Geostatistical software - geoR and geoRglm

The packages geoR and geoRglm are contributed packages to the statistical software system R, implementing methods for geostatistical data analysis. Diggle, Ribeiro Jr. and Christensen (2003) provides an introduction to the modelling and theory behind these two packages. In this paper we focus on the capabilities of the packages, the computational implementation and related issues, and indicate directions for future developments. geoR implements methods for models where the sampling distribution is Gaussian with extensions to the family of Box-Cox transformations. The package includes functions for reading and preparing the data, exploratory analysis, inference on model parameters including variogram based and likelihood based methods, and spatial interpolation. The generic term kriging is used in the geostatistical literature in connection with several methods of spatial interpolation/prediction. geoR implements some of the classical “kriging flavours” such as simple, ordinary, universal and external trend kriging and includes algorithms for conditional simulation. The package also implements Bayesian methods which take the parameter uncertainty into account when predicting at specified locations. The classical geostatistical model assumes that data are normally distributed, which may be an unrealistic assumption for some data sets. The package geoRglm is an extension of geoR for inference in generalised linear spatial models using Markov chain Monte Carlo (MCMC) methods. geoRglm implements conditional simulation and Bayesian inference for the Poisson and Binomial models. ∗Dept Estat́ıstica, Universidade Federal do Paraná, Brasil, E-mail: [email protected] †Bioinformatics Research Center, Aarhus Universitet, Denmark, E-mail: [email protected] ‡Dept. Mathematics and Statistics, Lancaster University, UK, E-mail: [email protected] DSC 2003 Working Papers 2