Industries / Technology / Mapping and geospatial
Open map data
Open map data is map data that anyone may copy, use and share, built by volunteers, governments and companies and published under open licences.
7
Milestones achieved, from 2002 to 2018.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
2002
Addresses made free
In 2002 Denmark's Minister for Finance and its municipalities agreed to make the official address data, kept by the municipalities, free of charge, so that users paid only the cost of delivery; the agreement took full effect in 2005 after a change to the rules of the national data server. A study by the consultancy COWI for the Danish Enterprise and Construction Authority, the agency responsible for road names and addresses, put the direct benefit to society from 2005 to 2009 at about €62 million, against total costs of about €2 million, and estimated that about 70% of the benefit went to the private sector.
Lessonin this case the gain from free use was about 30 times the cost of giving the data away.
Credit
Danish Ministry of Finance; Danish municipalities; Danish Enterprise and Construction Authority
2004
A free world map
OpenStreetMap is a map of the world drawn by volunteers and free for anyone to use. Steve Coast founded it in 2004 while at University College London, because Britain's government-run Ordnance Survey held detailed map data but did not distribute it freely. He registered openstreetmap.org on 9 August 2004 and entered the first street on 10 December 2004, after cycling around London's Regent's Park with a GPS receiver. The project counted 10 million registered users on 19 January 2023.
Credit
Steve Coast; OpenStreetMap's contributors
2008
US roads imported
An import is a bulk upload of an outside dataset. OpenStreetMap's first import of the US Census Bureau's TIGER road files started in 2005 but was shut off, and its data purged, in November 2006 because of data integrity problems. A second import, of the 2005 TIGER files, ran from September 2007 to 23 January 2008 and filled the nearly empty map of the United States. It also brought "a plethora of problems with data inaccuracy, some of which persist to this day", in the words of the project's wiki, which keeps a "TIGER fixup" page of things to repair.
Lessonan import fills a map in months, but its errors become the volunteers' work for years.
Credit
OpenStreetMap contributors; US Census Bureau
2010
A national agency opens up
On 1 April 2010 Ordnance Survey, Great Britain's national mapping agency, launched OS OpenData, a site giving free and unrestricted access to a range of its maps and boundary data for reuse. It followed the Prime Minister's announcement of 17 November 2009 under the "Making Public Data Public" effort led by Tim Berners-Lee and Nigel Shadbolt. The free set left out the 1:25,000 and 1:50,000 maps that walkers use. OpenStreetMap's history page says the project was partly responsible for bringing the release about.
Credit
Ordnance Survey; the UK government; Tim Berners-Lee; Nigel Shadbolt
2012
A licence change costs 1%
A licence sets what others may do with data. On 12 September 2012 OpenStreetMap moved from the Creative Commons CC-BY-SA 2.0 licence to the Open Database License (ODbL), a share-alike licence written for databases: anyone may use the data, but must share their improvements to the database on the same terms. Most contributors accepted the new terms. Data from those who did not was removed, and the project's wiki puts the loss at about 1% of the data, which volunteers then worked to replace.
Lessona database built by many people can change its terms only with each contributor's consent, or by deleting their work.
Credit
OpenStreetMap Foundation; Open Knowledge Foundation, which developed the ODbL
2017
Roads 83% mapped
Christopher Barrington-Leigh of McGill University and Adam Millard-Ball of the University of California, Santa Cruz estimated how much of the world's road network OpenStreetMap held. People compared a sample of places with satellite imagery, and a statistical model scaled the sample up to whole countries. They found the map about 83% complete, with more than 40% of countries fully mapped, and found that global road datasets used by the World Bank undercounted roads by more than 30%. An update with April 2017 data put completeness at about 89%.
Credit
Christopher Barrington-Leigh; Adam Millard-Ball
2018
AI traces a country's roads
From June 2017 to December 2018 Facebook's mapping team added roads to OpenStreetMap across Thailand, chosen because few of its roads were mapped, using AI that found roads in Maxar satellite imagery and trained mappers who checked the results. By Facebook's count on the project's wiki page, 515,306 km of roads were added in that period, including the community's own additions. In 2019 Facebook released its editor for this work to everyone as RapiD. The same page lists the AI's faults: streams taken for roads, road types guessed from neighbouring roads, and roads wrongly joined or split where the imagery is confusing.
