Artificial neural networks for predicting the sorption coefficient of S-metolachlor: a hypothetical alternative to mitigate environmental risks and enhance herbicide efficiency in weed management

The sorption coefficients of herbicides are directly linked to soil properties, making predictive mathematical modeling crucial for understanding their behavior in soil environments. This study aims to leverage artificial neural networks (ANNs) to predict the sorption coefficient of S-metolachlor in soils, facilitating the assessment of environmental contamination risk and recommending pre-emergent herbicides. The objectives are: (1) to identify predictive variables for ANNs that can estimate S-metolachlor sorption coefficients; (2) to assess the performance of ANNs in predicting S-metolachlor sorption coefficients; and (3) to determine the key soil attributes for predicting these coefficients using ANNs. The ANNs successfully estimated the sorption capacity of S-metolachlor in the soils under study. By carefully selecting representative variables and employing appropriate structures, the ANNs achieved high precision and accuracy in estimating the S-metolachlor sorption coefficient (Kfs) in soil. The predictive screening process, utilizing the bootstrap forest partitioning method to select ANN inputs, emerged as a crucial step in developing well-trained models with strong generalization capabilities during testing. Furthermore, predictive screening revealed that certain attributes, not typically correlated with herbicide sorption in soils, contribute significantly to predicting Kfs. These findings suggest a theoretical alternative for optimizing the application of S-metolachlor in soil, ensuring efficiency while minimizing environmental risks. Future research in this area should focus on biological assays in controlled environments and field settings to validate the effectiveness of the proposed method for agronomic recommendations regarding S-metolachlor.