Assessing typhoon damages to Taiwan in the recent decade: return period analysis and loss prediction

Devastating typhoons that induce enormous losses to various sectors of the economy underline the importance of an improved understanding of the regional hazard-to-loss relationship. This study utilizes the up-to-date loss data of typhoons in Taiwan from 2006 to 2015 to analyze the interannual variations in the annual aggregate losses (AALs) and develop a loss prediction model for the major administrative divisions. Return period analysis applied to the AALs identifies western-to-southwestern Taiwan as the high-risk region, among which Chiayi and Pingtung exhibit the highest 10-year AALs over 100 million. The gamma hurdle model (GHM) is adopted for loss prediction for its ability to stepwise model the loss occurrence and amount, leading to straightforward discussion regarding the explanatory power and statistical significance of meteorological predictors in their marginal and joint space. In the first part of the GHM, maximum daily rainfall and maximum gust wind are selected as the two most significant meteorological predictors for the logistic regression model of the loss occurrence, showing a remarkable model accuracy of $${\sim 0.9}$$ ∼ 0.9 . In the second part of the GHM, maximum sustained wind is added to the gamma generalized linear model of the loss amount, generating the cross-validated Nash–Sutcliffe efficiency (mean absolute error) values higher (lower) than 0.6 (3 million) for several southwestern cities. Event assessment for Typhoons Soudelor (2015) and Morakot (2009) further demonstrates the utility of the GHM and illustrates the essential for accounting for the combination effect of rainfall and wind on loss estimation.