Explainable AI Reveals Statistical Associations Between Industrial Activity and PFAS Contamination of Public Water Systems

Explainable AI Reveals Statistical Associations Between Industrial Activity and PFAS Contamination of Public Water Systems | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Explainable AI Reveals Statistical Associations Between Industrial Activity and PFAS Contamination of Public Water Systems Manish Kumar, Priyanshu Gupta, Hogan Nance, Khang Nguyen, Mateo Srivathanakul, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7935132/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Recent United States Environmental Protection Agency (USEPA) studies detected Per- and polyfluoroalkyl substances (PFAS) in ~45% of U.S. tap water highlighting the widespread environmental and public health concern. Although industrial activity in general and aqueous film forming foam (AFFF) usage are known contributors to the contamination, the specific industry sectors driving it remain unclear. Here, we apply eXplainable-AI (XAI) methods to move beyond coarse industrial categorizations and uncover the sectors most strongly associated with PFAS contamination. Using the national PFAS monitoring (UCMR5) data, industrial geolocations and socioeconomicfeatures we achieved strong predictive ability (F1-score = 0.84). SHAP analysis identified metal treatment, fabrication, and polymer manufacturing as dominant contributors, corroborating prior studies, while revealing specialty chemical manufacturing as a major yet previously overlooked predictor of PFAS contamination—often surpassing AFFF in influence. A paradoxical socioeconomic pattern also emerged: PFAS contamination was more likely in affluent regions (higher income, education, and professional employment). Earth and environmental sciences/Environmental sciences/Environmental impact Earth and environmental sciences/Environmental social sciences/Environmental impact Figures Figure 1 Figure 2 Figure 3 Figure 4 Main mated 45% of the nation’s tap water 1 . This widespread detection indicates that the scale of contamination facing drinking water sources in the United States – including public water systems (PWSs)- is both significant and concerning ( Figure 1a ). Over 90% of US residents get their drinking water from PWSs 2 . These PWSs often rely on groundwater or surface water sources that may already be contaminated with PFAS 3–5 . Although several technologies have demonstrated efficacy in reducing PFAS concentrations at the laboratory and pilot scales 6–11 , empirical data from operational drinking water plant indicate that, if present in raw water, PFAS compounds are typically detected in the finished/tap water at comparable concentrations 12 - indicating minimal removal of PFASs by most widely used drinking water treatment processes except advanced treatment processes including activated carbon adsorption, nanofiltration, or reverse osmosis 12,13 . Potential pathways for PFAS contamination of drinking water systems are diverse ( Figure 1b ) 13–15 . Direct sources include: a) industrial releases from sectors involved in the manufacture and/or use of PFAS compounds, b) the use of aqueous film forming foams (AFFF) for firefighting activities, and c) consumer products (eg. stain-resistant coatings) 16 . Indirect sources can include a) biosolids application, and b) discharges from wastewater treatment, septic, stormwater and landfill systems. 17 Although PFAS discharges from indirect sources have been studied in detail, these pathways generally act as aggregators and/or conduits rather than being the point of introduction of PFAS into the wastewater streams. The true entry points lie with the direct sources, with whom the decision to use or manufacture PFAS-containing materials ultimately lies. Notably, EPA’s 2022 decision to monitor PFAS at the point of discharge through the National Pollutant Discharge Elimination System (NPDES) program, rather than in downstream carriers, underscores a shift in regulatory focus 18 . While the U.S. Environmental Protection Agency (USEPA) has identified a broad range of industrial sectors potentially associated with PFAS use and release, much of the academic literature remains focused on a narrower set of sources. At the site level, numerous studies have investigated PFAS contamination in proximity to known AFFF users 15,19–21 or individual industrial factories. However, these studies are typically limited to isolated cases and rarely contextualize the contribution of a specific industry relative to other potential sources. In contrast to site-specific investigations, studies conducted at regional 20,22–28 , or national 29–34 often attribute sources of PFAS contamination using broad categorical labels such as “industry” or “manufacturing”. As a result, despite widespread recognition that PFAS are used across a diverse array of industries, the relative influence of different sectors remains poorly characterized in large-scale, data driven assessments. In this study, we move past the broad categorical label of “industry” to systematically identify the specific sectors within industry that could act as key drivers of PFAS contamination of drinking water. By integrating the geospatial distribution of 25 industry sectors identified by Salvatore et. al. 31 as “presumptive PFAS contamination sites” with AFFF users (airports, military bases and fire-fighting training facilities), and sociodemographic characteristics, it may be possible to capture the complex and sector-specific patterns underlying PFAS contamination. A novel aspect of this work includes the use of explainable AI (XAI) methods to uncover these patterns linking different industry sectors to PFAS contamination. We use publicly available drinking water concentration data for eight PFAS compounds (PFPeA, PFBA, PFHxA, PFBS, PFOS, PFOA, PFHxS, and PFHpA) from the US EPA’s fifth Unregulated Contaminant Monitoring Rule (UCMR5) ( see Figure 1a and Supplementary Information, Table S2 ). Given that PFAS fate and transport are governed by hydrological connectivity and mixing from multiple sources, individual PWS-specific drainage delineations would not be representative 35,36 . Instead of using arbitrary boundaries like counties or states, which often cut across natural stream networks, HUC8 subbasins are defined by the actual drainage patterns of a landscape 37 . Thus, we aggregated the PWS geolocations into Hydrological Unit Code 8 (HUC8) subbasins developed by the United States Geological Survey to capture watershed-level source-mixing and transport dynamics of the contaminants within these regions. ( Fig 1c ). For each subbasin, we engineered a set of 30 descriptors (or features) to characterize potential drivers of PFAS contamination ( Fig 1c ). The first 25 features each describe the geospatial distribution of a corresponding industry sector identified as “presumptive contamination sites” 31 , followed by three features- one for each type of AFFF discharge sites (i.e. airports, military bases and fire-fighting training facilities). The last 2 features describe sociodemographic characteristics of the subbasin. Machine learning (ML) models were then trained to learn from these descriptors in order to predict the likelihood of detecting PFAS in public water systems within each subbasin. A key advance of this study is the use of SHAP analysis to interpret model predictions and quantify the contribution of individual features ( Fig 1d,e ). This approach allowed us to uncover sector-specific patterns of association with PFAS contamination that remain hidden under coarser categorization schemes. Results The use of three different ensemble-based machine learning models - Random Forest (RF), XGBoost (XGB), and LightGBM (LGBM) – proved effective in learning from watershed level features and predicting PFAS detections in the PWSs in a watershed. All three architectures yielded comparable predictive and classification performance ( Supplementary Information, Table S3 and S4, Supplementary Fig S1 ) suggesting a strong and consistent predictive signal in the data. Across the three architectures, the PFOA- (AUROC:0.83, accuracy: 83.8%, F1: 0.825) and PFHxA- (AUROC:0.82, accuracy: 82%, F1: 0.77) specific models outperformed the other targets ( Supplementary Information, Table S4 ), indicating that these two compounds exhibit stronger associations with the input features used in this study, albeit only marginally. The SUMPFAS model, trained on the combined data of all 8 PFAS compounds in this study, also showed comparable performance (AUROC: 0.78, sensitivity: 73.5, specificity: 73.5). Optimizing the threshold for Youden’s J-max improved the F1-score by raising the precision at the expense of recall ( Supplementary Information, Table S4 ). Furthermore, re-training the RF models with updated data as of July 2025 yielded a minor improvement in AUROC, while preserving the overall feature importance rankings ( Fig. 2a ). Modest shifts the distribution of SHAP impacts indicate slight reweighting of feature contributions without altering the dominant drivers of model predictions ( Fig. 2b ). The minor changes in performance and |SHAP| value rankings on retraining with the updated July 2025 dataset indicates that the model captures stable underlying relationships between predictors and PFAS occurrence. Given comparable predictive performance across all three ensemble-based architectures, we focus our interpretability analysis on the RF model. RF is particularly well-suited for explainable AI applications, as it is composed of an ensemble of rule-based decision paths defined by cutoff splits (eg. if value of feature > cutoff ), making its internal logic more transparent than the gradient boosting-based methods 38 . Going forward, model performance is discussed in terms of the RF architecture-based SUMPFAS model trained on data up to October 2024. SHAP analysis enables us to interpret the underlying patterns captured by our predictive models, highlighting the most influential factors driving PFAS occurrence. SHAP is a game-theoretic approach that decomposes the model predictions into additive feature contributions 39 ( Fig. 1f ). Figure 3 presents analysis of the individual SHAP contributions for 2 exemplar HUC8 subbasins. The Upper Dry Watershed (HUC8 18030009) in California ( Fig. 3a ) characterized by high industrial density and a large population was classified as a “detects” subbasin by the model, consistent with the six PWSs in the region reporting PFAS contamination. SHAP analysis highlights how the large population size and multiple industrial sectors jointly drive the prediction toward detection, yielding a model probability near 1. In contrast, the Red Lake Watershed (HUC8 15010007) in Arizona, contains only six relevant industrial facilities and a sparse population. Here, SHAP illustrates how a limited presence of Paint and Coating Manufacturing, and Soap and Other Detergents Manufacturing industries is insufficient to offset the suppressive effect of low population and absence of other important industrial drivers, leading the model to correctly predict a non-detection. SHAP summary plots. Building on these individual case studies, we turn to a broader view of how individual features influence the model’s predictions across all HUC8 regions using SHAP summary plots. While the waterfall plots ( Figure 3b-c ), provide granular insights into how specific feature values drive model predictions for individual subbasins, they are inherently limited to single datapoints. To assess whether these patterns generalize across the entire dataset, we turn to SHAP beeswarm plots ( Figure 4a-g ), which provide a distributional view of feature contributions across all datapoints. These plots allow us to evaluate not only the magnitude, but also the directionality of each feature’s effect-i.e, whether higher or lower values of a feature consistently increase or decrease the predicted likelihood of PFAS detection. To simplify the subsequent discussions, Fig. 2b illustrates the mean of absolute SHAP values of all the features grouped by the thematic feature groups defined in Supplementary Information Table S1 . Among industrial sectors, three feature groups emerged as particularly influential in driving predicted PFAS detection probabilities, namely metal treatment & fabrication, plastics & polymer products, and chemical manufacturing. ML Model recovers strong association of metal treatment, fabrication, and coating facilities with PFAS contamination. According to Figure 3a , subbasins with higher counts (orange and red) of metal treatment, and fabrication, and metal coating facilities exhibit consistently positive SHAP values, indicating that the presence of these industrial sectors push the model output toward PFAS detection. Conversely, their absence (blue) is associated with negative SHAP values, contributing to predictions towards non-detection. This directional trend suggests a strong and systematic link between metal processing activities and PFAS contamination risk. Interestingly, this sector is known to employ PFAS across multiple surface finishing processes for chromium and other metals, including electroplating, electroless plating, anodizing, coating, etching and surface cleaning. 40,41 Examples include the use of polytetrafluoroethylene (PTFE) in electroless nickel plating and powder coating, PFAS-based deposition aids in copper plating, PFAS containing wetting agents for metal plating on plastics, and fluorochemical blocking agents in aluminum foil production 41 . PFAS compounds are valued in metal finishing for their ability to prevent corrosion, reduce mechanical wear, promote flow of metal coatings and prevent cracks during drying. However, prior site-specific studies 42,43 as well as white papers issued by state pollution control agencies 41,44 have documented their potential release into the environment via multiple pathways. Our model extends this understanding by identifying a persistent, positive association between this sector and PFAS contamination at the national scale. Notably, this product category ranks as the most influential industrial predictor in 7 of the 8 PFAS specific models we trained (Supplementary Figure S2 ). This recurring pattern across nearly all models underscores the widespread and systematic role of the metal treatment and fabrication sector in contributing to PFAS contamination across the nation. Similarly, the strong and directional effect of the plastics and polymer category aligns with prior evidence , particularly through the Plastics and Resin Manufacturing sector ( Fig. 3b ), which includes fluoropolymer resins manufacturing, fluorohydrocarbon resins manufacturing, and polytetrafluoroethylene resins manufacturing 45 - sectors previously linked to PFAS contamination of groundwater wells in close proximity 24,27,46–48 . The SHAP distribution shows that HUC8 regions with higher facility counts in this category (orange and red points) are consistently associated with positive contributions to the model’s prediction, suggesting that the presence of such facilities reliable elevates the risk of PFAS detection. These industrial operations often involve the use of PFAS as processing aids or raw materials, particularly in the extrusion and polymerization of fluorinated polymers 24,27,49 . Prior field investigations have found evidence of PFAS leaching and emissions from polymer manufacturing facilities into local water sources 24,27 . Our model generalizes these site-specific findings, demonstrating that even at the national scale, the presence of plastics and resin manufacturing is a robust and directional driver of predicted PFAS contamination. ML model reveals an understudied influence of the All Other Miscellaneous Chemical Product and Preparation Manufacturing sector ( hereafter referred to as the specialty chemical manufacturing sector ). As shown in Figure 3C , subbasins with a higher number of specialty chemical manufacturing facilities (orange/red points) are associated with consistently positive SHAP values, suggesting the presence of these facilities increase the predicted probability of PFAS detection. This clear directionality reveals a strong role of this sector as a non-negligible contributor to contamination risk. This sector encompasses the production of a wide range of chemical products- from surfactants to performance-enhancing additives like brake fluids, and hydraulic fluids- many of which rely on PFAS chemistries 49 . As early as 2005, Prevedourous et. al. 16 estimated that the emissions from manufacturing of perfluorooctyl sulfonyl fluoride- (POSF) and fluorotelomer-based specialty products accounted approximately 4 % of total historical global PFCA emissions. To our knowledge, the specialty chemical manufacturing in has received limited attention beyond a few site specific studies (for example 50–52 ) compared to other potential contamination sources. Notably, this sector also includes facilities engaged in the production of fire-extinguishing agents and related formulations 53 . While AFFF-using sectors such as airports, and fire training facilities have justifiably received significant scrutiny, our analysis reveals that upstream specialty chemical manufactures may play an even greater role ( Fig. 3c-d ). Furthermore, in all 9 of our models (8 PFAS-specific + 1 PFAS), this sector ranks higher than any of the well know AFFF users included in this study (Supplementary Figure S2 ). Hence, our results further challenge this omission, revealing strong and consistent contributions from the specialty chemical manufacturing sector to predicted risk of PFAS contamination. This contrast underscores the importance of broadening PFAS source attribution efforts to include upstream industrial producers. By highlighting the outsized and persistent influence of specialty chemical manufacturing, our findings point to a critical but underrecognized opportunity for advancing mitigation strategies through closer collaboration with chemical manufacturers. Greater engagement with this sector, paired with improved transparency around PFAS use in production, could inform the development of safer alternatives and support voluntary phase-outs. Even among AFFF users, our model presents a more