|
Abstract:
|
Areal data is an important subclass of spatial data, e.g. public health data at the county level or gridded climate model output. However, current models for spatial extremes which characterize spatial tail dependence, such as max-stable models, are geostatistical in nature and have proven to be difficult to fit to spatial datasets with many locations. In classical spatial statistics, the simultaneous autoregressive (SAR) model for areal data constructs a simple spatial model which captures spatial dependence given a neighborhood structure. We apply recent results on transformed linear operations for regularly varying random vectors with tail index ? = 2 (Cooley and Thibaud, 2016) to develop an analogous SAR model for extremes. We will describe the model and how to simulate from it, the resulting tail pairwise dependence matrix which depends on the neighborhood structure and autoregressive parameter, and discuss preliminary methods for estimation and inference.
|
