Spatial regression models for extremes (Q1966378): Difference between revisions

The authors propose a methodology of analysis of extremes in meteorological data which combines the well known extreme value approach and recent achievements in nonlinear spatial statistics. This is important because meteorological data are often recorded at a number of spatial locations. The model structure is the following: at each location, conditional on model parameters, a standard extreme value model is assumed to characterize the data-generating process at extreme levels; across locations the unobservable model parameters are themselves a realization of a smooth stochastic spatial process. Thus, a smooth, but possible non-linear, spatial structure is an intrinsic feature of the model. Markov chain Monte Carlo techniques are used for the necessary computations. A brief overview on a point process approach and details of Markov chain Monte Carlo schemes are given. A simulation study is carried out to illustrate the performance and advantages of spatial models. Finally, the model is applied to data generated from a climatological model in order to characterize the hurricane climate of Gulf and Atlantic coasts of the United States.