nuanced picture. Although only firefighting training facilities ranked in the top 10 most impacting features, both firefighting training facilities and military bases ( Supplementary Information, Figure S2 ) repeatedly show positive SHAP values at moderate to high facility densities, indicating a strong association with elevated PFAS risk. In contrast, the influence of airports appears more diffused across predictions ( Supplementary Information, Figure S2) , suggesting that while airports may be recognized as important sources of localized PFAS contamination, their contribution is less systematic at the national scale. Other industrial categories- namely textile, paper, & printing, waste management & chemical handling, electrical & electronics manufacturing, and petroleum & petrochemical products- demonstrated comparatively lower SHAP contributions in our national models ( Fig. 3f-g, FigS2 ), but their potential role in PFAS contamination should not be overlooked as discussed subsequently. Firstly, lower SHAP magnitudes do not imply irrelevance. In the case of textile, paper & printing category, SHAP plots ( Fig. 3f and Fig. S2 ) reveal that specific sectors like commercial printing show clear directionality in SHAP values, with higher facility densities associated with positive model contributions. The processes in this sector often utilize PFAS-based surfactants and coatings to impart durability and print quality, pointing to potential sources of environmental release. However, the lower overall contribution of the category could stem from limited spatial prevalence, or lower mean absolute contributions of other sectors within that category. Similarly, SHAP plots for waste management & chemical handling sectors ( Fig. 3g ) reveal that specific sectors like solid waste landfills and hazardous waste collection show a variable contributing at medium to high facility densities. The diffused but positive contribution at high facility densities may reflect the persistence of legacy PFAS-bearing materials in landfills as has been reported in past literature 54,54–57 . However, the diffused nature of the contributions could also stem from episodic or spatially dispersed emissions. In contrast, the electrical and electronics manufacturing, and petroleum related sectors displayed both low SHAP magnitude and were absent from top feature rankings . While these findings may indicate a limited role in PFAS violations nationally, it is also possible that PFAS usage in these sectors is better managed via closed-loop systems, or the facilities are not spatially well distributed ( Supplementary Information, Figure S2 ). Overall, muted national signals from specific sectors does not preclude their significance at regional or watershed scales. This study is designed to identify systemic drivers of PFAS violations at the national scale, and as such it emphasizes features with consistent influence across diverse geographies. However, this absence of national level feature importance does not imply that these sectors are benign; rather, it could highlight the need for spatially disaggregated or case-specific analyses to detect and respond to localized contamination risks. Beyond industrial and facility-based sources, our model also reveals surprising patterns in socio-demographic drivers of PFAS contamination ( Fig. 3h ). Total population emerged as the most influential feature across all models- acting as a proxy for the density of human activity, infrastructure and consumer product use. This finding suggests that PFAS exposure may not be limited to isolated industrial discharges but may also be a diffused outcome of widespread societal behaviors- like the use of stain resistant coatings in apparel and non-stick cookware 56 , and a range of other PFAS containing consumer products 58–61 . A striking and counterintuitive insight from the SHAP contributions is the role of affluence. The affluence metric used in this analysis reflects a composite of indicators of socio-demographic status- namely concentrations of adults with a bachelor’s degree or higher, with incomes >$125K, and employed in managerial and professional occupations. Our model suggests that subbasins with higher average affluence are more likely to be associated with PFAS detection in their drinking waters ( Fig. 3h ). We hypothesize that this observation may be partially driven by the increased use of specialized consumer products in more affluent neighborhoods 62,63 . While most consumer products do not contain any surfactants, PFAS are commonly used in products that serve specialized function like stain resistance, waterproofing, and non-stick performance, possibly making them more expensive and disproportionately consumed in wealthier communities. Additional empirical support for our hypothesis comes from studies reporting elevated PFAS concentrations in the blood serum of individuals from higher socioeconomic backgrounds 64–66 . Furthermore, preliminary results from Salawu et. al.’s characterization of PFAS concentration in the household wastewater across three different neighborhoods suggest that more affluent neighborhoods discharge more PFAS into the sewer systems, compared to less affluent ones 67 . While PFAS exposure is often framed as an environmental justice issue affecting communities with limited resources or systemic vulnerabilities, 64–66,68 our findings point to a more complex risk landscape- one in which patterns of economic activity and consumption may also contribute to elevated PFAS risk. Taken together, these patterns underscore the importance of moving beyond traditional assumptions about PFAS risk and considering a refined range of industrial and demographic factors. Despite the good-to-excellent model performance, limited regional mismatches between predicted and observed detections remain . These instances too offer additional insight and highlight opportunities for future research and targeted policy interventions. For instance, we found several watersheds where the model predicts a high PFAS risk- often due to dense clusters of industrial activity ( Supplementary Information, Fig. S3a ) – however, no PFAS were detected in the UCMR5 dataset. This discrepancy, called false positives, may reflect the presence of effective mitigation measures, such as advanced water treatment technologies, rigorous source water protection practices 69–71 or hydrological barriers that reduce contaminant transport 72 . These locations may represent examples of successful risk management and warrant closer examination as potential case studies for best practices. Conversely, there are watersheds ( Supplementary Information, Fig. S3b ) where the model predicts a low risk despite confirmed detections in UCMR5. The prediction of low risk in these regions is typically driven by the lack of obvious industrial sources. This pattern of false negatives suggests that contamination in these watersheds may be arise from diffuse, non-industrial pathways - such as legacy PFAS contamination 54 , long-range hydrological connectivity 73,74 , or unrepresented sources such as septic systems, industrial activities outside the modeled sectors, and land-applied biosolids - underscoring the need for broader consideration of upstream inputs and unmonitored contributors. For example, Peter and Lee 75 demonstrated that PFAS loads in agriculture-dominated watersheds can be dominated by diffuse legacy sources despite a lack of industrial activity. In their study, tile-drained catchments under sustained biosolids applications since the 1980s exhibited total PFAS concentrations of 79 - 170ng/L, whereas most agricultural streams remained below 2ng/L. The occurrence of false positives and false negatives highlights the inherent complexity of PFAS sources, fate and transport in the environment. These discrepancies underscore the influence of unmonitored or diffuse pathways that remain challenging to represent in current datasets. Together, these insights help advance the science of PFAS source attribution and help guide future monitoring and intervention strategies. Before concluding, we also acknowledge two important limitations inherent to our ML-based workflow and modeling framework, which arise from methodological choices rather than PFAS-specific factors. One key limitation of our approach that needs to be recognized is that SHAP analysis captures statistical associations, not causal mechanisms. Establishing causality for each PFAS detection would require detailed, site-specific investigations- which is beyond the scope of this national scale study. These associations should be viewed as hypothesis-generating, offering a starting point for more targeted investigations. Validating these findings will require facility-level data such as the Toxic Release Inventory (TRI), and the National Pollution Discharge Elimination Systems (NPDES) permits for each industrial facility. Such efforts could help distinguish between active emissions, legacy contamination and background levels. However, the requirement of reporting PFAS under either of systems is rather recent, and the databases are still in development 18,76 . Another limitation is that the use of the UCMR 5 database limits our study to contamination in public water systems, excluding private wells that serve an estimated 13-14% of the US population 1 . Previous site-specific studies as well as state-level analyses have studied the contamination of private wells in proximity to know sources. However, to the best of our knowledge, there is no publicly available dataset on the PFAS concentrations in the private wells across the nation. Conclusion This study demonstrates the utility of explainable AI in identifying and interpreting geospatial patterns underlying PFAS contamination in public water systems at the national scale. By integrating geospatial data about industry sectors and socio-demographic data, we provide a nuanced view of the factors most strongly associated with PFAS detections in drinking water. Among the key findings, the presence of three industrial categories- specialty chemical manufacturing, metal treatment and fabrication and plastics and polymer products- emerged as consistent patterns associated with PFAS detections, enabling our models to achieve AUC scores in the range of 0.75-0.82. These findings align with known PFAS use patterns while also bringing attention to production-process based sources that have received less focus in prior research. Our models further highlight important distinctions within AFFF use related sources: firefighting training stations and military installations show stronger and more consistent associations with PFAS detections than airports. Beyond point-source contributors, we find that total population is the most influential feature, indicating the broader role of human activity in shaping PFAS presence. Perhaps unexpectedly, our results also reveal that higher-income communities are more likely to be associated with detections, pointing to a more complex spatial risk profile than typically captured in existing frameworks. While the patterns learned by our models reveal robust statistical associations, they must not be mistaken to be establishing causality. Further validation using facility level discharge data (e.g. from TRI database, or NPDES permits) will be critical for confirming pathways of emissions into the environment. Additionally, the UCMR database does not exclude private well data or even the very small public water systems (serving <3,300 persons), thereby limiting the generalizability of our findings to the entire U.S. population. Despite these limitations, the strong performance of our models across multiple machine learning architectures supports the reliability of our results. The use of Random Forest model instead of more complex models allows easier interpretability. Together, these findings offer a scalable and interpretable framework for identifying high-risk areas, supporting monitoring priorities, and informing more targeted regulatory and policy responses. Methods a. PFAS DETECTION AT U.S. PUBLIC WATER SYSTEMS The foundation of this model is the fifth release of the Fifth Unregulated Contaminant Monitoring Rule (UCMR 5) dataset published by the U.S. Environmental Protection Agency (EPA) in October 2024. This data set provides 1,133,968 analytical results for measurement of 29 PFAS compounds and lithium measured at 7,237 public water systems (PWS) in the United States. Each PWS in the dataset is identified by a unique Public Water System ID (PWSID). This dataset represents ~55% of the data that the EPA expects to obtain under the UCMR 5 ruling by the completion of reporting in 2026. This study excluded PWS located in Alaska, Hawaii, and US territories due to differing geographical context- removing 131 PWSIDs from the datasets. This study focused on the detection of only 8 of the 29 PFAS compounds (PFOS, PFOA, PFNA, PFDA, PFHxS, PFHpA, PFBS, PFHxA) - all with the frequency of detection >300. 2,317 facilities detected at least one of these 8 contaminants to be above the proposed MRL threshold 77 , 4,722 sites did not detect any contaminants to be above the threshold, and 67 facilities did not report any measurements for any of the 8 contaminants. The 67 facilities were further excluded from downstream analysis. Finally, each PWSIDs was geocoded using the zipcodes of the service area provided by the UMCR 5. 215 facilities did not have any zipcode in the UMCR file and were excluded from further study. The resulting dataset had 6824 unique PWSIDs with geolocations and data on detection/non-detection of the PFASs of interest. An interactive map of the public water systems in continental United States colored by detection of PFAS, and the respective PFAS concentrations has been made available at priyanshurg.github.io/pfas_interactive_maps/industrial_densities.html. This map will be updated with the release of newer datasets under the UCMR 5 ruling. Preliminary data analysis of this dataset was conducted in Python using the NumPy 78 , Pandas 79 , GeoPandas 80 , Folium 81 , Upset 82 and Nominatim 83 . b. POTENTIAL PFAS SOURCES AND SOCIODEMOGRAPHIC FEATURES This study considered 28 point sources of PFAS – 25 different industrial sectors, and 3 AFFF foam users. The North American Industry Classification System (NAICS) codes categorize businesses establishments based on their primary type of economic activity, enabling uniform, comparable and standardized industry data analysis. NAICS codes have been employed by researchers, state environmental agencies as well as the EPA to identify facilities suspected of PFAS use. Salvatore et. al. 31 identified a list of 38 NAICS codes- around whose factories the existence of PFAS can be presumed. Self-reported data of the facilities within these 38 NAICS codes were downloaded from the EPA’s Facility Registry Service (FRS) EZ Query 84 . NAICS codes (hereon also referred to as industry sectors) with self-reported data on 500 or more factories were selected for model development. A total of 108,573 facilities across 25 different industry sectors were shortlisted as shown in Table 1. To simplify the analysis, the industry sectors were further clubbed by the authors based on the type of products/processes expected to be performed in the industry sector ( Table 1 ). Using the self-reported geospatial data on the potential point sources of PFAS, average density of the different industry sectors in each HUC8 region was calculated as: In addition to the manufacturing and industrial process, PFAS are also used extensively as a surfactant in Aqueous Film Forming Foam (AFFF). PFAS contamination is also expected wherever AFFF is discharged, particularly military sites, major airports and fire training areas. Geospatial datasets of airports, firefighting training facilities and military-installations across the continental US were obtained from publicly available sources. Sociodemographic features used as inputs for the model included: a combination of total population, and affluence from National Neighborhood Data Archive (NaNDA) 85 . Since these datasets are only available at a zipcode level instead of HUC8 level, a mean_affluence is calculated for each HUC8 region. The averaging is done by geocoding the zipcodes provided in the datasets, and then performing an average over the zipcodes that lie inside a certain HUC8 region. Maps showing the distribution of facilities under each NAICS code across the continental United States is available at: priyanshurg.github.io/pfas_interactive_maps/naics_codes_distributions.html. c. MODEL TRAINING A total of 9 random forest classifier models – 8 models for each of the 8 most occurring PFASs, and 1 model, called PFAS, encompassing all 8 compounds – were trained in this study. Each classifier was trained using 100 decision trees, each with a maximum depth of 10 to limit model complexity. The minimum number of samples required to split an internal node was set to 2, and each leaf node was required to contain at least one sample. For each split, a random subset of features equal to the square root of the total number of predictors was considered to introduce feature-level randomness and promote diversity among the decision trees. The models were evaluated using stratified 10-fold cross-validation to ensure robustness and repeated across five different random seeds (random_state ∈ {42, 11, 101, 25, 50}) to ensure reproducibility of the reported results. For each fold, features were standardized using a StandardScaler fitted on the training data in the fold. Model performance was assessed using the area under the receiver-operating-curver (AUC-ROC), with interpolated true positive rates (TPRs) and false positive rates (FPRs) computed to estimate a mean AUC-ROC and associated variability. Additional metrics like accuracy precision, recall, f1-score, and Youden’s J-Index were calculated at a default prediction probability threshold of 0.5 for each model. The prediction probability threshold was then optimized for each model by maximizing the Youden’s J-Index obtained the same metrics at the optimal threshold. Finally, Shapley Additive exPlanations (SHAP) 39 values were calculated for each of the 5 trainings of each of the 9 models. Mean absolute SHAP values for detection of PFAS were computed for each feature in the model’s input. To simplify analysis, we also add up the mean absolute SHAP values of features under each product category to derive a total impact of product category for each category based on categorization in Supplementary Table S1 . Declarations Acknowledgements: This work was supported by two specific endowments to MK – Doug N Williams Memorial Centennial Fellowship in Engneering, UT Austin and the Ashley Ashley H. Priddy Centennial Professorship in Engineering, UT Austin. Partial funding from the WoodNext Foundation is also acknowledged. The authors would also like to acknowledge the contributions of the students of course EVE 310: Sustainable Systems Engineering, Fall 2024, under the Maseeh Department for Civil, Architectural, and Environmental Engineering at UT Austin. Data Availability Statement : All the data and code used in this study are available on Github (https://github.com/priyanshurg/Predicting-PFAS-Drivers). References Smalling, K. L.; Romanok, K. M.; Bradley, P. M.; Morriss, M. C.; Gray, J. L.; Kanagy, L. K.; Gordon, S. E.; Williams, B. M.; Breitmeyer, S. E.; Jones, D. K.; others. Per-and Polyfluoroalkyl Substances (PFAS) in United States Tapwater: Comparison of Underserved Private-Well and Public-Supply Exposures and Associated Health Implications. Environ. Int. 2023 , 178 , 108033. Information about Public Water Systems | US EPA — Epa.Gov. https://www.epa.gov/dwreginfo/information-about-public-water-systems. Kurwadkar, S.; Dane, J.; Kanel, S. R.; Nadagouda, M. N.; Cawdrey, R. W.; Ambade, B.; Struckhoff, G. C.; Wilkin, R. Per - and Polyfluoroalkyl Substances in Water and Wastewater: A Critical Review of Their Global Occurrence and Distribution. Sci. Total Environ. 