01 CBO-TGRA
Confidence-Based Ownership Assessment with Targeted Geometric Risk Analysis
I originated and architected a four-stage geospatial intelligence methodology for assessing beneficial ownership, physical exploitation potential, and investigative priority for parcels near U.S. critical infrastructure.
USC–NGA Cooperative Research and Development Agreement · 2025–2026
Methodology Lead / Principal Architect
Pre-pilot specification Representative components prototyped · empirical calibration remains a pilot objective
Methods: Bayesian Belief Networks · Noisy-OR · Positive-Unlabeled Learning · Fellegi–Sunter · Graph ML · Dempster–Shafer / Yager · Monte Carlo Viewshed · Edge Betweenness · Spatial-Lag Models · Local Moran’s I · Remote-Sensing Change Detection · SHAP
Tools: Python · ArcPy · KNIME · ArcGIS Pro · Esri Geoprocessing · Figma
01 — The challenge
The initial requirement was to identify the ownership of parcels near critical infrastructure. I developed the methodology around the operational decision behind that requirement.
Where should limited counterintelligence and facility-security analytical capacity be directed first?
A proximity screen establishes a candidate area. Prioritization requires additional evidence: who plausibly controls each parcel, how reliable that attribution is, what physical opportunities the location creates, and how those findings relate to the mission.
The difficulty was connecting these questions across fragmented information. Parcel records, corporate registrations, regulatory information, commercial databases, terrain, transportation networks, and imagery differ in identifiers, geographic precision, completeness, update cycles, and evidentiary authority.
Identity introduces another layer of uncertainty. Before assessing ownership, the methodology must determine whether records refer to the same person, company, address, or parcel.
Spatial analysis carries its own uncertainty through facility coordinates, terrain elevation, observer position, and other modeling assumptions. I decomposed the problem into distinct analytical stages so that each claim could be inspected, tested, and traced to its supporting evidence.
02 — My contribution
I owned the analytical design from mission decomposition through the implementation and validation specifications.
Methodology architecture
Defined the four-stage methodology, ownership and geometric scoring structures, evidence model, uncertainty treatment, stage interfaces, and analyst workflow.
Representative implementation
Developed KNIME, Python, ArcPy, and Esri geoprocessing components for parcel screening, spatial joins, entity-resolution scaffolding, network accessibility, terrain and visibility analysis, change detection, scoring, and QA.
Specification & communication
Authored the technical specification and SOPs, developed analyst-interface concepts, scoped the Washington, D.C. metropolitan validation pilot, and briefed analyst, scientific, and leadership audiences.
The deliverables connected analytical reasoning to the practical requirements for implementation, evaluation, and analyst use.
03 — System architecture
Four stages connect spatial screening to mission-specific prioritization.
The escalation gate determines which parcels receive the more computationally expensive geometric analysis. Ownership and geometry remain separate analytical dimensions, with their outputs brought together at Terminal Fusion. Stage 1 ownership and evidence outputs are carried forward unchanged; mission preferences affect prioritization rather than rewriting the underlying ownership assessment.
Define the candidate parcel universe
Which parcels should enter the scoring cycle?
I designed three admission pathways: statutory screening, physics-informed screening, and a shadow zone addressing potential boundary-gaming. A parcel enters the analytical universe when it qualifies through any applicable pathway.
The physics-informed pathway uses the Maximum Theoretical Exploitation Bounding Box (MTEBB): the greatest modeled reach across four exploitation modalities, taken with a maximum operator so that any single modality can admit a parcel. Within that bound, parcel-level feasibility tests decide admission: a clear line of sight to the facility, an RF link that closes, road access inside the network service area, or a feasible UAS launch.
Assess ownership and the evidence behind it
What does the evidence support about ownership or control?
A knowledge-graph architecture represents ownership and control relationships alongside sources, dates, and identity-match confidence.
Entity resolution uses deterministic matching, Fellegi–Sunter probabilistic linkage, and graph-native machine learning. Four evidence families—Entity Opacity, Transaction Anomaly, Regulatory Signals, and Contextual Corroboration—feed Bayesian belief networks with Noisy-OR aggregation. The design also includes positive-unlabeled learning, propensity-aware estimation, probability calibration, and SHAP attribution.
Three outputs remain distinct: Ownership Confidence Score, Evidence Coverage Score, and Actor-of-Concern Score.
Assess physical exploitation potential
What could the parcel physically enable?
Probabilistic visibility uses Monte Carlo terrain realizations to evaluate line-of-sight support under elevation uncertainty.
Network and chokepoint analysis evaluates parcel frontage and nearby route importance using general and facility-specific edge betweenness. UAS feasibility considers launch and recovery conditions, operational reach, and loiter feasibility. Remote-sensing change detection examines vegetation, material, and structural change through time.
Translate findings into investigative priority
Where should limited analytical capacity go first?
Mission-specific weights combine ownership and geometric utility functions without rewriting either underlying assessment.
Dempster–Shafer belief and plausibility intervals represent evidential uncertainty, with Yager combination preserving unresolved conflict. The Coordinated Acquisition Flag uses spatial-lag residuals and Local Moran’s I with false-discovery control to add a spatial-pattern signal while preserving the ownership estimate.
