文献类型：期刊文献

中文题名：基于多水平非线性混合效应蒙古栎林单木断面积模型

英文题名：Multilevel Nonlinear Mixed-effects Basal Area Models for Individual Trees of Quercus mongolica

作者：符利勇[1] 唐守正[1] 张会儒[1] 雷相东[1]

第一作者：符利勇

机构：[1]中国林业科学研究院资源信息研究所

年份：2015

卷号：25

期号：1

起止页码：23-31

中文期刊名：林业科学研究

外文期刊名：Forest Research

收录：CSTPCD;;Scopus;北大核心:【北大核心2014】；CSCD:【CSCD2015_2016】；

基金：国家自然科学基金青年科学基金项目(31300534);中央级公益性科研院所基本科研业务费专项资金"随机效应的森林立地指数模型和应用研究"

语种：中文

中文关键词：蒙古栎林;混合模型;单木断面积生长模型

外文关键词：Quercus mongolica stand; mixed effects model; individual basal area increment model

分类号：S711

摘要：以吉林省汪清林业局184块样地中的10 111株蒙古栎为例,首先选用线性函数、Richards函数、Logistic函数、指数函数等7种常用函数形式,分析4个因变量(后期胸径、后期胸高断面积、直径增量和胸高断面积增量)与前期胸径的影响,确定一个用于构建混合效应模型的基础模型。然后确定同时考虑林场效应和林场与样地交互效应时基础模型中最优的形式参数构造形式,利用逐步回归方法确定模型中所包含的林分变量,并分析和比较用来消除异方差的3种常用残差方差函数(指数函数、幂函数和常数加幂函数),最后检验模型预测效果。结果表明:Wykoff模型且因变量为后期胸高断面积拟合效果较好,故作为基础模型;除前期胸高直径(D)外,当考虑坡度正切(ST),对象木胸高直径与样地算术平均直径的比(RAD),样地胸高总断面积(TBA),样地中大于对象木直径所有树木的胸高断面积和(GSBA),对象木胸高断面积与样地算术平均胸高断面积的比(RABA)和对象木胸高断面积与样地胸高总断面积的比(RBA)等林分变量时能进一步提高模型预测精度;对于残差方差,指数函数、幂函数和常数加幂函数都能消除异方差,但幂函数效果最好;当模型同时考虑林场效应和林场与样地交互效应时预测精度最高。
Taking 10 111 Q. mongolica individuals obtained from 184 plots in Wangqing Forestry Bureau in Jilin Province as test materials, this study is to develop individual basal area increment model for Quercus mongolica Fisch by mixed effects model approach. The relationship between four dependent variables （the later diameter at breast height, the later basal area at breast height, the diameter increment and the basal area increment） and earlier stage diameter at breast height were analyzed using seven functions commonly used, i.e. linear function, Richards function, logistic function, exponential function etc. The best model was selected as the base model to develop mixed effects model. And then, the best combination form of formal parameters in the base model was determined with considering both the random effects of forest farms and the plot simultaneously. Forest variables contained in the formal parameters of the model were determined by stepwise regression method. Three kinds of residual variance functions （ exponential function, power function and constant plus power function） that used to eliminate heteroske- dasticity were analyzed and compared and the prediction efficiency of model was tested. The results are as follows.The Wykoff model that the dependent variable was later basal area at breast height had a better fit effect and used as the base model. In addition to the early diameter at breast height （D） , the model included the stand variables, such as the tangent of slope （ST）, the ratio of D of target tree to arithmetical mean diameter of plot （RAD）, the to- tal basal area of plots （ TBA）, the sum of basal area of trees with diameter larger than target tree （ GSBA）, the ratio of basal area of target tree to arithmetical mean basal area of plot （RABA） and the ratio of basal area of target to to- tal basal area of plot （ RBA）, had a better prediction accuracy. For residual variance, the exponential function, power function and constant plus power function could eliminate the heteroskedasticity, but the power function was the best. The mixed effects model taking forest farm and plot effects into account has the highest prediction accuracy.