Prediction of the Mitigation Effect of Biochar on Greenhouse Gas Emissions from Farmland: An Elastic Net model Based on XGBoost

The mitigation effects of biochar on methane (CH₄) and nitrous oxide (N₂O) emissions are subject to considerable variation depending on soil properties, biochar characteristics, and crop type, making accurate assessment of emission-reduction efficacy in cropland systems inherently difficult. In this study, a database of CH₄ and N₂O emissions under biochar amendment was compiled through systematic literature screening. Key features and their interaction effects were identified using extreme gradient boosting (XGBoost) and shapley additive explanations (SHAP). Soil pH was adopted as the primary stratification criterion, and the Classification and Regression Tree (CART) algorithm was applied to determine the optimal split threshold of soil total nitrogen (TN) within each pH stratum. On this basis, pH–TN two-level stratified elastic net explicit predictive models were constructed. The results indicate that soil pH, biochar carbon content, biochar application rate, and soil TN are the key influential features, with significant interaction effects between soil pH and TN observed in the pH<6.5 and pH>7.5 ranges. The TN regulatory thresholds for CH₄ and N₂O emissions are 0.95 and 0.80 g kg⁻¹ in alkaline soils, and 1.99 and 1.86 g kg⁻¹ in acidic soils, respectively. Among the stratified elastic net models constructed, the acidic high-TN stratum model achieves the highest predictive accuracy for CH₄ emissions (R²=0.71), while the acidic low-TN stratum model performs best for N₂O emissions (R²=0.45). The pH–TN stratified predictive framework developed in this study provides a reference for assessing biochar-induced greenhouse gas mitigation effects and informing application strategies under varying soil conditions.