Integrated impacts of molecule and soil properties on the transport of PFAS in soils indicated by column experiments

Column experiments have been widely applied to investigate the transport of per- and polyfluoroalkyl substances (PFAS) in the soils and derive soil-water adsorption coefficient (K_d). However, the impacts of molecule and soil properties on K_d, especially their integrated effects, remain unknown. Therefore, K_d values of 13 PFAS in 10 soils were measured using miscible-displacement column experiments, and the quantitative relationships of K_d with both molecule and soil properties were investigated using machine learning (ML) techniques including multiple linear regression (MLR), random forest regression (RFR) and support vector regression (SVR). The result shows that <8C PFAS often present symmetric breakthrough curves, while >8C PFAS usually exhibit asymmetric breakthrough profiles. K_d values of a PFAS for different soils lie within a range of 2 to 3 orders of magnitude, while K_d of different PFAS for a soil can vary to 5 orders of magnitude. Compound-specific MLR suggests that higher content of minerals in soil significantly increased K_d of <6C PFAS, whereas organic matter mainly enhanced K_d of >6C PFAS. The ML analyses on the integrated effect of molecule and soil properties indicate that molecular volume and dipole moment of the molecules play vital roles together with minerals and organic matter of the soils in affecting K_d. ML techniques well describe the nonlinearity of PFAS transport in soils and provide powerful tools for accurate prediction.