Mining Flood Insurance Big Data to Incorporate Behavioural and Social Aspects in Flood Risk Modelling

Human behavior has shown to affect the exposure and vulnerability components of flood risk. Better understanding these human-flood dynamics is key to incorporate behavioural and social aspects in flood risk modelling. Yet, existing research on human behavior before, during, and after flood events is predominantly limited to site- and event-specific survey data or conceptual psychological theories. The recent availability of large-scale databases opened up new empirical bases to explore the interactions between individual/societal behaviors and flood risk at different scales. The US Federal Emergency Management Agency has recently released country-wide household-scale data on flood insurance policies in-force since 2009. Here, we consider flood insurance purchase as a proxy of flood resilience, awareness, and preparedness, which are drivers of flood vulnerability. We investigate relationships between people's socio-economic profiles and their probability of purchasing a flood insurance. We develop a gradient boosting framework with tree-based learning algorithms to model insurance coverage, on a census tract level, as a function of 390 socio-economic variables from the American Community Survey (e.g., household income, level of education, housing costs), as well as variables on previous flood experience based on flood insurance transaction data. SHAP (SHapley Additive exPlanations) are also analyzed to find the marginal contribution of each input and narrow down the vast range of features. Preliminary model predictions show heterogeneous performances with R values up to 0.75 during testing. Further analysis will enable us to identify the main determinants of flood insurance purchase throughout different states and social backgrounds. Understanding the factors driving peoples choices regarding flood insurance purchase is the first step to improve the National Flood Insurance Program's strategies and address societal inequalities in disaster risk management.