Using NIR Spectroscopy to Predict Sorption of Contaminants at Field Scale.

Models used to evaluate leaching of contaminants to groundwater are very sensitive to sorption coefficients (K_d). These models need reliable K_d data at field scale, but the number of samples needed makes the classic batch sorption experiments unappropriated for this propose. Since it is inexpensive and fast, near infrared spectroscopy (NIR) has been widely used for predicting soil properties like clay or organic carbon. These properties are related with the soil sorption capacity. In this study, we aim to predict the spatial variation of K_d from NIR spectra for two contaminants: phenanthrene (mainly sorbed on the organic fraction) and glyphosate (sorbed on the mineral fractions).

Forty five bulk soil samples were collected from an agricultural field in Estrup,Denmark, in a 15 m x 15 m grid. Samples were air dried and 2-mm sieved. Samples were analyzed for texture, iron and aluminium oxides, pH, and organic carbon content. Sorption coefficients were obtained from a batch equilibration experiment. Soil samples were scanned with a bench top NIR instrument covering the full visible and near infrared range between 350-2500 nm. Partial Least Squares (PLS) regression with full cross-validation was used to correlate the soil spectra with K_d values and soil properties. he sorption coefficients   ranged from 345 to 886 L kg^-1 and from 162 to 536 L kg^-1 for phenanthrene and glyphosate, respectively. By means of the NIR spectra we were able to predict phenanthre (r2 = 0.938, RMSECV = 36.93) and glyphosate sorption (r2 = 0.616, RMSECV = 61.89). The normalized regression coefficients showed that phenanthrene sorption is correlated with total organic carbon and aluminium oxides, and glyphosate sorption with clay content and iron oxides. Predicted K_d values caught the spatial variability within the field suggesting that NIR - K_d predictions can be used as input for leaching risk assessment models.