The World's 2.75B Buildings

Summary

The United Nations (UN) believes there are 4B buildings on Earth. This week, a dataset called "GlobalBuildingAtlas" (GBA) was published by researchers at the Technical University of Munich (TUM) that attempts to estimate this number at being closer to 2.75B. GBA is broken up into two datasets. The first is a 1.1 TB, level-of-detail 1 (LoD1) dataset made up of 922 uncompressed GeoJSON files. I spent the last few days downloading these and converting them into 210 GB of Parquet. These are now being hosted on AWS S3 thanks to the kind generosity of Source Cooperative and Taylor Geospatial Engine. Below is a heatmap of this dataset. The above basemap is mostly made up of vector data from Natural Earth and Overture. The second "Height" dataset is made up of 35 TB of raster height maps. Given the LoD1 files already contain building heights and downloading 35 TB off of TUM's servers will likely take months, I'll just show an individual raster in this post rather than run any dataset-wide operations on this dataset. Parts of these datasets were developed by running open source (code, not imagery or weights) deep learning models on Planet Labs satellite imagery. Planet Labs operate several constellations totalling hundreds of satellites in low earth orbit. They capture images of the entire Earth's landmasses every day. In this post, I'll walk through loading the LoD1 data from S3 into QGIS, describe the ETL process I ran to build the Parquet version of this dataset and discuss some observations of the footprints themselves. My Workstation I'm using a 5.7 GHz AMD Ryzen 9 9950X CPU. It has 16 cores and 32 threads and 1.2 MB of L1, 16 MB of L2 and 64 MB of L3 cache. It has a liquid cooler attached and is housed in a spacious, full-sized Cooler Master HAF 700 computer case. The system has 96 GB of DDR5 RAM clocked at 4,800 MT/s and a 5th-generation, Crucial T700 4 TB NVMe M.2 SSD which can read at speeds up to 12,400 MB/s. There is a heatsink on the SSD to help keep its temperat...