AI-generated buildings in OpenStreetMap: frequency of use and differences from non-AI-generated buildings
| Authors | |
|---|---|
| Year of publication | 2025 |
| Type | Article in Periodical |
| Magazine / Source | International Journal of Digital Earth |
| MU Faculty or unit | |
| Citation | |
| web | https://doi.org/10.1080/17538947.2025.2473637 |
| Doi | http://dx.doi.org/10.1080/17538947.2025.2473637 |
| Keywords | Open Street Map; buildings; AI-assisted mapping; ohsome; data quality |
| Description | AI-assisted mapping is an innovative approach to data production in OpenStreetMap (OSM), designed to add new buildings to maps using advanced editing tools based on deep learning techniques and recently released global-scale building datasets derived from satellite imagery. However, the identification of OSM data derived from AI-generated datasets remains challenging without a comprehensive global overview of the scale, magnitude, and impact of AI-assisted mapping in OSM. The present study examines the evolution of spatiotemporal mapping of buildings in OSM, applying the ohsome framework, a high-performance data analysis platform for full-history OSM data analysis. The study’s findings indicate that tags recommended by data providers are effective in identifying AI-generated buildings, and that the spatial distribution of AI-assisted mapping is highly uneven, with over 50 percent of all AI-generated buildings in OSM located in the United States and 75 percent concentrated in just five countries. A positive correlation is observed between the prevalence of AI-generated buildings in maps and both population size and natural disaster mortality rates per 100,000 people. In most countries, AI-generated buildings are modified less frequently than non-AI-generated buildings. A case study of a selected location to verify the quality of AI-generated buildings is also presented. |
| Related projects: |