Evaluating the impact of climate change on hurricane wind risk: A machine learning approach.

Abstract

In the residential sector, hurricane winds are a major contributor to storm-related losses, with substantial annual costs to the US economy. With the potential increase in hurricane intensity in changing climate conditions, hurricane impacts are expected to worsen. Current hurricane risk management practices are based on the hurricane risk assessment without considering climate impact, which would result in a higher level of risk for the built environment than expected. It is crucial to investigate the impact of climate change on hurricane risk to develop effective hurricane risk management strategies. However, investigation of future hurricane risk can be very time-consuming because of the high resolution of the models for climate-dependent hazard simulation and regional loss assessment. This study aims to investigate the climate change impact on hurricane wind risk on residential buildings across the southeastern US coastal states. To address the challenge of computational inefficiency, we develop surrogate models using machine learning techniques for evaluating wind and rain-ingress losses of simulated climate-dependent hurricane scenarios. We collect historical hurricane data and use selected climate variables to predict changing hurricane attributes under climate change. We build the surrogate loss model using data generated by the existing fragility-based loss model. The loss estimation of synthetic events using the surrogate model shows an accuracy with a 0.78 R-squared value compared to Hazard U.S. - Multi Hazard (HAZUS-MH) estimation. The results demonstrate the feasibility of utilizing surrogate models to predict risk changes and underline the increasing hurricane wind risk due to climate change.