Which weather model is most accurate in your city?
Every major model's forecast for each city, scored once the day is over; most days, in most places, it's close.
How to read the leaderboards
The Spread — the forecast spread — archives what every major model predicted, then scores those forecasts against what actually happened. Where no model leads by 0.22°C or more, the page says so and shows them all.
Each row is one city. The dots are every model's typical miss at the day picked above the verdict, one to seven days out, on one scale for every city and every day, so the field drifts right as the days go out. The verdict names a leader only where its margin is more than the noise, and otherwise gives the gap between the top two. Pick a city for its own scoreboard; how the scoring works is its own section further down the page.
Who has been right · typical miss one day out · last 30 days
In 11 of 30 cities one model is clearly ahead.
Five different models lead those 11, by 0.25 to 0.59°C. In the other 19, the top model is too close to call, but we show you how all the models perform relative to each other.
Top, or within 0.22°C of the top · cities of 30
- GFS18
- ICON16
- GEM9
- UKMO6
- AIFS5
- ECMWF5
- AIGFS4
Who has been right · typical miss two days out · last 30 days
In 12 of 30 cities one model is clearly ahead.
Five different models lead those 12, by 0.23 to 0.67°C. In the other 18, the top model is too close to call, but we show you how all the models perform relative to each other.
Top, or within 0.22°C of the top · cities of 30
- GFS18
- ICON17
- AIFS9
- AIGFS6
- ECMWF6
- GEM6
- UKMO5
Who has been right · typical miss three days out · last 30 days
In 11 of 30 cities one model is clearly ahead.
Five different models lead those 11, by 0.23 to 0.44°C. In the other 19, the top model is too close to call, but we show you how all the models perform relative to each other.
Top, or within 0.22°C of the top · cities of 30
- GFS19
- ICON15
- AIFS10
- AIGFS6
- ECMWF6
- UKMO5
- GEM3
Who has been right · typical miss four days out · last 30 days
In 6 of 30 cities one model is clearly ahead.
Three different models lead those six, by 0.23 to 0.58°C. In the other 24, the top model is too close to call, but we show you how all the models perform relative to each other.
Top, or within 0.22°C of the top · cities of 30
- GFS20
- AIFS15
- ICON12
- ECMWF8
- UKMO8
- AIGFS7
- GEM2
Who has been right · typical miss five days out · last 30 days
In 11 of 30 cities one model is clearly ahead.
Five different models lead those 11, by 0.24 to 0.73°C. In the other 19, the top model is too close to call, but we show you how all the models perform relative to each other.
Top, or within 0.22°C of the top · cities of 30
- AIFS14
- GFS14
- AIGFS9
- ECMWF8
- ICON7
- UKMO5
- GEM2
Who has been right · typical miss six days out · last 30 days
In 6 of 30 cities one model is clearly ahead.
Four different models lead those six, by 0.23 to 0.83°C. In the other 24, the top model is too close to call, but we show you how all the models perform relative to each other.
Top, or within 0.22°C of the top · cities of 30
- AIFS22
- AIGFS16
- GFS15
- ICON10
- GEM9
- UKMO4
- ECMWF2
Six models are scored seven days out; UKMO is not.
Who has been right · typical miss seven days out · last 30 days
In 8 of 30 cities one model is clearly ahead.
Three different models lead those eight, by 0.28 to 0.62°C. In the other 22, the top model is too close to call, but we show you how all the models perform relative to each other.
