Demand-oriented regionalization with local data

Climate adaptation planning often relies on census-based or neighborhood boundaries, yet such fixed units seldom match the diverse and evolving problems that interventions seek to resolve. This underscores the necessity of designing demand-oriented regionalization that is problem-specific and responsive to local priorities. Traditional regionalization methods, however, struggle to balance socioeconomic and environmental variables while maintaining spatial coherence to meet practical and physical constraints. To address these shortcomings, we present RepSC-SOM, a Representative-initialized, Spatially Constrained Self-Organizing Map that extends traditional SOM with representative-based initialization, adaptive geographic filtering, and region-growing refinement. The method maximizes within-region similarity and between-region dissimilarity while maintaining spatial coherence. Notably, the framework is designed to enhance transparency and interpretability by standardizing how planning regions are defined, reducing reliance on subjective or historically bounded spatial units. Applied to flood-induced water contamination (E. coli concentrations) in Jacksonville, FL, RepSC-SOM-generated regions provided a more accurate characterization of the problem than standard units of analysis. Specifically, these regions achieved a higher average pairwise difference (223 per 100mL) compared to census tracts (154 per 100mL, [Formula: see text]), traffic analysis zones (118 per 100mL, [Formula: see text]), and neighborhoods (145 per 100mL, [Formula: see text]), indicating accurate detection and more precise delineation of contamination hot spots. These results suggest a strong potential for applying RepSC-SOM in real planning contexts to guide targeted interventions, prioritize resource allocation, and support coordinated climate adaptation strategies.