In search of spatial extremes.

Abstract

In environmental studies attention increasingly focuses on identification of spatial extremes: locations with observations that are apparently higher than either a preset background threshold or neighbouring observations. We consider various procedures for identifying values and locations of these extremes: extreme value theory, conditionally simulated fields and disjunctive kriging. In a recent research project we studied the distribution of nine environmental pollutants (heavy metals, polyaromatic hydrocarbons and mineral oil) in a large industrial estate in the southern Netherlands. This area is interesting as it is influenced by three contaminating processes: inundation by the river Meuse, anthropogenic heightening with contaminated materials and local industrial activities. Concentrations are measured. In this paper the peaks over threshold method was useful for finding extremes for spatially uncorrelated variables, whereas conditional simulations applied in particular to areas with spatial dependence.