Combining hazard, exposure and vulnerability data to predict historical United States hurricane losses

Hurricanes are among the most destructive natural hazards globally. The widely used Saffir–Simpson scale is an effective public-communication tool, but it is based on a single hazard quantity (wind speed) and has low skill in representing historical economic losses. Accurate risk assessment requires hazard, exposure, and vulnerability information. We present a statistical model to predict losses from North Atlantic hurricanes making landfall in the United States using optimally weighted, normalised-rank quantities describing hazard, exposure, and vulnerability. The model significantly outperforms single-parameter predictions, including landfall wind-speed maxima and central-pressure minima. Root-mean-square error between observed losses and losses predicted from landfall wind speed alone is USD 35.6 billion, which our model reduces to USD 7.0 billion. To improve the characterisation of risk, we introduce a loss-based “Hurricane Predictive Loss Scale” to more directly link hurricane characteristics and landfall to financial impacts. These results demonstrate that integrating exposure and vulnerability data with hazard observations yields skilful estimates of historical hurricane losses, and our approach may help assess how loss from a forecast landfall may rank among historical events. This work is applicable to other cyclone-prone regions and highlights the critical need for open-source exposure and vulnerability data to advance climate risk understanding.