Predicting partitioning of radiolabelled ¹⁴C-PFOA in a range of soils using diffuse reflectance infrared spectroscopy

The aim of this study was to establish partitioning coefficients (K_d) of perfluorooctanoic acid (PFOA) in a wide range of soils and determine if those values can be predicted from soil properties using multiple linear regression (MLR) and from infrared spectra of soils using partial least squares regression (PLSR). For 100 different soils, the K_d values of spiked radiolabelled ¹⁴C-PFOA ranged from 0.6 to 14.8 L/kg and significantly decreased with soil depth (p < 0.05) due to soil properties that change with depth. The MLR modelling revealed that PFOA sorption was significantly (p < 0.05) influenced, in decreasing order, by organic carbon (OC) content, silt-plus-clay content and soil pH. Soils were partitioned into all soils and surface soils alone. The MLR models using OC, silt-plus-clay content and pH together explained most of the variation in sorption in all soils as well as surface soils alone (0-15 cm). However, correlations between soil properties and K_d values in some soils could not be explained by the MLR model. Modelling of K_d prediction in soils with PLSR and diffuse reflectance (mid) infrared Fourier transform spectroscopy (DRIFT) showed comparable success in explaining the predictions of K_d values, including some of the outliers identified in the MLR model. The PLSR loading weights suggested that quartz, and possibly pyrophyllite minerals, were inversely correlated with the K_d values. Given that MLR requires a-priori characterisation of a range of soil properties and PLSR-DRIFT is a method based on the direct relationship between spectra and soil components, mid-infrared spectroscopy may be a more economical and rapid technique to predict the solid-liquid partitioning of PFOA in soils.