VeriGeo: Runtime Validation and Repair for Projections, Areal Interpolation, Topology, and Spatial Statistics in Automated GIS
Year: 2026
Type: preprint
Abstract: Automated GIS scripts can complete without runtime exceptions while still producing spatially invalid outputs. We present VeriGeo (NS-VGP 1.0), a Python runtime checking and repair layer for coordinate reference systems, areal interpolation, polygon topology, and spatial-statistics preconditions. We evaluate the implementation on a deterministic synthetic benchmark comprising 70 tasks (60 fault-injected tasks and 10 clean controls); the repository records 535 method executions across these tasks and ablations. The unverified GeoPandas baseline passed 0/60 fault-injected tasks and 10/10 controls (14.3% overall), with a 35.00% mean geodesic area error and a 169.56% mean population overcount on the benchmark joins. Under the benchmark's stated evaluator, VeriGeo passed 57/60 fault-injected tasks (95.0%) and 10/10 controls (67/70 = 95.7% overall; 95% Wilson CI [88.1%, 98.5%]), with a 25.44 ms mean execution latency and a 0.024% mean geodesic area error. Across the nine Gaussian-hotspot joins, per-zone population NMAE was 3.84% for VeriGeo versus 7.97% for area-weighted interpolation; including the unmasked water-barrier join, the ten-join means were 10.10% and 12.80%, respectively. Three tasks remain failures: the unmasked water barrier (MAUP-14), an ambiguously named integer-valued income field (MAUP-15), and a legitimate narrow canal removed by the sliver filter (TOPO-15). The statistics benchmark evaluates finite Moran-style outputs and selected spatial preconditions; it does not establish that GWR, LISA, or Getis-Ord estimators or inferential workflows have been validated.
DOI: 10.31223/x5s81p
Author: Prince Prince
Language: English
Subfield: Geography, Planning and Development
Field: Social Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7234486526
