Jianming Yin, Tokio Marine Technologies LLC Bo Yu, Tokio Marine Technologies LLC
Severe thunderstorms can occur anywhere in Australia. From 1967 to 1999, severe thunderstorms have been estimated to represent more than one quarter of the average annual cost of natural disasters in Australia. Severe thunderstorm risk is a significant issue for the insurance industry: paid insurance claims for severe thunderstorm damage are greater than those for tropical cyclones, earthquakes, or bushfires in Australia. Therefore, Re/insurance companies need reliable models to analyze and quantify severe thunderstorm risk in Australia. After reviewing the quality of available Australian severe thunderstorm data required for simulation of severe thunderstorm risks, historical thunderstorm reports from 1999 through 2009 from the Australian Bureau of Meteorology (BoM) are used to model the risk. Monte Carlo technique is employed to simulate the stochastic severe thunderstorms. In this endeavor we consider three major perils associated with a severe thunderstorm, that is, hailstorm with hailstones of diameter of 20 mm or more, downburst straight-line wind, and tornado. To allow companies to assess their reinsurance needs at a portfolio level, the concept of severe thunderstorm event is introduced and defined as a congregation of individual hailstorms, straight-line winds and tornadoes spawned by the same convective precipitation system moving across Australia within a continuous 7-day timeframe and within a circular area of 1,000-km in radius. The simulation process starts with simulating the probable number of severe thunderstorm events in a simulation year by sampling from a Negative Binomial (NB) distribution fitted by the historical annual thunderstorm events. The starting date of a stochastic thunderstorm event is sampled with equal probability from the seasonality of the BoM historical events. The sampled candidate date is then perturbed using a Gaussian distribution with standard deviation of one day. This sampling approach reflects the seasonal patterns of BoM historical thunderstorms. The location of a stochastic thunderstorm is simulated by perturbing the location of the candidate historical thunderstorm in both longitude and latitude directions using a bivariate Gaussian distribution. The stochastic thunderstorm path length, width, orientation, and intensity are simulated using different statistical distributions fitted by the BoM historical thunderstorm data. The same simulation process repeats for all simulation years to simulate up to 100,000-year worth of future possible severe thunderstorm events in Australia. Comparisons show that the simulated climatology of severe thunderstorm risk matches reasonably well with the BoM historical data. The severe thunderstorm model developed in this study provides a reasonable quantification of the severe thunderstorm risk in Australia from Re/insurance perspectives. Coupled with the severe thunderstorm damage modeling and insurance portfolio exposures, the model can help re/insurance companies to quantify the severe thunderstorm loss potentials in Australia at both individual location and at portfolio levels.