European city data methodology

1 · What this dataset is

Cities of Europe is a source-literal comparison of European cities and functional urban areas built on official statistics. We report exactly what the source publishes — an indicator label, a value, a unit, a reference year and a geography type — and attach that provenance to every figure on every page.

It is a faithful presentation layer over public data, not a model. We do not estimate values for cities the source does not cover, and we do not blend indicators into composite scores.

2 · Sources

Every number on the site comes from one of these source families. Registered sources are documented but not yet built into published data.

How Eurostat produces these numbers

Eurostat's City statistics (the Urban Audit programme) collect comparable indicators for European cities and their functional urban areas from national statistical institutes. Reporting is voluntary for many indicators, so coverage varies by indicator and country. We retrieve the figures directly from Eurostat's JSON-stat dissemination API and never re-key them by hand.

How non-city-grain sources are handled

Eurostat regional and PPP values, EEA city-air-quality areas and Copernicus climate grid cells are useful context, but they are not all city-proper measurements. The pipeline keeps the source-native grain in value_grain and the source geography in value_geo_code, then labels the figure accordingly on the page.

3 · Geography & the grain rule

European city statistics exist at several distinct geographies. We keep them separate and always label which one a figure uses:

• City — the core municipality (Eurostat "city").
• Greater city — an approximation of the wider urban area where the city proper underbounds the real city (e.g. Paris, Madrid, Lisboa).
• Functional urban area (FUA) — the city plus its commuting zone. A different concept from the city; never presented as a city-proper ranking.

The geography spine also carries a NUTS3 region code for every city. We use it to attach regional context — GDP per inhabitant (NUTS3) and life expectancy at birth (NUTS2) from Eurostat regional statistics. These are regional values for the area the city sits in, shown in their own section, labelled with the NUTS code, and never presented as a city-measured figure. Population density and the 65-and-over share are derived from Eurostat population and GISCO area at the city's own grain.

The country locator map

Each city page shows a small country outline with a pulsing marker at the city. The outline is the official Eurostat GISCO country geometry (CNTR_RG, 1:20M generalization, 2024), and the marker reuses the GISCO Urban Audit label point for that city. Both use the same per-country cosine-corrected equirectangular transform. Two staggered rings make the point easier to find; browsers that request reduced motion show only the fixed marker. This is a generalized locator graphic, not a boundary or sovereignty claim. Overseas and non-contiguous territories are cropped so the inset shows the home landmass, and the marker is a labelled city point, not a precise address.

4 · How the data is fetched, stored and processed

We follow the same discipline as our sibling data projects: raw snapshots are evidence, processed files are deterministic build products, and this methodology defines what the site may claim.

• Fetch. Raw source responses are pulled by workflows/data_pipeline.py and stored immutably under data/raw/<authority>/, each with a committed MANIFEST.json recording URL, byte size, SHA-256 and fetch time. Bulk bytes stay out of version control; the manifest is the reproducibility contract.
• Build. Processed CSVs in data/processed/core/ are a deterministic function of those raw snapshots plus checked-in code. Each build writes a build_manifest.json with input hashes, source-update timestamps and row counts, and a field_dictionary.json defining every field.
• Validate. data/sources/update_registry.json declares per-output validators (row counts, required columns, unique keys, null-rate checks, normalized metric contracts and crosswalk contracts) enforced by python3 workflows/data_pipeline.py validate and the test suite before anything ships.
• Render. The site bundles one generated JSON (site/src/data/cities.generated.json) produced by workflows/build_site_data.py; pages render straight from it, so the figure on the page is exactly the figure in the processed data.

How city pages present the data

Each city page compares a metric only with records from the same reference year and the same geography grain. It shows rank, percentile, reporting median and denominator. City and greater-city values never share a comparison set. Ranks are never an unqualified “Nth largest”; the page states how many comparable records reported the field. The narrative summary is generated deterministically from the same values (no AI writing, a banned-phrase check fails the build). The 12-month climate chart uses the ERA5-Land monthly 1991–2020 normals.

Initial publishing wave

The full dataset and the indexable city-page set are separate. The initial release selects exactly 150 profiles through a checked-in publishing cohort. Selection rewards source-family depth, metric coverage, complete climate data, population, capital status, geography confidence and country diversity. Other profiles remain reachable through the explorer and tools but carry noindex,follow and stay out of the sitemap. A source refresh cannot silently add indexable profiles.

