Predicting and Communicating PFAS Exposure Risks from Rural Private WellsEPA Grant Number: R840081
Title: Predicting and Communicating PFAS Exposure Risks from Rural Private Wells
Investigators: MacDonald Gibson, Jacqueline , Salamova, Amina , Redmon, Jenny , Livanapatirana, Chamindu , de Bruin, Wandi Bruine
Institution: Indiana University - Bloomington , Research Triangle Institute , University of Southern California
Current Institution: Indiana University - Bloomington , University of Southern California , Research Triangle Institute
EPA Project Officer: Packard, Benjamin H
Project Period: September 1, 2020 through August 31, 2023
Project Amount: $1,584,420
RFA: National Priorities: Research on PFAS Impacts in Rural Communities and Agricultural Operations (2020) RFA Text | Recipients Lists
Research Category: Drinking Water , Water
Although per- and polyfluoroalkyl substances (PFAS) have been detected in groundwater nationwide, exposure risks in rural communities relying on private wells for their drinking water remain poorly understood. Our preliminary research in rural North Carolina (NC) suggests that machine-learned Bayesian networks can leverage routinely collected data to accurately predict PFAS contamination sources and occurrence in private wells. Such models also provide a transparent, interactive platform for risk communication and management. In this project, we propose to develop a scalable platform for predicting PFAS occurrence in private wells that integrates our Bayesian network platform with a conventional physical fate-and-transport model. The relative importance of various PFAS sources to private well contamination risk nationwide and the influence of soil amendments on exposure risks will be assessed. In addition, materials for training state and local decisionmakers on use of the integrated modeling platform and for communicating PFAS exposure risks in private well water will be developed.
We propose two overarching hypotheses: (1) integrating physical fate-and-transport model output with additional site-specific data using machine-learned Bayesian networks will provide more accurate risk predictions than either fate-andtransport or Bayesian network models alone and (2) local public health and environmental agency personnel will prefer our integrated model to a fate-and-transport model alone as a basis for decision-making and risk communication.
We will expand our preliminary research by integrating our existing Bayesian network model of PFAS risks in private wells with a physical fate-and-transport model reflecting the unique retentive properties of PFAS. Optimal models selected via cross validation will be tested by comparing their predictions to measured PFAS concentrations in new, national private well samples collected via a citizen science campaign supported by a web platform and mail-out test kits. The integrated modeling approach then will be scaled nationally to develop nationwide maps of PFAS exposure risks in rural private well water. Exposure from private well water will be compared to other exposure sources. Materials for training state and local agencies on model use and for communicating PFAS risks will be developed using approaches from cognitive psychology.
This project will be the first to build machine-learned models for predicting and communicating PFAS exposure risks in rural private well water, to validate such models at multiple sites, and to scale them for national use. It will be the first national citizen-science PFAS monitoring campaign in rural private wells. The proposed integrated modeling, PFAS testing, risk assessment, and risk communication approach is expected to substantially improve the accuracy of risk predictions and to facilitate informed risk management decisions.