| 摘要: |
| 【目的】利用线性混合效应模型分析杉木树高与胸径的关系，为杉木树高测量提供支持。【方法】收集688组有效杉木研究数据，利用最小二乘法构建树高（H）和胸径（DBH）的线性基础模型，同时考虑林分密度效应和海拔效应，在R 3.2.2软件中拟合混合模型,比较基础模型和2种混合模型的赤池信息规则（AIC）、贝叶斯信息规则（BIC）和-2倍对数似然值(-2log lik)，在此基础上，引入误差效应方差协方差矩阵及指数函数、幂函数和恒等式函数，筛选较好的混合模型；基于独立验证数据对模型进行验证，选取R2、|E|、RMSE 3个评价指标对模型精度进行评价。【结果】固定模型的AIC=2 089.731，BIC=2 102.151，-2log lik=2 083.732,均大于混合模型，即混合效应模型拟合效果优于固定模型；考虑模型误差效应方差协方差矩阵，加入恒等式异方差函数能够显著提高模型的精度，且含有不同随机参数的混合模型精度不同，引入海拔随机效应的混合模型拟合精度（R2=0.804 4，|E|=1.553 9，RMSE=2.143 0）高于含有林分密度效应的混合模型（R2=0.797 0，|E|=1.576 6，RMSE=2.183 0）。【结论】考虑随机效应的混合模型既能反映杉木树高的总体变化趋势，还能体现不同组分间的差异，在估测精度和通用性上均优于固定模型。 |
| 关键词: 杉木 树高-胸径方程 线性混合模型 随机效应 |
| DOI： |
| 分类号: |
| 基金项目:国家“973”计划项目(2012CB416905)；国家自然科学基金项目（31370626） |
|
| Height-diameter model for Chinese fir based on linear mixed model |
|
XU Chonghua,CUI Jun,HUANG Xingzhao,et al
|
| Abstract: |
| 【Objective】Relationship between height (H) and diameter at breast-height (DBH) of Chinese fir (Cunninghamia lanceolata) was studied using linear mixed model to provide support for height measurement.【Method】Data of 688 Chinese fir plants were collected and least square method was used to established the H-DBH linear model.Considering stand density and altitude effects,the Akaike information criterion (AIC),Bayesian information criterion (BIC) and the -2 time of logarithm likelihood value (-2log lik) are compared between the basic model and the 2 mixed models.Based on this,the exponential function (Exp),power function (Power),and identity heteroscedasticity function (Ident) were added to select the better mixing models.The accuracy of the model was evaluated using of R2,|E| and RMSE as indexes in R 3.2.2 software.【Result】The Akaike information criterion (2 089.731),Bayesian information criterion (2 102.151),and - 2 time of logarithm likelihood value (2 083.732) were higher than the mixed model,indicating that the mixed model was better.Considering the error variance covariance matrix and adding identity heteroscedasticity function (Ident) significantly improved the model accuracy.Compared with the density level mixed model (R2=0.797 0,|E|=1.576 6,RMSE=2.183 0),the altitude-level mixed model had better precision (R2=0.804 4,|E|=1.553 9, RMSE=2.143 0).【Conclusion】The mixed model with the consideration of random effects was better than fixed model since it can reveal both overall trends and differences between different groups of Chinese fir plantations. |
| Key words: Chinese fir H-DBH equation linear mixed model random-effects |