2022 , 809 , 151003. https://doi.org/10.1016/j.scitotenv.2021.151003. Sims, J. L.; Stroski, K. M.; Kim, S.; Killeen, G.; Ehalt, R.; Simcik, M. F.; Brooks, B. W. Global Occurrence and Probabilistic Environmental Health Hazard Assessment of Per- and Polyfluoroalkyl Substances (PFASs) in Groundwater and Surface Waters. Sci. Total Environ. 2022 , 816 , 151535. https://doi.org/10.1016/j.scitotenv.2021.151535. Masoner, J. R.; Kolpin, D. W.; Cozzarelli, I. M.; Barber, L. B.; Burden, D. S.; Foreman, W. T.; Forshay, K. J.; Furlong, E. T.; Groves, J. F.; Hladik, M. L.; Hopton, M. E.; Jaeschke, J. B.; Keefe, S. H.; Krabbenhoft, D. P.; Lowrance, R.; Romanok, K. M.; Rus, D. L.; Selbig, W. R.; Williams, B. H.; Bradley, P. M. Urban Stormwater: An Overlooked Pathway of Extensive Mixed Contaminants to Surface and Groundwaters in the United States. Environ. Sci. Technol. 2019 , 53 (17), 10070–10081. https://doi.org/10.1021/acs.est.9b02867. Shi, H.; Wang, Y.; Li, C.; Pierce, R.; Gao, S.; Huang, Q. Degradation of Perfluorooctanesulfonate by Reactive Electrochemical Membrane Composed of Magneli Phase Titanium Suboxide. Environ. Sci. Technol. 2019 , 53 (24), 14528–14537. Li, Z.; Li, S.; Tang, Y.; Li, X.; Wang, J.; Li, L. Highly Efficient Degradation of Perfluorooctanoic Acid: An Integrated Photo-Electrocatalytic Ozonation and Mechanism Study. Chem. Eng. J. 2020 , 391 , 123533. Lei, Y.-J.; Tian, Y.; Sobhani, Z.; Naidu, R.; Fang, C. Synergistic Degradation of PFAS in Water and Soil by Dual-Frequency Ultrasonic Activated Persulfate. Chem. Eng. J. 2020 , 388 , 124215. Le, T. X. H.; Haflich, H.; Shah, A. D.; Chaplin, B. P. Energy-Efficient Electrochemical Oxidation of Perfluoroalkyl Substances Using a Ti4O7 Reactive Electrochemical Membrane Anode. Environ. Sci. Technol. Lett. 2019 , 6 (8), 504–510. Soker, O.; Tajdini, B.; Abarca-Perez, A.; Wadia, A.; Bellona, C.; Hao, S.; Doudrick, K.; Strathmann, T. J. Reuse of Spent Granular Activated Carbon for PFAS Removal Following Hydrothermal Alkaline Treatment. Water Res. 2025 , 123794. Soriano, Á.; Gorri, D.; Urtiaga, A. Efficient Treatment of Perfluorohexanoic Acid by Nanofiltration Followed by Electrochemical Degradation of the NF Concentrate. Water Res. 2017 , 112 , 147–156. Rahman, M. F.; Peldszus, S.; Anderson, W. B. Behaviour and Fate of Perfluoroalkyl and Polyfluoroalkyl Substances (PFASs) in Drinking Water Treatment: A Review. Water Res. 2014 , 50 , 318–340. https://doi.org/10.1016/j.watres.2013.10.045. Wee, S. Y.; Aris, A. Z. Environmental Impacts, Exposure Pathways, and Health Effects of PFOA and PFOS. Ecotoxicol. Environ. Saf. 2023 , 267 , 115663. De Silva, A. O.; Armitage, J. M.; Bruton, T. A.; Dassuncao, C.; Heiger‐Bernays, W.; Hu, X. C.; Kärrman, A.; Kelly, B.; Ng, C.; Robuck, A.; Sun, M.; Webster, T. F.; Sunderland, E. M. PFAS Exposure Pathways for Humans and Wildlife: A Synthesis of Current Knowledge and Key Gaps in Understanding. Environ. Toxicol. Chem. 2021 , 40 (3), 631–657. https://doi.org/10.1002/etc.4935. Sunderland, E. M.; Hu, X. C.; Dassuncao, C.; Tokranov, A. K.; Wagner, C. C.; Allen, J. G. A Review of the Pathways of Human Exposure to Poly- and Perfluoroalkyl Substances (PFASs) and Present Understanding of Health Effects. J. Expo. Sci. Environ. Epidemiol. 2019 , 29 (2), 131–147. https://doi.org/10.1038/s41370-018-0094-1. Prevedouros, K.; Cousins, I. T.; Buck, R. C.; Korzeniowski, S. H. Sources, Fate and Transport of Perfluorocarboxylates. Environ. Sci. Technol. 2006 , 40 (1), 32–44. https://doi.org/10.1021/es0512475. Dasu, K.; Xia, X.; Siriwardena, D.; Klupinski, T. P.; Seay, B. Concentration Profiles of Per- and Polyfluoroalkyl Substances in Major Sources to the Environment. J. Environ. Manage. 2022 , 301 , 113879. https://doi.org/10.1016/j.jenvman.2021.113879. Fox, R. Addressing PFAS Discharges in NPDES Permits and Through the Pretreatment Program and Monitoring Programs. https://www.epa.gov/system/files/documents/2022-12/NPDES_PFAS_State%20Memo_December_2022.pdf (accessed 2025-07-11). Rotander, A.; Toms, L.-M. L.; Aylward, L.; Kay, M.; Mueller, J. F. Elevated Levels of PFOS and PFHxS in Firefighters Exposed to Aqueous Film Forming Foam (AFFF). Environ. Int. 2015 , 82 , 28–34. https://doi.org/10.1016/j.envint.2015.05.005. Weiß, O.; Wiesmüller, G. A.; Bunte, A.; Göen, T.; Schmidt, C. K.; Wilhelm, M.; Hölzer, J. Perfluorinated Compounds in the Vicinity of a Fire Training Area – Human Biomonitoring among 10 Persons Drinking Water from Contaminated Private Wells in Cologne, Germany. Int. J. Hyg. Environ. Health 2012 , 215 (2), 212–215. https://doi.org/10.1016/j.ijheh.2011.08.016. Barzen-Hanson, K. A.; Roberts, S. C.; Choyke, S.; Oetjen, K.; McAlees, A.; Riddell, N.; McCrindle, R.; Ferguson, P. L.; Higgins, C. P.; Field, J. A. Discovery of 40 Classes of Per- and Polyfluoroalkyl Substances in Historical Aqueous Film-Forming Foams (AFFFs) and AFFF-Impacted Groundwater. Environ. Sci. Technol. 2017 , 51 (4), 2047–2057. https://doi.org/10.1021/acs.est.6b05843. Hu, X. C.; Ge, B.; Ruyle, B. J.; Sun, J.; Sunderland, E. M. A Statistical Approach for Identifying Private Wells Susceptible to Perfluoroalkyl Substances (PFAS) Contamination. Env. Sci Technol Lett 2021 , 8 (7), 596–602. https://doi.org/10.1021/acs.estlett.1c00264. Zhang, X.; Lohmann, R.; Dassuncao, C.; Hu, X. C.; Weber, A. K.; Vecitis, C. D.; Sunderland, E. M. Source Attribution of Poly- and Perfluoroalkyl Substances (PFASs) in Surface Waters from Rhode Island and the New York Metropolitan Area. Env. Sci Technol Lett 2016 , 3 (9), 316–321. https://doi.org/10.1021/acs.estlett.6b00255. Hoffman, K.; Webster, T. F.; Bartell, S. M.; Weisskopf, M. G.; Fletcher, T.; Vieira, V. M. Private Drinking Water Wells as a Source of Exposure to Perfluorooctanoic Acid (PFOA) in Communities Surrounding a Fluoropolymer Production Facility. Environ. Health Perspect. 2011 , 119 (1), 92–97. https://doi.org/10.1289/ehp.1002503. Kosiarski, K.; Veith, T. L.; Kibuye, F. A.; Fetter, J.; Boser, S.; Vanden Heuvel, J. P.; Thompson, C. L.; Preisendanz, H. E. Geospatial and Socioeconomic Factors of PFAS Contamination in Private Drinking Water Wells: Insights for Monitoring and Management. J. Environ. Manage. 2025 , 388 , 125863. https://doi.org/10.1016/j.jenvman.2025.125863. Panikkar, B.; Lemmond, B.; Allen, L.; DiPirro, C.; Kasper, S. Making the Invisible Visible: Results of a Community-Led Health Survey Following PFAS Contamination of Drinking Water in Merrimack, New Hampshire. Environ. Health 2019 , 18 (1), 79. https://doi.org/10.1186/s12940-019-0513-3. Bach, C.; Dauchy, X.; Boiteux, V.; Colin, A.; Hemard, J.; Sagres, V.; Rosin, C.; Munoz, J.-F. The Impact of Two Fluoropolymer Manufacturing Facilities on Downstream Contamination of a River and Drinking Water Resources with Per- and Polyfluoroalkyl Substances. Environ. Sci. Pollut. Res. 2017 , 24 (5), 4916–4925. https://doi.org/10.1007/s11356-016-8243-3. McMahon, P. B.; Tokranov, A. K.; Bexfield, L. M.; Lindsey, B. D.; Johnson, T. D.; Lombard, M. A.; Watson, E. Perfluoroalkyl and Polyfluoroalkyl Substances in Groundwater Used as a Source of Drinking Water in the Eastern United States. Environ. Sci. Technol. 2022 , 56 (4), 2279–2288. https://doi.org/10.1021/acs.est.1c04795. Hu, X. C.; Andrews, D. Q.; Lindstrom, A. B.; Bruton, T. A.; Schaider, L. A.; Grandjean, P.; Lohmann, R.; Carignan, C. C.; Blum, A.; Balan, S. A.; Higgins, C. P.; Sunderland, E. M. Detection of Poly- and Perfluoroalkyl Substances (PFASs) in U.S. Drinking Water Linked to Industrial Sites, Military Fire Training Areas, and Wastewater Treatment Plants. Env. Sci Technol Lett 2016 , 3 (10), 344–350. https://doi.org/10.1021/acs.estlett.6b00260. Andrews, D. Q.; Hayes, J.; Stoiber, T.; Brewer, B.; Campbell, C.; Naidenko, O. V. Identification of Point Source Dischargers of Per-and Polyfluoroalkyl Substances in the United States. AWWA Water Sci. 2021 , 3 (5), e1252. Salvatore, D.; Mok, K.; Garrett, K. K.; Poudrier, G.; Brown, P.; Birnbaum, L. S.; Goldenman, G.; Miller, M. F.; Patton, S.; Poehlein, M.; others. Presumptive Contamination: A New Approach to PFAS Contamination Based on Likely Sources. Environ. Sci. Technol. Lett. 2022 , 9 (11), 983–990. Liddie, J. M.; Schaider, L. A.; Sunderland, E. M. Sociodemographic Factors Are Associated with the Abundance of PFAS Sources and Detection in U.S. Community Water Systems. Env. Sci Technol 2023 , 57 (21), 7902–7912. https://doi.org/10.1021/acs.est.2c07255. Maruzzo, A. J.; Hernandez, A. B.; Swartz, C. H.; Liddie, J. M.; Schaider, L. A. Socioeconomic Disparities in Exposures to PFAS and Other Unregulated Industrial Drinking Water Contaminants in US Public Water Systems. Environ. Health Perspect. 2025 , 133 (1), 017002. https://doi.org/10.1289/EHP14721. Tokranov, A. K.; Ransom, K. M.; Bexfield, L. M.; Lindsey, B. D.; Watson, E.; Dupuy, D. I.; Stackelberg, P. E.; Fram, M. S.; Voss, S. A.; Kingsbury, J. A.; Jurgens, B. C.; Smalling, K. L.; Bradley, P. M. Predictions of Groundwater PFAS Occurrence at Drinking Water Supply Depths in the United States. Science 2024 , 386 (6723), 748–755. https://doi.org/10.1126/science.ado6638. Rafiei, V.; Nejadhashemi, A. P. Watershed Scale PFAS Fate and Transport Model for Source Identification and Management Implications. Water Res. 2023 , 240 , 120073. https://doi.org/10.1016/j.watres.2023.120073. Advances and Research Gaps for PFAS Modeling in Watersheds and Receiving Waters . https://doi.org/10.1061/JOEEDU.EEENG-8137. Nasr, M.; Orwin, J. F. A Geospatial Approach to Identifying and Mapping Areas of Relative Environmental Pressure on Ecosystem Integrity. J. Environ. Manage. 2024 , 370 , 122445. https://doi.org/10.1016/j.jenvman.2024.122445. Farhadi, S.; Tatullo, S.; Ferrian, F. Comparative Analysis of Ensemble Learning Techniques for Enhanced Fatigue Life Prediction. Sci. Rep. 2025 , 15 (1), 11136. https://doi.org/10.1038/s41598-024-79476-y. Lundberg, S. M.; Lee, S.-I. A Unified Approach to Interpreting Model Predictions. In Advances in Neural Information Processing Systems ; Curran Associates, Inc., 2017; Vol. 30. Jiawei, T.; Yizhen, Z.; Jiajun, S.; Xuelu, S.; Chao, S.; Chunhui, Z. Occurrence and Characteristics of Perfluoroalkyl Substances (PFASs) in Electroplating Industrial Wastewater. Water Sci. Technol. 2019 , 79 (4), 731–740. https://doi.org/10.2166/wst.2019.092. Gilchrist, M. PFAS in the Metal Plating and Finishing Industry. Minnesota Pollution Control Agency , 2022. https://www.pca.state.mn.us/sites/default/files/gp3-05.pdf. Michigan Department of Environment, Great Lakes, and Energy. Targeted and Nontargeted Analysis of PFAS in Fume Suppressant Products at Chrome Plating Facilities. ; 2025. https://www.michigan.gov/-/media/Project/Websites/egle/Documents/Programs/WRD/IPP/pfas-ipp-intiative-identified-sources.pdf?rev=0f234a957d4947968ba3b44711a93e10 (accessed 2025-06-16). Qu, Y.; Huang, J.; Willand, W.; Weber, R. Occurrence, Removal and Emission of per- and Polyfluorinated Alkyl Substances (PFASs) from Chrome Plating Industry: A Case Study in Southeast China. Emerg. Contam. 2020 , 6 , 376–384. https://doi.org/10.1016/j.emcon.2020.10.001. U.S. ENVIRONMENTAL PROTECTION AGENCY-REGION 5. Chromium-Plating-PFOS-Study-EPA-Region-5-Sept-2009.Pdf ; 2009. https://cswab.org/wp-content/uploads/2018/09/Chromium-Plating-PFOS-Study-EPA-Region-5-Sept-2009.pdf (accessed 2025-06-16). North American Industry Classification System (NAICS) U.S. Census Bureau . https://www.census.gov/naics/?input=325211&year=2022&details=325211 (accessed 2025-06-16). Roostaei, J.; Colley, S.; Mulhern, R.; May, A. A.; Gibson, J. M. Predicting the Risk of GenX Contamination in Private Well Water Using a Machine-Learned Bayesian Network Model. J. Hazard. Mater. 2021 , 411 , 125075. https://doi.org/10.1016/j.jhazmat.2021.125075. Bao, J.; Yu, W.-J.; Liu, Y.; Wang, X.; Jin, Y.-H.; Dong, G.-H. Perfluoroalkyl Substances in Groundwater and Home-Produced Vegetables and Eggs around a Fluorochemical Industrial Park in China. Ecotoxicol. Environ. Saf. 2019 , 171 , 199–205. https://doi.org/10.1016/j.ecoenv.2018.12.086. Gebbink, W. A.; van Asseldonk, L.; van Leeuwen, S. P. J. Presence of Emerging Per- and Polyfluoroalkyl Substances (PFASs) in River and Drinking Water near a Fluorochemical Production Plant in the Netherlands. Environ. Sci. Technol. 2017 , 51 (19), 11057–11065. https://doi.org/10.1021/acs.est.7b02488. Glüge, J.; Scheringer, M.; Cousins, I. T.; DeWitt, J. C.; Goldenman, G.; Herzke, D.; Lohmann, R.; Ng, C. A.; Trier, X.; Wang, Z. An Overview of the Uses of Per-and Polyfluoroalkyl Substances (PFAS). Environ. Sci. Process. Impacts 2020 , 22 (12), 2345–2373. Agency for Toxic Substances and Disease Registry. Health Consultation: Perfluorochemical Releases at the 3M--Cottage Grove Facility ; U.S. Department of Health and Human Services, Public Health Service, Division of Health Assessment and Consultation, 2005. https://www.health.state.mn.us/communities/environment/hazardous/docs/sites/washington/3mcg0205.pdf?utm_source=chatgpt.com (accessed 2025-10-21). Sun, M.; Arevalo, E.; Strynar, M.; Lindstrom, A.; Richardson, M.; Kearns, B.; Pickett, A.; Smith, C.; Knappe, D. R. U. Legacy and Emerging Perfluoroalkyl Substances Are Important Drinking Water Contaminants in the Cape Fear River Watershed of North Carolina. Environ. Sci. Technol. Lett. 2016 , 3 (12), 415–419. https://doi.org/10.1021/acs.estlett.6b00398. Andres, K. L.; Olsen, G. W.; Krisko, R. M.; Nunnally, M. C.; Boeding, R. R.; Leniek, K. L.; Taiwo, O. A. An Investigation of 3M Cordova, IL Production Worker’s per- and Polyfluoroalkyl Substances Biomonitoring Results and Mortality Experience. Int. J. Hyg. Environ. Health 2024 , 256 , 114321. https://doi.org/10.1016/j.ijheh.2024.114321. North American Industry Classification System (NAICS) U.S. Census Bureau . https://www.census.gov/naics/?input=325998&year=2022&details=325998 (accessed 2025-07-15). Hepburn, E.; Madden, C.; Szabo, D.; Coggan, T. L.; Clarke, B.; Currell, M. Contamination of Groundwater with Per- and Polyfluoroalkyl Substances (PFAS) from Legacy Landfills in an Urban Re-Development Precinct. Environ. Pollut. 2019 , 248 , 101–113. https://doi.org/10.1016/j.envpol.2019.02.018. Harrad, S.; Drage, D. S.; Sharkey, M.; Berresheim, H. Perfluoroalkyl Substances and Brominated Flame Retardants in Landfill-Related Air, Soil, and Groundwater from Ireland. Sci. Total Environ. 2020 , 705 , 135834. Lang, J. R.; Allred, B. M.; Peaslee, G. F.; Field, J. A.; Barlaz, M. A. Release of Per- and Polyfluoroalkyl Substances (PFASs) from Carpet and Clothing in Model Anaerobic Landfill Reactors. Environ. Sci. Technol. 2016 , 50 (10), 5024–5032. https://doi.org/10.1021/acs.est.5b06237. Tolaymat, T.; Robey, N.; Krause, M.; Larson, J.; Weitz, K.; Parvathikar, S.; Phelps, L.; Linak, W.; Burden, S.; Speth, T.; Krug, J. A Critical Review of Perfluoroalkyl and Polyfluoroalkyl Substances (PFAS) Landfill Disposal in the United States. Sci. Total Environ. 2023 , 905 , 167185. https://doi.org/10.1016/j.scitotenv.2023.167185. Minet, L.; Wang, Z.; Shalin, A.; A. Bruton, T.; Blum, A.; F. Peaslee, G.; Schwartz-Narbonne, H.; Venier, M.; Whitehead, H.; Wu, Y.; L. Diamond, M. Use and Release of Per- and Polyfluoroalkyl Substances (PFASs) in Consumer Food Packaging in U.S. and Canada. 2022 . https://doi.org/10.1039/D2EM00166G. Herzke, D.; Olsson, E.; Posner, S. Perfluoroalkyl and Polyfluoroalkyl Substances (PFASs) in Consumer Products in Norway – A Pilot Study. Chemosphere 2012 , 88 (8), 980–987. https://doi.org/10.1016/j.chemosphere.2012.03.035. Kotthoff, M.; Müller, J.; Jürling, H.; Schlummer, M.; Fiedler, D. Perfluoroalkyl and Polyfluoroalkyl Substances in Consumer Products. Environ. Sci. Pollut. Res. 2015 , 22 (19), 14546–14559. https://doi.org/10.1007/s11356-015-4202-7. Ye, F.; Zushi, Y.; Masunaga, S. Survey of Perfluoroalkyl Acids (PFAAs) and Their Precursors Present in Japanese Consumer Products. Chemosphere 2015 , 127 , 262–268. https://doi.org/10.1016/j.chemosphere.2015.02.026. Neiman, B.; Vavra, J. S. The Rise of Niche Consumption. Reed, C.; Yu, T. E.; Hughes, D. Evaluating the Factors Influencing the Location Strategies of Specialty Grocers versus Traditional Supermarkets in the United States. Appl. Geogr. 2023 , 158 , 103034. https://doi.org/10.1016/j.apgeog.2023.103034. Chang, C.