Climate data and machine learning integration for evaluating flood insurance risk patterns

Flood events, exacerbated by climate variability, pose significant challenges to flood risk management and the insurance industry in the United States. To enhance flood risk modeling strategies, this study employs machine learning to predict regions prone to high flood insurance claims by integrating hydrological, meteorological, and socio-economic data. We combine the FEMA NFIP Redacted Claims dataset, detailing over 2.5 million flood-related insurance claims, with the US-CAMELS streamflow dataset, offering rich hydrological insights across numerous catchments in the USA. A key focus is the influence of climate indices, such as the El Niño-Southern Oscillation (ENSO), on flood patterns. Using the Oceanic Niño Index (ONI) as a quantitative metric, we explore the spatiotemporal relationship between ENSO phases, streamflow variability, and flood insurance claims. The analysis considers the geographic proximity of the study regions to hydrographic networks and coastal areas, where flood risks are often heightened due to complex interactions between inland and coastal processes. Furthermore, machine learning models are employed to identify the attributes driving flood vulnerability. Predictors include climate indices, basin characteristics, streamflow patterns, and historical claims data. This integrated approach aims to develop a predictive framework that enhances flood early warning systems and informs policy-making for targeted risk mitigation. By quantifying the connections between large-scale climate phenomena, regional hydrology, and localized flood risks, this research provides a pathway for advancing flood insurance risk assessment and improving resilience to hydroclimate-driven hazards. Results will be showcased with a case study from the USA, emphasizing the applicability of machine learning in data-driven flood risk management.