Was the forecast right for San Francisco on Thu Sep 3?
San Francisco, California · daily high temperature in °C · forecasts from each model's archived runs, verified against the SFOC1 observations
12 days out, 4 models put San Francisco's Thu Sep 3 high anywhere from 21° to 25°; by the final call the field had settled to a median of 20°. The day verified at 22° against the SFOC1 observations, with 0.8 mm of precipitation — drizzle.
GFS landed closest, off by 0.1°. ICON ran 3° cold.
What every model said, at every lead
Daily-high calls for Thu Sep 3 as each model's runs approached it — the last archived run per day. NBM and The Spread's own blend are consensus products and stay out of these ranges and rankings; the San Francisco scoreboard shows them, labelled.
| Lead | AIFS | AIGFS | ECMWF | GEM | GFS | ICON | UKMO | Range |
|---|---|---|---|---|---|---|---|---|
| 12d · 2026-08-22 | 23° | 21° | 25° | — | 21° | — | — | 21–25° |
| 11d · 2026-08-23 | 23° | 23° | 19° | — | 24° | — | — | 19–24° |
| 10d · 2026-08-24 | 20° | 24° | 18° | — | 20° | — | — | 18–24° |
| 9d · 2026-08-25 | 24° | 24° | 21° | — | 24° | — | — | 21–24° |
| 8d · 2026-08-26 | 20° | 27° | 18° | 19° | 25° | — | — | 18–27° |
| 7d · 2026-08-27 | 20° | 26° | 18° | 21° | 22° | — | — | 18–26° |
| 6d · 2026-08-28 | 21° | 22° | 18° | 18° | 21° | 18° | — | 18–22° |
| 5d · 2026-08-29 | 19° | 22° | 21° | 19° | 21° | 19° | 20° | 19–22° |
| 4d · 2026-08-30 | 20° | 22° | 20° | 20° | 22° | 18° | 21° | 18–22° |
| 3d · 2026-08-31 | 19° | 21° | 19° | 19° | 22° | 17° | 20° | 17–22° |
| 2d · 2026-09-01 | 18° | 19° | 20° | 20° | 18° | 18° | 20° | 18–20° |
| 1d · 2026-09-02 | 19° | 20° | 21° | 20° | 20° | 18° | 19° | 18–21° |
| 0d · 2026-09-03 | 20° | 21° | 20° | 20° | 22° | 18° | 21° | 18–22° |
| actual | 22° |
Read it honestly
This is one date, not a league table. A model can win a day it deserved to lose; rankings only mean something over a window. For that, see which model has been most accurate in San Francisco — or every city The Spread scores.
Forecasts are what each model's archived runs said at the time, captured at The Spread's sweep times — never reconstructed after the fact. Truth is the SFOC1 observations. Archive depth varies by city; receipts exist only for dates the archive watched arrive.