How accurate the fog forecast is
A fog chance of 20% is accurate if fog happens on about 20 of 100 mornings with that chance. This was checked for the forecast on 9,794 mornings since January 2024, in 22 cities in Europe, the United States and New Zealand.
| Mornings | Fog chance the forecast gave | Fog happened |
|---|---|---|
| 6,453 | about 1% | 1% |
| 1,383 | about 9% | 10% |
| 821 | about 22% | 24% |
| 712 | about 39% | 41% |
| 353 | about 58% | 52% |
| 72 | about 76% | 65% |
Up to a chance of about 40%, fog happened as often as the forecast said, within two percentage points. Fog is rare, so most mornings are in the first row: a chance near zero, and almost no fog measured.
Above 50% the forecast was too high: mornings with a chance of about 76% had fog 65% of the time.
How the forecast is made
The forecast is a formula (a logistic regression). Its inputs are the weather model's data for the morning, such as humidity and wind, and its result is a chance of fog. The formula was fitted on two and a half winters of forecasts from the ICON weather model, put next to the visibility measured at weather stations, mostly airports. Weather services use the same method for local forecasts, under the name model output statistics. The method and all test results are on GitHub.
The formula was fitted at airports, but the forecast on a city page is for the city centre. The weather model has humidity, wind and the rest for every point on the map, so for a city page the formula gets the numbers for the centre in place of those for the airport. The assumption is that the weather that makes fog at the airport makes fog in the centre too, a few kilometres away.
How it was tested
The test was done at the stations, because visibility is measured only there. The station is written on each city's page. For each morning, the forecast made the day before for the station was compared with the visibility measured there.
To keep the test fair, each winter was tested with a formula fitted on the other winters only, so every morning of the test was new to its formula.
A morning was counted as foggy when, in at least two of the six hours from 4 to 10 h, visibility was under 1 km with no rain, snow or drizzle. The mornings are from October to March, and from April to September in New Zealand.
A second test, against airport forecasts
In a second test the forecast was compared with the professional fog forecasts that airports have. Several times a day a forecaster writes down the visibility expected at the airport over the next day (a TAF). The forecast of this site was compared with them at the six US airports on the site, on 2,584 mornings, 243 of them with fog. Both were given the same number of fog warnings.
A day ahead the forecast of this site was a little ahead: it caught 90 of the fog mornings, and the airport forecast issued about 21 hours before caught 81. Airport forecasts are written for flight planning and renewed every six hours, and they are cautious about fog at their far end. The next airport forecast, issued about 15 hours before with six more hours of information, did about as well as the forecast of this site: it caught 116 of the fog mornings, and the forecast of this site 112.
For the same comparison in Europe, airport forecasts are saved daily since October 2026.
Which fog it covers
The forecast is made for fog that forms overnight in calm air. Fog that comes in from the sea is a different kind, so coastal cities such as Los Angeles and San Francisco have no page.
Delhi, Lahore and Amritsar
These three cities have a formula of their own and were tested separately. Fog on that plain comes in runs of days: after a foggy morning, the next morning had fog about three times in four. So their formula has one more input, whether the morning before had fog.
The lowest and the highest chances are accurate in all three cities: a chance under 5% or over 70% can be taken as it is. In between the chance is too high in Delhi and too low in Amritsar: mornings with a chance of about 60% had fog 43% of the time in Delhi, 58% in Lahore and 80% in Amritsar.