Multivariate analysis for geochemical process identification using stream sediment geochemical data: A perspective from compositional data

Identification of underlying geochemical processes based on samples of different types such as stream sediments, soils, and water is important for a range of applications including mineral exploration, land use planning, and environmental assessment of both natural and anthropogenic factors. However, almost all geochemical compositions of these samples are subject two limitations: outliers and the data closure effect. In the present study, bivariate relationships between selected major elements are examined to illustrate their spurious correlation by using centered log ratio (clr) transformation. In addition, robust factor analysis (FA) and compositional data analysis are used to prevent the effect of outliers and to reduce the influence of data closure in the identification of geochemical processes. First, a k-means algorithm is applied to partition geochemical data into three clusters to enhance the interpretation of the geochemical data. Then, robust FA is applied to log ratio-transformed geochemical data. The first five factors are extracted on the basis of the scree plot of eigenvalues. The results indicate that robust FA applied to log ratio-transformed data can be used to effectively identify geochemical processes and to determine the extent of anthropogenic and natural influences such as mineralization, weathering and diagenesis, heavy metal accumulation or contamination, or a combination of these factors. Several geochemical processes are indicated by the first five factors, explained as follows: (a) F1 reflects granitic rocks and natural or industrial contamination by Cu, Ni, Sb, As, Cd, and Cr; (b) F2 reflects W polymetallic mineralization; (c) F3 reflects Au anomalies and heavy metal contamination by Zn, Cd, Mn, and Pb; (d) F4 reflects Mo and Au anomalies; and (e) F5 reflects Ag-W-Be-La mineralization and heavy metal contamination by Hg and Sb.