Student Seminar Series: Robert Yuen, A Gauss-Pareto process model for spatial prediction of extreme precipitation
In order to develop adaptive strategies for dealing with consequences of extreme precipitation such as insufficient drainage and various aspects of flooding, it is necessary to be able to estimate extremes at unobserved sites. We introduce a hierarchical Gauss-Pareto model for spatial prediction of precipitation given nearby observations that are extreme. The model belongs to the max-domain of attraction of popular Brown-Resnick max-stable processes (Brown and Resnick, 1977; Kabluchko et al., 2009) and retains the essential dependence structure of their corresponding generalized Pareto processes (Ferreira and DeHaan, 2012). An MCMC algorithm is developed for inference. The algorithm allows for left censored data from precipitation that accumulates below instrument precision, which often happens despite nearby observations that are extreme. The model and methodology is applied to summer extreme 24 hour cumulative precipitation over south central Sweden. We discuss some extensions and challenges for future work.