04 — Interactive models
Explore selected mechanisms within the architecture through two complementary modeling environments.
Admission pathways made inspectable
Compare statutory, physics-informed, and shadow pathways in a synthetic terrain scene. Change scenario inputs, inspect the resulting envelope, and examine which parcels enter the candidate universe.
Model context: synthetic terrain and parcels; interactive screening demonstration.Open the MTEBB Sandbox →Terrain-aware geometric reasoning
Explore how real elevation changes modeled visibility, propagation, access, and selected UAS scenarios around a notional facility.
Model context: real elevation with notional facilities and routes; separate from a parcel-ownership assessment.Open Terrain Lab →05 — Engineering decisions
The integration framework gives each source a defined analytical purpose, pipeline location, and requirements for normalization, provenance, and quality assessment. The knowledge graph represents ownership, control, management, shared-address, shared-agent, and financing relationships while preserving source documents and match confidence.
A bi-temporal evidence substrate
The architecture records both when a relationship was valid and when the system learned about it. That distinction supports reconstruction of an assessment using the evidence available at a particular time.
Explicit tradeoffs
| Design decision | Reason | Tradeoff to manage |
|---|---|---|
| Screen broadly before deeper analysis. | A missed admission can remove a relevant parcel from the scoring cycle. | Additional candidates increase research and review workload. |
| Preserve ownership and geometry as distinct outputs. | Each answers a different analytical question. | The interface must make several dimensions easy to interpret. |
| Carry identity-match confidence into the evidence model. | Incorrect record merges can distort downstream attribution. | Ambiguous matches require adjudication and review-queue management. |
| Propagate terrain uncertainty through visibility analysis. | Geometric conclusions depend on uncertain inputs. | Repeated simulation increases compute requirements. |
| Version sources, parameters, and stage outputs. | Assessments need to be reproducible and explainable over time. | Re-scoring, storage, and configuration management become architectural requirements. |
Controlled stage handoffs
Each stage has a defined input and output package. Completed scores are preserved within a scoring cycle, and superseded values enter the historical record when evidence changes. Source provenance, transformation lineage, parameter versions, and analyst decisions travel with the assessment. Quality checks at stage boundaries address schema conformity, completeness, positional suitability, and traceability before downstream processing.
06 — Validation and development status
The methodology is specified and representative components have been prototyped. Empirical calibration remains a pilot objective.
I scoped a validation pilot using Regrid Premium parcel data across the Washington, D.C. metropolitan area. Its purpose is to establish how the methodology behaves against real parcel inventories and how its thresholds translate into analyst workload.
| Area | Validation focus | Question to resolve |
|---|---|---|
| Spatial screening | Positional quality, admission coverage, parameter sensitivity. | Does the candidate universe capture relevant parcels at manageable scope? |
| Entity resolution | False matches, missed matches, adjudication consistency, review volume. | Are linked records reliable enough to support ownership inference? |
| Ownership inference | Calibration curves, Brier scores, geographic holdouts, label-selection effects. | Do confidence estimates behave as intended on held-out cases? |
| Geometric analysis | Input suitability, uncertainty propagation, assumption sensitivity. | Which geometric findings remain stable across plausible conditions? |
| Prioritization | Score distributions, threshold effects, ranking stability, mission-weight sensitivity. | How do settings change review tiers and analytical workload? |
| Reproducibility | Archived sources, parameters, model versions, provenance. | Can an assessment be reconstructed and its decisions explained? |
The specification also identifies open calibration questions involving evidence dependence, escalation thresholds, modality weights, and uncertainty adjustments. RF and electromagnetic propagation remain deferred within the operational methodology, distinct from the simplified mechanisms illustrated in the public models.
07 — Delivered work
Technical specification & operating procedures
Mission decomposition, analytical architecture, evidence model, formulas, schemas, stage interfaces, geospatial workflows, uncertainty treatment, governance, validation requirements, analyst review, and decision documentation.
Representative prototype components
Workflow and geoprocessing components in KNIME, Python, ArcPy, and Esri tools, alongside analyst-interface concepts.
Analyst, scientific & leadership communication
Workflow and evidence interpretation for analysts; assumptions, uncertainty, and validation for scientific reviewers; operational decision and development priorities for leadership.
Pilot design
Proposed study area, parcel-data foundation, calibration questions, and evaluation priorities for progressing from specification and prototypes to empirical assessment.
Together, these deliverables established an implementation and evaluation baseline for the methodology.
Technical Analyses
- 01Ownership InferenceBayesian networks · Positive-unlabeled learning · Evidential reasoning
- 02Physics-Informed Buffers & MTEBBSpatial screening · Terrain modeling · Sensitivity analysis
- 03Probabilistic ViewshedMonte Carlo simulation · Correlated terrain error · Line of sight
- 04Entity Resolution & Knowledge GraphFellegi–Sunter · Graph-based machine learning · Bi-temporal provenance
- 05Coordinated Acquisition DetectionSpatial-lag models · Local Moran’s I · False-discovery control
- 06Remote-Sensing Change DetectionMultitemporal imagery · Change detection · Baseline calibration