Geostatistical Methods in Geography: Applications in Human Geography
Kerry, Ruth, Haining, Robert P., Oliver, Margaret A., Geographical Analysis
The special issue on geostatistical methods in geography has been split into two issues of Geographical Analysis. They comprise applications in human and physical geography.
These issues are timely in that they follow another special issue entitled "Celebrating 40 Years of Scientific Impacts by Cliff and Ord." Andrew Cliff and Keith Ord's (1969) paper, "The Problem of Spatial Autocorrelation," was a forerunner of part of the large field now known as spatial statistics, which today includes geostatistics (Cressie 1993). Interestingly, Cliff and Ord's work was fairly contemporaneous with George Matheron's development of geostatistics. These three journal issues, taken together, explore different methods of dealing with second-order behavior in spatial data; Cliff and Ord regard it as essentially a problem for classical statistical analyses, for example, in regression modeling, whereas for geostatisticians it lies at the heart of their analyses.
This double special issue came into being as the result of several discussions among ourselves as to why geographers do not make more use of geostatistical methods. Our aim in putting together this special issue therefore is to bring to the attention of geographers and spatial scientists the wide range of geostatistical techniques that have been developed, often in other disciplines, in order to show how these techniques can be used in both human and physical geography research and with different types of spatial data.
The first article in this issue reviews the history and development of geostatistical theory and methods, showing its relationship to other areas of spatial statistics more commonly used by human geographers and giving a researcher who is unfamiliar with geostatistics some important background. It highlights an academic lineage that is generally different from the spatial autocorrelation work of Cliff and Ord, but it also examines the similarities in the approaches. The "different lineage," notions about what kinds of data are suitable for geostatistical analysis, and an impression that geostatistics is synonymous with kriging, for example, are probably some of the reasons why geographers, particularly human geographers, have not paid much attention to geostatistics. Rather than spatial autocorrelation being regarded as a nuisance embodied in a single error term, in geostatistics, sampling schemes are designed so that spatial autocorrelation in georeferenced data can be quantified and modeled. The importance of the variogram for providing insight into the structure of spatial variation is illustrated, something that is not always appreciated when geostatistics is considered synonymous with kriging for interpolation. …