Interpretable GeoAI for smart urbanism in data-scarce environments: Spatial assessment of cardiometabolic resilience in heat-vulnerable cities

Abstract

Advancing anticipatory smart urbanism in data-scarce environments requires interpretable methodologies capable of translating limited urban data into actionable insights. This study proposes a GeoAI-driven framework for the spatial modelling of cardiometabolic resilience in heat-vulnerable cities, integrating unsupervised machine learning with knowledge-based reasoning to decode the urban exposome. Applied to San José, Costa Rica, the framework reveals pronounced spatial inequalities, identifying a North-South polarisation driven by a “geography of double exclusion,” where tropical thermal stress (characterised by land surface temperatures exceeding 36 °C, high humidity, and socio-environmental vulnerability) intensifies cardiometabolic risk. Moving beyond conventional descriptive analytics, the approach establishes an evidence-based, anticipatory decision-support system that operates effectively under data scarcity. By constructing a foundational layer of mathematical infrastructure, the methodology combines Fuzzy-AHP logic, K-means clustering, and Principal Component Analysis (PCA) to integrate heterogeneous indicators, reduce dimensionality, and identify critical urban hotspots. This hybrid GeoAI architecture enables the anticipation of risk patterns and supports evidence-based urban interventions. The results demonstrate how diagnostic spatial modelling can inform optimal sensor placement strategies (AI Readiness), guide targeted mitigation policies, and enhance resilience planning in resource-constrained settings. The proposed framework positions GeoAI as a strategic enabler of equitable and sustainable smart city transitions, particularly in the Global South, where data limitations and climate vulnerability intersect most acutely.