Key points are not available for this paper at this time.
Summary Data for 85 soil profiles, located at random within 17 physiographic units in a 1,000 km 2 area near Oxford, and each characterized by 37 properties, were analysed by two sets of multivariate techniques. Principal component analysis yielded two components accounting for 44 per cent of the total variation, and the distribution of the profiles along these components is plotted as a scatter diagram (ordination). Similarity analysis and hierarchical grouping (classification) were performed using three different similarity indices, two different standardization techniques, and three different sorting strategies. Before analysis the profiles were classified by two pedologists, independently using their judgement. For the Brown Earths there was moderately good agreement between the results of the two prior classifications and between them and those produced by the several numerical procedures. The degree of clustering was slight and much less than imagined prior to analysis. For the Gleys there was no evident clustering. Different numerical procedures gave different results, as also did the two prior classifications. Pedologists faced with difficult classificatory decisions cannot look to any hierarchical clustering strategy as arbiter, though they should get guidance from a principal component plot.
He et al. (Tue,) studied this question.