Hybrid Supervised-Unsupervised Modeling for Post-Hurricane Private Well Contamination Risk Score Using Empirical...

Hurricane-related flooding can mobilize microbial and chemical contaminants, while limited well testing and uneven disaster response capacity leave many households relying on private wells at elevated risk, particularly during the early recovery phase when contamination risk is acute. To address this challenge, we propose a data-driven framework for quantifying post-hurricane private well contamination risk. The framework produces a numerical risk score by integrating 78 geospatial variables across three modules representing hazard, physical vulnerability, and social capacity. The score is constructed using a hybrid approach that combines supervised and unsupervised learning to generate interpretable indices grounded in theory and calibrated to real-world data. We applied the framework to western North Carolina following Hurricane Helene and evaluated its performance using post-hurricane well testing data from the North Carolina Department of Health and Human Services together with community-informed assessments in two counties. Higher risk scores were significantly associated with increased total coliform contamination (p = 0.006, Wilcoxon rank-sum test), demonstrating the value of the framework for identifying areas with elevated contamination risk. These findings suggest that the framework can help identify areas with elevated contamination risk following extreme weather events, although predictive performance remains moderate and further systematic evaluation is needed. The framework is designed to be transferable and can be adapted to other storms and regions where geospatial and well testing data are available. Overall, this work provides a practical, data-informed tool to support disaster preparedness, prioritize well testing, and protect private well users after extreme weather events.