Application of singular value decomposition (SVD) and semi-discrete decomposition (SDD) techniques in clustering of geochemical data: an environmental study in central Iran

Common multivariate clustering techniques are ineffective in identifying subtle patterns of correlation, and clustering of variables or samples within complex geochemical datasets. This study compares the combination of singular value decomposition (SVD) and semi discrete decomposition (SDD), with that of hierarchical cluster analysis (HCA), to examine patterns within a multielement soil geochemical dataset from an agricultural area in the vicinity of Pb–Zn mining operations in central Iran. SVD was used to both identify patterns of correlation between variables and samples and to “denoise” the data, and SDD to simultaneously cluster the samples and variables. The results reveal various spatial associations of mining waste-associated metals As, Ba, Pb and Zn, and within the remaining elements whose distribution is largely controlled by the major oxides. SVD–SDD was found to be superior to HCA, in its ability to detect subtle clusters in soil geochemistry indicative of mine-related contamination in the study area.