Every forecast on this site gets graded. When a forecast is made, by us or by the National Weather Service, it is written down and locked. When the hour it predicted arrives, the station measures what actually happened. The difference between the two is that forecast's miss.
The number beside each source below is its average miss over the last 30 days. A score of 2 °F on overnight lows means that source has typically been off by about two degrees, one way or the other. Smaller is better.
One contender needs introducing: persistence. It is the forecast that says tomorrow will be the same as today. It requires no model, no data, and no skill, which makes it the honesty test. Any forecast worth running has to beat it. When ours doesn't, this page says so.
Probability of precipitation
Last 30 days · 704 scored forecasts across 1 source
Bias and sample counts
| Source | n | MAE | Bias |
|---|---|---|---|
| Our Zambretti | 704 | 0.421 | +0.421 |
Bias is signed: positive means that source runs too wet on average. A consistent bias is worth knowing because it can simply be subtracted out.
Overnight minimum temperature
Last 30 days · 1461 scored forecasts across 3 sources
Bias and sample counts
| Source | n | MAE (°F) | Bias (°F) |
|---|---|---|---|
| National Weather Service | 696 | 3.40 | +3.18 |
| Persistence (same as yesterday) | 709 | 3.99 | +2.18 |
| Our dew-point model | 56 | 4.98 | -4.31 |
Bias is signed: positive means that source runs too warm on average. A consistent bias is worth knowing because it can simply be subtracted out.