VeriGeo: Runtime Validation and Repair for Projections, Areal Interpolation, Topology, and Spatial Statistics in Automated GIS
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.
Authors
- Prince (ORCID: https://orcid.org/0009-0008-5483-2837)
Publication Details
- Journal
- EarthArXiv (California Digital Library)
- Published
- 2026-11-16
- DOI
- https://doi.org/10.31223/x5s81p
- Primary Topic
- Geographic Information Systems Studies
- Type
- preprint