Spatial and Temporal Analysis, and Machine Learning-Based Prediction of PCB Water Concentrations in U.S. Natural Water Systems.

Data on dissolved phase water concentrations of polychlorinated biphenyls (PCBs) from 32 locations across the U.S. were compiled from reports, Web sites, and peer-reviewed papers, spanning 1979-2020, resulting in 5132 individual samples. Data wrangling enabled the organization and analysis of this extensive data set. Most samples originated from PCB Superfund sites like the Fox, Hudson, and Kalamazoo rivers, New Bedford Harbor, and Lake Michigan. ΣPCB concentrations ranged from 10°-10 pg/L, while individual congener medians ranged from nondetected to 380 pg/L. Non-Aroclor congeners, e.g., PCBs 11, 67, and 68, were also reported. Using a machine learning technique, a Random Forest model accurately predicted the temporal and spatial occurrence of dissolved PCBs, achieving Pearson correlations greater than 0.87 for the Anacostia, Fox, Hudson, Kalamazoo, Passaic, and Spokane rivers, Chesapeake Bay, and New Bedford Harbor. These models can be used to forecast PCB concentrations. Through a linear mixed-effects model, half-lives of approximately 8 years for ΣPCB and individual congeners were determined, but the resulting half-lives showed considerable variability. An interactive map of ΣPCB was created. This investigation highlights the need for additional sampling in PCB-contaminated sites that may expose communities to airborne PCBs, and in other locations, to enhance our understanding of PCB occurrence and distribution.