Open data

A city open-data programme is the routine publication of public records in machine-readable form, under a licence that permits reuse, maintained on a schedule someone is accountable for. A portal is the visible part. The licence, the update cycle and the named owner are what decide whether anyone can build on it.

This guide connects the plain-language open data definition with the documented records in the atlas: more than 75 entries across over 40 cities in more than 25 countries, on six continents. They range from an air-quality series Berlin has published since 1975 to roadworks feeds launched in 2026 — which makes this one of the few smart-city fields with enough history to judge what lasts.

The portal wave, and what came after

Municipal portals arrived in a tight cluster. London and Helsinki's transit API opened in 2010; Berlin, Paris, Vienna, Helsinki Region Infoshare and Barcelona followed in 2011. Nearly all of them are still running, which is unusual in this atlas — most smart-city genres show a much higher failure rate.

What changed is the ambition. The second decade produced narrower, sharper instruments: an open standard written by a city rather than bought from a vendor (Los Angeles' Mobility Data Specification, now used by well over 100 cities and the basis on which Denver regulates its scooter operators), and transparency registers that publish what the administration itself runs — Amsterdam's algorithm register, its public-space sensor register, London's privacy register.

"Open" is not one thing

The word covers arrangements that differ fundamentally in what a reuser may actually do. Three documented examples mark the range:

All three are legitimate policy choices and all three get called open data. Only the first lets a developer, a journalist or a rival city do anything unsupervised with the material. When comparing programmes, read the licence before the dataset count.

What the evidence actually shows

The most-cited economic case is Transport for London's: a 2017 Deloitte study put the value of the open-data policy at up to £130m a year against publishing costs of roughly £1m a year, alongside 13,000+ registered developers and apps used by 42% of Londoners. It is a strong figure with an honest caveat — TfL commissioned the study itself.

That caveat generalises. Of the open-data entries in the atlas that carry an evidence grade, the large majority are graded official, meaning the publishing body's own account; fewer than ten are independent or peer-reviewed. Portals are unusually good at reporting inputs (datasets, downloads, API calls) and unusually quiet about outcomes. A city planning a programme should decide in advance which of the two it intends to be judged on.

Where it went wrong

Two failures in the atlas are worth more than a dozen success stories, because they failed for structural reasons rather than bad luck.

Copenhagen's City Data Exchange (2016–2018) tried to make urban data a market: a platform where public and private parties would buy and sell traffic, parking, weather and energy data. It could not attract enough sellers or buyers and the partners closed it, citing weak financial sustainability. Publishing data as infrastructure has survived for fifteen years; selling it has repeatedly not.

Decide Madrid shows the other failure mode — political rather than technical. Its open-source engine was reused by 250+ institutions worldwide, but a change of administration in 2019 pared the platform back, discarded approved proposals and halved its budget. The software was never the fragile part.

Data the city did not collect

Some of the best-evidenced entries were not published by an administration at all. Ramani Huria trained students and Red Cross volunteers to map roughly half of Dar es Salaam — some 450,000 buildings — into OpenStreetMap; the dataset outlived its funding and fed into national statistics and disaster planning. Stuttgart's Sensor.Community grew from a local grievance about particulate matter into one of the world's largest citizen monitoring networks, and its cheap sensors were independently tested by the state environment agency against a certified reference analyser. Nairobi's Digital Matatus mapped an informal minibus network no official dataset described.

Questions to ask before launching a portal

• Which licence? If reuse needs an application, say so plainly rather than calling it open.
• Who owns each dataset's refresh? A named team and an interval, per dataset — not a portal-wide promise.
• What is the first real reuse? One committed consumer beats a hundred untouched datasets.
• Publishing or selling? The atlas has no documented case of a data marketplace surviving; it has many of publishing surviving.
• Which outcome will be measured? Downloads are activity. Decisions changed, services built, costs avoided are outcomes.
• What is published about the administration itself? Algorithm and sensor registers are the field's newest genre and its most demanding.

The practical takeaway: open data is one of the few smart-city investments with a fifteen-year survival record — but the portals that survived were treated as infrastructure with owners and update contracts, not as launches. The licence decides who can build on it; the refresh cycle decides whether they still can next year.