AI for Cadastre: How We Automated Property Validation

Cadastral validation is the silent bottleneck of multipurpose cadastral updating in Colombia. For every property captured in the field, a professional must verify the geometric, topological, and alphanumeric consistency of the information before it can be packaged into an XTF file and delivered to IGAC. In a 20,000-property project, this means thousands of hours of manual review. At GEOSAT we decided that artificial intelligence could solve this problem.

The manual validation problem

In a traditional cadastral workflow, validation occurs at three levels:

Geometric validation

Each property polygon must comply with basic rules: no self-intersections, no duplicate vertices, a reasonable minimum area, coherence between the geometry and the area reported in the record.

A professional reviewing manually can verify between 80 and 120 properties per day. With 20,000 properties, this implies 170 to 250 person-days solely for geometric validation.

Topological validation

Properties do not exist in isolation — they form a continuous cadastral mesh. Each property must:

Not overlap with neighboring properties.
Not leave gaps with adjacent properties (or gaps must correspond to roads, rivers, or other identified features).
Share boundaries exactly with neighboring properties, without displacements or gaps.

This validation is especially complex in dense urban areas where thousands of properties interconnect.

Alphanumeric validation

Property attributes — address, economic classification, terrain area, built area, owner data — must be consistent with each other and with external sources. Examples of inconsistencies:

A property classified as "residential" with a built area of 0 m².
An owner with an identity document that does not match the expected format.
An address that does not exist in the municipality's official nomenclature.

What we automate with AI

In Terraes, our AI validation module addresses each of these levels:

Automated geometric verification

Spatial analysis algorithms that review each polygon in milliseconds:

Self-intersection detection: The system identifies polygons whose boundary lines cross and suggests corrections.
Duplicate and excessive vertices: Detects redundant points that can cause problems in XTF export.
Area-geometry coherence: Compares the area calculated from geometry with the area reported in the record. Differences exceeding 5% are flagged for review.
Degenerate geometries: Polygons with fewer than 3 unique vertices, zero-area lines, isolated points.

Intelligent topological validation

This is the most sophisticated module. It uses topological analysis algorithms that go beyond binary checks (overlaps/does not overlap):

Overlap quantification: Not only detects that two properties overlap, but calculates the overlap area and classifies it as "probable digitization error" (< 1 m²) or "real boundary conflict" (> 1 m²).
Gap analysis: Identifies voids in the cadastral mesh and automatically classifies them: is it a road? A water body? A capture error?
Shared boundary consistency: Verifies that boundary segments between adjacent properties are geometrically identical, with configurable tolerances.

Cross-reference attribute validation

Machine learning models trained on data from previous projects that detect statistical anomalies:

Atypical area values: An urban property of 50,000 m² in a zone where the average is 200 m² is automatically flagged.
Classification-area inconsistencies: Cross-references economic classification with built area to detect improbable combinations.
Document validation: Verifies that identity document numbers comply with expected formats and lengths for each type (CC, NIT, CE, TI).
Comparison with previous cadastre: When prior cadastral information exists, compares changes and flags differences exceeding reasonable thresholds.

Suspicious pattern detection

The system identifies patterns that, individually, might go unnoticed but together suggest problems:

Cloned properties: Records with suspiciously identical attributes that could indicate accidental duplication during capture.
Error sequences: When a property recognizer makes the same type of error repeatedly, the system alerts for corrective training.
High-inconsistency zones: Geographic areas where error density is abnormally high, suggesting systematic capture problems.

Measurable results

Implementation of the AI module in Terraes has produced quantifiable improvements:

Validation time reduced by 70%. What previously took weeks now executes in hours.
IGAC rejection rate reduced to less than 3%. Before AI, first-submission rejection rates oscillated between 10% and 15%.
Early detection. Errors are identified during capture, not at the end of the project. This drastically reduces correction costs.
Consistency. AI applies the same rules to all properties without fatigue or bias. A human validator after 8 hours of work inevitably reduces their attention level.

What AI does not replace

It is important to be clear: AI does not eliminate the need for cadastral professionals. What it does is automate repetitive and predictable verifications so that professionals can focus on cases that truly require human judgment:

Properties with complex legal situations.
Boundary conflicts between owners.
Zones with informal tenure where clear documentation does not exist.
Decisions about economic classification in ambiguous cases.

The cadastral professional remains indispensable. But instead of spending 80% of their time verifying that polygons do not overlap, they now dedicate that time to solving problems that truly need their expertise.

Integration with the cadastral workflow

The AI module is not an isolated component — it is integrated at every stage of the Terraes workflow:

During capture: Real-time validations on the property recognizer's tablet.
In the office: Automated bulk validations with detailed reports by zone and operator.
Pre-export: Complete verification before generating the XTF file, replicating IGAC validator rules.
Post-delivery: Analysis of IGAC validation results to feed back into the models.

The future of AI in cadastre

We are working on the next generation of AI capabilities for cadastre, which include computer vision for automatic boundary extraction from orthoimages and predictive models for cadastral value estimation based on spatial and contextual characteristics.

Cadastral updating in Colombia is a national-scale project. AI is not a luxury — it is an operational necessity to meet the deadlines and standards the country has set for itself.