velometer

Measures protected cycling infrastructure in 21 European cities from OpenStreetMap, year by year since 2018, with every rule and number reproducible.

Auf Deutsch: Ergebnisse und Hintergründe für Berlin im Zählwerk Berlin Datalab.

Explore the results · Methodology · Decisions · Datalab (Deutsch)

Note

All lengths are mapped km: what OpenStreetMap contributors have mapped, not what a city has built (ADR 0006). Mapping is more complete in some cities and years than in others, so a gap between two cities can be a gap in the map.

Why

Published cycling rankings disagree, and most of them can't be re-run: some mix figures the cities submit with OSM data and publish no code, others are expert scores. velometer is a pipeline anyone can run. It produces per-city numbers with explicit definitions, and it shows why rankings disagree (definitions, denominators, boundaries) instead of hiding it.

What it does

• Classifies every side of every road in OSM into one of five tiers, from A (separated cycle track) to E (cycle street or bus lane). The rules derive from the ECF's QECIO tag logic and live as data in config/classifier/; a rule change is a new classifier version.
• Measures lengths in each city's projected coordinate system, never in degrees, inside two boundaries: the city proper (LAU) and its functional urban area (FUA).
• Goes back in time with yearly snapshots on 1 January from 2018, cut from OSM's full history, plus a current snapshot. Every result row records its classifier version and OSM timestamp.
• Compares its own numbers with official city figures and published rankings, which stay a separate layer and are never blended in (ADR 0001).
• Publishes a static, bilingual reader site: per-city pages with charts and tier maps, a compare view, stories, sources, and CSV and GeoJSON downloads, each carrying its attribution.

The headline metric is A+B km per 100 km of main roads (separated tracks plus shared foot and cycle paths, against the main road network without motorways; ADR 0003).

_{The compare view: headline metric, city proper, OSM snapshot of 3 October 2026, classifier v4.}

Quick start

You need uv and osmium-tool (apt install osmium-tool).

git clone https://github.com/senzelden/velometer.git && cd velometer
uv sync
uv run velometer init                   # SQLite schema and the 21 cities
uv run velometer run --cities ljubljana # download, clip, classify, store
uv run velometer results                # one row per city

The run takes about a minute, most of it the Slovenia download. results then prints (abridged):

city       osm_date    headline  ecf_qecio  clean_cities  A_km   B_km  C_km  D_km   E_km  main_km
ljubljana  2026-10-27  72.7      41.8       20.7          205.7  36.7  0.0   101.7  1.8   333.2

headline is A+B km per 100 km of main roads; ecf_qecio and clean_cities are the same city measured by those rankings' definitions.

A full run downloads one Geofabrik extract per country. The yearly history needs a free OSM account for Geofabrik's full-history files (velometer auth). uv run velometer --help lists everything else: reconciliation with external figures, time series, trust diagnostics, counters, plan networks and the static export (velometer app export).

Cities and sources

Cities	Amsterdam, Barcelona, Berlin, Bordeaux, Bratislava, Budapest, København, Helsinki, Kraków, Ljubljana, Lyon, Madrid, Oslo, Paris, Praha, Stockholm, Strasbourg, Utrecht, Warszawa, Wien, Wrocław
Infrastructure	OpenStreetMap via Geofabrik extracts, current and full history
Boundaries	Eurostat GISCO, LAU 2024 and FUA
Population	Eurostat, INSEE (France), INE (Spain)
Context	Open bike-counter data (Berlin, Paris, Wien, Helsinki, Madrid, Barcelona, Oslo), Berlin's cycling plan network
Comparison only	Official city figures, Clean Cities 2025, Copenhagenize, ECF QECIO, BikeNetKit; cited with source and date, never redistributed
Basemap	Protomaps extract, served from our own host

How to read the numbers

• Mapped, not built. A rise can be new infrastructure or new mapping. velometer diagnoses mapping effort per country (velometer trust) instead of correcting it away.
• Flagged years. Jumps that an automatic check catches are flagged in every chart, and each flag carries a written explanation.
• Boundaries matter. The city proper and the wider urban area tell different stories; both are computed, the city proper is the default.
• Definitions are presets. "Protected" means different tier sums in different rankings; the presets in config/presets.yaml reproduce them side by side.

The full method, generated from the config and the database and checked for staleness, is in docs/methodology.md. The decisions behind it are the ADRs.

Development

uv run ruff format --check . && uv run ruff check . && uv run pytest \
  && node --test "src/velometer/app/static/js/tests/*.test.mjs"

Network tests are marked slow and deselected by default. Deployment of the static edition is in docs/deploy.md. The images in this README are rendered from the site by tools/readme_shots.py.

Citation

Use GitHub's "Cite this repository" button (CITATION.cff), and name the classifier version and OSM snapshot of the figures you cite: the Cite block under each chart on the site gives both, ready to copy.

• Software, all versions: doi.org/10.5281/zenodo.23945133
• Data, all snapshots: doi.org/10.5281/zenodo.23932324

Licences and attribution

Part	Licence
Code and classifier rules	MIT
Derived data: tier lengths, metrics, CSV and GeoJSON downloads	ODbL 1.0, map data © OpenStreetMap contributors
Administrative boundaries	© EuroGeographics for the administrative boundaries (non-commercial use)
Counter data	Each publisher's licence (dl-de-zero-2.0, ODbL, CC BY 4.0, NLOD), credited on every page that shows it
External comparison figures	Cited with source URL and retrieval date, not relicensed

Each licence, its credit line and where it appears are recorded in ADR 0008.