Controls on Flood Insurance Penetration over the contiguous United States

Flooding is among the most devastating natural hazards globally, inflicting widespread economic damage and affecting numerous communities each year. Despite the availability of the National Flood Insurance Program in the United States, challenges such as low participation, poor risk assessment, and financial unsustainability persist. Understanding the drivers behind flood insurance penetration is thus crucial for resilience planning and risk mitigation. To comprehensively assess the factors influencing the willingness to purchase flood insurance, we utilized a dataset comprising 37 features. These features fall into four main categories: geo-hydrology, socio-economic, claim-related, and property-specific, and span the years 2010 to 2019. However, traditional one-step models prioritize geo-hydrological factors like property elevation and water proximity, thus underrepresenting socio-economic aspects. Here we introduce a two-step Random Forest model: the first model focuses on geo-hydrological features, and its residuals inform the second model, which includes all features. The proposed model demonstrated satisfactory performance, as evidenced by an RMSE of 63 and an R-squared value of 0.85 on an independent dataset. We further quantified model uncertainty through additional metrics, including confidence intervals for the predictions. Moreover, we conducted a quantitative assessment of each predictor's contribution to the overall model. Notably, we identified specific features with substantial potential to enhance flood insurance penetration rates, such as historical claim frequency, house age, and household income. These results underscore the model's robustness to adequately weigh the role of socio-economic drivers that were previously overshadowed. Such insights are instrumental for the development of tailored insurance products, particularly benefiting marginalized populations.