-J.; Ryan, P. B.; Smarr, M. M.; Kannan, K.; Panuwet, P.; Dunlop, A. L.; Corwin, E. J.; Barr, D. B. Serum Per- and Polyfluoroalkyl Substance (PFAS) Concentrations and Predictors of Exposure among Pregnant African American Women in the Atlanta Area, Georgia. Environ. Res. 2021 , 198 , 110445. https://doi.org/10.1016/j.envres.2020.110445. Glynn, A.; Kotova, N.; Dahlgren, E.; Lindh, C.; Jakobsson, K.; Gyllenhammar, I.; Lignell, S.; Nälsén, C. Determinants of Serum Concentrations of Perfluoroalkyl Acids (PFAAs) in School Children and the Contribution of Low-Level PFAA-Contaminated Drinking Water. Environ. Sci. Process. Impacts 2020 , 22 (4), 930–944. https://doi.org/10.1039/C9EM00497A. Schildroth, S.; Rodgers, K. M.; Strynar, M.; McCord, J.; Poma, G.; Covaci, A.; Dodson, R. E. Per-and Polyfluoroalkyl Substances (PFAS) and Persistent Chemical Mixtures in Dust from US Colleges. Environ. Res. 2022 , 206 , 112530. https://doi.org/10.1016/j.envres.2021.112530. Salawu, O. A.; Han, Z.; Senehi, N. L.; Roback, S. L.; Adeleye, A. S. Determination of Household Wastewater PFAS Composition and Concentrations via Neighborhood Sampling. J. Hazard. Mater. 2025 , 498 , 139834. https://doi.org/10.1016/j.jhazmat.2025.139834. Schlummer, M.; Gruber, L.; Fiedler, D.; Kizlauskas, M.; Mueller, J. Detection of Fluorotelomer Alcohols in Indoor Environments and Their Relevance for Human Exposure. Environ. Int. 2013 , 57–58 , 42–49. https://doi.org/10.1016/j.envint.2013.03.010. Merrett, H. C.; Chen, W. T.; Horng, J. J. A Systems Analysis Approach to Identifying Critical Success Factors in Drinking Water Source Protection Programs. SUSTAINABILITY 2019 , 11 (9). https://doi.org/10.3390/su11092606. Schlea, S. H.; Wilkes, W. Using the Toxic Substance Control Act for Source Water Protection. J. Am. WATER WORKS Assoc. 2021 , 113 (6), 26–36. https://doi.org/10.1002/awwa.1745. US EPA, O. Basic Information about Source Water Protection . https://www.epa.gov/sourcewaterprotection/basic-information-about-source-water-protection (accessed 2025-07-16). Yu, J.; Fu, Y.; Zhu, J.; Chen, X.; Zhong, Y.; Pan, Y.; Ma, Y. Marine Transport Barrier for Traditional and Emerging Per- and Polyfluoroalkyl Substances in the Southeast Indian Ocean and Antarctic Marginal Seas. Environ. Sci. Technol. 2025 , 59 (10), 5283–5292. https://doi.org/10.1021/acs.est.4c13574. Alam, M. S.; Abbasi, A.; Chen, G. Fate, Distribution, and Transport Dynamics of Per- and Polyfluoroalkyl Substances (PFASs) in the Environment. J. Environ. Manage. 2024 , 371 , 123163. https://doi.org/10.1016/j.jenvman.2024.123163. Lyu, X.; Xiao, F.; Shen, C.; Chen, J.; Park, C. M.; Sun, Y.; Flury, M.; Wang, D. Per- and Polyfluoroalkyl Substances (PFAS) in Subsurface Environments: Occurrence, Fate, Transport, and Research Prospect. Rev. Geophys. 2022 , 60 (3), e2021RG000765. https://doi.org/10.1029/2021RG000765. Peter, L. G.; Lee, L. S. Sources and Pathways of PFAS Occurrence in Water Sources: Relative Contribution of Land-Applied Biosolids in an Agricultural Dominated Watershed. Environ. Sci. Technol. 2025 , 59 (2), 1344–1353. US EPA, O. Addition of Certain PFAS to the TRI by the National Defense Authorization Act . https://www.epa.gov/toxics-release-inventory-tri-program/addition-certain-pfas-tri-national-defense-authorization-act (accessed 2025-07-11). The Fifth Unregulated Contaminant Monitoring Rule (UCMR 5) Data Summary: July 2025. Harris, C. R.; Millman, K. J.; Walt, S. J. van der; Gommers, R.; Virtanen, P.; Cournapeau, D.; Wieser, E.; Taylor, J.; Berg, S.; Smith, N. J.; Kern, R.; Picus, M.; Hoyer, S.; Kerkwijk, M. H. van; Brett, M.; Haldane, A.; Río, J. F. del; Wiebe, M.; Peterson, P.; Gérard-Marchant, P.; Sheppard, K.; Reddy, T.; Weckesser, W.; Abbasi, H.; Gohlke, C.; Oliphant, T. E. Array Programming with NumPy. Nature 2020 , 585 (7825), 357–362. https://doi.org/10.1038/s41586-020-2649-2. team, T. pandas development. Pandas-Dev/Pandas: Pandas, 2020. https://doi.org/10.5281/zenodo.3509134. Jordahl, K.; Bossche, J. V. den; Fleischmann, M.; Wasserman, J.; McBride, J.; Gerard, J.; Tratner, J.; Perry, M.; Badaracco, A. G.; Farmer, C.; Hjelle, G. A.; Snow, A. D.; Cochran, M.; Gillies, S.; Culbertson, L.; Bartos, M.; Eubank, N.; maxalbert; Bilogur, A.; Rey, S.; Ren, C.; Arribas-Bel, D.; Wasser, L.; Wolf, L. J.; Journois, M.; Wilson, J.; Greenhall, A.; Holdgraf, C.; Filipe; Leblanc, F. Geopandas/Geopandas: V0.8.1, 2020. https://doi.org/10.5281/zenodo.3946761. python-visualization. Folium, 2020. https://python-visualization.github.io/folium/. Lex, A.; Gehlenborg, N.; Strobelt, H.; Vuillemot, R.; Pfister, H. UpSet: Visualization of Intersecting Sets. IEEE Trans. Vis. Comput. Graph. 2014 , 20 (12), 1983–1992. https://doi.org/10.1109/TVCG.2014.2346248. python-nominatim. Nominatim. https://nominatim.org. US EPA, O. FRS EZ Query . https://www.epa.gov/frs/frs-ez-query (accessed 2025-06-20). Nanda (National Neighborhood Data Archive) | . https://nanda.isr.umich.edu/ (accessed 2025-06-20). Additional Declarations There is NO Competing Interest. Supplementary Files SIXAIforcontaminationsourceattributionPFAS.pdf Supplementary Information Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7935132","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":534152907,"identity":"d39dffa0-ebe6-4f99-ab06-78c906e5613b","order_by":0,"name":"Manish Kumar","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA9ElEQVRIiWNgGAWjYJCCA4wNDAZg1ge4GBuRWhhnQLUS1MIA08LMQ4wW3fazBw/+3MFgbHDt8LPPtm1/5AyOH3/A8KHsME4tZmfyEg7znmEwM7idZjw7t83AWLInx4Bxxjk8Wg7kGBxmbGOwMbidYMwM1JLYz5DDwMzbhkfL+TcGB3+CtaR/ZrZsM6hv43/+gPkvPi03cgwO8LaBHJZjzMzYZpDAL5FgAGTg0/LG4DBvm4Sx5O2cYsaec8aGM2cA7e05l47HYTnGH3+22Rj23U7fzPCjTE7e4Hz6wwc/yqxxaoECCVTuAULqR8EoGAWjYBTgBwDLEleUzcoowQAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0000-0001-5545-3793","institution":"University of Texas at Austin","correspondingAuthor":true,"prefix":"","firstName":"Manish","middleName":"","lastName":"Kumar","suffix":""},{"id":534152908,"identity":"6073d650-a6c9-47ab-9843-76872301a495","order_by":1,"name":"Priyanshu Gupta","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Priyanshu","middleName":"","lastName":"Gupta","suffix":""},{"id":534152909,"identity":"7b4c77ff-e2f0-4eb6-96de-6d6364fa8d86","order_by":2,"name":"Hogan Nance","email":"","orcid":"","institution":"University of Texas at Austin","correspondingAuthor":false,"prefix":"","firstName":"Hogan","middleName":"","lastName":"Nance","suffix":""},{"id":534152910,"identity":"f78763bb-ab91-46cb-96c2-039a52035c0b","order_by":3,"name":"Khang Nguyen","email":"","orcid":"","institution":"University of Texas at Austin","correspondingAuthor":false,"prefix":"","firstName":"Khang","middleName":"","lastName":"Nguyen","suffix":""},{"id":534152911,"identity":"d1561335-7c08-4f64-9f4e-261a8df196fd","order_by":4,"name":"Mateo Srivathanakul","email":"","orcid":"","institution":"University of Texas at Austin","correspondingAuthor":false,"prefix":"","firstName":"Mateo","middleName":"","lastName":"Srivathanakul","suffix":""},{"id":534152912,"identity":"c0807466-fbc6-4f47-88e6-31558d2b315e","order_by":5,"name":"Emilie Tang","email":"","orcid":"","institution":"University of Texas at Austin","correspondingAuthor":false,"prefix":"","firstName":"Emilie","middleName":"","lastName":"Tang","suffix":""},{"id":534152913,"identity":"a55aeecc-abe2-451d-bce8-06f6ed77a126","order_by":6,"name":"Jaden Deegan","email":"","orcid":"","institution":"University of Texas at Austin","correspondingAuthor":false,"prefix":"","firstName":"Jaden","middleName":"","lastName":"Deegan","suffix":""},{"id":534152914,"identity":"da6314bd-e03d-4325-b843-13f0ea9e9022","order_by":7,"name":"Raman Dhiman","email":"","orcid":"","institution":"University of Texas at Austin","correspondingAuthor":false,"prefix":"","firstName":"Raman","middleName":"","lastName":"Dhiman","suffix":""}],"badges":[],"createdAt":"2025-10-23 20:40:28","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7935132/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7935132/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":95267403,"identity":"3f31c3a4-1c43-4ee5-aa42-3af589fc0021","added_by":"auto","created_at":"2025-11-06 06:20:05","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":428790,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eOverview of PFAS contamination pathways, feature engineering, modeling framework, and interpretability.\u003c/strong\u003e(a) Conceptual diagram illustrating the major pathways through which per- and polyfluoroalkyl substances (PFAS) enter drinking water supplies, including industrial discharges, aqueous film-forming foam (AFFF) usage, landfills, and historical releases with transport via surface water (SW) and groundwater (GW) to public water systems (PWSs). (b) National distribution of PFAS detections (red) and non-detections (blue) in drinking water from U.S. EPA’s UCMR5 (as of October 2024) (c) Feature engineering strategy: 25 industrial/landfill indicators, 3 AFFF use indicators, and 2 sociodemographic variables aggregated to the Hydrologic Unit Code 8 (HUC8) watershed level, yielding 30-dimensional signatures. (d) Machine learning workflow: a Random Forest classifier trained with 10-fold cross-validation to predict PFAS detection, with model performance evaluated by mean receiver operating characteristic (ROC) curves (green line, shaded area ±1 s.d.). (e) SHAP analysis of the trained model showing feature contributions to PFAS detection probability, with impact \u0026gt; 0 indicating positive contributions, and impact \u0026lt; 0 indicating negative contribution towards PFAS detection. Features’ values are colored on red/blue scale and the direction of contribution is colored by salmon/teal scale. (f) Additive decomposition of SHAP values illustrating how different industrial, AFFF and sociodemographic features contribute towards the final predicted probability of PFAS detection.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7935132/v1/ed00a02816e6a9f464204355.png"},{"id":95267401,"identity":"c588868b-a7a1-46b2-b6de-5e6b6c4858d6","added_by":"auto","created_at":"2025-11-06 06:20:05","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":121712,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eReceiver Operating Characteristic (ROC) curves and mean absolute SHAP values and rankings demonstrate the robustness of the Random Forest (RF) models. \u003c/strong\u003eRetraining the model with the July 2025 UCMR5 update (which added ~2,000 public water systems) yields nearly identical predictive performance and feature importance patterns compared to the October 2024 model.\u003cstrong\u003e \u003c/strong\u003e(a) ROC curves showing mean area under the ROC (AUROC ± 1 s.d.) across 10 cross-validation folds for RF models trained on the UCMR5 data up to the respective cutoff dates. (b) Mean absolute SHAP values (± 1 s.d.) summarizing the relative importance of top predictors aggregated by thematic feature groups. Sociodemographic factors, metal treatment and fabrication and plastics and polymer manufacturing emerge as the top contributors. The rankings also reveal the stronger impact of chemical manufacturing sector (driven by misc. chemical product and preparation manufacturing) relative to the AFFF users.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7935132/v1/b7e64344ca6f7a0f64d1e16c.png"},{"id":95313460,"identity":"aeae841b-bd16-45ec-b436-643af61aa731","added_by":"auto","created_at":"2025-11-06 15:51:28","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":301892,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSHAP-based interpretability for exemplar HUC8 regions highlights distinct industrial and demographic drivers of PFAS detection. \u003c/strong\u003ea) Geographic distribution of industrial facilities (colored by sector, see Table S1) and Public Water Systems (PWSs) for two exemplar HUC8 subbasins: Upper Dry Watershed, CA (HUC8 180300009) and Red Lake Watershed, AZ (HUC8 15010007). Red pins indicate PWSs with reported PFAS detections (True label = 1), while blue pins indicate no detections (label = 0). b) SHAP waterfall plots showing the additive feature contributions to the model’s predicted probability of PFAS detection for each subbasin. The model prediction begins with a baseline probability of 0.498, and the red and blue lines push the model prediction towards 1 and 0 respectively. Arrows indicate the magnitude and direction of each feature’s contribution to the prediction.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-7935132/v1/ac17b3a9c8957f5235a66baa.png"},{"id":95267404,"identity":"be3d34ae-457e-4e6c-b73f-804ca55d351c","added_by":"auto","created_at":"2025-11-06 06:20:05","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":236817,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eInterpretability analysis reveals key industrial and sociodemographic drivers of PFAS contamination.\u003c/strong\u003e(b-h) SHAP impact distributions for the top 10 most impactful features distributed by thematic feature-groups. \u0026nbsp;Positive values increase the likelihood of a detection classification, while negative values push the prediction toward non-detection. Each point corresponds to an individual subbasin, with slight vertical jitter added for visualization; the blue–red gradient reflects the underlying magnitude of the feature value across subbasins. Features not displayed here show mixed and/or weaker influence (Supplementary Figure S2).\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-7935132/v1/843b924ddae229957091b0c8.png"},{"id":96916158,"identity":"a0353162-475f-4379-852b-c779a8ce1a9c","added_by":"auto","created_at":"2025-11-27 14:08:07","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3069377,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7935132/v1/8549ec61-d99e-4da1-8c72-3a6fd4953a54.pdf"},{"id":95267405,"identity":"7d8a9f01-7127-4d30-931e-99c01069a5ac","added_by":"auto","created_at":"2025-11-06 06:20:05","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":1538658,"visible":true,"origin":"","legend":"Supplementary Information","description":"","filename":"SIXAIforcontaminationsourceattributionPFAS.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7935132/v1/bc26d8866661f423ecbd27e4.pdf"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Explainable AI Reveals Statistical Associations Between Industrial Activity and PFAS Contamination of Public Water Systems","fulltext":[{"header":"Main","content":"\u003cp\u003emated 45% of the nation\u0026rsquo;s tap water\u003csup\u003e1\u003c/sup\u003e. This widespread detection indicates that the scale of contamination facing drinking water sources in the United States \u0026ndash; including public water systems (PWSs)- is both significant and concerning (\u003cstrong\u003eFigure 1a\u003c/strong\u003e). Over 90% of US residents get their drinking water from PWSs\u003csup\u003e2\u003c/sup\u003e. These PWSs often rely on groundwater or surface water sources that may already be contaminated with PFAS\u003csup\u003e3\u0026ndash;5\u003c/sup\u003e. Although several technologies have demonstrated efficacy in reducing PFAS concentrations at the laboratory and pilot scales\u003csup\u003e6\u0026ndash;11\u003c/sup\u003e, empirical data from operational drinking water plant indicate that, if present in raw water, PFAS compounds are typically detected in the finished/tap water at comparable concentrations\u003csup\u003e12\u003c/sup\u003e- indicating minimal removal of PFASs by most widely used drinking water treatment processes except advanced treatment processes including activated carbon adsorption, nanofiltration, or reverse osmosis\u003csup\u003e12,13\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003ePotential pathways for PFAS contamination of drinking water systems are diverse (\u003cstrong\u003eFigure 1b\u003c/strong\u003e) \u003csup\u003e13\u0026ndash;15\u003c/sup\u003e. Direct sources include: a) industrial releases from sectors involved in the manufacture and/or use of PFAS compounds, b) the use of aqueous film forming foams (AFFF) for firefighting activities, and c) consumer products (eg. stain-resistant coatings) \u003csup\u003e16\u003c/sup\u003e. Indirect sources can include a) biosolids application, and b) discharges from wastewater treatment, septic, stormwater and landfill systems.\u003csup\u003e17\u003c/sup\u003e Although PFAS discharges from indirect sources have been studied in detail, these pathways generally act as aggregators and/or conduits rather than being the point of introduction of PFAS into the wastewater streams. The true entry points lie with the direct sources, with whom the decision to use or manufacture PFAS-containing materials ultimately lies. Notably, EPA\u0026rsquo;s 2022 decision to monitor PFAS at the point of discharge through the National Pollutant Discharge Elimination System (NPDES) program, rather than in downstream carriers, underscores a shift in regulatory focus\u003csup\u003e18\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eWhile the U.S. Environmental Protection Agency (USEPA) has identified a broad range of industrial sectors potentially associated with PFAS use and release, much of the academic literature remains focused on a narrower set of sources. At the site level, numerous studies have investigated PFAS contamination in proximity to known AFFF users \u003csup\u003e15,19\u0026ndash;21\u003c/sup\u003e or individual industrial factories. However, these studies are typically limited to isolated cases and rarely contextualize the contribution of a specific industry relative to other potential sources. In contrast to site-specific investigations, studies conducted at regional\u003csup\u003e20,22\u0026ndash;28\u003c/sup\u003e, or national\u003csup\u003e29\u0026ndash;34\u003c/sup\u003e often attribute sources of PFAS contamination using broad categorical labels such as \u0026ldquo;industry\u0026rdquo; or \u0026ldquo;manufacturing\u0026rdquo;. As a result, despite widespread recognition that PFAS are used across a diverse array of industries, the relative influence of different sectors remains poorly characterized in large-scale, data driven assessments.\u003c/p\u003e\n\u003cp\u003eIn this study, we move past the broad categorical label of \u0026ldquo;industry\u0026rdquo; to systematically identify the specific sectors within industry that could act as key drivers of PFAS contamination of drinking water. By integrating the geospatial distribution of 25 industry sectors identified by Salvatore et. al.