Skip to main content

Global Building Atlas

TOOL
๐Ÿ“–6 min read

Introductionโ€‹

The Global Building Atlas is a 3D dataset containing over 2.75 billion building footprints, developed by the Technical University of Munich (TUM). It covers areas that standard digital maps miss, including rural regions and parts of the Global South.

Unlike standard OpenStreetMap data, it includes estimated heights (Level of Detail 1, or LoD1) derived from satellite imagery and machine learning. This page covers how to stream the dataset directly into QGIS via a Web Feature Service (WFS) connection, and how to bring the result into Blender.

Key Takeawaysโ€‹

Global Coverage

Access 2.75 billion building footprints worldwide.

3D Ready

Includes height data for immediate 3D extrusion.

WFS Stream

Stream data directly without downloading large files.

Usageโ€‹

Connect the WFS serverยท TUM endpoint

Open QGIS. Navigate to Layer > Add Layer > Add WFS/OGC API - Features Layer... Under Server Connections click New and enter:

  • Name: Global Building Atlas
  • URL: https://tubvsig-so2sat-vm1.srv.mwn.de/geoserver/ows?

Click OK.

Add the LoD1 layerยท filter to the view extent

Request the layer, but only for the area on screen.

Handle with care

This dataset covers the entire world. Loading the full dataset at once can freeze QGIS. Filter the request to your current view extent.

  1. In the Data Source Manager, ensure Global Building Atlas is selected.
  2. Click Connect.
  3. Locate the layer named global3D:lod1_global.
  4. Prepare your view: close the manager temporarily and zoom the map canvas to your specific area of interest (e.g., a specific neighborhood).
  5. Re-open the WFS window (Layer > Add Layer > Add WFS...).
  6. Select global3D:lod1_global.
  7. Important: check the box "Only request features overlapping the view extent". This limits the download to what you currently see.
  8. Click Add.

Verify the coordinate systemยท EPSG:3857

The native data is in EPSG:3857 (Web Mercator). QGIS handles reprojection automatically, but it is useful to verify.

  • Check the bottom right corner of the QGIS window.
  • Ensure your Project CRS is set to EPSG:3857 or a local UTM zone for accurate visualization.

Visualize in 3Dยท extrude the heights

"LoD1" means buildings are simple extruded blocks. You can visualize them directly in QGIS.

  1. Check Attributes: right-click the layer and open the Attribute Table. Look for the height field (often height, elevation, or h_est).
  2. Open 3D View: go to View > 3D Map Views > New 3D Map View.
  3. Style the Layer: right-click the layer > Properties > 3D View. Set "No Symbol" to Single Symbol. Set Altitude Clamping to "Absolute" (or "Terrain" if you have a DEM). Click the data override icon (ฮต) next to Extrusion, select Field type, and choose the height field found in step 1. Click OK.

Export a shapefileยท UTM zone

To use this data in Blender, export it as a Shapefile.

  1. Right-click the global3D:lod1_global layer.
  2. Select Export > Save Features As....
  3. Format: select ESRI Shapefile.
  4. CRS: do not keep the project default. Find the UTM zone for your location on EPSG, then in QGIS click the globe next to CRS, search under Universal Transverse Mercator, and pick the correct zone. The dialog shows a preview of the zone's extent.
  5. Save the file.

Import into Blenderยท BlenderGIS

To bring this data into Blender, use the BlenderGIS add-on.

  1. Open Blender.
  2. Navigate to GIS > Import.
  3. Choose Shapefile (.shp).
  4. Extrusion from field: select the height attribute (e.g., height or h_est). This tells BlenderGIS to automatically extrude the flat footprints into 3D volumes.
  5. Elevation Source: leave it at none unless you want the buildings lifted to a ground altitude. This setting controls the base level the buildings sit on, not the building height.
  6. Separate Objects: check it to keep each building as its own object, or uncheck it to merge the buildings into a single mesh (better for performance).
  7. Click OK.

The buildings will appear in your scene, extruded to the heights the Global Building Atlas estimated.

In an Investigationโ€‹

Use the block below to document how you used the Global Building Atlas.

Citing the dataset

Citationโ€‹

When using this data in an investigation, cite the original research:

Zhu, X. X., Chen, S., Zhang, F., Shi, Y., & Wang, Y. (2025). GlobalBuildingAtlas: An Open Global and Complete Dataset of Building Polygons, Heights and LoD1 3D Models. Technical University of Munich. https://doi.org/10.14459/2025mp1782307

Methodology exampleโ€‹

"Building footprints and height data were obtained from the Global Building Atlas (Zhu et al., 2025) via a WFS connection in QGIS. The data was filtered to the investigation area and exported as a shapefile for 3D reconstruction."

Into Blenderโ€‹

The output of this workflow is an ESRI shapefile: LoD1 footprints with a height field, projected to your local UTM zone.

  • Import path: in Blender, open GIS > Import > Shapefile with the BlenderGIS add-on. Set Extrusion from field to the height attribute. Leave Elevation Source at none unless the buildings should sit on a ground altitude.
  • Separate objects: keep each building as its own object while you check the data. Merge into one mesh before you do anything that costs per-object overhead.
  • Keep one CRS per scene. If your base layer was exported under the custom orthographic CRS from the baseline setup, project this layer to that CRS before export (or run the whole workflow in UTM). Mixed projections are the most common cause of layers landing in the wrong place.
  • Heights are estimates. Extruded volumes from this dataset are reference geometry. Do not present an estimated height as a measured one (see Limitations).

Limitationsโ€‹

The dataset is strong on coverage (quantity) and weak on precision (quality). Use it for reference. Do not use it when you need accurate heights.

Coverage: highโ€‹

  • Strength: The dataset covers areas that maps have historically missed, such as rural parts of Africa and South America. It likely contains buildings that Google Maps and OpenStreetMap do not.
  • Weakness: The data is auto-generated from satellite imagery, so it can include buildings that do not exist.

Height: lowโ€‹

  • Heights come from monocular estimation: the model infers height from a single flat image (shadows, apparent size), not from laser scanning (LiDAR).
  • Expect estimates. A 10 m building may be listed as 8 or 12 m. The typical error is ยฑ3โ€“9 m (about 1โ€“3 floors).

Shape: blockyโ€‹

  • LoD1 footprints are extruded blocks. No slanted roofs, domes, or spires.
  • The polygonization algorithm smooths complex shapes. An irregular apartment block can end up as a simple rectangle.

Resourcesโ€‹

Sourcesโ€‹

  1. GlobalBuildingAtlas (TUM), dataset repository and web viewer
  2. Zhu et al. (2025), the dataset paper
  3. TUM WFS server, the streaming endpoint
  4. EPSG, UTM zone lookup for the export step
Tags:
Last updated on by blender-visual-investigation