AI-powered GlobalBuildingAtlas maps 2.75 billion buildings in 3D, researchers reveal

A new 3D map of the world’s buildings, called GlobalBuildingAtlas, offers a global catalog of 2.75 billion structures visible from space. Developed by researchers at the Technical University of Munich in Germany, the free web viewer merges satellite imagery and artificial intelligence to render building footprints and heights with unprecedented scale.

The atlas relies on nearly 800,000 satellite images and an AI algorithm that predicts the size of missing or occluded buildings by analyzing the heights of neighboring structures. The project provides location and size data for each building, enabling applications ranging from urban planning and disaster response to climate modeling and infrastructure monitoring. The team says the tool is designed as an open resource for researchers, policymakers, and educators, with the goal of offering a global view of urban density, growth patterns, and built-environment dynamics.

The GlobalBuildingAtlas project represents a milestone in scaling satellite-derived information about the built environment. By fusing hundreds of thousands of satellite observations with an AI-based inference engine, the researchers can estimate the height and extent of buildings across diverse urban forms and terrains. The result is a continually updating, freely accessible platform that can support planning, risk assessment, and research at a planetary scale.

In addition to the technical achievement, the atlas has entered the public imagination through a separate feature by a major news outlet that asked readers to guess seven famous landmarks drawn from the map. The exercise illustrated how the same 3D data used by scientists can reveal recognizable silhouettes and architectural signatures across continents, even as some structures require inference where direct measurements are incomplete. The landmarks highlighted in that feature were: Tower of London, England; Sydney Opera House, Australia; Brandenburg Gate, Berlin, Germany; Space Needle, Seattle, USA; Marina Bay Sands, Singapore; Burj Khalifa, Dubai, UAE; and Petronas Towers, Kuala Lumpur, Malaysia.

The seven-image quiz also underscored the role of color shading in the atlas’s visualizations. In the viewer, different shades of blue indicate height, with taller buildings shown in darker tones. That visual cue helps users interpret the data quickly, whether they are students, policymakers, or members of the public exploring urban form.

Experts behind GlobalBuildingAtlas say the project is not only a map but a tool for understanding how cities grow, where infrastructure is concentrated, and how neighborhoods evolve over time. By making the data freely available, the team hopes to spur collaboration across disciplines and geographies, fostering new research into topics such as urban resilience, energy use, and equitable access to services. The atlas is designed to be updated as new satellite imagery becomes available and as the AI model refines its estimates through ongoing validation against ground-truth measurements.

The GlobalBuildingAtlas project also illustrates the broader potential of AI-assisted mapping. Combining vast image datasets with machine learning enables researchers to fill gaps where direct measurements are sparse or unavailable, offering a scalable approach to monitor the world’s rapidly expanding built environment. While the platform is intended for legitimate research and educational purposes, the developers emphasize responsible use, data transparency, and ongoing validation to keep the model aligned with real-world conditions.

The technical team notes that the atlas currently catalogues buildings by location and estimated size, not by ownership or other sensitive attributes. As the project evolves, they plan to incorporate additional data layers, such as land cover, usage type, and temporal change, to provide deeper insights while maintaining a focus on privacy and ethical considerations. Researchers also stress that the platform is a complement to, not a replacement for, traditional surveying and local knowledge.