\u003csup\u003e31\u003c/sup\u003e as \u0026ldquo;presumptive PFAS contamination sites\u0026rdquo; with AFFF users (airports, military bases and fire-fighting training facilities), and sociodemographic characteristics, it may be possible to capture the complex and sector-specific patterns underlying PFAS contamination. A novel aspect of this work includes the use of explainable AI (XAI) methods to uncover these patterns linking different industry sectors to PFAS contamination.\u003c/p\u003e\n\u003cp\u003eWe use publicly available drinking water concentration data for eight PFAS compounds (PFPeA, PFBA, PFHxA, PFBS, PFOS, PFOA, PFHxS, and PFHpA) from the US EPA\u0026rsquo;s fifth Unregulated Contaminant Monitoring Rule (UCMR5) (\u003cstrong\u003esee\u003c/strong\u003e \u003cstrong\u003eFigure 1a\u003c/strong\u003e \u003cstrong\u003eand Supplementary Information, Table S2\u003c/strong\u003e). Given that PFAS fate and transport are governed by hydrological connectivity and mixing from multiple sources, individual PWS-specific drainage delineations would not be representative\u003csup\u003e35,36\u003c/sup\u003e. Instead of using arbitrary boundaries like counties or states, which often cut across natural stream networks, HUC8 subbasins are defined by the actual drainage patterns of a landscape\u003csup\u003e37\u003c/sup\u003e. Thus, we aggregated the PWS geolocations into Hydrological Unit Code 8 (HUC8) subbasins developed by the United States Geological Survey to capture watershed-level source-mixing and transport dynamics of the contaminants within these regions. (\u003cstrong\u003eFig 1c\u003c/strong\u003e). For each subbasin, we engineered a set of 30 descriptors (or features) to characterize potential drivers of PFAS contamination (\u003cstrong\u003eFig 1c\u003c/strong\u003e). The first 25 features each describe the geospatial distribution of a corresponding industry sector identified as \u0026ldquo;presumptive contamination sites\u0026rdquo;\u003csup\u003e31\u003c/sup\u003e, followed by three features- one for each type of AFFF discharge sites (i.e. airports, military bases and fire-fighting training facilities). The last 2 features describe sociodemographic characteristics of the subbasin. Machine learning (ML) models were then trained to learn from these descriptors in order to predict the likelihood of detecting PFAS in public water systems within each subbasin.\u003c/p\u003e\n\u003cp\u003eA key advance of this study is the use of SHAP analysis to interpret model predictions and quantify the contribution of individual features (\u003cstrong\u003eFig 1d,e\u003c/strong\u003e). This approach allowed us to uncover sector-specific patterns of association with PFAS contamination that remain hidden under coarser categorization schemes.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cstrong\u003eThe use of three different ensemble-based machine learning models - Random Forest (RF), XGBoost (XGB), and LightGBM (LGBM) \u0026ndash; proved effective in learning from watershed level features and predicting PFAS detections in the PWSs in a watershed.\u0026nbsp;\u003c/strong\u003eAll three architectures yielded comparable predictive and classification performance (\u003cstrong\u003eSupplementary Information,\u003c/strong\u003e \u003cstrong\u003eTable S3\u0026nbsp;\u003c/strong\u003eand \u003cstrong\u003eS4, Supplementary\u003c/strong\u003e \u003cstrong\u003eFig S1\u003c/strong\u003e) suggesting a strong and consistent predictive signal in the data. Across the three architectures, the PFOA- (AUROC:0.83, accuracy: 83.8%, F1: 0.825) and PFHxA- (AUROC:0.82, accuracy: 82%, F1: 0.77) specific models outperformed the other targets (\u003cstrong\u003eSupplementary Information,\u003c/strong\u003e \u003cstrong\u003eTable S4\u003c/strong\u003e), indicating that these two compounds exhibit stronger associations with the input features used in this study, albeit only marginally. The SUMPFAS model, trained on the combined data of all 8 PFAS compounds in this study, also showed comparable performance (AUROC: 0.78, sensitivity: 73.5, specificity: 73.5). Optimizing the threshold for Youden\u0026rsquo;s J-max improved the F1-score by raising the precision at the expense of recall (\u003cstrong\u003eSupplementary Information,\u003c/strong\u003e \u003cstrong\u003eTable S4\u003c/strong\u003e). Furthermore, re-training the RF models with updated data as of July 2025 yielded a minor improvement in AUROC, while preserving the overall feature importance rankings (\u003cstrong\u003eFig. 2a\u003c/strong\u003e). Modest shifts the distribution of SHAP impacts indicate slight reweighting of feature contributions without altering the dominant drivers of model predictions (\u003cstrong\u003eFig. 2b\u003c/strong\u003e). The minor changes in performance and |SHAP| value rankings on retraining with the updated July 2025 dataset indicates that the model captures stable underlying relationships between predictors and PFAS occurrence.\u003c/p\u003e\n\u003cp\u003eGiven comparable predictive performance across all three ensemble-based architectures, we focus our interpretability analysis on the RF model. RF is particularly well-suited for explainable AI applications, as it is composed of an ensemble of rule-based decision paths defined by cutoff splits (eg. \u003cem\u003eif value of feature \u0026gt; cutoff\u003c/em\u003e), making its internal logic more transparent than the gradient boosting-based methods\u003csup\u003e38\u003c/sup\u003e. Going forward, model performance is discussed in terms of the RF architecture-based SUMPFAS model trained on data up to October 2024.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSHAP analysis enables us to interpret the underlying patterns captured by our predictive models, highlighting the most influential factors driving PFAS occurrence.\u0026nbsp;\u003c/strong\u003eSHAP is a game-theoretic approach that decomposes the model predictions into additive feature contributions \u003csup\u003e39\u003c/sup\u003e (\u003cstrong\u003eFig. 1f\u003c/strong\u003e). \u003cstrong\u003eFigure 3 presents\u0026nbsp;\u003c/strong\u003eanalysis of the individual SHAP contributions for 2 exemplar HUC8 subbasins. The Upper Dry Watershed (HUC8 18030009) in California (\u003cstrong\u003eFig. 3a\u003c/strong\u003e) characterized by high industrial density and a large population was classified as a \u0026ldquo;detects\u0026rdquo; subbasin by the model, consistent with the six PWSs in the region reporting PFAS contamination. SHAP analysis highlights how the large population size and multiple industrial sectors jointly drive the prediction toward detection, yielding a model probability near 1.\u003c/p\u003e\n\u003cp\u003eIn contrast, the Red Lake Watershed (HUC8 15010007) in Arizona, contains only six relevant industrial facilities and a sparse population. Here, SHAP illustrates how a limited presence of Paint and Coating Manufacturing, and Soap and Other Detergents Manufacturing industries is insufficient to offset the suppressive effect of low population and absence of other important industrial drivers, leading the model to correctly predict a non-detection.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSHAP summary plots.\u0026nbsp;\u003c/strong\u003eBuilding on these individual case studies, we turn to a broader view of how individual features influence the model\u0026rsquo;s predictions across all HUC8 regions using SHAP summary plots. While the waterfall plots (\u003cstrong\u003eFigure 3b-c\u003c/strong\u003e), provide granular insights into how specific feature values drive model predictions for individual subbasins, they are inherently limited to single datapoints. To assess whether these patterns generalize across the entire dataset, we turn to SHAP beeswarm plots (\u003cstrong\u003eFigure 4a-g\u003c/strong\u003e), which provide a distributional view of feature contributions across all datapoints. These plots allow us to evaluate not only the magnitude, but also the directionality of each feature\u0026rsquo;s effect-i.e, whether higher or lower values of a feature consistently increase or decrease the predicted likelihood of PFAS detection. To simplify the subsequent discussions, \u003cstrong\u003eFig. 2b\u003c/strong\u003e illustrates the mean of absolute SHAP values of all the features grouped by the thematic feature groups defined in \u003cstrong\u003eSupplementary Information Table S1\u003c/strong\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAmong industrial sectors, three feature groups emerged as particularly influential in driving predicted PFAS detection probabilities, namely metal treatment \u0026amp; fabrication, plastics \u0026amp; polymer products, and chemical manufacturing. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eML Model recovers strong association of metal treatment, fabrication, and coating facilities\u003c/strong\u003e \u003cstrong\u003ewith PFAS contamination.\u003c/strong\u003e According to \u003cstrong\u003eFigure 3a\u003c/strong\u003e, subbasins with higher counts (orange and red) of metal treatment, and fabrication, and metal coating facilities exhibit consistently positive SHAP values, indicating that the presence of these industrial sectors push the model output toward PFAS detection. Conversely, their absence (blue) is associated with negative SHAP values, contributing to predictions towards non-detection. This directional trend suggests a strong and systematic link between metal processing activities and PFAS contamination risk.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eInterestingly, this sector is known to employ PFAS across multiple surface finishing processes for chromium and other metals, including electroplating, electroless plating, anodizing, coating, etching and surface cleaning.\u003csup\u003e40,41\u003c/sup\u003e Examples include the use of polytetrafluoroethylene (PTFE) in electroless nickel plating and powder coating, PFAS-based deposition aids in copper plating, PFAS containing wetting agents for metal plating on plastics, and fluorochemical blocking agents in aluminum foil production\u003csup\u003e41\u003c/sup\u003e. PFAS compounds are valued in metal finishing for their ability to prevent corrosion, reduce mechanical wear, promote flow of metal coatings and prevent cracks during drying. However, prior site-specific studies\u003csup\u003e42,43\u003c/sup\u003e as well as white papers issued by state pollution control agencies\u003csup\u003e41,44\u003c/sup\u003e have documented their potential release into the environment via multiple pathways. Our model extends this understanding by identifying a persistent, positive association between this sector and PFAS contamination at the national scale. Notably, this product category ranks as the most influential industrial predictor in 7 of the 8 PFAS specific models we trained (Supplementary \u003cstrong\u003eFigure S2\u003c/strong\u003e). This recurring pattern across nearly all models underscores the widespread and systematic role of the metal treatment and fabrication sector in contributing to PFAS contamination across the nation.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSimilarly, the strong and directional effect of the plastics and polymer category aligns with prior evidence\u003c/strong\u003e, particularly through the Plastics and Resin Manufacturing sector (\u003cstrong\u003eFig. 3b\u003c/strong\u003e), which includes fluoropolymer resins manufacturing, fluorohydrocarbon resins manufacturing, and polytetrafluoroethylene resins manufacturing\u003csup\u003e45\u003c/sup\u003e\u003cem\u003e\u0026nbsp;-\u003c/em\u003e sectors previously linked to PFAS contamination of groundwater wells in close proximity\u003csup\u003e24,27,46\u0026ndash;48\u003c/sup\u003e. The SHAP distribution shows that HUC8 regions with higher facility counts in this category (orange and red points) are consistently associated with positive contributions to the model\u0026rsquo;s prediction, suggesting that the presence of such facilities reliable elevates the risk of PFAS detection. These industrial operations often involve the use of PFAS as processing aids or raw materials, particularly in the extrusion and polymerization of fluorinated polymers\u003csup\u003e24,27,49\u003c/sup\u003e. Prior field investigations have found evidence of PFAS leaching and emissions from polymer manufacturing facilities into local water sources\u003csup\u003e24,27\u003c/sup\u003e. Our model generalizes these site-specific findings, demonstrating that even at the national scale, the presence of plastics and resin manufacturing is a robust and directional driver of predicted PFAS contamination.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eML model reveals an understudied influence of the All Other Miscellaneous Chemical Product and Preparation Manufacturing sector\u0026nbsp;\u003c/strong\u003e(\u003cstrong\u003ehereafter referred to as the specialty chemical manufacturing sector\u003c/strong\u003e). As shown in \u003cstrong\u003eFigure 3C\u003c/strong\u003e, subbasins with a higher number of specialty chemical manufacturing facilities (orange/red points) are associated with consistently positive SHAP values, suggesting the presence of these facilities increase the predicted probability of PFAS detection. This clear directionality reveals a strong role of this sector as a non-negligible contributor to contamination risk.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;This sector encompasses the production of a wide range of chemical products- from surfactants to performance-enhancing additives like brake fluids, and hydraulic fluids- many of which rely on PFAS chemistries\u003csup\u003e49\u003c/sup\u003e. As early as 2005, Prevedourous et. al.\u003csup\u003e16\u003c/sup\u003e estimated that the emissions from manufacturing of perfluorooctyl sulfonyl fluoride- (POSF) and fluorotelomer-based specialty products accounted approximately 4 % of total historical global PFCA emissions. \u0026nbsp;To our knowledge, the specialty chemical manufacturing in has received limited attention beyond a few site specific studies (for example \u003csup\u003e50\u0026ndash;52\u003c/sup\u003e ) compared to other potential contamination sources.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eNotably, this sector also includes facilities engaged in the production of fire-extinguishing agents and related formulations\u003csup\u003e53\u003c/sup\u003e. While AFFF-using sectors such as airports, and fire training facilities have justifiably received significant scrutiny, our analysis reveals that upstream specialty chemical manufactures may play an even greater role (\u003cstrong\u003eFig. 3c-d\u003c/strong\u003e). Furthermore, in all 9 of our models (8 PFAS-specific + 1 \u003cimg width=\"9\" height=\"22\" src=\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAA0AAAAhCAMAAAA1W9EDAAAAAXNSR0IArs4c6QAAAEhQTFRFAAAAAAAAAAA6AABmADqQAGa2OgAAOpDbZgAAZjoAZrb/kDoAkGaQkNv/tmYAtv//25A627Zm2////7Zm/9uQ/9u2//+2///bxw7W0gAAAAF0Uk5TAEDm2GYAAAAJcEhZcwAAFiUAABYlAUlSJPAAAAAZdEVYdFNvZnR3YXJlAE1pY3Jvc29mdCBPZmZpY2V/7TVxAAAAaElEQVQoU8WPSRKAIAwEE1cURZTt/z/VJKJ5gfatq4bMAPAxccBKdwB4bDdasCNZsdiTFUsGAXEm9WzgOFPJBke1PWCzvnod0tmgo9lwx42TfsHLjbRQTxy4XNrrozSROf0H/4iM/5ETQCoDbryujXgAAAAASUVORK5CYII=\" alt=\"image\"\u003ePFAS), this sector ranks higher than any of the well know AFFF users included in this study (Supplementary \u003cstrong\u003eFigure S2\u003c/strong\u003e). Hence, our results further challenge this omission, revealing strong and consistent contributions from the specialty chemical manufacturing sector to predicted risk of PFAS contamination.\u003c/p\u003e\n\u003cp\u003eThis contrast underscores the importance of broadening PFAS source attribution efforts to include upstream industrial producers. By highlighting the outsized and persistent influence of specialty chemical manufacturing, our findings point to a critical but underrecognized opportunity for advancing mitigation strategies through closer collaboration with chemical manufacturers. Greater engagement with this sector, paired with improved transparency around PFAS use in production, could inform the development of safer alternatives and support voluntary phase-outs.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEven among AFFF users, our model presents a more nuanced picture.\u003c/strong\u003e Although only firefighting training facilities ranked in the top 10 most impacting features, both firefighting training facilities and military bases (\u003cstrong\u003eSupplementary Information, Figure S2\u003c/strong\u003e) repeatedly show positive SHAP values at moderate to high facility densities, indicating a strong association with elevated PFAS risk. In contrast, the influence of airports appears more diffused across predictions (\u003cstrong\u003eSupplementary Information, Figure S2)\u003c/strong\u003e, suggesting that while airports may be recognized as important sources of localized PFAS contamination, their contribution is less systematic at the national scale.\u003c/p\u003e\n\u003cp\u003eOther industrial categories- namely textile, paper, \u0026amp; printing, waste management \u0026amp; chemical handling, electrical \u0026amp; electronics manufacturing, and petroleum \u0026amp; petrochemical products- demonstrated comparatively lower SHAP contributions in our national models (\u003cstrong\u003eFig. 3f-g, FigS2\u003c/strong\u003e), but their potential role in PFAS contamination should not be overlooked as discussed subsequently.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFirstly, lower SHAP magnitudes do not imply irrelevance.\u003c/strong\u003e In the case of textile, paper \u0026amp; printing category, SHAP plots (\u003cstrong\u003eFig. 3f\u003c/strong\u003e and \u003cstrong\u003eFig. S2\u003c/strong\u003e) reveal that specific sectors like commercial printing show clear directionality in SHAP values, with higher facility densities associated with positive model contributions. The processes in this sector often utilize PFAS-based surfactants and coatings to impart durability and print quality, pointing to potential sources of environmental release. However, the lower overall contribution of the category could stem from limited spatial prevalence, or lower mean absolute contributions of other sectors within that category.