- The authority lacked reliable building height data, and manual field surveys were too expensive to perform at scale.
- While raw LiDAR data was available, the agency lacked the specialized expertise to process it into actionable property attributes.
.png){kind=small}
National LiDAR processing programme for property assessment aligned with U.S. valuation standards
VE3 utilized PDAL to process point clouds into high-resolution 0.5m DTM and DSM surfaces. Normalised DSMs were then intersected with cadastral data to isolate property-specific height statistics.
Initial estimates based on architectural type-specific floor heights were refined by a Random Forest classifier. This hybrid model achieved 89.4% accuracy using LiDAR geometry, footprint area, and property age.
A profile analysis module fitted piecewise linear models to roof slopes to estimate true eave height. This corrected habitable height prevented the overcounting of floors common in properties with steep roof pitches.
The system identified habitable lofts via a 1.5m headroom model and excluded low-height integral garages from calculations. These refinements significantly improved accuracy across pre-war urban and post-war suburban housing stocks.
The pipeline used Spark and GeoMesa on AWS to process 1.1 million properties in parallel. Full provenance tracking ensured each estimate remained linked to its specific source data and model version.
Overall Floor Count Accuracy
Validated against building permit records; exceeded the 85% accuracy target
Two-storey Stock Accuracy
Highest accuracy for the dominant residential typology
Loft Conversion Undercounting
LiDAR headroom model materially improved habitable area estimates for converted attic spaces
Cost Reduction vs Field Survey
Automated pipeline replaced planned field survey programme at a fraction of the cost
Properties Processed
Full coverage of three-state assessment territory within 12-month contract term
.png)
This engagement proves VE3’s ability to deliver highly accurate, automated height estimation at a scale of 1.1 million properties, directly de-risking the VOA trial through field-tested architectural correction methodologies.
.png)
