My primary research interest is statistical and mathematical modeling of environmental systems. I am also interested in multi-criteria decision analysis, and quantitative health and environmental policy analyses.

My thesis research aims to provide insight into the risks contaminated river sediments pose to the environment and human health. This work focuses on polychlorinated biphenyls (PCBs), which are anthropogenic and carcinogenic chemicals. PCBs were routinely released into the environment before scientists became aware of their harmful properties. Since this time, significant research efforts and funds have been put forth to reduce human exposure to PCBs. Despite four decades of experimental work, limited progress has been made in our understanding of how PCBs degrade in the environment. Thus, I have taken a nontraditional approach utilizing statistical and mathematical models to lend additional insight.

Through biologically mediated anaerobic degradation, PCBs-contaminated sites can become less risky. Biodegradation occurs through sequential losses of a chlorine atom, following 840 pathways from higher- chlorinated to lesser-chlorinated congeners or biphenyl. Eight patterns of 6 to 33 dechlorination pathways, called processes, have been identified through the analysis of shifts in congener masses in field and laboratory studies.  While processes were originally described in terms of explicitly reported pathways, generalizations of each process were qualitatively extrapolated.  Due to analytical limitations and a paucity of studies, the explicitly reported pathways and generalizations likely do not represent all of the pathways that may be observed at contaminated sites. 

A novel application of classification trees is an alternative, quantitative and replicable approach to the identification of candidate pathways for inclusion in process generalizations.  It expands on the accepted structure-based premise that was originally taken to identify generalizations while providing a basis for exploring the implications of including or excluding alternative pathways in a given dechlorination process.  The classification trees performed very well, with correct classification rates ranging from 0.90 to 0.99, and predicted the inclusion of 486 pathways in addition to the 108 explicitly reported pathways. While many of the attributes used in the original generalizations were also selected as predictors for dechlorination process inclusion by the classification trees, the trees suggest that the small number of attributes used to identify the original generalizations were not able to fully capture the microbial specificity represented by dechlorination processes. This work is published in Environmental Science & Technology's special issue on PCBs.