Ginninderra revisited: the contribution of B. E. Butler to statistical pedology

Abstract

The 1960s were a decade of controversy like no other in soil systematics. Bruce E. Butler had recently moved to Canberra and was perplexed by the apparent lack of ‘orderliness’ in the soil patterns on the Tablelands there. He sought elucidation by the then newly developing methods in multivariate and spatial statistics. Two surveys of soil on the Ginninderra Experiment Station, 12 km north of Canberra, were planned to provide data for analysis. The first was made on a grid at ~180-m intervals, at the nodes of which the soil profiles were described and samples taken for chemical analysis. Correlations between variables were weak, a principal coordinate analysis showed the scatter of sampling points as a single cloud, and classification proved to be of little value in predicting individual variables. The second survey had a spatially nested design with distances 5, 18, 56, and 180 m, leading to the estimation of the components of variance of several topsoil properties at those distances. The accumulated components provided rough variograms with contrasting patterns; there was little coherence. Almost all the variance in soil potassium (K), for example, occurs within 180 m, whereas some 35% of the variance in phosphorus (P) occurs over longer distances. This paper revisits that research, emphasising its innovation and adding a modern geostatistical analysis from the original grid data. The conventional variogram of P has an estimated correlation range of 728 m, and a map made by kriging shows a coarse pattern of variation. In contrast, the variogram of K appears as all nugget at the working scale, and so kriging or any form of interpolation from 180-m grids is not sensible. Denser sampling is needed, and proximal sensing by gamma-ray spectrometry seems a promising approach.