Top, or within 0.22°C of the top · cities of 30
- AIFS17
- AIGFS14
- GEM13
- GFS8
- ECMWF7
- ICON2 of 3
| City | The race | Verdict | Spread today |
|---|---|---|---|
| Atlanta, Georgia | AIFS & GEM within 0.14°C | — | |
| Berlin, State of Berlin | AIGFS & GFS within 0.11°C | today's spread 2°C | |
| Boston, Massachusetts | AIGFS & GFS within 0.05°C | — | |
| Buffalo, New York | GFS & AIFS within 0.14°C | — | |
| Charlotte, North Carolina | AIFS by 0.49°C over AIGFS | — | |
| Chicago, Illinois | AIGFS by 0.24°C over AIFS | — | |
| Dallas, Texas | GFS & AIGFS within 0.22°C | — | |
| Denver, Colorado | GEM & GFS within 0.08°C | today's spread 2°C | |
| Detroit, Michigan | GFS & AIFS within 0.1°C | — | |
| Houston, Texas | ICON & AIFS within 0.02°C | — | |
| London, England | ICON & GEM within 0.05°C | today's spread 1°C | |
| Los Angeles, California | GEM by 0.48°C over ICON | today's spread 2°C | |
| Miami, Florida | GFS & AIGFS within 0.16°C | — | |
| Milan, Lombardy | ICON & UKMO within 0.06°C | today's spread 2°C | |
| Minneapolis, Minnesota | AIFS by 0.83°C over ECMWF | — | |
| New Orleans, Louisiana | AIGFS & ICON within 0.1°C | — | |
| New York, New York | AIFS & GFS within 0.08°C | — | |
| Oklahoma City, Oklahoma | AIGFS & AIFS within 0.11°C | today's spread 2°C | |
| Orlando, Florida | AIFS & AIGFS level | today's spread 3°C | |
| Paris, Île-de-France Region | GEM & UKMO within 0.08°C | today's spread 1°C | |
| Philadelphia, Pennsylvania | AIFS by 0.47°C over GFS | — | |
| Phoenix, Arizona | GFS by 0.23°C over ECMWF | today's spread 2°C | |
| Salt Lake City, Utah | AIFS & ICON within 0.11°C | today's spread 2°C | |
| San Antonio, Texas | AIFS & ECMWF within 0.12°C | — | |
| San Diego, California | ICON & GEM within 0.02°C | today's spread 2°C | |
| San Francisco, California | AIGFS & AIFS within 0.04°C | today's spread 1°C | |
| Seattle, Washington | AIFS & UKMO within 0.1°C | today's spread 3°C | |
| St. Louis, Missouri | GFS & AIFS within 0.1°C | — | |
| Tampa, Florida | GFS & AIGFS within 0.17°C | — | |
| Washington, District of Columbia | AIFS & GFS within 0.21°C | — |
What one scored day looks like
The most recent day we scored, whichever way it went.
14 days out, 4 models put Oklahoma City's Sun Nov 1 high anywhere from 17° to 26°; by the final call, the day before, they had settled to a median of 22°. It hit 23°. ICON landed closest, off 0.2°.
- 14d
- 13d
- 12d
- 11d
- 10d
- 9d
- 8d
- 7d
- 6d
- 5d
- 4d
- 3d
- 2d
- 1d
11° 29°
23° actual
Each bar is the full range of the models still forecasting that day, in °C; the rule is what the OKC observations recorded. This one opened 9° wide and closed to 3°. Across every day we have scored, the range narrows even though more models arrive as the day approaches: a median 11.6° at 13 days out with four models, 5.8° on the day itself with seven. See how this day verified →
More scored days
The next most recent days we scored. Each bar is the models' opening range; the rule is what happened.
-
Washington · Sun Nov 1
4 models opened 19–28° at 14 days out; it hit 19°; GEM closest, off 0.4°.
-
St. Louis · Sun Nov 1
4 models opened 18–25° at 14 days out; it hit 18°; ECMWF closest, off 1.1°.
-
Seattle · Sun Nov 1
4 models opened 7–11° at 14 days out; it hit 9°; AIGFS closest, off 1.2°.
-
San Francisco · Sun Nov 1
4 models opened 15–20° at 14 days out; it hit 17°; AIFS closest, off 0.1°.
How the scoring works
How the scoring works
For every day, each model's forecast from one to seven days ahead is the last one The Spread archived on that day, the forecast the model was showing at the time (nothing is reconstructed after the fact). Each is verified against the city's own weather station once the day has happened; a day the station did not report is left out rather than filled in. Every model is compared on the same days. Each city's own page scores the same numbers. Bias is the mean signed error of the predicted daily high — positive means the model ran warm. Where a station is the truth, a model's high is the warmest of its hourly values and the station's is the warmest it got at any moment, so a model's high reads a little cool, and its low a little warm, even on a day it gets every hour right. ECMWF IFS gives a value only every 3 hours (every 6 past day 6), and ECMWF AIFS (AI) and NOAA AIGFS (AI) only every 6, with the hours between filled in, so their highs read cooler still and their lows warmer. In the forecasts scored here GFS also steps to every 3 hours from day 5, ICON and GEM from day 4, and UKMO from day 3, every 6 by day 6. The typical miss (MAE) is the average absolute error. Cities join these pages as they join The Spread's archive.
Each model's forecasts one to seven days ahead, as we archived them at the time, scored against each city's weather station over the last 30 days.
About the yardstick. Every city here is pinned to its own weather station, and models are verified against that thermometer's observed highs outright; each city page names its station. A day the station did not report is left out, never filled in from the ECMWF IFS analysis, which is ECMWF's own operational product: two of the seven models scored here are ECMWF's, and scoring them against it would be scoring them against themselves.
The Spread itself is in closed beta. These verification pages and the Updates log are public. If you want in, ask for an invite: the forecast, the Model Lab and Verification for any place you choose, and it's free during the beta.