An Interactive Tool for Residual Diagnostics for Fitting Spatial Dependencies (with Implementation in R)

There exists a growing number of spatial statistics software packages covering a broad range of methods including the classical technique of kriging following variogram fitting (for a review see e.g. Bivand and Gebhardt, 2000). However, most of these programs do not contain tools for assessing the fit of particular models of the spatial dependencies. Based upon a variogram cloud estimation method proposed in Müller, 1999 we present various techniques for that purpose and their implementations in an R-package called vardiag. The paper concentrates on the interactive aspect of the problem worked out in the dissertation of Glatzer, 2002. For the interactive exploration of the fit several plots (a map view, the square-root-differences cloud and two types of residual plots) are linked. The proposed tool allows brushing of single points or sets of points in these plots. The selection of points in one plot is reflected by an automatic selection of the corresponding parts in the other plots. The program is exemplified on a data set of chlorid concentrations in the Südliche Tullnerfeld.