Sunshine Atlas: monthly sunshine scores and climate normals for 3,833 airport-served destinations

Monthly 0–100 Sunshine Scores, average day/night temperatures, rainfall and sea-surface temperature for 3,833 destinations worldwide — every destination served by its own airport (one primary airport per metro area). Derived from long-term climate normals, not forecasts, by Sunshine Atlas. Files: sunshine-atlas-destinations.csv — one row per destination (identity, coordinates, elevation, population, climate band, destination type, annual sunshine hours, annual rainfall mm, best month, year + 12 monthly Sunshine Scores); sunshine-atlas-monthly-climate.csv — one row per destination-month (45,996 rows: score, day high °C, night low °C, rain mm, sea temperature °C); sunshine-atlas-destinations.json — both combined, with per-destination page URLs. Join key: iata. The Sunshine Score combines daytime warmth (ideal 20–32 °C), dryness and annual sunshine hours into one 0–100 number per destination-month: methodology. The canonical, always-current copy is sunshineatlas.com/data/; this record is the 2026.08 edition. Also on Hugging Face and Kaggle, and queryable live by AI assistants via MCP. Changed in 2026.08: temperature, rainfall and sunshine now derive from NASA POWER's January 2001 – December 2020 climatology throughout. Earlier editions anchored temperatures to, and modelled sunshine from, CRU CL v2.0 — a 1961–1990 normal — producing a hybrid series with no single reference period. Temperatures are now corrected for the difference between a grid cell's mean elevation and the destination's own using the standard 6.5 °C/km lapse rate; sunshine hours are modelled from POWER's all-sky clearness index via the Ångström–Prescott relation and each month's true astronomical daylength. The rainy_days_per_year column is replaced by annual_rainfall_mm: POWER reports precipitation depth, not wet-day frequency, and that column was not worth guessing. Provenance: every value is either a measurement from the sources below or computed from them by a formula documented on the methodology page. Nothing in this dataset is written, estimated or filled in by a language model. Three caveats: annual_sunshine_hours is modelled, not observed — validated against 404 published national met-service figures at r = 0.92, MAE 212 h/yr, bias +22 h/yr, and it reads high where persistent low marine cloud or dust sits under the satellite's view (Lima and Cape Verde ~35–40% over) and low in the overcast Sichuan basin, a known limitation of satellite-derived sunshine (Kothe et al. 2017, Remote Sensing 9(5), 429); sea_temp_c is a single year (2024), not a normal; climate and destination_type are rule-based classifications and sunshine_score is an opinionated index, not measurements. Also corrected in 2026.08: the continent column. It previously came from Wikidata per-airport, was absent for 25% of destinations and silently defaulted to Asia — Fiji, Greenland, the Bahamas and 157 Canadian airports all shipped as "Asia" — and was unreliable where present (Bilbao and Girona as Africa, Chaniá on Crete as Asia, Novosibirsk as Europe). It is now derived from OurAirports' country-to-continent table plus explicit geographic rules for the six transcontinental countries (Urals for Russia, Bosphorus for Turkey, Ural river for Kazakhstan, Canaries and Melilla to Africa, Sinai to Asia, Hawaii to Oceania). Upstream sources: temperature, rainfall and sunshine from NASA POWER climatology, 2001–2020 (CC BY 4.0); place data © GeoNames & OurAirports (CC BY); elevations from the Copernicus DEM via Open-Meteo; sea temperatures via Open-Meteo Marine (CC BY 4.0). Every upstream source permits commercial use and redistribution.