Predicting water contamination with Bayesian networks — watch Javad Roostaei's talk on this topic at the recent BayesiaLab Conference in Durham.
Javad Roostaei, Ph.D., Postdoctoral Research Associate at University of North Carolina at Chapel Hill
During the past year, per- and polyfluoroalkyl substances (PFAS), including GenX, have been detected in more than 75% of 769 private water supply wells located near the Chemours Company Fayetteville Works in North Carolina. GenX concentrations exceeded the North Carolina provisional public health goal of 140 ng/L in nearly 25% of the wells. High geographic variation in PFAS occurrence has been observed in multiple areas; properties with highly contaminated wells neighbor properties where no PFAS have been detected. The causes of this variation are not understood. A wide variety of factors—from fine-scale geologic heterogeneity to well depth and age to wind direction relative to the Chemours facility—could influence contamination risk. However, the relative importance of such factors and how they interact to influence whether a specific drinking water well will be contaminated are not understood. This presentation will describe a detailed spatial data set and a machine-learned Bayesian network model for risk assessment of GenX—one type of PFAS—in private drinking water wells in North Carolina. Accuracy of the model has been verified by 10-fold cross-validation. The Bayesian network model will be useful for predicting which unsampled wells may be at risk, not only in North Carolina but also potentially in other locations struggling with PFAS contamination of groundwater.
Dr. Roostaei received his Ph.D. in Civil and Environmental Engineering with a master’s degree in Computer Science at Wayne State University in Detroit, Michigan. Currently, he is working as a machine learning postdoctoral research associate at the University of North Carolina at Chapel Hill. His research involves developing Bayesian network models for environmental risk evaluation in private water wells. He is applying machine learning methods to a variety of public health and civil and environmental engineering problems ranging from emerging contaminants and lead in drinking water to the development of harmful algal blooms in surface waters.