Dedicated population and climate pages render their complete source tables in HTML, then add filtering, sorting, pagination and CSV export in the browser. Filter combinations do not become separate indexable pages. Saved random-city draws use the clean tool page as canonical and carry noindex,follow.

Missing values stay missing. We never fill a city from a country or regional average unless the rendered label explicitly says the value is country- or region-grain context.

5 · Coverage & missingness

Reporting is voluntary, so we publish how many cities actually report each indicator rather than implying complete coverage. Counts below are from the current build.

6 · Field recipes

Every published field is source-literal: either a verbatim official indicator, a documented grain-preserving attachment, or a documented derivation from named official inputs. We do not impute missing values or blend fields into composite scores. Each field carries a guardrail describing what it may not be used to claim.

7 · Eurostat observation-status flags

Where Eurostat marks an observation with a status flag, we preserve it next to the value and explain it on the page.

8 · What we do not claim

To keep every page defensible, the project forbids these claims regardless of demand:

• No "best", "most livable" or "safest" city as an objective verdict.
• No "cheapest" city without naming the exact metric, unit and year.
• No full cost-of-living score — our housing and transport fields are named component inputs, not a basket.
• No city cost-of-living claim from country PPP; PPP is country-grain context only.
• No crime-rate claim from perception data; perception fields are respondent shares to a specific survey question.
• No forecast, current condition or legal-compliance claim from EEA annualized air-quality values.
• No mixing of city, greater-city and FUA values inside one comparison.
• No station weather or current conditions from ERA5-Land reanalysis.

9 · Limits & roadmap

Current limits. Population, housing, employment, education, demography, transport, tourism and culture indicators are reference year 2022; the perception survey is 2023; EEA air-quality values are annualized from 2024 monthly city averages. Many indicators are reported by only a subset of cities (see Coverage).

Live now. Eurostat city employment, education and basic demographic indicators; regional GDP and life expectancy joined by NUTS; PPP country price-level context; derived population density and 65-and-over share; ERA5-Land climate normals; and EEA PM2.5, NO2 and O3 annualized city-air values. Registered but gated. OECD functional urban areas and Copernicus Urban Atlas are documented and fetchable, but they do not emit city metric rows until crosswalk or geospatial-overlay validators pass.

Climate (ERA5-Land, 1991–2020). Mean annual, January and July air temperature, mean annual precipitation, and mean surface solar radiation are taken from the Copernicus ERA5-Land monthly-means reanalysis, averaged over 1991–2020 and assigned to each city's nearest 0.1° grid cell by its GISCO centroid. These are reanalysis values for that grid cell — not station observations, not forecasts, and not current weather. Solar is mean surface solar radiation (irradiance), not sunshine hours. 10 m wind is captured in the dataset but not shown on pages, because monthly-mean components yield only a mean-vector speed that understates true wind speed.

10 · Licensing & attribution

• Eurostat — City statistics (Urban Audit). © European Union. Eurostat data are reusable with source acknowledgement (Commission Decision 2011/833/EU). Source ↗
• Eurostat GISCO — Urban Audit 2024 geography. © EuroGeographics for administrative boundaries; Eurostat GISCO distribution terms apply. Source ↗
• Eurostat GISCO — Country boundaries (CNTR_RG). © EuroGeographics for administrative boundaries; Eurostat GISCO distribution terms apply. Source ↗
• Eurostat — Regional statistics (NUTS). © European Union. Eurostat data are reusable with source acknowledgement (Commission Decision 2011/833/EU). Source ↗
• Eurostat — Purchasing power parities. © European Union. Eurostat data are reusable with source acknowledgement (Commission Decision 2011/833/EU). Source ↗
• EEA — Monthly AQ statistics in European cities. European Environment Agency data reuse terms apply; cite EEA as source. Source ↗
• Copernicus ERA5-Land monthly means. Generated using Copernicus Climate Change Service information; ERA5-Land, licence CC-BY-4.0, ECMWF/Copernicus attribution required. Source ↗

This methodology page is kept in step with the source registry and the data pipeline; when a source, field or vintage changes, this page changes in the same update.