Chemometric mapping of polychlorinated dibenzo-p-dioxin (PCDD) and dibenzofuran (PCDF) congeners from the Passaic River, NJ: Integrated application of RSIMCA, PVA, and t-SNE

Abstract

Robust Independent Modelling of Class Analogy (RSIMCA) was applied to classify over 2,800 Passaic River sediment samples into seven groups or not assigned to any group after an initial screening of a 3,255 sample dataset. This multivariate statistical output was compressed from seven latent dimensions into two interpretable dimensions using t-Distributed Stochastic Neighbor Embedding (t-SNE) graphics. Polytopic Vector Analysis (PVA) was then used to identify distinct source end-members based on PCDD/F characteristics of the classified samples. Among several advantages, the integrated chemometrics approach 1) applies emerging data visualization tools in this “Big Data” era to retain the fidelity of high-dimensional data attributes of a chemical dataset spanning over two decades of sample collection; 2) employs a classification technique undisturbed by compositional outliers yet tracks those for subsequent investigations; 3) provides an intuitive reduced-dimensional data visualization map for the PVA mixing polytope solution; 4) fills a data gap in the contextual inventory of PCDD/F source dynamics in a complex river system; and 5) serves as a backdrop for further forensics investigations of the finer structure of less dominant point sources and potential upland source end-members in sediments. This tiered chemometrics strategy provides a strong weight-of-evidence approach to the interpretation of sediment data.