AEG Event 12/5/24: Data Mining for PFAS

ABSTRACT
Per- and polyfluoroalkyl substances (PFAS) are a class of anthropogenic fluorinated organic chemicals with varying toxic effects and diminutive cleanup goals. The mobility of PFAS in atmospheric particulate matter and water can result in diffuse non-point contributions to environmental matrices, necessitating estimation of site-specific recontamination risk due to ambient concentration levels. The traditional approach to estimating ambient levels involves targeted sampling at an established reference area that is demonstrably unaffected by site activities. Hypothesis tests and comparisons are typically used to address extreme values (i.e., potential outliers), and values below the detection limit (BDL), after which confidence limits are calculated to represent statistical thresholds. If a clear reference area is lacking or does not reflect the full range of nonpoint-source variability, alternative approaches must be considered.

This study estimates ambient levels for multiple PFAS compounds simultaneously using a sophisticated data-mining technique (Gaussian Mixture Modeling – GMM) that is BDL-aware. The findings distinguish PFAS distributions for rural / headwater streams from urban areas. Nonlinear relationships with hydroclimate and anthropogenic variables clarify potential controls on PFAS occurrence and inform a conceptual model for the PFAS distribution in Pennsylvania surface waters. These insights have the potential to strengthen conceptual site models, inform remediation conversations early in the project life cycle, and improve engineering designs. The algorithms used here are also part of a larger machine-learning framework now widely available to practitioners. This presentation will also discuss that framework, for additional context for remediation practitioners.
 

BIOGRAPHY

Skyler is a hydrogeologist at WSP with a technical focus on groundwater modeling, multivariate statistical assessments, and PFAS. He is a member of the WSP Data Science and Emerging Contaminants practice areas and supports projects requiring an environmental data science approach to extract complex patterns from large datasets. He also works with the ITRC (Interstate Technology and Regulatory Council) PFAS Team, including the Source Identification subgroup.