A new satellite-derived atlas gives a height to nearly every building on Earth. Adding up their volume in the centres of fifty cities, and dividing by the people who live there, gives a measure of how much building each resident has.
FindingAcross 50 city centres the median is about 200 m³ of building per resident; a resident of central New York has seven and a half times the building volume of one in central Dhaka.
How to citeAl, Stefan. 2026. “Built Mass.” Formlume. https://research.stefanal.com/built-mass/
DataGlobalBuildingAtlas (Zhu et al.), with footprints including © OpenStreetMap contributors; European Commission JRC: GHS-POP. Please also cite the original data when you use these figures.
UseFigures on white may be reused for non-commercial purposes with credit (CC BY-NC 4.0). The page design is © Stefan Al, all rights reserved; for commercial use, please get in touch.
Cubic metres of building per person living in the same 400 km²; colour shows mean building height, as on the map. Click a city to map it.
Residents per km² against building volume per m² of land, both for the whole square. Diagonals mark equal volume per person; cities above a line have more building per resident.
Building volume per m² of land (bars) and residents per km² (line), ring by ring from the centre.
Buildings. GlobalBuildingAtlas (Zhu et al. 2025, TU Munich), LoD1 release via Source Cooperative: 2.75 billion footprints from Google, Microsoft, OpenStreetMap and the authors' own model, each with a height estimated from PlanetScope imagery (reported errors 1.5–8.9 m by continent; reference year 2019). Volume is footprint area times height.
People. European Commission JRC, GHS-POP R2023A, residents in 2020 at 1 km. Residents only: offices, shops and hotels add volume but no residents, so business districts score high.
Area. A 20 × 20 km square centred on each city's traditional centre (the same centres as the density cones), water included. Floor area per person is estimated as volume ÷ 3.2 m storeys.
Reading it. Heights from satellite are least reliable for low, dense, informal fabric and for very tall towers; the atlas is consistent across cities, which is what makes the comparison possible, not exact for any one building.
Tools. Python (PyArrow reading the building atlas straight from cloud storage, GeoPandas, Rasterio) and D3.js.