【摘要】 构建杉木（Cunninghamia lanceolata）人工林单木冠幅模型，可以提高杉木冠幅的预测精度，并为杉木人工林的科学经营提供参考依据。以杉木为研究对象，基于江西省崇义县杉木人工林57个样地的7 198株单木数据，对基础模型进行扩展，并构建含林分因子和单木因子的冠幅非线性混合效应模型;采用十折交叉验证方法，以调整后的决定系数(R²_adj)、均方根误差(E_RMS)和平均绝对误差(E_MA)模型评价指标对各模型的预测效果进行评价。结果表明:（1）备选模型中，Logistic模型的预测效果最佳。（2）除胸径以外，有6个指标对冠幅有显著影响（P<0.05）。6个指标分别是地位指数、林龄、基尼系数（Gini系数）、林分密度指数、枝下高和大于对象木的断面积之和。添加协变量构建的广义模型较基础模型更优，R²_adj增加了28.53%,E_RMS和E_MA分别下降了17.80%和15.72%。（3）考虑样地水平随机效应的混合效应模型优于广义模型，R²_adj增加了27.86%,E_RMS和E_MA分别降低了35.73%和51.87%。协变量和随机效应的加入能有效提高冠幅模型的预测精度，且反映林木大小多样性的Gini系数在冠幅模拟中表现较好，值得进一步探讨该类指标在其他类型人工林模拟中的应用。更多还原

【Abstract】 Constructing a single tree crown width model for Chinese fir （Cunninghamia lanceolata） artificial forests can improve the prediction accuracy of Chinese fir crown width and provide a reference basis for the scientific management of artificial Chinese fir forests.The basic model was extended,and a nonlinear mixed effect model of crown width was constructed,including stand and individual tree factors and using data from 7 198 individual trees in 57 plots of Chinese fir plantations in Chongyi County,Jiangxi Province.Using a 10-fold cross-validation method,adjusted coefficient of determination (R²_adj),root mean square error (E_RMS),and mean absolute error (E_MA) were used as model evaluation indicators to evaluate the predictive performance of each model.The results indicate that:（1） The logistic model had the best fitting and prediction performance among the candidate models.（2） In addition to the diameter at breast height,there were 6 indicators that had significant effects on crown width （P<0.05）.And,the six indexes were site index,stand age,Gini coefficient,stand density index,height to crown base and basal area of trees larger than the subject tree.The generalized model constructed with covariates added was more effective than the basic model;R²_adjincreased by28.53%,whereas E_RMSand E_MAdecreased by 17.80%and 15.72%,respectively.（3） The mixed effects model that considered the random effects at the sampling site level was better than the generalized model;R²_adjincreased by 27.86%,whereas E_RMSand E_MA decreased by 35.73%and 51.87%,respectively.The addition of covariates and random effects could effectively improve the prediction accuracy of crown width models,and the Gini coefficient,which reflects the diversity of tree size,performed well in crown width simulations.It is worth exploring the application of this type of indicator in other types of artificial forest simulations.更多还原