Mickey Warner and Bruno Sanso
07/24/2018 10:36 AM
Statistics
Climate models predict an intensification of extreme weather pat- terns, thus it is important to assess how similar the extreme behavior in climate model simulations is to that of observations. We fit a Bayesian hierarchical threshold exceedance model to simulations from the climate model CanCM4. Three simulation classes are analyzed and compared— decadal, historical, and pre-industrial control—as well as an observation product. To assess the extremes of the series considered we fit a gener- alized Pareto model to the exceedances over a threshold. Our method includes a likelihood-based hierarchical model for declustering. We find that in most domains, the distributions of the simulations have a tail behavior in agreement among themselves and with the observations. In order to study the joint tail behavior of simulations and observations, we perform a bivariate extreme value analysis using simple Pareto pro- cesses in conjunction with a Bayesian non-parametric model of an an- gular measure. The results show weak to moderate tail dependence in nearly every comparison made.
