How we verify forecasts and calculate our data

Weather forecasts: what we compare

Every day we collect the daily forecasts of five weather services for Italy's 107 provincial capitals. Forecasts are stored before the day they refer to, in some cases several days ahead, and compared with the observed weather once that day is over.

The services we compare are Meteo Aeronautica Militare, 3BMeteo, Apple Weather, Meteo.it and iLMeteo.it.

The observed weather, our reference, comes from Visual Crossing historical weather data for the same 107 capitals. Visual Crossing is only used as the reference and is not part of the ranking.

The accuracy score

For each service and each province we calculate a score from 0 to 100, using the forecasts verified over the last 14 days:

score = 0.25 × temperature + 0.75 × rain (F2)

When there is a winner

• The window moves every day: today's score covers the last 14 verified days, not the whole year.
• If too little rain fell during those 14 days, we do not publish a ranking for that province: the rain comparison would not be meaningful. In that case we show the service that was most accurate on temperature, labelled as such on the map and on the province page.
• On rain, we name a service as the most accurate only if it beats the runner-up by at least 2 points. Below that threshold the lead has not proved stable over time: the map shows the province as a tie, and the province page marks the leader as almost tied with the runner-up. This threshold does not apply to temperature.

Yesterday's report card uses a different mark: it covers a single day and the forecast issued the day before for the provincial capital, and weighs temperature at 50%, rain at 35% (whether it rained and how many millimetres) and sky conditions at 15%.

From provincial capitals to towns

Forecasts and observed data only exist for the 107 provincial capitals. Every town's score, provincial capitals included, is an estimate built from provincial data:

• the town's own province weighs from 90%, for a town near the centre of the province, down to 50%, for a town 40 km or more from the centre;
• the rest comes from the nearest provinces (at most 8), weighted by the inverse square of the distance: the closer a province, the more it counts;
• the centre of a province is the average of its towns' coordinates.

A town's score should therefore be read as the accuracy of forecasts in its area, not as a measurement taken in the town itself. Neighbouring towns often have the same result.

Other weather data have their own sources: current rain comes from the Italian Civil Protection radar (Radar-DPC), updated every 5 minutes; alerts come from the Civil Protection risk bulletin; comparisons with normal temperatures use the 1991-2020 climate normals from Copernicus (ERA5-Land).

Earthquakes

Earthquakes come from the web service of INGV (Italy's National Institute of Geophysics and Volcanology), which we read every 5 minutes; the history starts in 2015. Magnitude, depth and location are those published by INGV and may be revised in the following hours.

Each earthquake is linked to the town closest to the epicentre. If the epicentre is more than 25 km from the nearest town, meaning it is at sea or abroad, we show the area described by INGV.

On town pages, “surroundings” always means epicentres within 30 km of the town centre: the same distance applies to the statistics, the list of earthquakes, the map and the chart over time.

For each town we also show data that do not change from day to day:

• earthquakes felt in the past, from INGV's Italian Macroseismic Database (DBMI15) and historical catalogue (CPTI15);
• the seismic zone according to the Civil Protection classification, updated May 2025;
• the expected ground shaking according to INGV's MPS04 seismic hazard map;
• the nearest seismogenic structure listed in INGV's DISS 3.3 database.

Some values are our own estimates based on INGV data and models: for example how far away an earthquake can be felt, or how often, on average, earthquakes of a given strength occur in an area. They are statistical indications: earthquakes cannot be predicted.

Every earthquake has its own page. We only point search engines to the pages of earthquakes of magnitude 3.5 or higher with an epicentre in Italy, and of magnitude 4.0 or higher anywhere; the others remain available on the site.

Air quality

Measurements come from the monitoring stations of Italy's regional environmental agencies (ARPA) and are published in near real time by the European Environment Agency (EEA). They are provisional, not yet validated. We update them every hour.

• A town's value is an average of the 5 nearest stations within 75 km, weighted by the inverse square of the distance. We prefer background stations, which are less affected by the traffic of a single street.
• Where there are no stations close enough we use the European Copernicus CAMS model, with cells of about 10 km, and say so on the page.
• The index is the EEA European Air Quality Index, with the thresholds updated in 2025: each pollutant gets a band, and the town's index is the worst band.
• The reliability of the value depends on the distance to the nearest station for the main pollutant: high within 10 km, medium within 30 km, low beyond. It drops one level if measurement and model differ by more than 50%.
• Pollen and the UV index are forecasts from the Copernicus CAMS model; pollen levels follow the EAACI clinical thresholds.

Limits

• Weather forecasts are verified at 107 points: every town's score, provincial capitals included, is an estimate for the area in which the town's own province weighs most.
• Observed data also have a margin of error, and 14 days is a short period: the ranking can change from one week to the next.
• Real-time earthquake and air station data are provisional and may be corrected by their sources.