Portfolio

Flashcards and slides do not make world geography stick, and the test format rewards short-term memorization over real geographic intuition. Geo 412 turns the problem into a game. Players are prompted with a country and must type the answer or find it on the world map, get instant feedback, and work through all of the countries in short, self-paced sessions. Because the answers are organized by geographic combatant command, cadets learn the operational geography they will actually be tested on and use as officers.

The game is a hand-built front end in JavaScript, HTML, and CSS with an interactive SVG world map, developed with AI-assisted coding in Claude Code, version-controlled on GitHub, and deployed to Microsoft Azure at its own domain. It is live and in use in the classroom, and it also supports my doctoral research into how AI can help educators design learning aids and effective teaching tools.

The first version of the game, a click-the-country map quiz, is open source on GitHub under the MIT License.

GEOINT education has no shared language for capability development. Programs define what to teach, not how thinking grows, and competency models are linear while GEOINT roles are not. The framework extends Jo and Bednarz's spatial thinking taxonomy from the classroom into a professional capability space with three axes:

• Analytic Complexity — Descriptive → Relational / Patterned → Predictive / Anticipatory
• Representational Engagement (the Practitioner pathway) — Consume → Manipulate → Design
• Decision Support Level (the Consumer pathway) — Data Exploitation → Analytic Judgment → Decision Advantage

Together they form 27 distinct capability contexts. Every GEOINT role occupies a region of the cube; the framework does not prescribe where a program should be, it shows where it is. It was built from a systematic review of GEOINT education literature and workforce models, refined through expert consultation, and piloted against the USGIF certificate curriculum at the Air Force Academy.

The poster also walks through a four-step program assessment (inventory, map, read the gaps, define your audience) and is candid about limits: the cube is still conceptual. Next steps are cross-institutional curriculum mapping, a learner self-assessment instrument, and empirical validation.

Each scene was classified into water and land with a supervised K-nearest-neighbor classifier in ArcGIS Pro, checked with a stratified accuracy assessment, converted to polygons, and measured.

The classified surface area fell 47.5% across the study period, and the key finding is the timing: the reservoir was stable through 1999 and has been in continuous decline since.

Originally submitted for GEOG 883; revised for portfolio presentation with the analysis and figures unchanged.

Two variables from Colorado Department of Public Health & Environment open data were analyzed independently across all 620 census tracts: age-adjusted opioid and heroin mortality per 100,000 and the share of residents below the federal poverty line, both for 2015–2019. In GeoDa I built a queen-contiguity spatial weights matrix, ran global Moran's I with 999 permutations for each variable, and used Local Indicators of Spatial Association (LISA) to locate the hot spots, cold spots, and outliers driving each result.

Poverty proved roughly twice as spatially structured as overdose mortality (Moran's I of 0.445 versus 0.184). Poverty forms a coherent high-poverty core through central Denver and a broad low-poverty suburban ring, with 255 of 620 tracts in a significant local cluster. Mortality is far more diffuse: only 121 tracts are significant, and nearly four fifths of the metro reaches no significance threshold at all. To show where the two burdens coincide, each layer was reclassified to a one-to-five scale and summed with map algebra in ArcGIS Pro, producing a composite index that peaks tightly in central Denver.

The report is explicit about what it does not establish. Univariate Moran's I and LISA describe each variable on its own terms and do not test the poverty–mortality relationship, and tract-level co-occurrence cannot support claims about individuals. It closes with the appropriate next step, bivariate Moran's I and LISA on the same weights matrix, plus Empirical Bayes smoothing to tame unstable small-population rates and a re-run restricted to the urbanized core.

Originally submitted for GEO 586; revised for portfolio presentation with framing, statistics reporting, and cartography corrected and the underlying analysis unchanged.