Weed control efficiency and the environmental contamination potential of herbicides depend on soil sorption and desorption. Among the indexes that evaluate the soil adsorption processes, the coefficients sorption (Kfs) and desorption (Kfd) obtained by Freundlich isotherms can provide accurate information about the behavior of an herbicide in the soil. The values of Kfs and Kfd of an herbicide vary according to the physicochemical characteristics of the soil, so it is possible to estimate these coefficients with high precision if good predictive mathematical models are constructed. Therefore, our objective aimed to evaluate the use of multiple regression models (MLR) associated with multivariate techniques to estimate the coefficient Kfs and Kfd for the hexazinone based on the chemical and physical attributes of soils. The correlation analyses, principal components, and clustering analysis allowed the multiple linear regression technique to generate models with higher adjustment coefficient (R2) for Kfs (0.73 to 0.99) and Kfd (0.94 to 0.99), and lower root mean squared error (RMSE) for Kfs (0.003 to 0.065) and Kfd (0.018 to 0.120). Regression models created from groups of soils showed greater prediction performance for Kfs and Kfd. The organic matter followed by the cation exchange capacity was the most important attributes of soils in sorption and desorption processes of hexazinone.
Keywords: Cluster analysis; Discriminant analysis; Environmental impact; Factor analysis.