A Two-Axis Validation Framework for Open Building Footprints: Position, Provenance, and Height Reliability for National 3D Cadastre Development

Authors:
Open Access
Publisher website
Google Scholar
Abstract
Level of Detail 1 (LOD1) building models underpin national 3D cadastre development, yet OpenStreetMap (OSM) and Global Building Atlas (GBA) have rarely been validated against authoritative cadastral data using designs that separate positional, geometric, and completeness errors. Both were evaluated against Integrated Complete Systematic Land Registration (PTSL Terintegrasi) footprints in Banjarsari, a low-rise semi-urban Indonesian settlement, using a stratified cluster sample of 189 reference buildings. A polygon-level provenance audit showed that 73.1% of sampled GBA polygons were geometrically identical to OSM, leaving only Google-sourced polygons independent. OSM carried a systematic 4.0 m offset; correcting it raised mean Intersection over Union from 0.29 to 0.61 and reversed the shape ranking, with OSM overtaking Google (0.42). Fusion lowered design-weighted omission from 31.3% to 21.1%. GBA heights were nearly unbiased (+0.23 m) yet less accurate than guessing the sample mean, beating that baseline only after local recalibration. Findings held under radius, threshold, and quality-assurance sensitivity tests. Neither source is directly adequate as LOD1 cadastral geometry without source-specific remediation. Methodologically, the study contributes a two-axis framework partitioning error by remediability, provenance auditing that prevents circular comparison, and design-based estimation. Practically, classifying defects by remediation cost leaves 21% field capture as the alternative to digitizing 100%.

This publication has 44 references indexed in Scilit: