Analyzing California’s PFAS data: AI insights in source tracking

Per- and polyfluoroalkyl substances (PFAS) are among the most persistent and complex contaminants facing the water sector today. With increasing attention from regulators and the public, understanding where PFAS come from and how they behave in treatment processes has become a top priority for water utilities.

In a new article published in CWEA’s Clean Water magazine, Carollo’s Tulasi Ravindran, Yoko Koyama, and Kyle Thompson explore how artificial intelligence (AI) can uncover new insights from California’s PFAS monitoring data. Their co-authored article, Analyzing California’s PFAS Data: AI Insights in Source Tracking, analyzes information collected under the State Water Board’s 2020 investigative orders to better understand PFAS presence in wastewater influent, effluent, and biosolids.

PFAS Detection Trends and What They Tell Us

The study reveals that short-chain PFAS like PFPeA and PFHxA are among the most frequently detected compounds in California wastewater, likely due to improved analytical methods and ongoing transformation of PFAS precursors. Meanwhile, long-chain PFAS such as PFDA are more likely to accumulate in biosolids due to their hydrophobic nature.

The researchers found close alignment between California’s wastewater data and national PFAS trends in drinking water. This could indicate potential connections between treated effluent and drinking water sources, either through reuse or broader environmental contamination.

Machine Learning Clusters Reveal Potential PFAS Sources

One of the most innovative aspects of the analysis is the use of unsupervised machine learning to identify clusters of PFAS signatures in biosolids. These clusters were then compared with proximity to landfills and airports, two known sources of PFAS contamination. Facilities with higher levels of 5:3 FTCA, a PFAS commonly associated with landfill leachate, were more likely to be located near landfills.

While a distinct airport signature was not confirmed, the clustering method offers a promising approach for tracking PFAS sources using publicly available data and AI-based analytics.

Long-Chain PFAS Reduction: A Positive Trend

The study ends on an encouraging note. By combining California’s data with national datasets, the authors observed a steady decline in long-chain PFAS like PFOA and PFOS, about 9% and 7% per year, respectively. These reductions reflect the ongoing impacts of product phaseouts and targeted source control efforts.

To learn more about how AI and data analysis can support PFAS monitoring and source identification in wastewater systems, read the full article in Clean Water magazine.