A comparison of methods for estimating extreme hurricane winds

A statistical model from extreme value theory has been used to estimate the return period of catastrophic coastal winds conditioned on global temperature (Jagger and Elsner 2006). The occurrence of a hurricane above a specified threshold intensity level is assumed to follow a Poisson distribution and the distribution of the maximum wind is assumed to follow a generalized Pareto distribution. This method is useful because it provides a parametric distribution for the return level (storm intensity) as a function of return period. However, the method can be difficult to implement because the multiple parameters have to be regressed on the predictor set and there is not a straightforward interpretation of the parameter values with regard to the predictors. Here we provide an alternative approach using conditional quantiles. This method allows us to more easily interpret the return level in terms of the parameters. Since the parameters are functions of the return periods, the method allows us to see how the relationship changes as the return period varies. The disadvantage is that the method is distribution-free and does not take advantage of extreme value theory in determining the parameters. The talk will compare and contrast the results of these two methods for a set of data from the U.S. coast. A Bayesian model that exploits the advantages of both methods will be demonstrated.