Machine learning just helped researchers create the biggest 3D map of buildings around the world

Researchers publish GlobalBuildingAtlas, mapping 2.75 billion buildings in 3D

by · TechSpot

Serving tech enthusiasts for over 25 years.
TechSpot means tech analysis and advice you can trust.

Connecting the dots: By applying machine learning techniques to satellite imagery, researchers have built an unprecedented database of man-made structures across the globe. The data could reshape research in fields ranging from urban planning to climate science, while providing a flexible, updatable tool for tracking human development over time.

A team of Germany-based researchers recently unveiled GlobalBuildingAtlas, which is likely the most detailed 3D map of buildings ever developed. Introduced earlier this month, the database covers 97 percent of the buildings on Earth's surface, with tiny three-dimensional polygons depicting more than 2.75 billion different man-made structures.

GlobalBuildingAtlas was created by a team co-led by Xiaoxiang Zhu at the Technical University of Munich. The publicly available dataset was made possible by detailed satellite imagery, deep learning algorithms, and laser-scanning techniques designed to predict building height. First, the scientists trained the system using reference data collected from 168 cities across Europe, North America, and Oceania.

The researchers fed the tool around 800,000 satellite images taken in 2019, tasking the algorithms with predicting the height, volume, and area of every building. GlobalBuildingAtlas has a spatial resolution of 3 meters by 3 meters and can be regularly updated with new data.

The database already provides insights into buildings in different parts of the world. A study published in the Earth System Science Data journal states that Asia accounts for half of the world's total building "population," or around 1.22 billion structures.

The continent also has the largest total building volume – 1.27 trillion cubic meters – reflecting rapid urbanization in China, India, and other densely populated countries in the region.

// Related Stories

Africa is the second-largest continent in terms of building count, with 540 million structures, the study says, but with a much lower combined volume of 117 billion cubic meters. African countries tend to host smaller, lower-rise buildings. GlobalBuildingAtlas can also help analyze the correlation between population density and economic development, making comparisons between countries within the same continent easier.

The study confirms, for example, that Finland has six times more building volume than Greece, and that Niger's per-capita building volume is 27 times smaller than the global average. Such measurements are essentially impossible with 2D maps, where differences in building stock and living conditions are far harder to infer.

Several researchers working on urbanization issues have welcomed GlobalBuildingAtlas as a potentially significant advancement in the field. Dorina Pojani, an urban planning researcher at the University of Queensland in Brisbane, Australia, said the database could help her track how urban areas develop over time.

Liton Kamruzzaman, a transport and urban planner at Monash University in Melbourne, also praised the new tool for its potential to track urbanization on a global scale. Many parts of the world, he noted, do not provide reliable data on how cities are growing.