Multivariate statistical “unmixing” of Indian and Pacific Ocean sediment provenance

Abstract

The geochemistry of marine sediment is a massive archive of (paleo)oceanographic information. Accessing that information requires “unmixing” the various influences on marine sediment geochemistry to understand individual sources and marine geochemical processes. Q-mode factor analysis (QFA) and independent component analysis (ICA) are multivariate statistical techniques that have successfully been applied to large datasets of marine sediment element concentrations to identify the number and composition of marine sediment sources or end-members. In this study, we apply both techniques to two datasets of marine sediment geochemistry, compare the output, and discuss the advantages of each approach. In both datasets, ICA identified a mixing trend between carbonates and dust, whereas QFA represented the end-members as two separate factors. In the Pacific and Indian Oceans dataset, both techniques produced three factors or independent components involving rare earth elements, but two of the QFA factors explained a small, almost negligible, amount of the variability of the dataset. Also, QFA identified more aluminosilicate end-members (dust or volcanic ash) than ICA. In the Indian Ocean Sites 738 and 752 dataset, ICA identified two processes affecting Sr and Ba concentrations as separate independent components, while QFA created a factor representing the covariation of Sr and Ba over intervals of the site's paleoceanographic history. The results of this study exemplify that QFA identifies covariances and finds discrete end-members contributing to the bulk mass of sediment. ICA works best with non-Gaussian element distributions and finds geochemical signals and mixing trends that constitute the characteristic structure of the multielemental data.