70% came within 3°F
Local temperature forecast performance
How accurate are temperature forecasts in San Francisco, California?
Daytime highs were the harder part of the next-day temperature forecast in San Francisco. Highs missed by 2.5°F on average, compared with 1.9°F for nighttime lows.
Opens ZIP 94102 using the nearest current station, which may differ from the archive station below.
What stands out
San Francisco forecast findings
- The strongest monthly directional tendency appeared in November lows, when forecasts averaged 2.3°F cooler than observed.
- By forecast day 7, the typical miss increased to 3.7°F for highs and 2.7°F for lows, showing how precision changed farther ahead.
- On forecast day 1, June lows had the smallest typical miss (1.2°F), while December highs had the largest (4.3°F).
- Across days 1, 3, 5, and 7, San Francisco's typical error was 0.8°F lower than the national archive average.
At a glance
Forecast day 1 accuracy
Highs Lows
San Francisco: 2.7°F typical miss National archive average: 3.5°F
Combined high and low results across forecast days 1, 3, 5, and 7.86% came within 3°F
Error generally grows farther into the forecast
The dots show the typical miss on each forecast day.
See within-3°F percentages and window timing
High and low windows represent the same numbered forecast day, but the high window ends 12 hours later and therefore has 12 more hours of lead time.
How forecasts change through the year
Each tile is forecast day 1’s typical miss.
High temperatures
Low temperatures
Based on archived raw NBM forecasts compared with observed daily highs and lows near San Francisco. Results describe this model archive, not every forecast provider.
How these numbers are calculated
The forecasts are raw NBM median temperature guidance initialized at 12Z. Results include multiple NBM model eras and are retrospective, so they describe this archive rather than guaranteeing future performance.
“Typical miss” is mean absolute error: the average size of the error regardless of direction. Warmer or cooler tendency is signed bias, which can be small even when individual errors are larger.