PCB or PAH sources and degradation in aquatic sediments determined by positive matrix factorization

Polychlorinated biphenyl (PCB) and polycyclic aromatic hydrocarbon (PAH) sources were determined using a new method of positive matrix factorization (PMF). Results from the new PMF method were compared to results from a factor analysis (FA) model and different PMF models. Two artificial data sets were developed to test and compare the different models. The three study areas were Lake Calumet (PAHs), Illinois, the Sheboygan River Inner Harbor (PCBs), Wisconsin, and Lake Hartwell (PCBs), South Carolina. PCBs in Sheboygan River sediments were determined in our lab. The primary sources of sediment PAHs to Lake Calumet were coke oven and traffic. The sediments of the Sheboygan River and Lake Hartwell were found to have dechlorinated versions of the original Aroclor mixtures, especially at high PCB concentrations.; The new PMF method is based on a method given in the literature except penalty terms are replaced by non-negative least squares (NNLS) rotations. The reasoning behind the method is to reproduce zero values in the source profile and contribution matrices more accurately. The method first calculates the source profile and source contribution matrices, generally containing negative elements, using an iterative least squares method until the change in the actual data matrix and the reproduced data matrix is small. Then the source profile and source contribution matrices are rotated using NNLS until the source profile matrix is not changing. Additional procedures such as scaling, backscaling, and normalization are performed. The source profile matrix can then be compared with literature source profiles to determine the actual sources. Diagnostic tools are incorporated to aid in determining the number of significant factors.; Two artificial data sets were developed to test the receptor models. Applications of all models to the artificial data sets were successful. For the first data set FA reproduced the source profiles the best. For the second data set, PMF with NNLS was able to reproduce zero value elements better than a different version of PMF that considered penalty terms.; This study is the first application of PMF to apportion PAHs and PCBs in sediments. The Lake Hartwell study is the first application of receptor modeling to distinguish a dechlorination plateau. The use of a preferential reaction order for dechlorination modeling is a novel and useful concept. These tools and applications are very useful for developing remediation strategies. (Abstract shortened by UMI.)...