Exploring Machine Learning Algorithms to Support Municipal Wastewater Treatment PFAS Source Identification and Control Efforts (WRF 5276)

PFAS are present in the influent of many water resource recovery facilities (WRRFs) due to their widespread use in manufacturing, commercial operations, and household products. In January 2023, the United States Environmental Protection Agency announced a plan to introduce PFAS limits for WRRFs as part of the PFAS Strategic Roadmap. Although no limits have been identified federally for WRRFs, regulatory agencies' current focus is on source identification within WRRF collection systems. However, source identification is resource intensive for WRRFs, both in staff time and budget. South Platte Renew (SPR), the third largest WRRF in Colorado, is required to complete a PFAS source identification study as part of additional monitoring required by the Colorado Department of Public Health and Environment (CDPHE). This effort has required hundreds of hours of staff time and tens of thousands of dollars to conduct sampling, lab analysis, and reporting. Similar source identification investigations by WRRFs are anticipated to be required in the future.

#The University of Oklahoma (OU) and Brown and Caldwell (BC) have been working collaboratively on identifying alternative methods that could be used in PFAS source identification-specifically, using machine learning algorithms and signatures found in PFAS samples (Kibbey et al. 2024). To date, this algorithm has only been used for environmental groundwater samples. The project team of OU, BC, and SPR - funded in part by the Water Research Foundation (WRF) project 5276 - aim to determine if existing machine learning applications can be expanded to identify PFAS sources in WRRF influent and if machine learning can identify correlations between commonly measured influent water quality parameters and PFAS. PFAS source classification through pattern identification would allow WRRFs to collaborate and develop a robust database that narrows down sources of the specific PFAS present in WRRFs' plant influent, marking a significant advancement in the industry. This ability has the potential to significantly reduce costs and time associated with meeting PFAS source identification requirements and the cost associated with future treatment if source control is implemented.

#The overall goal of this project is to identify the applicability of using machine learning to support PFAS source investigations conducted by municipal wastewater treatment facilities. A primary focus of the research is expanding an existing database and machine learning algorithm from environmental samples to those specifically found in wastewater collection systems and plant influent streams. To this end, SPR is conducting comprehensive PFAS sampling at industrial dischargers throughout their collection area. The team will analyze these results to identify possible distinct signatures from different types of dischargers, e.g. carpet cleaner, metal finisher, carwash, etc. Figure 1 displays example plots from the first half of sampling, showing distinct PFAS component results from different types of dischargers that may contribute industrial wastewater to a WRRF influent stream.

#The second component of the project will evaluate the relationship between commonly measured influent water quality parameters and influent PFAS concentrations. The project team will use machine learning to identify trends that may indicate a surrogate water quality parameter that could be used to monitor PFAS. This will be accomplished through the deployment of a spectrophotometer at the influent of SPR's wastewater treatment facility. The sensor will continuously measure UV light absorption across a broad spectrum. While PFAS substances are not expected to return a specific UV absorption, there may be co-contaminants in the waste stream that do return a distinct UV signature and can be used as sentinel compounds for high PFAS concentrations entering the plant. Figure 2 shows an example spectrophotometer result. The project team will develop repeatable routines to efficiently process and interpret large amounts of sensor data that can be shared with others in the industry looking to apply similar sensor-driven detection techniques.

#This research project serves as both a continuation of the previous source identification studies using machine learning completed by Kibbey et al. as well as a starting point for using advanced data sensing and analysis techniques to assist WRRFs with meeting evolving PFAS regulatory requirements. The presentation will include the complete 18-month WRF study results and will explain the key findings as they apply directly to wastewater utilities' response to PFAS regulations across the United States.