Microsoft Buildings

The Microsoft Building Density & Height Dataset offers quarterly, global estimates of building density and height. These estimates are generated by running machine learning models on Planet's quarterly PlanetScope basemaps. The dataset includes a global public layer (~100 m/px) for 2023 Q4, distributed as Cloud-Optimized GeoTIFFs (COGs) referenced by a GeoPackage tile index. It also provides more detailed data (~40 m/px) for five high-growth locations, covering quarterly estimates from 2020 Q2 through 2025 Q2.

This GitHub repository serves two main purposes: data exploration and model training. Most users will find value in accessing and working with the public datasets through provided tutorial notebooks and programmatic interfaces. For those interested in reproducing the analysis, the repository also contains resources and guidelines to train custom models using Planet imagery, enabling reproduction of the methodologies outlined in their associated paper. A web visualizer is available for interactive exploration.

The dataset leverages machine learning and remote sensing techniques to provide valuable insights into urban development and change over time. It is a significant open-data resource for researchers, developers, and analysts focused on global building patterns.