Patterns in US flood vulnerability revealed from flood insurance "big data"

Vulnerability functions are applied in flood risk models to calculate the losses incurred when a flood interacts with the built environment. Typically, these take the form of relative depth–damage relationships: flood depths at the location of a particular asset translate to a damage expressed as a certain percentage of its value. Vulnerability functions are a core component of risk and insurance industry catastrophe (CAT) models, permitting physical models of flood inundation under different scenarios (e.g. certain probabilities) to be translated to more tangible and useful estimates of loss. Much attention is devoted to the physical hazard component of flood risk models, but the final vulnerability component has historically received less attention — despite quantifications of risk being highly sensitive to these uncertain depth–damage functions. For the case of US flood risk models, ‘off-the-shelf’ functions from the US Army Corps of Engineers (USACE) are commonly used. In an analysis of roughly 2 million flood claims under the US National Flood Insurance Programme (NFIP), we find these ubiquitous USACE functions are not reflective of real damages at specified flood depths experienced by policy holders of the NFIP. Particularly for smaller flood depths (<1m), the majority of structural damages are of <10% of the building value compared to the 30–50% stipulated by the USACE functions. A deterministic relationship between depth and damage is shown to be invalid, with the claims data indicating damages at a certain depth form a beta distribution. Most reported damages are either <10% or >90% of building values, with the proportion of >90% damages increasing with water depth. The NFIP data also reveal that newer buildings tend to be more resilient (lower damages for a given depth), surface water flooding to be more damaging than fluvial flooding for a given depth, vulnerability to vary dramatically across space, and even the concept of a relative damage to be untenable in its application to expensive properties (e.g. even for depths >1m, properties worth >$250k rarely experience losses >20% of their value). The findings of this study have significant implications for developers of flood risk models, suggesting current estimates of US flood risk (in $ terms) may be substantial over-estimates.