Which AI Map of Every Building on Earth Is Best? New Scorecard
The maps that decide where disaster aid goes are often wrong.
Researchers benchmarked seven global or near-global building and settlement datasets — including Overture Maps, Google Open Buildings 2.5D Temporal, Microsoft TEMPO, GHSL, and WSF Tracker — against harmonized reference footprints across 135 study areas, measuring detection, geometric agreement, and aggregate quantity accuracy. Overture achieved the highest median city-level vector F1 (0.786). Raster rankings depended on resolution: OBT led at 10m (0.642) and WSF Tracker at 100m (0.862), though WSF Tracker substantially overestimated built-up area, so users need to know whether a raster identifies only buildings or includes additional impervious surfaces. Raster accuracy increased with building density, small candidate buildings were disproportionately associated with false positives in vector products, and temporally aligning WSF Tracker with reference imagery raised mean F1 by 0.060. The study offers a reproducible benchmark for these datasets, which increasingly support population mapping, exposure assessment, and urban monitoring.
- Seven free global building maps were compared in 135 places — Overture Maps was the most accurate in cities, with a 0.786 score out of 1.0.
- One map with a top score, WSF Tracker, was partly cheating: it counted roads and pavement as buildings, so it looked more accurate than it was.
- Maps were weakest in rural areas with small, scattered homes — exactly where disaster aid and insurance decisions matter most.
Why It Matters
Better building maps mean faster disaster aid, fairer insurance prices, and smarter spending on roads and services.