Formatted Title
Strategies for the Identification of Distinct PFAS Source Areas at DoD Installations
Background/Objectives
The ubiquity and recalcitrance of poly- and perfluoroalkyl substances (PFAS) has resulted in source areas that are often difficult to distinguish or attribute to a single source. There has been increased attention on forensic approaches to identify distinct sources. Several techniques are used for source appointment with varying degrees of complexity and applicability depending on the data richness and availability. These include identification of indicator compounds/isomers, ratios of specific PFAS concentrations, or multivariate techniques. Principal component analysis (PCA) is a particularly powerful multivariate tool as it can indicate where dissimilarity within a dataset is statistically significant. The inherent compositional nature of concentration data can result in misinterpretation of PCA. The correct data transformation technique should be applied prior to performing PCA. The work presented will discuss the methods used for source distinction and summarize forensics results from PFAS source areas located at two Department of Defense (DoD) Installations located in the western United States. Beyond the fundamental research value of this approach, accurate analysis and interpretation of forensic data supported the DoD client in resource allocation for future supplemental investigation at non-AFFF source areas. Broadly, this work has been valuable to DoD clients by differentiating AFFF-specific PFAS signatures from those influenced by non-AFFF PFAS sources as they migrate off base.
Approach/Activities
Analytical data from PFAS-impacted sites thought to be inclusive of multiple PFAS sources will be compiled. These data can be parsed out or grouped in several ways according to their collection method, depth interval, time domain, or perceived PFAS impact. Each compositional data set will be transformed using Z-score normalized, centered log ratio, additive log ratio, and isomeric log ratio. PCA will be performed for the transformed datasets and the raw compositional dataset. Each of the PCA analyses will be presented exhibiting the impact of data transformation on the interpretation of the percentage of variance explained. The results will be presented as bi- or tri-plots depending on the percent variance explained by each component.
Results/Lessons Learned
Preliminary results from one of the installations to be included in this study are suggestive of a PFAS-impacted site that has at least two potential sources: one associated with aqueous film forming foam (AFFF) and another distinct source potentially associated with automotive repair and maintenance. This distinction was previously suspected to be the case, but the Z-score transformed PCA Analysis revealed that this source contains groups of samples with significantly different signatures. Another PFAS source zone under evaluation is suspected of being influenced by landfill leachate. We anticipate that the results of this analysis will result in groupings characteristic of AFFF-impacts only, AFFF-landfill leachate commingled, and landfill leachate impacts only. The methodology and results presented will provide guidance on parsing data into meaningful clusters and informing how transformation methods can influence interpretation of multivariate analysis. It is hypothesized that the data transformation resulting in the most accurate interpretation will not be conserved between the sites. This finding would suggest that transformation techniques be carefully considered when performing PCA and interpreting multivariate results on compositional data.