GlobalBuildingAtlas

Introduction

In this project, we provide the level of detail 1 (LoD1) data of buildings across the globe.

A overview of the dataset is illustrated bellow:

License Notice:

⚠️This dataset is provided in three parts: ODbL-licensed polygons (GBA.ODbLPolygon), CC BY-NC 4.0 polygons and LoD1 building models (GBA.Polygon and GBA.LoD1), and CC BY-NC 4.0 height maps (GBA.Height).
⚠️Users may combine these datasets for analysis or downstream applications, but doing so may create license implications.
⚠️It is the responsibility of each user to ensure that their use complies with the respective licenses.
⚠️This repository does not provide legal advice; users should review the original licenses and, if necessary, consult legal counsel.

FAQ

Is Canada / Türkiye / ... in the dataset?

The dataset aims at global coverage, so all countries, territories, and cities should be included. However, due to data quality limitations, some areas may be absent or may not have height attributes. You can check availability using our web viewer.

Why are the height values incorrect? Why are some buildings missing?

This is a machine-learning–derived product. Errors may occur. Please refer to the publication for validation results and further details.

I cannot access the web viewer.

High traffic can occasionally affect the web viewer. We restart the server as needed to maintain access. Please check the bulletin board for maintenance updates.

Why are there no LoD1 GeoJSON files anymore? Why is the dataset split?

We used some building footprints from ODbL-licensed sources (OSM and Microsoft Global ML Building Footprints). According to ODbL, derivatives must also be ODbL-licensed, which conflicts with our PLANET-derived BY-NC data. Therefore, the dataset is split:

• Part I – HuggingFace: GBA.ODbLPolygon contains only building polygons derived from ODbL-licensed sources.
• Part II – HuggingFace: GBA.LoD1 contains additional building footprints from other sources, and LoD1 JSON files linking all building polygon features from GBA.ODbLPolygon and GBA.Polygon.
• GBA.Height – mediaTUM.

I still want the LoD1 GeoJSON files. What should I do?

You can follow the README instructions under this repository, or HuggingFace to derive the final LoD1 GeoJSON files.

⚠️ By deriving the final LoD1 GeoJSON files, you acknowledge the License Notice.

What is the CRS of the dataset?

All building polygons are recorded in EPSG:3857. Some files in GBA.ODbLPolygon on HuggingFace may appear in EPSG:4326 — please treat them as EPSG:3857.

Access to the Data

Web Feature Service (WFS)

A WFS is provided so that one can access the data using other websites or GIS softwares such as QGIS and ArcGIS.

Url: https://tubvsig-so2sat-vm1.srv.mwn.de/geoserver/ows?

Web Viewer

A web interface for viewing the data is available at: website.

Note: Over the past few days, our web viewer has received nearly 280,000 access requests. Due to this unusually high traffic, some data may not load completely, which may result in a significant portion of buildings not being displayed.

Full Data Download

⚠️ By downloading the data, you agree to the Terms of Use and acknowledge the License Notice.

The full data can be downloaded as follows:

• Part I - HuggingFace, GBA.ODbLPolygon contains ONLY building polygons derived from ODbL-licensed data sources.
• Part II - HuggingFace, GBA.LoD1 contains additional building footprints from other data sources, and LoD1 JSON files linking all building polygons features from GBA.ODbLPolygon and GBA.Polygon.
• GBA.Height - mediaTUM.

How to Use the Data

1. Access the representative dataset

   • Either in the representative/ folder from HuggingFace folder of this repository, or via mediaTUM.

2. Identify tiles overlapping your region of interest (RoI)

   • For polygons and LoD1 models: use lod1.geojson to find the intersecting tiles in GBA.Polygon or GBA.LoD1.
   • For heights: use height_zip.geojson and height_tif.geojson to find intersecting tiles in GBA.Height.

3. Download required tiles

   • ODbL polygons: download from HuggingFace and save under:

     ./ODbLPolygon
     

   • LoD1 polygons and other data: download from this repository, and save under:

     ./Polygon
     ./LoD1
     

4. Run the enrichment script provided at HuggingFace

   python produce_lod1.py

   • You may also specify folder paths at your choice.

   python produce_lod1.py \
       --odbl_root /path/to/odbl \
       --polygon_root /path/to/polygon \
       --json_root /path/to/json \
       --output_root /path/to/output

   • The script will read the two GeoJSON folders and the JSON folder, merge the properties, and add height and var fields.

5. Output

   • LoD1 GeoJSON files will be written under

   ./LoD1_GeoJSON
   

   or the folder you specified.

   • If you are only interested in GBA.Polygon, you can ignore the height and var fields.

Development Code

Global Building Polygon Generation using Satellite Data (Sec. 4.3)

For codes related to building map extraction, regularization, polygonization, and simplification, i.e., generating building polygons from satellite images (Sec. 4.3.2, Sec. 4.3.3, and Sec. 4.3.4), please refer to ./im2bf.

Global Building Height Estimation (Sec. 4.4)

1. For codes related to monocular height estimation using HTC-DC Net (Sec. 4.4.2), please refer to ./im2bh.
2. For codes related to the global inference and uncertainty quantification (Sec. 4.4.3), please refer to ./infer_height

Global LoD1 Building Model Generation (Sec. 4.5)

1. For codes related to quality-guided building polygon fusion (Sec. 4.5.1), please refer to ./fuse_bf.
2. For codes related to LoD1 building model generation (Sec. 4.5.2), please refer to ./make_lod1.

Visualization Code

For codes to reproduce the plots in the manuscript, please refer to ./make_plots.

Code License

MIT with Commons Clause (no commercial use allowed). See LICENSE.

How to cite

If you find this dataset helpful in your work, please cite the following paper.

@Article{essd-17-6647-2025,
AUTHOR = {Zhu, X. X. and Chen, S. and Zhang, F. and Shi, Y. and Wang, Y.},
TITLE = {GlobalBuildingAtlas: an open global and complete dataset of building polygons, heights and LoD1 3D models},
JOURNAL = {Earth System Science Data},
VOLUME = {17},
YEAR = {2025},
NUMBER = {12},
PAGES = {6647--6668},
URL = {https://essd.copernicus.org/articles/17/6647/2025/},
DOI = {10.5194/essd-17-6647-2025}
}