【摘要】 机器学习算法在单木胸径生长预测中的应用已较为广泛，但由于机器学习模型的“黑箱”特性，缺乏对模型的解释分析。基于此，以机器学习模型为基础，结合SHAP算法，分析各因子与单木胸径生长的关系及其影响。以金沟岭林场连续15 a云杉调查数据为研究对象，候选单木、林分、立地和气候因子共36个，采用主成分分析PCA方法对气候因子进行特征降维，利用K-近邻法（K-NN）、支持向量回归（SVR）和随机森林（RF）算法建立云杉单木胸径生长模型并进行比较，基于十折交叉验证网格搜索法寻找最优超参数并训练，通过决定系数（R²）、均方根误差(E_RMS)和平均绝对误差(E_MA)3种指标对3种模型算法进行评价，并用测试集数据检验各模型的泛化能力。同时，利用SHAP的可解释性分析，对模型特征重要性进行具体分析。在3种机器学习模型中，SVR模型优于RF和K-NN模型，具有更高的拟合精度和较小的预测误差，SVR模型的预测精度R²、E_RMS和E_MA分别为0.751 6、0.666 1和0.496 7 cm。林木期初胸径（D₁）的倒数（1/D₁）、大于对象木的林木断面积之和(L_BA)和林分中大于对象木的所有林木平方直径和（D_H）与胸径生长呈现出明显的负相关关系；林木期初胸径的对数值(L_ND1)、对象木胸径与林分断面积平均胸径之比（R_D）、林木期初胸径（D₁）与胸径生长呈正相关关系。气候因子与胸径生长呈负相关关系，立地因子与胸径生长呈正相关关系。机器学习算法对于胸径生长模型的构建均取得了一定的效果，而SVR模型的泛化能力更优，且SHAP的分析方法能够较好的解析各特征变量与胸径生长的关系。更多还原

【Abstract】 The application of machine learning algorithms in individual tree DBH（Diameter at Breast Height） growth prediction has been relatively extensive, but due to the “black-box” nature of machine learning models, there is a lack of interpretive analysis of the models. Based on this, this study uses machine learning models combined with the SHAP algorithm to analyze the relationships and influences of various factors on individual DBH growth. Using 15 consecutive years of spruce survey data from the Jingouling Forest Farm, 36 candidate factors were selected, including individual tree, stand, site, and climatic factors. Principal component analysis（PCA） was applied to reduce the dimensionality of climatic factors. Models for spruce individual DBH growth were established and compared using K-nearest neighbors（K-NN）, support vector regression（SVR）, and random forest（RF） algorithms. Optimal hyperparameters were identified through 10-fold cross-validation with grid search, and models were trained accordingly. The models were evaluated using three metrics: coefficient of determination（R²）, root mean square error(E_RMS), and mean absolute error(E_MA), and their generalization capabilities were tested using a test dataset. Additionally, SHAP explainability analysis was used to specifically analyze feature importance in the models. Among the three machine learning models, the SVR model outperformed RF and K-NN, demonstrating higher fitting accuracy and smaller prediction errors, with R², E_RMS, and E_MA values of 0.751 6, 0.666 1, and 0.496 7 cm, respectively. The reciprocal of initial DBH（1/D₁）, the sum of basal areas of trees larger than the target tree(L_BA), and the sum of squared diameters of trees larger than the target tree（D_H） showed significant negative correlations with DBH growth, while the natural logarithm of initial DBH(L_ND1), the ratio of target tree DBH to stand basal area mean DBH（R_D）, and initial DBH（D₁） showed positive correlations. Climatic factors exhibited negative correlations with DBH growth, while site factors showed positive correlations. Machine learning algorithms have achieved certain effects in constructing DBH growth models, with the SVR model demonstrating superior generalization ability. The SHAP analysis method effectively interprets the relationships between feature variables and DBH growth.更多还原