The World's 2.75B Buildings
5 days ago
- #geospatial
- #building-footprints
- #data-processing
- The UN estimates 4B buildings on Earth, but the GlobalBuildingAtlas (GBA) dataset suggests 2.75B.
- GBA consists of two datasets: a 1.1 TB LoD1 GeoJSON dataset (converted to 210 GB Parquet) and a 35 TB raster height dataset.
- The LoD1 dataset was processed using deep learning models on Planet Labs satellite imagery.
- The author details their high-performance workstation setup for handling large datasets.
- Steps for downloading and processing the GBA dataset using QGIS and DuckDB are outlined.
- The dataset includes 2.75B buildings, with sources like Google, OSM, Microsoft, and AI-generated data.
- Comparison with Overture's dataset shows GBA has 208M more buildings.
- Accuracy improvements in AI-detected buildings in Beijing are highlighted, though some inconsistencies remain.
- The dataset lacks timestamps, which could affect data reliability for rapidly changing areas.
- Building counts by country are provided, with India leading at 467M buildings.