Natural Hazards Research Summit 2022: Data-driven machine learning predictions of multi-hazard hurricane damage to buildings

The study presented here uses a data-driven machine learning framework that incorporates loading onto buildings via hurricane wind and surge intensity, load propagation through geospatial features, and load resistance dependent on building components. This framework was applied to a case study of hurricanes Harvey, Irma, Michael, and Laura, which draws from reconnaissance data and observed hazard intensities to hindcast categorical damage states to buildings affected by these hurricanes with 76% accuracy. Variations of the case study further demonstrate favorable performance compared to existing frameworks, identify which features enable accurate predictions, and demonstrate the capacity for increased performance as more reconnaissance data becomes available.