Scholarlink

<p>Soil organic C (SOC) and microbial biomass (MBC, MBN) are crucial for the health and climate mitigation potential of forest plantations. However, their dynamics are driven by complex, depth-dependent interactions that remain poorly quantified. This study integrates explainable machine learning (Random Forest and XGBoost) coupled with SHapley Additive exPlanations (SHAP) analysis to quantitatively disentangle the underlying mechanisms and non-linear silvicultural thresholds driving SOC, MBC, and MBN in 13-year-old subtropical plantations. Predictor variables encompassed tree species, planting pattern (monoculture vs. intercropping), stand structural attributes, and soil depth. Both models demonstrated high predictive accuracy (RMSE 1.450&#xa0;g kg<sup>− 1</sup>, MAE of 1.120&#xa0;g/kg and R² up to 0.86 for SOC concentration). SHAP analysis revealed a depth-dependent shifts in controlling factors with soil depth: tree species identity was the most important predictor in surface soils (0–10&#xa0;cm), whereas stand structural attributes, specifically diameter at breast height, were paramount in subsurface soils (30–50&#xa0;cm). MBC was strongly associated with labile C pools (POC, EOC), while MBN was primarily driven by soil N availability (TN, NH₄⁺-N). Significant synergistic interactions between specific tree species and intercropping patterns were identified. Our findings provide preliminary evidence for depth-specific forest management, indicating that strategies such as selecting species like <i>Cinnamomum camphora</i> for topsoil C and optimizing stand structures to support larger diameter at breast height for subsoil C could be explored in similar subtropical contexts.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

From biological to structural controls: explainable machine learning reveals depth-dependent shifts in predictors of soil organic carbon and microbial biomass across depth in subtropical plantations

  • 期刊论文

  • 发表日期：

    2026年5月2日

查看全文
  • Jingjing Zhuang,
  • Cuina Yang,
  • Yixuan Wang,
  • Linbo Ma,
  • Cui Cheng

摘要

Soil organic C (SOC) and microbial biomass (MBC, MBN) are crucial for the health and climate mitigation potential of forest plantations. However, their dynamics are driven by complex, depth-dependent interactions that remain poorly quantified. This study integrates explainable machine learning (Random Forest and XGBoost) coupled with SHapley Additive exPlanations (SHAP) analysis to quantitatively disentangle the underlying mechanisms and non-linear silvicultural thresholds driving SOC, MBC, and MBN in 13-year-old subtropical plantations. Predictor variables encompassed tree species, planting pattern (monoculture vs. intercropping), stand structural attributes, and soil depth. Both models demonstrated high predictive accuracy (RMSE 1.450 g kg− 1, MAE of 1.120 g/kg and R² up to 0.86 for SOC concentration). SHAP analysis revealed a depth-dependent shifts in controlling factors with soil depth: tree species identity was the most important predictor in surface soils (0–10 cm), whereas stand structural attributes, specifically diameter at breast height, were paramount in subsurface soils (30–50 cm). MBC was strongly associated with labile C pools (POC, EOC), while MBN was primarily driven by soil N availability (TN, NH₄⁺-N). Significant synergistic interactions between specific tree species and intercropping patterns were identified. Our findings provide preliminary evidence for depth-specific forest management, indicating that strategies such as selecting species like Cinnamomum camphora for topsoil C and optimizing stand structures to support larger diameter at breast height for subsoil C could be explored in similar subtropical contexts.