A probabilistic framework to construct tropical cyclone loss models for building portfolios

Abstract

Tropical cyclones (TCs) evolve over time and space and can cause substantial damage to building portfolios. Therefore, timely and accurate TC damage assessment is essential for effective risk management. One practical approach is to establish a relationship between hazard intensity (e.g., wind speeds) and regional damage. However, when the study area is large, spatial heterogeneity, such as clustered building distributions, terrain variability, and spatial variations in wind speeds, can hinder accurate modelling of the hazard-damage relationship. To address this challenge, the present study employs a spatial clustering algorithm to divide the entire area into multiple sub-regions with relatively homogeneous characteristics. For each sub-region, a TC loss model is developed as a function of wind speed at the sub-regional centroid and the corresponding building portfolio loss ratio. In practice, losses in all sub-regions are first assessed individually and then aggregated to estimate the total regional loss. This divide-and-aggregate approach significantly improves the accuracy and applicability of TC loss modelling and can be readily applied to various contexts, such as long-term risk management in large-scale communities.