source and from the National Weather Service, and a final observation. A rank starts at 90 matched days. That status is provisional, shown here as Early read. At 180 matched days the status is established, shown here as Proven. Below the first line the row is still collecting and is not a rank.
How sure is a 95% confidence interval: a percentile bootstrap of the score, resampling the matched days 2,000 times. Overlapping intervals are marked tied.
The headline score uses days 1 through 5, with lead weights 0.30, 0.25, 0.20, 0.15, and 0.10. Days 6 and 7 stay out of it. Variable weights are high 0.35, low 0.30, and rain chance 0.35. Observations come from ACIS, with the National Weather Service climate report when ACIS has a hole. Model numbers come from each provider’s public files.
The methodology page is the full reference. How we grade says the same rules in plain English.
Includes archived forecasts back to March 31, 2026.
- 1 HRRR Behind the NWS by 18% 2.6° high error
- 2 ECMWF IFS Behind the NWS by 47% 2.9° high error
- 3 NAM Behind the NWS by 48% 3.0° high error
| Place on this list. 1 was closest to the weather that happened. A tie shares this number, and the next place skips ahead. Early read until the longer day count, then proven. | The forecast source this row scores. | Mean skillBeats the National Weather Service, or falls behind it, averaged across markets with a published rank. 0 is the same, positive is better, negative is worse. The technical name is mean skill. | High errorAverage miss on daily highs, in degrees F. Lower is better. | How many markets this model has a published rank in. A national row needs at least 30. | Matched daysDays scored. How many climate days this number is averaged over. The technical name is matched days. | StatusEarly read means at least 90 days scored. Proven means 180 or more. Still collecting means it is too soon to rank. The technical names are provisional and established. |
|---|---|---|---|---|---|---|
| 1 | HRRR | Behind the NWS by 18% | 206 | 209 | Proven | |
| 2 | ECMWF IFS | Behind the NWS by 47% | 210 | 209 | Proven | |
| 3 | NAM | Behind the NWS by 48% | 207 | 209 | Proven | |
| 4 | GFS | Behind the NWS by 49% | 210 | 210 | Proven | |
| 5 | ECMWF AIFS | Behind the NWS by 69% | 210 | 209 | Proven |
The bar is the average miss on highs. A shorter bar is closer to the high that happened. Beats the NWS means a smaller miss than the National Weather Service on the same days.