\u003c/p\u003e\n\u003cp\u003eSimilarly, SHAP plots for waste management \u0026amp; chemical handling sectors (\u003cstrong\u003eFig. 3g\u003c/strong\u003e) reveal that specific sectors like solid waste landfills and hazardous waste collection show a variable contributing at medium to high facility densities. The diffused but positive contribution at high facility densities may reflect the persistence of legacy PFAS-bearing materials in landfills as has been reported in past literature\u003csup\u003e54,54\u0026ndash;57\u003c/sup\u003e. However, the diffused nature of the contributions could also stem from episodic or spatially dispersed emissions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eIn contrast, the electrical and electronics manufacturing, and petroleum related sectors displayed both low SHAP magnitude and were absent from top feature rankings\u003c/strong\u003e. While these findings may indicate a limited role in PFAS violations nationally, it is also possible that PFAS usage in these sectors is better managed via closed-loop systems, or the facilities are not spatially well distributed (\u003cstrong\u003eSupplementary Information, Figure S2\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eOverall, muted national signals from specific sectors does not preclude their significance at regional or watershed scales.\u003c/strong\u003e This study is designed to identify systemic drivers of PFAS violations at the national scale, and as such it emphasizes features with consistent influence across diverse geographies. However, this absence of national level feature importance does not imply that these sectors are benign; rather, it could highlight the need for spatially disaggregated or case-specific analyses to detect and respond to localized contamination risks.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eBeyond industrial and facility-based sources, our model also reveals surprising patterns in socio-demographic drivers of PFAS contamination\u003c/strong\u003e (\u003cstrong\u003eFig. 3h\u003c/strong\u003e). Total population emerged as the most influential feature across all models- acting as a proxy for the density of human activity, infrastructure and consumer product use. This finding suggests that PFAS exposure may not be limited to isolated industrial discharges but may also be a diffused outcome of widespread societal behaviors- like the use of stain resistant coatings in apparel and non-stick cookware\u003csup\u003e56\u003c/sup\u003e, and a range of other PFAS containing consumer products\u003csup\u003e58\u0026ndash;61\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eA striking and counterintuitive insight from the SHAP contributions is the role of affluence.\u003c/strong\u003e The affluence metric used in this analysis reflects a composite of indicators of socio-demographic status- namely concentrations of adults with a bachelor\u0026rsquo;s degree or higher, with incomes \u0026gt;$125K, and employed in managerial and professional occupations. Our model suggests that subbasins with higher average affluence are more likely to be associated with PFAS detection in their drinking waters (\u003cstrong\u003eFig. 3h\u003c/strong\u003e). We hypothesize that this observation may be partially driven by the increased use of specialized consumer products in more affluent neighborhoods\u003csup\u003e62,63\u003c/sup\u003e.\u0026nbsp;While most consumer products do not contain any surfactants, PFAS are commonly used in products that serve specialized function like stain resistance, waterproofing, and non-stick performance, possibly making them more expensive and disproportionately consumed in wealthier communities. Additional empirical support for our hypothesis comes from studies reporting elevated PFAS concentrations in the blood serum of individuals from higher socioeconomic backgrounds\u003csup\u003e64\u0026ndash;66\u003c/sup\u003e. Furthermore, preliminary results from Salawu et. al.\u0026rsquo;s characterization of PFAS concentration in the household wastewater across three different neighborhoods suggest that more affluent neighborhoods discharge more PFAS into the sewer systems, compared to less affluent ones\u003csup\u003e67\u003c/sup\u003e. While PFAS exposure is often framed as an environmental justice issue affecting communities with limited resources or systemic vulnerabilities,\u003csup\u003e64\u0026ndash;66,68\u003c/sup\u003e our findings point to a more complex risk landscape- one in which patterns of economic activity and consumption may also contribute to elevated PFAS risk.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTaken together, these patterns underscore the importance of moving beyond traditional assumptions about PFAS risk and considering a refined range of industrial and demographic factors.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDespite the good-to-excellent model performance, limited regional mismatches between predicted and observed detections remain\u003c/strong\u003e. These instances too offer additional insight and highlight opportunities for future research and targeted policy interventions.\u003c/p\u003e\n\u003cp\u003eFor instance, we found several watersheds where the model predicts a high PFAS risk- often due to dense clusters of industrial activity (\u003cstrong\u003eSupplementary Information, Fig. S3a\u003c/strong\u003e) \u0026ndash; however, no PFAS were detected in the UCMR5 dataset. This discrepancy, called false positives, may reflect the presence of effective mitigation measures, such as advanced water treatment technologies, rigorous source water protection practices\u003csup\u003e69\u0026ndash;71\u003c/sup\u003e or hydrological barriers that reduce contaminant transport\u003csup\u003e72\u003c/sup\u003e. These locations may represent examples of successful risk management and warrant closer examination as potential case studies for best practices.\u003c/p\u003e\n\u003cp\u003eConversely, there are watersheds (\u003cstrong\u003eSupplementary Information, Fig. S3b\u003c/strong\u003e) where the model predicts a low risk despite confirmed detections in UCMR5. The prediction of low risk in these regions is typically driven by the lack of obvious industrial sources. This pattern of false negatives suggests that contamination in these watersheds may be arise from diffuse, non-industrial pathways - such as legacy PFAS contamination \u003csup\u003e54\u003c/sup\u003e, long-range hydrological connectivity\u003csup\u003e73,74\u003c/sup\u003e, or unrepresented sources such as septic systems, industrial activities outside the modeled sectors, and land-applied biosolids - underscoring the need for broader consideration of upstream inputs and unmonitored contributors. For example, Peter and Lee \u003csup\u003e75\u003c/sup\u003e demonstrated that PFAS loads in agriculture-dominated watersheds can be dominated by diffuse legacy sources despite a lack of industrial activity. In their study, tile-drained catchments under sustained biosolids applications since the 1980s exhibited total PFAS concentrations of 79 - 170ng/L, whereas most agricultural streams remained below 2ng/L.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe occurrence of false positives and false negatives highlights the inherent complexity of PFAS sources, fate and transport in the environment. These discrepancies underscore the influence of unmonitored or diffuse pathways that remain challenging to represent in current datasets. Together, these insights help advance the science of PFAS source attribution and help guide future monitoring and intervention strategies. Before concluding, we also acknowledge two important limitations inherent to our ML-based workflow and modeling framework, which arise from methodological choices rather than PFAS-specific factors.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eOne key limitation of our approach that needs to be recognized is that SHAP analysis captures statistical associations, not causal mechanisms.\u003c/strong\u003e Establishing causality for each PFAS detection would require detailed, site-specific investigations- which is beyond the scope of this national scale study. These associations should be viewed as hypothesis-generating, offering a starting point for more targeted investigations. Validating these findings will require facility-level data such as the Toxic Release Inventory (TRI), and the National Pollution Discharge Elimination Systems (NPDES) permits for each industrial facility. Such efforts could help distinguish between active emissions, legacy contamination and background levels. However, the requirement of reporting PFAS under either of systems is rather recent, and the databases are still in development\u003csup\u003e18,76\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAnother limitation is that the use of the UCMR 5 database limits our study to contamination in public water systems, excluding private wells that serve an estimated 13-14% of the US population\u003c/strong\u003e\u003cstrong\u003e\u003csup\u003e1\u003c/sup\u003e\u003c/strong\u003e\u003cstrong\u003e.\u003c/strong\u003e Previous site-specific studies as well as state-level analyses have studied the contamination of private wells in proximity to know sources. However, to the best of our knowledge, there is no publicly available dataset on the PFAS concentrations in the private wells across the nation.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis study demonstrates the utility of explainable AI in identifying and interpreting geospatial patterns underlying PFAS contamination in public water systems at the national scale. By integrating geospatial data about industry sectors and socio-demographic data, we provide a nuanced view of the factors most strongly associated with PFAS detections in drinking water. Among the key findings, the presence of three industrial categories- specialty chemical manufacturing, metal treatment and fabrication and plastics and polymer products- emerged as consistent patterns associated with PFAS detections, enabling our models to achieve AUC scores in the range of 0.75-0.82. These findings align with known PFAS use patterns while also bringing attention to production-process based sources that have received less focus in prior research.\u003c/p\u003e\n\u003cp\u003eOur models further highlight important distinctions within AFFF use related sources: firefighting training stations and military installations show stronger and more consistent associations with PFAS detections than airports. Beyond point-source contributors, we find that total population is the most influential feature, indicating the broader role of human activity in shaping PFAS presence. Perhaps unexpectedly, our results also reveal that higher-income communities are more likely to be associated with detections, pointing to a more complex spatial risk profile than typically captured in existing frameworks.\u003c/p\u003e\n\u003cp\u003eWhile the patterns learned by our models reveal robust statistical associations, they must not be mistaken to be establishing causality. Further validation using facility level discharge data (e.g. from TRI database, or NPDES permits) will be critical for confirming pathways of emissions into the environment. Additionally, the UCMR database does not exclude private well data or even the very small public water systems (serving \u0026lt;3,300 persons), thereby limiting the generalizability of our findings to the entire U.S. population. Despite these limitations, the strong performance of our models across multiple machine learning architectures supports the reliability of our results. The use of Random Forest model instead of more complex models allows easier interpretability.\u003c/p\u003e\n\u003cp\u003eTogether, these findings offer a scalable and interpretable framework for identifying high-risk areas, supporting monitoring priorities, and informing more targeted regulatory and policy responses.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e\u003cstrong\u003ea.\u0026nbsp; PFAS DETECTION AT U.S. PUBLIC WATER SYSTEMS\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe foundation of this model is the fifth release of the Fifth Unregulated Contaminant Monitoring Rule (UCMR 5) dataset published by the U.S. Environmental Protection Agency (EPA) in October 2024. This data set provides 1,133,968 analytical results for measurement of 29 PFAS compounds and lithium measured at 7,237 public water systems (PWS) in the United States. Each PWS in the dataset is identified by a unique Public Water System ID (PWSID). This dataset represents ~55% of the data that the EPA expects to obtain under the UCMR 5 ruling by the completion of reporting in 2026.\u003c/p\u003e\n\u003cp\u003eThis study excluded PWS located in Alaska, Hawaii, and US territories due to differing geographical context- removing 131 PWSIDs from the datasets.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThis study focused on the detection of only 8 of the 29 PFAS compounds (PFOS, PFOA, PFNA, PFDA, PFHxS, PFHpA, PFBS, PFHxA) - all with the frequency of detection \u0026gt;300. 2,317 facilities detected at least one of these 8 contaminants to be above the proposed MRL threshold\u003csup\u003e77\u003c/sup\u003e, 4,722 sites did not detect any contaminants to be above the threshold, and 67 facilities did not report any measurements for any of the 8 contaminants. The 67 facilities were further excluded from downstream analysis.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFinally, each PWSIDs was geocoded using the zipcodes of the service area provided by the UMCR 5. 215 facilities did not have any zipcode in the UMCR file and were excluded from further study. The resulting dataset had 6824 unique PWSIDs with geolocations and data on detection/non-detection of the PFASs of interest.\u003c/p\u003e\n\u003cp\u003eAn interactive map of the public water systems in continental United States colored by detection of PFAS, and the respective PFAS concentrations has been made available at priyanshurg.github.io/pfas_interactive_maps/industrial_densities.html. This map will be updated with the release of newer datasets under the UCMR 5 ruling.\u003c/p\u003e\n\u003cp\u003ePreliminary data analysis of this dataset was conducted in Python using the NumPy\u003csup\u003e78\u003c/sup\u003e, Pandas\u003csup\u003e79\u003c/sup\u003e, GeoPandas\u003csup\u003e80\u003c/sup\u003e, Folium\u003csup\u003e81\u003c/sup\u003e, Upset\u003csup\u003e82\u003c/sup\u003e and Nominatim\u003csup\u003e83\u003c/sup\u003e.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eb. POTENTIAL PFAS SOURCES AND SOCIODEMOGRAPHIC FEATURES\u003c/p\u003e\n\u003cp\u003eThis study considered 28 point sources of PFAS \u0026ndash; 25 different industrial sectors, and 3 AFFF foam users. The North American Industry Classification System (NAICS) codes categorize businesses establishments based on their primary type of economic activity, enabling uniform, comparable and standardized industry data analysis. NAICS codes have been employed by researchers, state environmental agencies as well as the EPA to identify facilities suspected of PFAS use. Salvatore et. al.\u003csup\u003e31\u003c/sup\u003e identified a list of 38 NAICS codes- around whose factories the existence of PFAS can be presumed. Self-reported data of the facilities within these 38 NAICS codes were downloaded from the EPA\u0026rsquo;s Facility Registry Service (FRS) EZ Query\u003csup\u003e84\u003c/sup\u003e. NAICS codes (hereon also referred to as industry sectors) with self-reported data on 500 or more factories were selected for model development. A total of 108,573 facilities across 25 different industry sectors were shortlisted as shown in Table 1. To simplify the analysis, the industry sectors were further clubbed by the authors based on the type of products/processes expected to be performed in the industry sector (\u003cstrong\u003eTable 1\u003c/strong\u003e).\u003c/p\u003e\n\u003cp\u003eUsing the self-reported geospatial data on the potential point sources of PFAS, average density of the different industry sectors in each HUC8 region was calculated as:\u003cbr\u003e\u0026nbsp;\u0026nbsp;\u003cimg width=\"624\" height=\"45\" src=\"data:image/png;base64,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\" alt=\"image\"\u003e\u003c/p\u003e\n\u003cp\u003eIn addition to the manufacturing and industrial process, PFAS are also used extensively as a surfactant in Aqueous Film Forming Foam (AFFF). PFAS contamination is also expected wherever AFFF is discharged, particularly military sites, major airports and fire training areas. Geospatial datasets of airports, firefighting training facilities and military-installations across the continental US were obtained from publicly available sources.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSociodemographic features used as inputs for the model included: a combination of total population, and affluence from National Neighborhood Data Archive (NaNDA)\u003csup\u003e85\u003c/sup\u003e. Since these datasets are only available at a zipcode level instead of HUC8 level, a mean_affluence is calculated for each HUC8 region. The averaging is done by geocoding the zipcodes provided in the datasets, and then performing an average over the zipcodes that lie inside a certain HUC8 region.\u003c/p\u003e\n\u003cp\u003eMaps showing the distribution of facilities under each NAICS code across the continental United States is available at: priyanshurg.github.io/pfas_interactive_maps/naics_codes_distributions.html.\u003c/p\u003e\n\u003cp\u003ec. MODEL TRAINING\u003c/p\u003e\n\u003cp\u003eA total of 9 random forest classifier models \u0026ndash; 8 models for each of the 8 most occurring PFASs, and 1 model, called \u0026nbsp;\u003cimg width=\"9\" height=\"22\" src=\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAA0AAAAhCAMAAAA1W9EDAAAAAXNSR0IArs4c6QAAAEhQTFRFAAAAAAAAAAA6AABmADqQAGa2OgAAOpDbZgAAZjoAZrb/kDoAkGaQkNv/tmYAtv//25A627Zm2////7Zm/9uQ/9u2//+2///bxw7W0gAAAAF0Uk5TAEDm2GYAAAAJcEhZcwAAFiUAABYlAUlSJPAAAAAZdEVYdFNvZnR3YXJlAE1pY3Jvc29mdCBPZmZpY2V/7TVxAAAAaElEQVQoU8WPSRKAIAwEE1cURZTt/z/VJKJ5gfatq4bMAPAxccBKdwB4bDdasCNZsdiTFUsGAXEm9WzgOFPJBke1PWCzvnod0tmgo9lwx42TfsHLjbRQTxy4XNrrozSROf0H/4iM/5ETQCoDbryujXgAAAAASUVORK5CYII=\" alt=\"image\"\u003ePFAS, encompassing all 8 compounds \u0026ndash; were trained in this study. Each classifier was trained using 100 decision trees, each with a maximum depth of 10 to limit model complexity. The minimum number of samples required to split an internal node was set to 2, and each leaf node was required to contain at least one sample. For each split, a random subset of features equal to the square root of the total number of predictors was considered to introduce feature-level randomness and promote diversity among the decision trees.