How we measure accuracy
Every morning at 06:10 UTC we freeze each provider's forecast for the next 5 days. The next day we compare it with real observations (METAR from the reference airport, public domain). Metrics follow WMO/TD 1023: mean absolute error and bias of daily high/low, share of days within ±2 °C, contingency table for rain yes/no (≥ 0.1 mm or reported phenomenon), wind error. The baseline is persistence ("tomorrow like today"). 30-day scores need 30 days of data; rain scores need 90. No AI touches any value.
Cities
- Forecast accuracy for Moscow
- Forecast accuracy for London
- Forecast accuracy for New York City
- Forecast accuracy for St Petersburg
- Forecast accuracy for Los Angeles
- Forecast accuracy for Berlin
- Forecast accuracy for Madrid
- Forecast accuracy for Kyiv
- Forecast accuracy for Brooklyn
- Forecast accuracy for Chicago
- Forecast accuracy for Rome
- Forecast accuracy for Queens
- Forecast accuracy for Houston
- Forecast accuracy for Paris
- Forecast accuracy for Hamburg
- Forecast accuracy for Bucharest
- Forecast accuracy for Budapest
- Forecast accuracy for Warsaw
- Forecast accuracy for Vienna
- Forecast accuracy for Barcelona
Articles
How to read a 30-day forecastDays 1–7 are a forecast; days 17–30 are a trend. What each one means and how much you can rely on it.What the weather warning colours meanYellow, orange, red: the Meteoalarm system used by Europe’s weather services, explained simply.How we measure forecast accuracyEvery morning we freeze the forecasts; the next day we compare them with real airport observations. The score is on every city page.Feels-like temperature, wind and the UV indexWhy “feels like” differs from the thermometer, what gusts mean and when the UV index matters.