Prediction of the glyphosate sorption coefficient across two loamy agricultural fields

Sorption is considered one of the most important processes controlling pesticide mobility in agricultural soils. Accurate predictions of sorption coefficients are needed for reliable risk assessments of groundwater contamination from pesticides. In this work, we aim to estimate the glyphosate sorption coefficient, K_d, from easily measurable soil properties in two loamy, agricultural fields in Denmark: Estrup and Silstrup. Forty-five soil samples in Estrup and 65 in Silstrup were collected from the surface in a rectangular grid of 15×15-m from each field, and selected soil properties and glyphosate sorption coefficients were determined. Multiple linear regression (MLR) analyses were performed using nine geo-referenced soil properties as variables to identify the parameters related with glyphosate sorption. Scenarios considered in the analyses included: (i) each field separately, (ii) both fields together, and (iii) northern and southern sections of the field in Silstrup. Considering correlations with all possible sets of the same nine geo-referenced properties, a best-four set of parameters was identified for each model scenario. The best-four set for the field in Estrup included clay, oxalate-extractable Fe, Olsen P and pH, while the best-four set for Silstrup included clay, OC, Olsen P and EC. When the field in Silstrup was separated in a northern and southern section, the northern section included EC, and oxalate-extractable Fe, Al and P, whereas the southern part included pH, clay, OC and Olsen P. The best-four set for both fields together included clay, sand, pH and EC. Thus, the most common parameters repeated in the best-four sets included clay and pH as also reported previously in the literature, but in general, the composition of the best-four set differed for each scenario, suggesting that different properties control glyphosate sorption in different locations and at different scales of analysis. Better predictions were obtained for the best-four set for the field in Estrup (R²=0.87) and for both fields (R²=0.70), while the field in Silstrup showed a lower predictability (R²=0.36). Possibly, the low predictability for the field in Silstrup originated from opposing gradients in clay and oxalate-extractable Fe across the field. Also, whereas a lower clay content in Estrup may be the limiting variable for glyphosate sorption, the field in Silstrup has a higher clay content not limiting the sorption, but introducing more variability in K_d due to changes in other soil properties.