Text

theses

This dissertation explored the sorption of persistent organic pollutants onto different types of sediments at two Superfund sites. The compounds investigated were Polycyclic Aromatic Hydrocarbons (PAHs) and Polychlorinated Biphenyls (PCBs) which are some of the primary contaminants of potential concern (COPCs) driving the choice of remedy at the two sites, the Portland Harbor Superfund Site (PHSS) in Oregon and the Upper Hudson River (UHR) portion of the Hudson River Superfund Site in New York. This dissertation was a data mining exercise utilizing the large databases on contaminant concentrations and sediment characteristics at these two sites. The data on contaminant concentrations in sediment had previously been analyzed via Positive Matrix Factorization (PMF) to identify source types. The sediment characteristics considered were percent solids, fines, sand, and gravel, as well as total organic carbon (TOC), water content, and porosity. The concentrations of the contaminant source types generated by PMF were compared with the sediment characteristics using statistical Bayesian inferences. To our knowledge, this is the first study to combine PMF and Bayesian analysis in this way. Multivariate and Seemingly Unrelated Regression (SUR) Bayesian models were used. The dissertation began with the simplest case of PAHs at the PHSS (Chapter 2). The PHSS is a single hydraulic unit, so all data from this site could be analyzed in one data set. PAHs were chosen as the first contaminant because their concentrations are high and their sources are ubiquitous. The Bayesian analysis suggested that sorption of lower molecular weight PAH source types was strongly driven by TOC, but the sorption behavior of PAHs arising from urban background/runoff was more complicated. Chapter 3 examined the slightly more complicated case of PCBs at the PHSS, since PCBs were present there at lower concentrations than the PAHs and they were used at a smaller number of sites along the river. The Bayesian inference results indicated that none of the PCB source types are significantly correlated with TOC. Instead, the fines and sand particles were negatively correlated with most of the PCB source types. Bayesian inference therefore suggested that TOC was not the master variables that determines the extent of sorption for PCBs at the PHSS. Chapter 4 considered a much more complicated scenario: PCBs in the UHR. This analysis was more complicated due to the lack of hydraulic connection between the various reaches of the UHR, which necessitated the division of the data set into subsets representing various reaches. Because these reaches are not truly independent of each other, Seemingly Unrelated Regression was used to investigate PCB distribution across the various sediment types in the reaches. The results supported the prevailing understanding that PCB sorption was driven primarily by TOC. No obvious differences in sorption behavior were observed for dechlorination products versus unweathered source PCBs, suggesting that TOC is not an important driver of dechlorination. Also, there were no obvious differences in the distribution of the non-GE PCB source term across the sediment types relative to GE-derived PCBs, suggesting that they enter the river and then redistribute across the sediment fractions. Taken as a whole, this dissertation demonstrates that the combination of field data, PMF, and Bayesian statistics is an effective approach in analyzing these data sets and did yield useful insights about sorption of hydrophobic organic contaminants to sediments at these sites. However, this approach works best when the data sets are large, i.e. more than about 200 discrete sediment samples in which all the parameters were measured. This condition was met for the PHSS, and the analysis of this data successfully identified sources and processes at that site. In contrast, the number of samples available for the UHR was smaller, with only about 100 samples, and these samples were divided across eight hydraulically distinct reaches. This meant that data set was further subdivided, resulting in data sets with only about 20 samples, which limited the utility of this analysis. Generally, the results of this analysis were consistent with the prevailing understanding that sorption of hydrophobic contaminants is driven by TOC. However, some of the results suggested that TOC was not the only determinant of sorption. Other factors may also be relevant. Some of these were measured as part of the routine Superfund investigations, including grain size distribution, total solids, and porosity. Other factors that might be important to sorption were not measured. The results from Chapter 2, which examined the distribution of PAHs across sediments in the PHSS, hinted that perhaps black carbon (also called soot carbon or elemental carbon) may be an important sorbent, but this is not typically measured.

http://dissertations.umi.com/gsnb.rutgers:11897

Ph.D.

Includes bibliographical references

doi:10.7282/t3-0ch6-jk80