湘西杉木人工林树高-胸径混合效应模型构建

Construction of mixed effect model of height-diameter at breast height of Cunninghamia lanceolata plantation in Xiangxi Tujia and Miao Autonomous Prefecture

建立湘西杉木树高-胸径模型,分析定量、定性因子对林分树高生长的影响,为区域尺度树高估算提供参考。以湘西土家族苗族自治州7个县的杉木人工林为研究对象,基于55块杉木人工林样地数据,采用随机森林法,筛选出显著性影响因子分别作为再参数化变量与随机效应;在6个树高-胸径基础模型中选择拟合程度最好的模型作为最优基础模型;采用再参数化的方法引入密度(ρ)/1000变量,构建含林分密度的最优再参数化模型;在最优再参数化模型的基础上,结合混合效应模型分析显著性定性因子对杉木人工林树高的影响,建立湘西杉木人工林最优混合效应模型。结果表明:湘西土家族苗族自治州杉木人工林树高的显著性影响因子为胸径(P<0.01)、龄组(P<0.01)、林分密度(ρ<0.05)。6组候选基础模型中,N slund(Model 1)模型最优,其AIC、BIC值均最小,分别为154.7417、159.6544,R 2=0.5727;其参数a、b均极显著,具有统计学意义。3种再参数化模型的拟合效果均优于最优基础模型(Model 1)的。Model 1.1各参数均显著,确定系数R 2=0.6202,均方根误差RMSE=1.6159。3个混合效应模型的拟合效果均优于最优基础模型、最优再参数化模型的;与最优基础模型、最优再参数化模型相比,nlme 1.1、nlme 1.3的确定系数(R 2)分别提高了19.6%、10.4%;均方根误差(RMSE)分别降低了14.1%、8.9%。考虑到模型的简易程度,将nlme 1.1作为湘西土家族苗族自治州杉木人工林的最优混合效应模型。与传统回归模型相比,采用再参数化方法、非线性混合效应法拟合的树高-胸径模型,其预测效果更具有优越性,模型精度更高、误差更小。研究结果可为湘西土家族苗族自治州林业生产提供参考。

This study aimed to establish a model for height-diameter at breast height relationship of Cunninghamia lanceolata plantaion in Xiangxi Tujia and Miao Autonomous Prefecture,analyzed the influences of both quantitative factor and qualitative factor on the height growth of forest stands,and provided a model and theoretical basis for regional-scale height estimation.The study focused on C.lanceolata plantations in seven counties of Xiangxi Tujia and Miao Autonomous Prefecture.Based on the data collected from 55 sample plots in C.lanceolata plantations,the random forest method was employed to select significant influencing factors as reparameterization variables and random effects.Among six basic height-diameter models,the best-fitting function was chosen as the optimal base model.The optimal reparameterization model,incorporating the variable SD/1000 to represent stand densitywas constructed.On the basis of the optimal reparameterization model,the significant qualitative factors influencing the height of C.lanceolata plantations were analyzed by a mixed effects model,leading to the establishment of the optimal mixed effects model for C.lanceolata plantations in Xiangxi Tujia and Miao Autonomous Prefecture.The results showed that the significant factors influencing the height of C.lanceolata plantations in Xiangxi Tujia and Miao Autonomous Prefecture included diameter at breast height(P<0.01),age group(P<0.01)and stand density(P<0.05).Among the six candidate models,the N slund model(Model 1)was found to be the optimal candidate with the smallest AIC and BIC values of 154.7417 and 159.6544,respectively and R 2=0.5727.Its parameters a and b were extremely significant and had statistical significance.The optimal reparametrization model(Model1.1)has a determined coefficient of R 2=0.6202 and a root mean square error of RMSE=1.6159.The three mixed effect models which contain qualitative factors were found to be the optimal models and have better goodness-of-fit performance than the previous models.The nlme 1.1 model and nlme 1.3 model had the best performance with an increase of 19.6%and 10.4%in the R 2 and a decrease of 14.1%and 8.9%in the RMSE compared to the optimal baseline model,respectively.Considering the simplicity of the model,the nlme 1.1 was chosen as the optimal mixed effects model for C.lanceolate plantations in Xiangxi Tujia and Miao Autonomous Prefecture.Compared with traditional regression models,the tree height-diameter model fitted using reparameterization and nonlinear mixed effects methods has superior predictive performance,it was higher model accuracy,and smaller errors.This study provided theoretical bases for forestry management and practice in Xiangxi Tujia and Miao Autonomous Prefecture.