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe models were evaluated using stratified 10-fold cross-validation to ensure robustness and repeated across five different random seeds (random_state\u0026nbsp;\u0026isin;\u0026nbsp;{42, 11, 101, 25, 50}) to ensure reproducibility of the reported results. For each fold, features were standardized using a StandardScaler fitted on the training data in the fold.\u003c/p\u003e\n\u003cp\u003eModel performance was assessed using the area under the receiver-operating-curver (AUC-ROC), with interpolated true positive rates (TPRs) and false positive rates (FPRs) computed to estimate a mean AUC-ROC and associated variability. Additional metrics like accuracy precision, recall, f1-score, and Youden\u0026rsquo;s J-Index were calculated at a default prediction probability threshold of 0.5 for each model. The prediction probability threshold was then optimized for each model by maximizing the Youden\u0026rsquo;s J-Index obtained the same metrics at the optimal threshold.\u003c/p\u003e\n\u003cp\u003eFinally, Shapley Additive exPlanations (SHAP)\u003csup\u003e39\u003c/sup\u003e values were calculated for each of the 5 trainings of each of the 9 models. Mean absolute SHAP values for detection of PFAS were computed for each feature in the model\u0026rsquo;s input. To simplify analysis, we also add up the mean absolute SHAP values of features under each product category to derive a \u003cem\u003etotal impact of product category\u003c/em\u003e for each category based on categorization in \u003cstrong\u003eSupplementary\u003c/strong\u003e \u003cstrong\u003eTable S1\u003c/strong\u003e.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements:\u003c/strong\u003e This work was supported by two specific endowments to MK – Doug N Williams Memorial Centennial Fellowship in Engneering, UT Austin and the Ashley Ashley H. Priddy Centennial Professorship in Engineering, UT Austin. Partial funding from the WoodNext Foundation is also acknowledged. The authors would also like to acknowledge the contributions of the students of course EVE 310: Sustainable Systems Engineering, Fall 2024, under the Maseeh Department for Civil, Architectural, and Environmental Engineering at UT Austin.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability Statement\u003c/strong\u003e: All the data and code used in this study are available on Github (https://github.com/priyanshurg/Predicting-PFAS-Drivers).\u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eSmalling, K. L.; Romanok, K. M.; Bradley, P. M.; Morriss, M. C.; Gray, J. L.; Kanagy, L. K.; Gordon, S. E.; Williams, B. M.; Breitmeyer, S. E.; Jones, D. K.; others. Per-and Polyfluoroalkyl Substances (PFAS) in United States Tapwater: Comparison of Underserved Private-Well and Public-Supply Exposures and Associated Health Implications. \u003cem\u003eEnviron. Int.\u003c/em\u003e\u003cstrong\u003e2023\u003c/strong\u003e, \u003cem\u003e178\u003c/em\u003e, 108033.\u003c/li\u003e\n\u003cli\u003eInformation about Public Water Systems | US EPA \u0026mdash; Epa.Gov. https://www.epa.gov/dwreginfo/information-about-public-water-systems.\u003c/li\u003e\n\u003cli\u003eKurwadkar, S.; Dane, J.; Kanel, S. R.; Nadagouda, M. N.; Cawdrey, R. W.; Ambade, B.; Struckhoff, G. C.; Wilkin, R. \u003cem\u003ePer\u003c/em\u003e- and Polyfluoroalkyl Substances in Water and Wastewater: A Critical Review of Their Global Occurrence and Distribution. \u003cem\u003eSci. Total Environ.\u003c/em\u003e\u003cstrong\u003e2022\u003c/strong\u003e, \u003cem\u003e809\u003c/em\u003e, 151003. https://doi.org/10.1016/j.scitotenv.2021.151003.\u003c/li\u003e\n\u003cli\u003eSims, J. L.; Stroski, K. M.; Kim, S.; Killeen, G.; Ehalt, R.; Simcik, M. F.; Brooks, B. W. Global Occurrence and Probabilistic Environmental Health Hazard Assessment of Per- and Polyfluoroalkyl Substances (PFASs) in Groundwater and Surface Waters. \u003cem\u003eSci. Total Environ.\u003c/em\u003e\u003cstrong\u003e2022\u003c/strong\u003e, \u003cem\u003e816\u003c/em\u003e, 151535. https://doi.org/10.1016/j.scitotenv.2021.151535.\u003c/li\u003e\n\u003cli\u003eMasoner, J. R.; Kolpin, D. W.; Cozzarelli, I. M.; Barber, L. B.; Burden, D. S.; Foreman, W. T.; Forshay, K. J.; Furlong, E. T.; Groves, J. F.; Hladik, M. L.; Hopton, M. E.; Jaeschke, J. B.; Keefe, S. H.; Krabbenhoft, D. P.; Lowrance, R.; Romanok, K. M.; Rus, D. L.; Selbig, W. R.; Williams, B. H.; Bradley, P. M. Urban Stormwater: An Overlooked Pathway of Extensive Mixed Contaminants to Surface and Groundwaters in the United States. \u003cem\u003eEnviron. Sci. Technol.\u003c/em\u003e\u003cstrong\u003e2019\u003c/strong\u003e, \u003cem\u003e53\u003c/em\u003e (17), 10070\u0026ndash;10081. https://doi.org/10.1021/acs.est.9b02867.\u003c/li\u003e\n\u003cli\u003eShi, H.; Wang, Y.; Li, C.; Pierce, R.; Gao, S.; Huang, Q. Degradation of Perfluorooctanesulfonate by Reactive Electrochemical Membrane Composed of Magneli Phase Titanium Suboxide. \u003cem\u003eEnviron. Sci. Technol.\u003c/em\u003e\u003cstrong\u003e2019\u003c/strong\u003e, \u003cem\u003e53\u003c/em\u003e (24), 14528\u0026ndash;14537.\u003c/li\u003e\n\u003cli\u003eLi, Z.; Li, S.; Tang, Y.; Li, X.; Wang, J.; Li, L. Highly Efficient Degradation of Perfluorooctanoic Acid: An Integrated Photo-Electrocatalytic Ozonation and Mechanism Study. \u003cem\u003eChem. Eng. J.\u003c/em\u003e\u003cstrong\u003e2020\u003c/strong\u003e, \u003cem\u003e391\u003c/em\u003e, 123533.\u003c/li\u003e\n\u003cli\u003eLei, Y.-J.; Tian, Y.; Sobhani, Z.; Naidu, R.; Fang, C. Synergistic Degradation of PFAS in Water and Soil by Dual-Frequency Ultrasonic Activated Persulfate. \u003cem\u003eChem. Eng. J.\u003c/em\u003e\u003cstrong\u003e2020\u003c/strong\u003e, \u003cem\u003e388\u003c/em\u003e, 124215.\u003c/li\u003e\n\u003cli\u003eLe, T. X. H.; Haflich, H.; Shah, A. D.; Chaplin, B. P. Energy-Efficient Electrochemical Oxidation of Perfluoroalkyl Substances Using a Ti4O7 Reactive Electrochemical Membrane Anode. \u003cem\u003eEnviron. Sci. Technol. Lett.\u003c/em\u003e\u003cstrong\u003e2019\u003c/strong\u003e, \u003cem\u003e6\u003c/em\u003e (8), 504\u0026ndash;510.\u003c/li\u003e\n\u003cli\u003e Soker, O.; Tajdini, B.; Abarca-Perez, A.; Wadia, A.; Bellona, C.; Hao, S.; Doudrick, K.; Strathmann, T. J. Reuse of Spent Granular Activated Carbon for PFAS Removal Following Hydrothermal Alkaline Treatment. \u003cem\u003eWater Res.\u003c/em\u003e\u003cstrong\u003e2025\u003c/strong\u003e, 123794.\u003c/li\u003e\n\u003cli\u003e Soriano, \u0026Aacute;.; Gorri, D.; Urtiaga, A. Efficient Treatment of Perfluorohexanoic Acid by Nanofiltration Followed by Electrochemical Degradation of the NF Concentrate. \u003cem\u003eWater Res.\u003c/em\u003e\u003cstrong\u003e2017\u003c/strong\u003e, \u003cem\u003e112\u003c/em\u003e, 147\u0026ndash;156.\u003c/li\u003e\n\u003cli\u003e Rahman, M. F.; Peldszus, S.; Anderson, W. B. Behaviour and Fate of Perfluoroalkyl and Polyfluoroalkyl Substances (PFASs) in Drinking Water Treatment: A Review. \u003cem\u003eWater Res.\u003c/em\u003e\u003cstrong\u003e2014\u003c/strong\u003e, \u003cem\u003e50\u003c/em\u003e, 318\u0026ndash;340. https://doi.org/10.1016/j.watres.2013.10.045.\u003c/li\u003e\n\u003cli\u003e Wee, S. Y.; Aris, A. Z. Environmental Impacts, Exposure Pathways, and Health Effects of PFOA and PFOS. \u003cem\u003eEcotoxicol. Environ. Saf.\u003c/em\u003e\u003cstrong\u003e2023\u003c/strong\u003e, \u003cem\u003e267\u003c/em\u003e, 115663.\u003c/li\u003e\n\u003cli\u003e De Silva, A. O.; Armitage, J. M.; Bruton, T. A.; Dassuncao, C.; Heiger‐Bernays, W.; Hu, X. C.; K\u0026auml;rrman, A.; Kelly, B.; Ng, C.; Robuck, A.; Sun, M.; Webster, T. F.; Sunderland, E. M. PFAS Exposure Pathways for Humans and Wildlife: A Synthesis of Current Knowledge and Key Gaps in Understanding. \u003cem\u003eEnviron. Toxicol. Chem.\u003c/em\u003e\u003cstrong\u003e2021\u003c/strong\u003e, \u003cem\u003e40\u003c/em\u003e (3), 631\u0026ndash;657. https://doi.org/10.1002/etc.4935.\u003c/li\u003e\n\u003cli\u003e Sunderland, E. M.; Hu, X. C.; Dassuncao, C.; Tokranov, A. K.; Wagner, C. C.; Allen, J. G. A Review of the Pathways of Human Exposure to Poly- and Perfluoroalkyl Substances (PFASs) and Present Understanding of Health Effects. \u003cem\u003eJ. Expo. Sci. Environ. Epidemiol.\u003c/em\u003e\u003cstrong\u003e2019\u003c/strong\u003e, \u003cem\u003e29\u003c/em\u003e (2), 131\u0026ndash;147. https://doi.org/10.1038/s41370-018-0094-1.\u003c/li\u003e\n\u003cli\u003e Prevedouros, K.; Cousins, I. T.; Buck, R. C.; Korzeniowski, S. H. Sources, Fate and Transport of Perfluorocarboxylates. \u003cem\u003eEnviron. Sci. Technol.\u003c/em\u003e\u003cstrong\u003e2006\u003c/strong\u003e, \u003cem\u003e40\u003c/em\u003e (1), 32\u0026ndash;44. https://doi.org/10.1021/es0512475.\u003c/li\u003e\n\u003cli\u003e Dasu, K.; Xia, X.; Siriwardena, D.; Klupinski, T. P.; Seay, B. Concentration Profiles of Per- and Polyfluoroalkyl Substances in Major Sources to the Environment. \u003cem\u003eJ. Environ. Manage.\u003c/em\u003e\u003cstrong\u003e2022\u003c/strong\u003e, \u003cem\u003e301\u003c/em\u003e, 113879. https://doi.org/10.1016/j.jenvman.2021.113879.\u003c/li\u003e\n\u003cli\u003e Fox, R. Addressing PFAS Discharges in NPDES Permits and Through the Pretreatment Program and Monitoring Programs. https://www.epa.gov/system/files/documents/2022-12/NPDES_PFAS_State%20Memo_December_2022.pdf (accessed 2025-07-11).\u003c/li\u003e\n\u003cli\u003e Rotander, A.; Toms, L.-M. L.; Aylward, L.; Kay, M.; Mueller, J. F. Elevated Levels of PFOS and PFHxS in Firefighters Exposed to Aqueous Film Forming Foam (AFFF). \u003cem\u003eEnviron. Int.\u003c/em\u003e\u003cstrong\u003e2015\u003c/strong\u003e, \u003cem\u003e82\u003c/em\u003e, 28\u0026ndash;34. https://doi.org/10.1016/j.envint.2015.05.005.\u003c/li\u003e\n\u003cli\u003e Wei\u0026szlig;, O.; Wiesm\u0026uuml;ller, G. A.; Bunte, A.; G\u0026ouml;en, T.; Schmidt, C. K.; Wilhelm, M.; H\u0026ouml;lzer, J. Perfluorinated Compounds in the Vicinity of a Fire Training Area \u0026ndash; Human Biomonitoring among 10 Persons Drinking Water from Contaminated Private Wells in Cologne, Germany. \u003cem\u003eInt. J. Hyg. Environ. Health\u003c/em\u003e\u003cstrong\u003e2012\u003c/strong\u003e, \u003cem\u003e215\u003c/em\u003e (2), 212\u0026ndash;215. https://doi.org/10.1016/j.ijheh.2011.08.016.\u003c/li\u003e\n\u003cli\u003e Barzen-Hanson, K. A.; Roberts, S. C.; Choyke, S.; Oetjen, K.; McAlees, A.; Riddell, N.; McCrindle, R.; Ferguson, P. L.; Higgins, C. P.; Field, J. A. Discovery of 40 Classes of Per- and Polyfluoroalkyl Substances in Historical Aqueous Film-Forming Foams (AFFFs) and AFFF-Impacted Groundwater. \u003cem\u003eEnviron. Sci. Technol.\u003c/em\u003e\u003cstrong\u003e2017\u003c/strong\u003e, \u003cem\u003e51\u003c/em\u003e (4), 2047\u0026ndash;2057. https://doi.org/10.1021/acs.est.6b05843.\u003c/li\u003e\n\u003cli\u003e Hu, X. C.; Ge, B.; Ruyle, B. J.; Sun, J.; Sunderland, E. M. A Statistical Approach for Identifying Private Wells Susceptible to Perfluoroalkyl Substances (PFAS) Contamination. \u003cem\u003eEnv. Sci Technol Lett\u003c/em\u003e\u003cstrong\u003e2021\u003c/strong\u003e, \u003cem\u003e8\u003c/em\u003e (7), 596\u0026ndash;602. https://doi.org/10.1021/acs.estlett.1c00264.\u003c/li\u003e\n\u003cli\u003e Zhang, X.; Lohmann, R.; Dassuncao, C.; Hu, X. C.; Weber, A. K.; Vecitis, C. D.; Sunderland, E. M. Source Attribution of Poly- and Perfluoroalkyl Substances (PFASs) in Surface Waters from Rhode Island and the New York Metropolitan Area. \u003cem\u003eEnv. Sci Technol Lett\u003c/em\u003e\u003cstrong\u003e2016\u003c/strong\u003e, \u003cem\u003e3\u003c/em\u003e (9), 316\u0026ndash;321. https://doi.org/10.1021/acs.estlett.6b00255.\u003c/li\u003e\n\u003cli\u003e Hoffman, K.; Webster, T. F.; Bartell, S. M.; Weisskopf, M. G.; Fletcher, T.; Vieira, V. M. Private Drinking Water Wells as a Source of Exposure to Perfluorooctanoic Acid (PFOA) in Communities Surrounding a Fluoropolymer Production Facility. \u003cem\u003eEnviron. Health Perspect.\u003c/em\u003e\u003cstrong\u003e2011\u003c/strong\u003e, \u003cem\u003e119\u003c/em\u003e (1), 92\u0026ndash;97. https://doi.org/10.1289/ehp.1002503.\u003c/li\u003e\n\u003cli\u003e Kosiarski, K.; Veith, T. L.; Kibuye, F. A.; Fetter, J.; Boser, S.; Vanden Heuvel, J. P.; Thompson, C. L.; Preisendanz, H. E. Geospatial and Socioeconomic Factors of PFAS Contamination in Private Drinking Water Wells: Insights for Monitoring and Management. \u003cem\u003eJ. Environ. Manage.\u003c/em\u003e\u003cstrong\u003e2025\u003c/strong\u003e, \u003cem\u003e388\u003c/em\u003e, 125863. https://doi.org/10.1016/j.jenvman.2025.125863.\u003c/li\u003e\n\u003cli\u003e Panikkar, B.; Lemmond, B.; Allen, L.; DiPirro, C.; Kasper, S. Making the Invisible Visible: Results of a Community-Led Health Survey Following PFAS Contamination of Drinking Water in Merrimack, New Hampshire. \u003cem\u003eEnviron. Health\u003c/em\u003e\u003cstrong\u003e2019\u003c/strong\u003e, \u003cem\u003e18\u003c/em\u003e (1), 79. https://doi.org/10.1186/s12940-019-0513-3.\u003c/li\u003e\n\u003cli\u003e Bach, C.; Dauchy, X.; Boiteux, V.; Colin, A.; Hemard, J.; Sagres, V.; Rosin, C.; Munoz, J.-F. The Impact of Two Fluoropolymer Manufacturing Facilities on Downstream Contamination of a River and Drinking Water Resources with Per- and Polyfluoroalkyl Substances. \u003cem\u003eEnviron. Sci. Pollut. Res.\u003c/em\u003e\u003cstrong\u003e2017\u003c/strong\u003e, \u003cem\u003e24\u003c/em\u003e (5), 4916\u0026ndash;4925. https://doi.org/10.1007/s11356-016-8243-3.\u003c/li\u003e\n\u003cli\u003e McMahon, P. B.; Tokranov, A. K.; Bexfield, L. M.; Lindsey, B. D.; Johnson, T. D.; Lombard, M. A.; Watson, E. Perfluoroalkyl and Polyfluoroalkyl Substances in Groundwater Used as a Source of Drinking Water in the Eastern United States. \u003cem\u003eEnviron. Sci. Technol.\u003c/em\u003e\u003cstrong\u003e2022\u003c/strong\u003e, \u003cem\u003e56\u003c/em\u003e (4), 2279\u0026ndash;2288. https://doi.org/10.1021/acs.est.1c04795.\u003c/li\u003e\n\u003cli\u003e Hu, X. C.; Andrews, D. Q.; Lindstrom, A. B.; Bruton, T. A.; Schaider, L. A.; Grandjean, P.; Lohmann, R.; Carignan, C. C.; Blum, A.; Balan, S. A.; Higgins, C. P.; Sunderland, E. M. Detection of Poly- and Perfluoroalkyl Substances (PFASs) in U.S. Drinking Water Linked to Industrial Sites, Military Fire Training Areas, and Wastewater Treatment Plants. \u003cem\u003eEnv. Sci Technol Lett\u003c/em\u003e\u003cstrong\u003e2016\u003c/strong\u003e, \u003cem\u003e3\u003c/em\u003e (10), 344\u0026ndash;350. https://doi.org/10.1021/acs.estlett.6b00260.\u003c/li\u003e\n\u003cli\u003e Andrews, D. Q.; Hayes, J.; Stoiber, T.; Brewer, B.; Campbell, C.; Naidenko, O. V. Identification of Point Source Dischargers of Per-and Polyfluoroalkyl Substances in the United States. \u003cem\u003eAWWA Water Sci.\u003c/em\u003e\u003cstrong\u003e2021\u003c/strong\u003e, \u003cem\u003e3\u003c/em\u003e (5), e1252.\u003c/li\u003e\n\u003cli\u003e Salvatore, D.; Mok, K.; Garrett, K. K.; Poudrier, G.; Brown, P.; Birnbaum, L. S.; Goldenman, G.; Miller, M. F.; Patton, S.; Poehlein, M.; others. Presumptive Contamination: A New Approach to PFAS Contamination Based on Likely Sources. \u003cem\u003eEnviron. Sci. Technol. Lett.\u003c/em\u003e\u003cstrong\u003e2022\u003c/strong\u003e, \u003cem\u003e9\u003c/em\u003e (11), 983\u0026ndash;990.\u003c/li\u003e\n\u003cli\u003e Liddie, J. M.; Schaider, L. A.; Sunderland, E. M. Sociodemographic Factors Are Associated with the Abundance of PFAS Sources and Detection in U.S. Community Water Systems. \u003cem\u003eEnv. Sci Technol\u003c/em\u003e\u003cstrong\u003e2023\u003c/strong\u003e, \u003cem\u003e57\u003c/em\u003e (21), 7902\u0026ndash;7912. https://doi.org/10.1021/acs.est.2c07255.\u003c/li\u003e\n\u003cli\u003e Maruzzo, A. J.; Hernandez, A. B.; Swartz, C. H.; Liddie, J. M.; Schaider, L. A. Socioeconomic Disparities in Exposures to PFAS and Other Unregulated Industrial Drinking Water Contaminants in US Public Water Systems. \u003cem\u003eEnviron. Health Perspect.\u003c/em\u003e\u003cstrong\u003e2025\u003c/strong\u003e, \u003cem\u003e133\u003c/em\u003e (1), 017002. https://doi.org/10.1289/EHP14721.\u003c/li\u003e\n\u003cli\u003e Tokranov, A. K.; Ransom, K. M.; Bexfield, L. M.; Lindsey, B. D.; Watson, E.; Dupuy, D. I.; Stackelberg, P. E.; Fram, M. S.; Voss, S. A.; Kingsbury, J. A.; Jurgens, B. C.; Smalling, K. L.; Bradley, P. M. Predictions of Groundwater PFAS Occurrence at Drinking Water Supply Depths in the United States. \u003cem\u003eScience\u003c/em\u003e\u003cstrong\u003e2024\u003c/strong\u003e, \u003cem\u003e386\u003c/em\u003e (6723), 748\u0026ndash;755. https://doi.org/10.1126/science.ado6638.\u003c/li\u003e\n\u003cli\u003e Rafiei, V.; Nejadhashemi, A. P. Watershed Scale PFAS Fate and Transport Model for Source Identification and Management Implications. \u003cem\u003eWater Res.\u003c/em\u003e\u003cstrong\u003e2023\u003c/strong\u003e, \u003cem\u003e240\u003c/em\u003e, 120073. https://doi.org/10.1016/j.watres.2023.120073.\u003c/li\u003e\n\u003cli\u003e \u003cem\u003eAdvances and Research Gaps for PFAS Modeling in Watersheds and Receiving Waters\u003c/em\u003e. https://doi.org/10.1061/JOEEDU.EEENG-8137.\u003c/li\u003e\n\u003cli\u003e Nasr, M.; Orwin, J. F. A Geospatial Approach to Identifying and Mapping Areas of Relative Environmental Pressure on Ecosystem Integrity. \u003cem\u003eJ. Environ. Manage.\u003c/em\u003e\u003cstrong\u003e2024\u003c/strong\u003e, \u003cem\u003e370\u003c/em\u003e, 122445. https://doi.org/10.1016/j.jenvman.2024.122445.\u003c/li\u003e\n\u003cli\u003e Farhadi, S.; Tatullo, S.; Ferrian, F. Comparative Analysis of Ensemble Learning Techniques for Enhanced Fatigue Life Prediction. \u003cem\u003eSci. Rep.\u003c/em\u003e\u003cstrong\u003e2025\u003c/strong\u003e, \u003cem\u003e15\u003c/em\u003e (1), 11136. https://doi.org/10.1038/s41598-024-79476-y.\u003c/li\u003e\n\u003cli\u003e Lundberg, S. M.; Lee, S.-I. A Unified Approach to Interpreting Model Predictions. In \u003cem\u003eAdvances in Neural Information Processing Systems\u003c/em\u003e; Curran Associates, Inc., 2017; Vol. 30.