Tobacco is a key economic crop, with leaf oil content serving as a critical determinant of leaf quality. To address the limited understanding of mechanisms underlying oil content improvement and the decline in flue-cured tobacco quality caused by long-term reliance on chemical fertilizers, this study integrated field experiments with a machine learning approach. Five treatments with varying organic nitrogen ratios (0%, 10%, 20%, 30%, and 40%) were evaluated at a single experimental site in Hengyang, Hunan. Results indicated that a 30% organic nitrogen ratio significantly enhanced the activity of key lipid metabolism enzymes, promoted the accumulation of lipid metabolites (including cembratriene-diol and sucrose esters), increased glandular trichome density, and improved leaf physical properties such as softness, tensile strength, and thickness, ultimately achieving the highest oil content. Using a robust data augmentation strategy and Recursive Feature Elimination, a Random Forest model was constructed to dissect the complex regulatory network. The model achieved a high predictive accuracy (CV R² = 0.819) on the augmented dataset, significantly outperforming the model based on original small-sample data. Feature importance analysis identified petroleum ether extract, cembratriene-diol, leaf softness, reducing sugar, and glandular trichome density as the primary predictors. Significant interactions among these features were also revealed by SHAP dependence plots. These findings provide a theoretical basis for optimizing organic nitrogen application to enhance tobacco leaf oil content and quality in agricultural production.
Sun et al. (Mon,) studied this question.