LessonAI made finding roads fast, but judging their type and connections still needed a person.
Credit
Facebook (now Meta); Thailand's OpenStreetMap community
Watch for
475 million open addresses
Overture Maps Foundation, which publishes open map data, says it maintains more than 475 million address points, compiled from more than 175 sources, mostly the community collections OpenAddresses and AddressForAll plus open data from public agencies. The addresses theme is still in alpha, an early test stage.
Achieved whenthe theme leaves alpha and an outside check of a random sample reports what share of its addresses exist and sit in the right place.
Claimed by
Overture Maps Foundation, November 2026
Open
01
AI-added roads, checked
A quality audit of roads added to OpenStreetMap with AI help, set against roads traced by hand in the same areas. Two reviewers check a random sample of each against current imagery: whether the road exists, whether it lies within a few meters of where it is drawn, whether it joins the right roads, and whether its class is right.
The briefCredit · unclaimed02
Road completeness, nine years on
A new estimate of how complete OpenStreetMap's road network is in every country, repeating the 2017 method: reviewers compare a random sample of places with satellite imagery, and a statistical model scales the sample up to each country.
The briefCredit · unclaimed
Open 01 · the brief
AI-added roads, checked
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A quality audit of roads added to OpenStreetMap with AI help, set against roads traced by hand in the same areas. Two reviewers check a random sample of each against current imagery: whether the road exists, whether it lies within a few meters of where it is drawn, whether it joins the right roads, and whether its class is right.
Why it is still open
Facebook's own page lists the AI's typical faults but no error rates. Fila, Štampach and Herfort (2025) used OpenStreetMap's full edit history to show that AI-generated buildings are edited less often than other buildings in most countries, and checked quality in only one case study. In a 2018 user study at MIT, people adding roads in Indonesia with a machine-assisted editor covered 1.7 times as much as people tracing by hand, with comparable error, but that measured a test session, not the roads now in OpenStreetMap. Andorful and colleagues (2025) tested whether AI-assisted roads can even be picked out in OpenStreetMap and found it harder as later edits blur where the data came from. No audit of AI-assisted roads against imagery was found.
wiki.openstreetmap.orgdoi.orgheigit.orgmapster.csail.mit.edu
Built with what exists
Facebook's changeset hashtags, such as #mapwithai, to find AI-assisted edits; OpenStreetMap's full history through HeiGIT's ohsome tools; Esri World Imagery, which Esri lets OpenStreetMap mappers use without restriction; the JOSM editor or Rapid; Python. Results are reported by method of mapping, never by individual mapper.
First step
Pull the roads in one Thai province added in Facebook's tagged changesets and an equal number of other roads there, and draw 100 of each at random for a trial review.
Done looks like
1,000 AI-assisted and 1,000 hand-traced road segments in Thailand and Indonesia, each judged by two reviewers, with error rates by type and the rate at which the reviewers agreed, published with the sample list.
Open 02 · the brief
Road completeness, nine years on
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A new estimate of how complete OpenStreetMap's road network is in every country, repeating the 2017 method: reviewers compare a random sample of places with satellite imagery, and a statistical model scales the sample up to each country.
Why it is still open
The last such estimate, from the authors' April 2017 data, put completeness at about 89%, and since then companies have added large amounts of road with AI and imports. Kontur's road completeness layer compares OpenStreetMap with the AI-detected roads Facebook published in 2019, so it measures agreement with one AI dataset, not with the ground. No newer sample-based estimate was found.
Built with what exists
The 2017 study's published code; current OpenStreetMap country extracts; Esri World Imagery for the visual check; Python or R for the statistical model; volunteer or student reviewers.
First step
Run the published code on one country's current OpenStreetMap extract and review 100 sample places against imagery.
Done looks like
Completeness for every country for 2026 with confidence intervals, from at least as many sample places as the 2017 study, published with the code and the reviewed sample and set beside Kontur's figures.