基于分位数回归的马尾松青冈栎混交林树高-胸径模型  

Height-diameter model of Pinus massoniana and Cyclobalanopsis glauca mixed forest based on Quantile Regression

作　　者：孙拥康[1] 汤景明[1] 王怡[1] SUN Yongkang;TANG Jingming;WANG Yi(Hubei Academy of Forestry,Wuhan 430075,Hubei,China)

机构地区：[1]湖北省林业科学研究院,湖北武汉430075

出　　处：《中南林业科技大学学报》2021年第12期18-25,共8页Journal of Central South University of Forestry & Technology

基　　金：中德财政合作林业重大科研项目(zdhz2017ky02);中央财政林业科技推广示范资金项目(鄂〔2020〕TG07);湖北省科技支撑计划项目(2015BBA212)。

摘　　要：【目的】采用分位数回归方法构建马尾松青冈栎混交林树高-胸径模型,并对比分析不同分位数模型与传统非线性回归模型的拟合与检验结果,以期提高模型预测精度,为混交林树高-胸径模型构建及科学经营提供新的方法和思路。【方法】以湖北省林科院九峰试验林场马尾松青冈栎混交林为研究对象,利用模型优选法确定出最优基础模型,并在此基础上,通过引入树种哑变量和分位数回归方法,构建不同树种不同分位点的树高-胸径分位数回归模型,选取平均绝对误差(MAE)、均方根误差(RMSE)、确定系数(R2)、T检验对不同模型进行对比分析。【结果】1)选取的6个代表性非线性树高曲线模型中,Richard模型综合表现最好,确定为最优基础模型。2)除个别分位点外,分位数回归模型整体拟合结果和预测能力优于哑变量模型和基础模型;马尾松和青冈栎最优分位数回归模型分别为τ=0.5和τ=0.7时;相同立地条件下,青冈栎生长势高于马尾松。3)模型独立检验结果表明分位数回归模型在描述树高曲线分布范围、变化规律以及稳健性上优于哑变量模型和基础模型。【结论】分位数回归方法在模拟混交林树高-胸径关系上表现出了较好的预测效果,将其应用到混交林或天然林树高-胸径关系等的研究中是一个可行思路。鉴于研究样本数据还较有限,兼顾数据整体和个体关联性的方法还有待进一步研究。【Objective】The height-diameter model of mixed forest of Pinus massoniana and Cyclobalanopsis glauca was established by quantile regression method,and the fitting and testing results of different quantile models and traditional nonlinear regression models were compared and analyzed,so as to improve the prediction accuracy of the model and provide new methods and ideas for the construction of height-diameter model and scientific management of mixed forest.【Method】Taking the mixed forest of Pinus massoniana and Cyclobalanopsis glauca in Jiufeng experimental forest farm of Hubei Academy of Forestry as the research object,the optimal basic model was determined by the model optimization method,and on this basis,the tree height-diameter quantile regression model of different tree species and quantiles were constructed by introducing tree species dummy variables and quantile regression methods,the average absolute error(MAE),root mean square error(RMSE),coefficient of determination(R2)and T test were used to compare and analyze the different models.【Result】1)Among the six representative nonlinear tree height curve models selected,Richard model has the best comprehensive performance and is determined as the optimal basic model;2)Except for individual quantiles,the overall fitting results and prediction ability of quantile regression model are better than those of dummy variable model and basic model;the optimal quantile regression models of Pinus massoniana and Cyclobalanopsis glauca areτ=0.5 andτ=0.7;under the same site conditions,the growth potential of Cyclobalanopsis glauca is higher than that of Pinus massoniana;3)The results of model independent test show that quantile regression model is better than dummy variable model and basic model in describing the distribution range,variation law and robustness of tree height curve.【Conclusion】The quantile regression method shows a good predictive effect in simulating the relationship between height and diameter in mixed forests.It is a feasible idea to appl

关 键 词：分位数回归 马尾松 青冈栎 混交林 树高-胸径模型 

分 类 号：S757[农业科学—森林经理学]