Probabilistic cyclone damage assessment on large spatially distributed civil infrastructure systems

Abstract

Tropical cyclones (TCs) cause severe economic losses worldwide each year, making accurate damage assessment essential for disaster response and long-term risk reduction. However, because TCs evolve dynamically across space and time, efficiently evaluating TC-induced damage at the regional scale remains challenging. In addition, ongoing climate change is expected to influence TC behaviour and associated coastal risks significantly.

This thesis addresses these challenges through three studies. The first study proposes a probabilistic framework for constructing TC loss models, defined as a function of regional wind speeds and the corresponding building loss ratios. To account for spatial heterogeneity in large study areas, a clustering-based spatial division approach is employed. Losses are first estimated for each sub-region and then aggregated to obtain total regional losses. The second study develops a Bayesian network (BN) model for predicting the spatial distribution of building losses. The model takes key TC parameters as inputs and outputs loss ratios for each subdivision. To improve predictive performance, supervised discretisation is applied to TC variables, and a clustering-based discretisation scheme is adopted for loss variables. The third study evaluates the potential impacts of climate change on future TC activity. A framework is developed to quantify projected changes in TC frequency, genesis location, and intensity using high-resolution global climate models. These projected characteristics are then used to assess coastal risk under future climate scenarios, and the individual and combined influences of the modified TC characteristics are analysed.