Public API

• GET/api/export/citta.json
  All cities, one row each: 1961–1990 and 1991–2020 normals, warming, per-decade trend with its 95% confidence interval, R².
• GET/api/export/citta.csv
  The same aggregates, as CSV.
• GET/api/export/citta/roma
  One city in full: yearly series, anomalies, decade means, monthly climatology, the per-(year, month) series behind /mese, records and trend. This is the endpoint for recomputing everything from scratch.
• GET/api/export/mese.json
  The month just ended: national anomaly, its rank among every same-calendar month since 1940, the full series and a city-by-city breakdown.
• GET/api/export/mari.json
  The daily sea surface temperature archive (mean, max, min) for six open-water points, since 24 November 2022.
• GET/api/export/mari.csv
  The same sea archive, one row per sea per day.
• GET/api/openapi.json
  The OpenAPI 3.1 specification for everything above.

Copy and paste: no authentication, no special headers.

curl https://italiarovente.app/api/export/citta.json
curl https://italiarovente.app/api/export/citta/roma
curl https://italiarovente.app/api/export/mese.json
curl https://italiarovente.app/api/export/mari.csv -o mari.csv

Every response carries dataset_generated_at (when the snapshot was recomputed, not the time of your request) and dataset_sha256, the fingerprint of the per-city aggregates. Re-run aggregate() on the same raw Open-Meteo data and you must get exactly these numbers and this fingerprint. If they disagree, one of us is wrong — and we want to know. Current fingerprint: a678395fd9419be45878f352342afc07dc2580b21e3af426ae565c7af32f39db.

• Warming = the 1991–2020 normal minus the 1961–1990 normal (two WMO thirty-year periods). Never a recent year against the past.
• Trend = ordinary least squares on the annual means of complete years only, with a 95% confidence interval on the slope.
• Monthly anomaly = deviation from the 1961–1990 normal of that same calendar month, not from the whole-series average.
• The seas have no trend. The series begins in November 2022: three years measure interannual noise, not climate. We publish no slope, and neither should you.

• Days missing upstream are not interpolated: the sea archive genuinely lacks 29 January – 11 February 2025, and keeps lacking it.
• Absolute daily records ignore the last 120 days: recent ERA5 (ERA5T) is preliminary and may be revised.
• Snapshot coverage grows build after build (Open-Meteo rate limits): the national monthly figure always states how many cities back it.
• This is not an official weather service. Informational use, not for critical decisions.

Endpoints under /api/export/ are explicitly allowed in robots.txt. The model-facing index is llms.txt, the machine-readable spec is openapi.json. Everything else under /api/ is disallowed for crawlers: it writes, it costs money, or it is the internal cron.

Free to use, commercially too, with attribution to Italia Rovente (italiarovente.app)and the original source (ERA5 — ECMWF/Copernicus C3S, via Open-Meteo). No rate limit beyond ordinary CDN fair use. If you build something with this data, we'd love to hear about it.