Airfree Research Lab publishes breakthrough in satellite-derived building height estimation
Illustrative preview: this is a sample announcement for illustration only, not a record of a past event.
A new method from the Airfree Research Lab estimates building heights directly from freely available satellite imagery, removing the need for costly LiDAR flights over large areas.
The Airfree Research Lab today published results from a new method that estimates the height of individual buildings directly from freely available multispectral satellite imagery. The approach substantially reduces the cost of producing the 3D city models that underpin digital twins, flood modelling, and solar-potential analysis.
The problem with height
Accurate building footprints are now widely available, but reliable heights have traditionally required airborne LiDAR — accurate, but expensive and infrequent over large or remote areas. Many regions therefore have good 2D data and no vertical dimension at all.
The method
The Research Lab’s model combines shadow-length cues, off-nadir viewing geometry, and seasonal illumination differences across a stack of Sentinel-2 acquisitions, using a convolutional network trained against LiDAR-derived ground truth. The result is a per-building height estimate with quantified uncertainty, produced entirely from open imagery.
- Inputs are open Sentinel-2 analysis-ready data — no proprietary imagery required.
- Each estimate carries an uncertainty band, so downstream models can weight it appropriately.
- Outputs slot directly into the Airfree 3D Tiles pipeline for immediate visualisation.
Open science
Consistent with the Lab’s commitment to reproducible research, the evaluation methodology and dataset parameters are documented in the accompanying technical report. Airfree welcomes collaboration with mapping agencies and universities holding reference LiDAR for further validation.