Scholarlink

<p>Tree height is a key indicator of forest growth, yet identifying its drivers remains challenging owing to complex interactions and limited data. We analyzed longitudinal data (2010–2021) from ten permanent 20 × 20&#xa0;m plots at the Jiangle State-Owned Forest Farm. To address the limitations of small sample sizes, this study proposes a novel integrated framework combining Bayesian hyperparameter-optimized Random Forest (RF) and Bootstrap Bayesian Networks (BN). Based on 23 variables retained after multicollinearity screening, the RF model identified 21 potential predictors, such as diameter at breast height (Dg), stand basal area (BA), and stand density (N). A key methodological contribution was the implementation of the Bootstrap BN with permutation tests, which significantly enhanced model robustness and edge reliability. The results revealed significant interactions: Dg and MAP exerted direct effects on tree height, whereas SDI influenced tree height indirectly through mediation effects. Notably, the proposed Bootstrap BN model achieved an 84% improvement in the Coefficient of Determination (R<sup>2</sup>) compared to the traditional BN model. These findings demonstrate the effectiveness of the proposed framework in unraveling multidimensional growth mechanisms and provide a solid scientific foundation for implementing density control strategies adapted to local hydrothermal conditions.</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

Exploring factors influencing the height growth of Cunninghamia lanceolata based on bootstrap bayesian network and random forest

  • 期刊论文

  • 发表日期：

    2026年2月19日

查看全文
  • Baoying Wang,
  • Rui Li,
  • Yujun Sun,
  • Ruiting Liang,
  • Zhaoting Zhu

摘要

Tree height is a key indicator of forest growth, yet identifying its drivers remains challenging owing to complex interactions and limited data. We analyzed longitudinal data (2010–2021) from ten permanent 20 × 20 m plots at the Jiangle State-Owned Forest Farm. To address the limitations of small sample sizes, this study proposes a novel integrated framework combining Bayesian hyperparameter-optimized Random Forest (RF) and Bootstrap Bayesian Networks (BN). Based on 23 variables retained after multicollinearity screening, the RF model identified 21 potential predictors, such as diameter at breast height (Dg), stand basal area (BA), and stand density (N). A key methodological contribution was the implementation of the Bootstrap BN with permutation tests, which significantly enhanced model robustness and edge reliability. The results revealed significant interactions: Dg and MAP exerted direct effects on tree height, whereas SDI influenced tree height indirectly through mediation effects. Notably, the proposed Bootstrap BN model achieved an 84% improvement in the Coefficient of Determination (R2) compared to the traditional BN model. These findings demonstrate the effectiveness of the proposed framework in unraveling multidimensional growth mechanisms and provide a solid scientific foundation for implementing density control strategies adapted to local hydrothermal conditions.