Sources. Weather data comes from public numerical weather-prediction models and from long-term observational records. We do not own meteorological stations and we do not claim to produce forecasts ourselves — we aggregate and interpret model output for a specific city.

A day is four readings, not twenty-four. Each day is described by four readings: night, morning, day and evening. The daytime temperature is the reading of the warmest part of the day, the night-time temperature is the reading of the coldest part. That is why MeteoCalendar never advertises an “hourly forecast”: the underlying data is not hourly.

Forecast vs climate estimate. The near-term part of any outlook (roughly the first two weeks) is a model forecast and is refreshed as new model runs arrive. Dates beyond that — and everything shown as “climate-based” or marked with hatching in the 30-day calendar — are statistical estimates built from the averages of the last 10 years of observations for that city and date. They describe what is typical, not what is certain, and are always labelled as such on the page.

Monthly pages. A monthly page answers the question “what is the weather in this city usually like in this month”. Averages are calculated over a rolling 10-year window, not a 30-year climate normal, and the page says so. Rainy days are days with measurable precipitation; sunny days are days without precipitation and with mostly clear skies.

Missing data. If a value is not available for a city and a date, the page shows nothing rather than a zero or a made-up number. A unit is never displayed without a value.

Updates. Forecast data is refreshed periodically in the background; monthly aggregates are recalculated when the underlying daily records change. A source outage never deletes the last successfully stored version of the data.

Publication. Pages are published in small verified batches: a city or a month appears in navigation, internal links and the sitemap only when its data is complete and its translations are ready.