Cluster and Principal Component Analysis of Soil Sites Along a Transect

Abstract

The grouping of soils with similar properties is one of the primary objectives of soil surveyors. We compared the groupings of 89 soil sites as classified by the U.S. system of soil taxonomy and by principal component and cluster analysis. The sample sites, examined at four depth intervals, were equally spaced on a 2,700-m transect in southern New Mexico. Ten laboratory-determined variables: clay, calcium carbonate, soil pH, coarse fragments, organic carbon (OC) and five sand fractions, were used to characterize each sample. The principal component analysis produced two components that accounted for 60, 53, 60, and 60% of total variation for the 0 to 30, 30 to 60, 60 to 90, and 90 to 120 cm depths, respectively. It was found that clay and sand contents were the major contributors to the first two principal components. Four cluster sorting strategies were used: (i) centroid, (ii) median, (iii) group average, and (iv) flexible. These sorting strategies all gave the same results by grouping the soil sites into two groups. We identified eight soil mapping units along the transect (one Vertisol, one Mollisol, and six Aridisols) using the traditional field methods of the Soil Survey Staff. Both the cluster analysis and the principal component analysis identified only two groups. The numerical methods separated the soils classified as Vertisols as one group, but grouped the soil classified as Mollisols and Aridisols together. Our sampling scheme was not designed to recognize the diagnostic horizons and other properties used as differentiating characteristics in the U.S. system of soil taxonomy. The results of numerical classification depend more on the strategy for data analysis and variables selected for analysis than on the particular numerical method.