Giant 3D map shows almost every building in the world (www.nature.com)

Researchers have unveiled the GlobalBuildingAtlas, a groundbreaking 3D map that details nearly every building on Earth, capturing 97% of structures with remarkable precision. Utilizing satellite imagery and advanced machine learning techniques, this comprehensive dataset encompasses 2.75 billion buildings, recorded with footprints and heights at a resolution of 3 meters by 3 meters. Key to this innovation is a deep learning model trained on LiDAR data from 168 cities, which enables the accurate estimation of building dimensions from around 800,000 satellite scenes captured in 2019. The significance of this project extends beyond mere mapping; it presents new opportunities for disaster risk assessment, climate modeling, and urban planning. Co-author Xiaoxiang Zhu points out that the insights gained from this dataset could enhance monitoring of the UN's Sustainable Development Goals related to urban development. Additionally, this work reveals critical disparities in building volume across different regions, highlighting the need for more nuanced analyses of urban growth than traditional 2D measurements can provide. As urbanization accelerates, especially in Asia, such detailed insights are vital for effective policy-making and sustainable development.