Influence of Soil Mineralogy and Organic Carbon Content on PFAS Sorption Behavior: A Machine Learning-Based Investigation

Abstract

Per- and polyfluoroalkyl substances (PFAS) are persistent organic contaminants of global environmental concern, exhibiting high stability and potential for bioaccumulation in terrestrial and aquatic ecosystems. Understanding the sorption behavior of PFAS in soils is critical for predicting their environmental fate, transport, and risk. This study investigates the influence of soil mineralogical composition (quartz, kaolinite, illite, goethite, montmorillonite) and organic carbon content (foc) on the sorption of three representative PFAS compounds  perfluorooctanoic acid (PFOA), perfluorooctane sulfonate (PFOS), and perfluorohexane sulfonate (PFHxS)  across 120 soil samples spanning five soil textural classes. A synthetic dataset grounded in published literature ranges was constructed to evaluate physicochemical controls on PFAS sorption, parameterized as the log-transformed soil–water distribution coefficient (log Kd, L/kg). Four machine learning algorithms.  Multiple Linear Regression (MLR), Random Forest (RF), Support Vector Regression (SVR), and Gradient Boosting (GB)  were trained and evaluated using R², RMSE, MAE, and 5-fold cross-validation. Feature importance analysis consistently identified foc (RF importance: PFOA = 0.507, PFOS = 0.505, PFHxS = 0.642) and specific surface area (SSA) as the dominant sorption controls, while pH exerted a significant negative effect. MLR achieved the highest test R² for PFOS (0.834), and ensemble methods demonstrated competitive performance for non-linear compound–mineral interactions. Response surface analysis revealed synergistic amplification of sorption at high foc–SSA combinations. These findings provide a quantitative framework for predicting PFAS fate in soils and informing remediation strategies.