\u003c/li\u003e\n\u003cli\u003e Jiawei, T.; Yizhen, Z.; Jiajun, S.; Xuelu, S.; Chao, S.; Chunhui, Z. Occurrence and Characteristics of Perfluoroalkyl Substances (PFASs) in Electroplating Industrial Wastewater. \u003cem\u003eWater Sci. Technol.\u003c/em\u003e\u003cstrong\u003e2019\u003c/strong\u003e, \u003cem\u003e79\u003c/em\u003e (4), 731\u0026ndash;740. https://doi.org/10.2166/wst.2019.092.\u003c/li\u003e\n\u003cli\u003e Gilchrist, M. PFAS in the Metal Plating and Finishing Industry. \u003cem\u003eMinnesota Pollution Control Agency\u003c/em\u003e, 2022. https://www.pca.state.mn.us/sites/default/files/gp3-05.pdf.\u003c/li\u003e\n\u003cli\u003e Michigan Department of Environment, Great Lakes, and Energy. \u003cem\u003eTargeted and Nontargeted Analysis of PFAS in Fume Suppressant Products at Chrome Plating Facilities.\u003c/em\u003e; 2025. https://www.michigan.gov/-/media/Project/Websites/egle/Documents/Programs/WRD/IPP/pfas-ipp-intiative-identified-sources.pdf?rev=0f234a957d4947968ba3b44711a93e10 (accessed 2025-06-16).\u003c/li\u003e\n\u003cli\u003e Qu, Y.; Huang, J.; Willand, W.; Weber, R. Occurrence, Removal and Emission of per- and Polyfluorinated Alkyl Substances (PFASs) from Chrome Plating Industry: A Case Study in Southeast China. \u003cem\u003eEmerg. Contam.\u003c/em\u003e\u003cstrong\u003e2020\u003c/strong\u003e, \u003cem\u003e6\u003c/em\u003e, 376\u0026ndash;384. https://doi.org/10.1016/j.emcon.2020.10.001.\u003c/li\u003e\n\u003cli\u003e U.S. ENVIRONMENTAL PROTECTION AGENCY-REGION 5. \u003cem\u003eChromium-Plating-PFOS-Study-EPA-Region-5-Sept-2009.Pdf\u003c/em\u003e; 2009. https://cswab.org/wp-content/uploads/2018/09/Chromium-Plating-PFOS-Study-EPA-Region-5-Sept-2009.pdf (accessed 2025-06-16).\u003c/li\u003e\n\u003cli\u003e \u003cem\u003eNorth American Industry Classification System (NAICS) U.S. Census Bureau\u003c/em\u003e. https://www.census.gov/naics/?input=325211\u0026amp;year=2022\u0026amp;details=325211 (accessed 2025-06-16).\u003c/li\u003e\n\u003cli\u003e Roostaei, J.; Colley, S.; Mulhern, R.; May, A. A.; Gibson, J. M. Predicting the Risk of GenX Contamination in Private Well Water Using a Machine-Learned Bayesian Network Model. \u003cem\u003eJ. Hazard. Mater.\u003c/em\u003e\u003cstrong\u003e2021\u003c/strong\u003e, \u003cem\u003e411\u003c/em\u003e, 125075. https://doi.org/10.1016/j.jhazmat.2021.125075.\u003c/li\u003e\n\u003cli\u003e Bao, J.; Yu, W.-J.; Liu, Y.; Wang, X.; Jin, Y.-H.; Dong, G.-H. Perfluoroalkyl Substances in Groundwater and Home-Produced Vegetables and Eggs around a Fluorochemical Industrial Park in China. \u003cem\u003eEcotoxicol. Environ. Saf.\u003c/em\u003e\u003cstrong\u003e2019\u003c/strong\u003e, \u003cem\u003e171\u003c/em\u003e, 199\u0026ndash;205. https://doi.org/10.1016/j.ecoenv.2018.12.086.\u003c/li\u003e\n\u003cli\u003e Gebbink, W. A.; van Asseldonk, L.; van Leeuwen, S. P. J. Presence of Emerging Per- and Polyfluoroalkyl Substances (PFASs) in River and Drinking Water near a Fluorochemical Production Plant in the Netherlands. \u003cem\u003eEnviron. Sci. Technol.\u003c/em\u003e\u003cstrong\u003e2017\u003c/strong\u003e, \u003cem\u003e51\u003c/em\u003e (19), 11057\u0026ndash;11065. https://doi.org/10.1021/acs.est.7b02488.\u003c/li\u003e\n\u003cli\u003e Gl\u0026uuml;ge, J.; Scheringer, M.; Cousins, I. T.; DeWitt, J. C.; Goldenman, G.; Herzke, D.; Lohmann, R.; Ng, C. A.; Trier, X.; Wang, Z. An Overview of the Uses of Per-and Polyfluoroalkyl Substances (PFAS). \u003cem\u003eEnviron. Sci. Process. Impacts\u003c/em\u003e\u003cstrong\u003e2020\u003c/strong\u003e, \u003cem\u003e22\u003c/em\u003e (12), 2345\u0026ndash;2373.\u003c/li\u003e\n\u003cli\u003e Agency for Toxic Substances and Disease Registry. \u003cem\u003eHealth Consultation: Perfluorochemical Releases at the 3M--Cottage Grove Facility\u003c/em\u003e; U.S. Department of Health and Human Services, Public Health Service, Division of Health Assessment and Consultation, 2005. https://www.health.state.mn.us/communities/environment/hazardous/docs/sites/washington/3mcg0205.pdf?utm_source=chatgpt.com (accessed 2025-10-21).\u003c/li\u003e\n\u003cli\u003e Sun, M.; Arevalo, E.; Strynar, M.; Lindstrom, A.; Richardson, M.; Kearns, B.; Pickett, A.; Smith, C.; Knappe, D. R. U. Legacy and Emerging Perfluoroalkyl Substances Are Important Drinking Water Contaminants in the Cape Fear River Watershed of North Carolina. \u003cem\u003eEnviron. Sci. Technol. Lett.\u003c/em\u003e\u003cstrong\u003e2016\u003c/strong\u003e, \u003cem\u003e3\u003c/em\u003e (12), 415\u0026ndash;419. https://doi.org/10.1021/acs.estlett.6b00398.\u003c/li\u003e\n\u003cli\u003e Andres, K. L.; Olsen, G. W.; Krisko, R. M.; Nunnally, M. C.; Boeding, R. R.; Leniek, K. L.; Taiwo, O. A. An Investigation of 3M Cordova, IL Production Worker\u0026rsquo;s per- and Polyfluoroalkyl Substances Biomonitoring Results and Mortality Experience. \u003cem\u003eInt. J. Hyg. Environ. Health\u003c/em\u003e\u003cstrong\u003e2024\u003c/strong\u003e, \u003cem\u003e256\u003c/em\u003e, 114321. https://doi.org/10.1016/j.ijheh.2024.114321.\u003c/li\u003e\n\u003cli\u003e \u003cem\u003eNorth American Industry Classification System (NAICS) U.S. Census Bureau\u003c/em\u003e. https://www.census.gov/naics/?input=325998\u0026amp;year=2022\u0026amp;details=325998 (accessed 2025-07-15).\u003c/li\u003e\n\u003cli\u003e Hepburn, E.; Madden, C.; Szabo, D.; Coggan, T. L.; Clarke, B.; Currell, M. Contamination of Groundwater with Per- and Polyfluoroalkyl Substances (PFAS) from Legacy Landfills in an Urban Re-Development Precinct. \u003cem\u003eEnviron. Pollut.\u003c/em\u003e\u003cstrong\u003e2019\u003c/strong\u003e, \u003cem\u003e248\u003c/em\u003e, 101\u0026ndash;113. https://doi.org/10.1016/j.envpol.2019.02.018.\u003c/li\u003e\n\u003cli\u003e Harrad, S.; Drage, D. S.; Sharkey, M.; Berresheim, H. Perfluoroalkyl Substances and Brominated Flame Retardants in Landfill-Related Air, Soil, and Groundwater from Ireland. \u003cem\u003eSci. Total Environ.\u003c/em\u003e\u003cstrong\u003e2020\u003c/strong\u003e, \u003cem\u003e705\u003c/em\u003e, 135834.\u003c/li\u003e\n\u003cli\u003e Lang, J. R.; Allred, B. M.; Peaslee, G. F.; Field, J. A.; Barlaz, M. A. Release of Per- and Polyfluoroalkyl Substances (PFASs) from Carpet and Clothing in Model Anaerobic Landfill Reactors. \u003cem\u003eEnviron. Sci. Technol.\u003c/em\u003e\u003cstrong\u003e2016\u003c/strong\u003e, \u003cem\u003e50\u003c/em\u003e (10), 5024\u0026ndash;5032. https://doi.org/10.1021/acs.est.5b06237.\u003c/li\u003e\n\u003cli\u003e Tolaymat, T.; Robey, N.; Krause, M.; Larson, J.; Weitz, K.; Parvathikar, S.; Phelps, L.; Linak, W.; Burden, S.; Speth, T.; Krug, J. A Critical Review of Perfluoroalkyl and Polyfluoroalkyl Substances (PFAS) Landfill Disposal in the United States. \u003cem\u003eSci. Total Environ.\u003c/em\u003e\u003cstrong\u003e2023\u003c/strong\u003e, \u003cem\u003e905\u003c/em\u003e, 167185. https://doi.org/10.1016/j.scitotenv.2023.167185.\u003c/li\u003e\n\u003cli\u003e Minet, L.; Wang, Z.; Shalin, A.; A.\u0026nbsp;Bruton, T.; Blum, A.; F.\u0026nbsp;Peaslee, G.; Schwartz-Narbonne, H.; Venier, M.; Whitehead, H.; Wu, Y.; L.\u0026nbsp;Diamond, M. Use and Release of Per- and Polyfluoroalkyl Substances (PFASs) in Consumer Food Packaging in U.S. and Canada. \u003cstrong\u003e2022\u003c/strong\u003e. https://doi.org/10.1039/D2EM00166G.\u003c/li\u003e\n\u003cli\u003e Herzke, D.; Olsson, E.; Posner, S. Perfluoroalkyl and Polyfluoroalkyl Substances (PFASs) in Consumer Products in Norway \u0026ndash; A Pilot Study. \u003cem\u003eChemosphere\u003c/em\u003e\u003cstrong\u003e2012\u003c/strong\u003e, \u003cem\u003e88\u003c/em\u003e (8), 980\u0026ndash;987. https://doi.org/10.1016/j.chemosphere.2012.03.035.\u003c/li\u003e\n\u003cli\u003e Kotthoff, M.; M\u0026uuml;ller, J.; J\u0026uuml;rling, H.; Schlummer, M.; Fiedler, D. Perfluoroalkyl and Polyfluoroalkyl Substances in Consumer Products. \u003cem\u003eEnviron. Sci. Pollut. Res.\u003c/em\u003e\u003cstrong\u003e2015\u003c/strong\u003e, \u003cem\u003e22\u003c/em\u003e (19), 14546\u0026ndash;14559. https://doi.org/10.1007/s11356-015-4202-7.\u003c/li\u003e\n\u003cli\u003e Ye, F.; Zushi, Y.; Masunaga, S. Survey of Perfluoroalkyl Acids (PFAAs) and Their Precursors Present in Japanese Consumer Products. \u003cem\u003eChemosphere\u003c/em\u003e\u003cstrong\u003e2015\u003c/strong\u003e, \u003cem\u003e127\u003c/em\u003e, 262\u0026ndash;268. https://doi.org/10.1016/j.chemosphere.2015.02.026.\u003c/li\u003e\n\u003cli\u003e Neiman, B.; Vavra, J. S. The Rise of Niche Consumption.\u003c/li\u003e\n\u003cli\u003e Reed, C.; Yu, T. E.; Hughes, D. Evaluating the Factors Influencing the Location Strategies of Specialty Grocers versus Traditional Supermarkets in the United States. \u003cem\u003eAppl. Geogr.\u003c/em\u003e\u003cstrong\u003e2023\u003c/strong\u003e, \u003cem\u003e158\u003c/em\u003e, 103034. https://doi.org/10.1016/j.apgeog.2023.103034.\u003c/li\u003e\n\u003cli\u003e Chang, C.-J.; Ryan, P. B.; Smarr, M. M.; Kannan, K.; Panuwet, P.; Dunlop, A. L.; Corwin, E. J.; Barr, D. B. Serum Per- and Polyfluoroalkyl Substance (PFAS) Concentrations and Predictors of Exposure among Pregnant African American Women in the Atlanta Area, Georgia. \u003cem\u003eEnviron. Res.\u003c/em\u003e\u003cstrong\u003e2021\u003c/strong\u003e, \u003cem\u003e198\u003c/em\u003e, 110445. https://doi.org/10.1016/j.envres.2020.110445.\u003c/li\u003e\n\u003cli\u003e Glynn, A.; Kotova, N.; Dahlgren, E.; Lindh, C.; Jakobsson, K.; Gyllenhammar, I.; Lignell, S.; N\u0026auml;ls\u0026eacute;n, C. Determinants of Serum Concentrations of Perfluoroalkyl Acids (PFAAs) in School Children and the Contribution of Low-Level PFAA-Contaminated Drinking Water. \u003cem\u003eEnviron. Sci. Process. Impacts\u003c/em\u003e\u003cstrong\u003e2020\u003c/strong\u003e, \u003cem\u003e22\u003c/em\u003e (4), 930\u0026ndash;944. https://doi.org/10.1039/C9EM00497A.\u003c/li\u003e\n\u003cli\u003e Schildroth, S.; Rodgers, K. M.; Strynar, M.; McCord, J.; Poma, G.; Covaci, A.; Dodson, R. E. Per-and Polyfluoroalkyl Substances (PFAS) and Persistent Chemical Mixtures in Dust from US Colleges. \u003cem\u003eEnviron. Res.\u003c/em\u003e\u003cstrong\u003e2022\u003c/strong\u003e, \u003cem\u003e206\u003c/em\u003e, 112530. https://doi.org/10.1016/j.envres.2021.112530.\u003c/li\u003e\n\u003cli\u003e Salawu, O. A.; Han, Z.; Senehi, N. L.; Roback, S. L.; Adeleye, A. S. Determination of Household Wastewater PFAS Composition and Concentrations via Neighborhood Sampling. \u003cem\u003eJ. Hazard. Mater.\u003c/em\u003e\u003cstrong\u003e2025\u003c/strong\u003e, \u003cem\u003e498\u003c/em\u003e, 139834. https://doi.org/10.1016/j.jhazmat.2025.139834.\u003c/li\u003e\n\u003cli\u003e Schlummer, M.; Gruber, L.; Fiedler, D.; Kizlauskas, M.; Mueller, J. Detection of Fluorotelomer Alcohols in Indoor Environments and Their Relevance for Human Exposure. \u003cem\u003eEnviron. Int.\u003c/em\u003e\u003cstrong\u003e2013\u003c/strong\u003e, \u003cem\u003e57\u0026ndash;58\u003c/em\u003e, 42\u0026ndash;49. https://doi.org/10.1016/j.envint.2013.03.010.\u003c/li\u003e\n\u003cli\u003e Merrett, H. C.; Chen, W. T.; Horng, J. J. A Systems Analysis Approach to Identifying Critical Success Factors in Drinking Water Source Protection Programs. \u003cem\u003eSUSTAINABILITY\u003c/em\u003e\u003cstrong\u003e2019\u003c/strong\u003e, \u003cem\u003e11\u003c/em\u003e (9). https://doi.org/10.3390/su11092606.\u003c/li\u003e\n\u003cli\u003e Schlea, S. H.; Wilkes, W. Using the Toxic Substance Control Act for Source Water Protection. \u003cem\u003eJ. Am. WATER WORKS Assoc.\u003c/em\u003e\u003cstrong\u003e2021\u003c/strong\u003e, \u003cem\u003e113\u003c/em\u003e (6), 26\u0026ndash;36. https://doi.org/10.1002/awwa.1745.\u003c/li\u003e\n\u003cli\u003e US EPA, O. \u003cem\u003eBasic Information about Source Water Protection\u003c/em\u003e. https://www.epa.gov/sourcewaterprotection/basic-information-about-source-water-protection (accessed 2025-07-16).\u003c/li\u003e\n\u003cli\u003e Yu, J.; Fu, Y.; Zhu, J.; Chen, X.; Zhong, Y.; Pan, Y.; Ma, Y. Marine Transport Barrier for Traditional and Emerging Per- and Polyfluoroalkyl Substances in the Southeast Indian Ocean and Antarctic Marginal Seas. \u003cem\u003eEnviron. Sci. Technol.\u003c/em\u003e\u003cstrong\u003e2025\u003c/strong\u003e, \u003cem\u003e59\u003c/em\u003e (10), 5283\u0026ndash;5292. https://doi.org/10.1021/acs.est.4c13574.\u003c/li\u003e\n\u003cli\u003e Alam, M. S.; Abbasi, A.; Chen, G. Fate, Distribution, and Transport Dynamics of Per- and Polyfluoroalkyl Substances (PFASs) in the Environment. \u003cem\u003eJ. Environ. Manage.\u003c/em\u003e\u003cstrong\u003e2024\u003c/strong\u003e, \u003cem\u003e371\u003c/em\u003e, 123163. https://doi.org/10.1016/j.jenvman.2024.123163.\u003c/li\u003e\n\u003cli\u003e Lyu, X.; Xiao, F.; Shen, C.; Chen, J.; Park, C. M.; Sun, Y.; Flury, M.; Wang, D. Per- and Polyfluoroalkyl Substances (PFAS) in Subsurface Environments: Occurrence, Fate, Transport, and Research Prospect. \u003cem\u003eRev. Geophys.\u003c/em\u003e\u003cstrong\u003e2022\u003c/strong\u003e, \u003cem\u003e60\u003c/em\u003e (3), e2021RG000765. https://doi.org/10.1029/2021RG000765.\u003c/li\u003e\n\u003cli\u003e Peter, L. G.; Lee, L. S. Sources and Pathways of PFAS Occurrence in Water Sources: Relative Contribution of Land-Applied Biosolids in an Agricultural Dominated Watershed. \u003cem\u003eEnviron. Sci. Technol.\u003c/em\u003e\u003cstrong\u003e2025\u003c/strong\u003e, \u003cem\u003e59\u003c/em\u003e (2), 1344\u0026ndash;1353.\u003c/li\u003e\n\u003cli\u003e US EPA, O. \u003cem\u003eAddition of Certain PFAS to the TRI by the National Defense Authorization Act\u003c/em\u003e. https://www.epa.gov/toxics-release-inventory-tri-program/addition-certain-pfas-tri-national-defense-authorization-act (accessed 2025-07-11).\u003c/li\u003e\n\u003cli\u003e The Fifth Unregulated Contaminant Monitoring Rule (UCMR 5) Data Summary: July 2025.\u003c/li\u003e\n\u003cli\u003e Harris, C. R.; Millman, K. J.; Walt, S. J. van der; Gommers, R.; Virtanen, P.; Cournapeau, D.; Wieser, E.; Taylor, J.; Berg, S.; Smith, N. J.; Kern, R.; Picus, M.; Hoyer, S.; Kerkwijk, M. H. van; Brett, M.; Haldane, A.; R\u0026iacute;o, J. F. del; Wiebe, M.; Peterson, P.; G\u0026eacute;rard-Marchant, P.; Sheppard, K.; Reddy, T.; Weckesser, W.; Abbasi, H.; Gohlke, C.; Oliphant, T. E. Array Programming with NumPy. \u003cem\u003eNature\u003c/em\u003e\u003cstrong\u003e2020\u003c/strong\u003e, \u003cem\u003e585\u003c/em\u003e (7825), 357\u0026ndash;362. https://doi.org/10.1038/s41586-020-2649-2.\u003c/li\u003e\n\u003cli\u003e team, T. pandas development. Pandas-Dev/Pandas: Pandas, 2020. https://doi.org/10.5281/zenodo.3509134.\u003c/li\u003e\n\u003cli\u003e Jordahl, K.; Bossche, J. V. den; Fleischmann, M.; Wasserman, J.; McBride, J.; Gerard, J.; Tratner, J.; Perry, M.; Badaracco, A. G.; Farmer, C.; Hjelle, G. A.; Snow, A. D.; Cochran, M.; Gillies, S.; Culbertson, L.; Bartos, M.; Eubank, N.; maxalbert; Bilogur, A.; Rey, S.; Ren, C.; Arribas-Bel, D.; Wasser, L.; Wolf, L. J.; Journois, M.; Wilson, J.; Greenhall, A.; Holdgraf, C.; Filipe; Leblanc, F. Geopandas/Geopandas: V0.8.1, 2020. https://doi.org/10.5281/zenodo.3946761.\u003c/li\u003e\n\u003cli\u003e python-visualization. Folium, 2020. https://python-visualization.github.io/folium/.\u003c/li\u003e\n\u003cli\u003e Lex, A.; Gehlenborg, N.; Strobelt, H.; Vuillemot, R.; Pfister, H. UpSet: Visualization of Intersecting Sets. \u003cem\u003eIEEE Trans. Vis. Comput. Graph.\u003c/em\u003e\u003cstrong\u003e2014\u003c/strong\u003e, \u003cem\u003e20\u003c/em\u003e (12), 1983\u0026ndash;1992. https://doi.org/10.1109/TVCG.2014.2346248.\u003c/li\u003e\n\u003cli\u003e python-nominatim. Nominatim. https://nominatim.org.\u003c/li\u003e\n\u003cli\u003e US EPA, O. \u003cem\u003eFRS EZ Query\u003c/em\u003e. https://www.epa.gov/frs/frs-ez-query (accessed 2025-06-20).\u003c/li\u003e\n\u003cli\u003e \u003cem\u003eNanda (National Neighborhood Data Archive) |\u003c/em\u003e. https://nanda.isr.umich.edu/ (accessed 2025-06-20).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-7935132/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7935132/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"Recent United States Environmental Protection Agency (USEPA) studies detected Per- and polyfluoroalkyl substances (PFAS) in ~45% of U.S. tap water highlighting the widespread environmental and public health concern. Although industrial activity in general and aqueous film forming foam (AFFF) usage are known contributors to the contamination, the specific industry sectors driving it remain unclear. Here, we apply eXplainable-AI (XAI) methods to move beyond coarse industrial categorizations and uncover the sectors most strongly associated with PFAS contamination. Using the national PFAS monitoring (UCMR5) data, industrial geolocations and socioeconomicfeatures we achieved strong predictive ability (F1-score = 0.84). SHAP analysis identified metal treatment, fabrication, and polymer manufacturing as dominant contributors, corroborating prior studies, while revealing specialty chemical manufacturing as a major yet previously overlooked predictor of PFAS contamination—often surpassing AFFF in influence. A paradoxical socioeconomic pattern also emerged: PFAS contamination was more likely in affluent regions (higher income, education, and professional employment).","manuscriptTitle":"Explainable AI Reveals Statistical Associations Between Industrial Activity and PFAS Contamination of Public Water Systems","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-11-06 06:20:01","doi":"10.21203/rs.3.rs-7935132/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"cbab96be-da52-42d6-82ea-c6624b6c2e81","owner":[],"postedDate":"November 6th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":56797694,"name":"Earth and environmental sciences/Environmental sciences/Environmental impact"},{"id":56797695,"name":"Earth and environmental sciences/Environmental social sciences/Environmental impact"}],"tags":[],"updatedAt":"2026-01-27T17:07:19+00:00","versionOfRecord":[],"versionCreatedAt":"2025-11-06 06:20:01","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7935132","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7935132","identity":"rs-7935132","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}