
"Scientists have produced the most detailed 3D map of almost all buildings in the world . The map, called GlobalBuildingAtlas, combines satellite imagery and machine learning to generate 3D models for 97% of buildings on Earth."
"Conventionally, creating detailed 3D maps at a global scale has been difficult, say Zhu, because it usually requires either laser scanning or high‑resolution stereo imagery. The team's solution was to combine deep learning with laser scanning techniques to predict building heights. The tool was trained on reference data obtained using light detection and ranging (LiDAR) from 168 cities, mostly in Europe, North America and Oceania."
"The researchers created the 3D maps from approximately 800,000 satellite scenes captured in 2019, using the deep-learning tool to predict building heights, volumes and areas. The study found that Asia accounts for nearly half of all mapped buildings in the world - approximately 1.22 billion structures. Asia also dominates the total building volume at 1.27 trillion cubic metres, reflecting rapid urbanization and dense metropolitan clusters in China, India and southeast Asia."
GlobalBuildingAtlas provides 3D models for 97% of Earth’s buildings, covering 2.75 billion structures with footprints and heights at a 3-metre by 3-metre spatial resolution. The dataset combines satellite imagery and machine learning, using a deep-learning tool trained on LiDAR reference data from 168 cities and applied to approximately 800,000 satellite scenes captured in 2019 to estimate building heights, volumes and areas. Asia contains roughly 1.22 billion structures and about 1.27 trillion cubic metres of building volume; Africa contains about 540 million buildings and 117 billion cubic metres, reflecting predominantly small low-rise structures. Applications include disaster risk assessment, climate modelling, urban planning and Sustainable Development Goal monitoring.
Read at Nature
Unable to calculate read time
Collection
[
|
...
]