【目的】探讨竞争指标和建模方法对天然闽楠(Phoebe bournei)单木冠幅预测模型的影响,以期为精准预测天然闽楠单木的冠幅提供参考。【方法】以江西省赣中25块闽楠天然次生林典型样地中的1 011株闽楠为研究对象,采用普通最小二乘法(OLS模型)、仅考虑样地水平的混合效应模型、增强回归树和随机森林4种建模方法建立单木冠幅模型,分别添加林分每公顷断面积、大于对象木的断面积之和、林分密度指数和简单竞争指数4种竞争指标,分析竞争指标对冠幅模型的影响,采用决定系数(R²)、均方根误差(RMSE)、平均相对误差绝对值(RMA)和平均绝对误差(MAE)确定最佳模型。【结果】不添加竞争指标时,模型的预测能力表现为:混合效应模型>OLS模型>增强回归树>随机森林;添加竞争指标时,最优模型表现为:混合效应模型>OLS模型>随机森林>增强回归树。OLS模型中添加大于对象木的断面积之和竞争指标时预测能力最佳;增强回归树中添加固定半径为7 m的简单竞争指数时预测能力最佳;随机森林中添加林分每公顷断面积竞争指标时预测能力最佳;混合效应模型不添加竞争指标时预测能力最佳(RMSE为0.846 0,RMA为0.211 1,MAE为0.650 1),并且都优于其他模型。【结论】研究结果可对天然闽楠单木冠幅生长进行精准预测,并为提高闽楠天然次生林的林分质量提供理论依据。

【Objective】 The influence of competitive indices and modeling methods on the prediction of individual tree crowns of natural Phoebe bournei in Jiangxi Province was discussed to provide a reference for accurately predicting the crown width of natural Phoebe bournei.【Method】An individual crown width model was developed based on a dataset comprising 1 011 Phoebe bournei from 25 typical plots of natural secondary Phoebe bournei forest located in Jiangxi Province. The individual tree crown width model was established using four modeling methods: ordinary least squares (OLS) model, mixed effects model, boosted regression trees, and random forest. Four competitive indices, namely, basal area, basal area of larger trees, stand density, and simple competition, were added to reflect the influence of competition on the crown width model. The model evaluation criteria included determination coefficient (R²), root-mean-square error (RMSE), relative mean absolute error (RMA) and mean absolute error (MAE).【Result】The results demonstrated that when no competition index was added, the order of model accuracy was mixed effects model > OLS model > boosted regression trees > random forest; when the competition index was added, the optimal model was as follows: mixed effects model > OLS model> random forest > boosted regression trees (BRT). The OLS model had the best predictive ability when the basal area of larger trees was added; the BRT had the best predictive ability when a simple competition index with a fixed radius of 7 m was added; the random forest had the best predictive ability when the basal area was added. The mixed effects model had the best predictive ability when no competitive index was added (RMSE is 0.846 0, RMA is 0.211 1, MAE is 0.650 1), and it was superior to the other models. 【Conclusion】The results of this study accurately predicted the growth of Phoebe bournei individual tree crowns, thereby providing a theoretical basis for improving the forest quality of natural secondary Phoebe bournei forest.