Modeling the feasibility of Se-rich corn cultivation in Se-deficient agricultural fields using random forest algorithm

In this study, we explore the factors influencing the Se bioaccumulation coefficient in corn based on a land quality geochemical survey at a 1:50,000 scale and establish predictive models for corn seed Se content using random forest and multiple linear regression approaches. The results indicate that the surface soil in the study area is deficient in Se (0.18 –1.21 mg/kg), but 54% of the corn grain samples met the standards for Se-rich products (0.02–0.30 mg/kg). The factors influencing the Se biological enrichment coefficient in corn seeds are soil pH and CaO and MgO content, with impact levels of 0.54, 0.42, and 0.35, respectively. Compared to mu ltiple linear regression models, the RF model provides more accurate and reliable predictions of corn Se content. The random forest model indicates that approximately 41% of the farmland within the study area is conducive to the cultivation of naturally Se-rich corn, which is a 26% increase in the p lanting area compared to recommendations based solely on soil Se content. In this research, we introduce an innovative methodological framework for organically cultivating naturally Se-rich corn within regions affected by Se deficiency.Graphical abstract
Source: Environmental Geochemistry and Health - Category: Environmental Health Source Type: research