浙江省毛竹竹秆材积模型

Stem Volume Models of Phyllostachys edulis in Zhejiang Province

【目的】在精准测定毛竹样竹竹秆材积的基础上,研建浙江省毛竹竹秆材积模型,准确估计毛竹竹秆材积,为毛竹林经营管理提供依据。【方法】在浙江临安、庆元、武义、常山、宁海、安吉、诸暨、余姚、黄岩和泰顺10个县(区、市)调查216株毛竹样竹,采用可测量不规则形状物体体积的排水法测定竹秆材积。首先,基于异速生长方程和寺崎渡方程,以胸径(D)和胸高节长(L)为自变量,选择1个一元模型(M1)和4个二元模型(M2~M5),利用全部样本建立5个毛竹竹秆材积模型;然后,采用似然函数法分析模型误差结构,确定应当采用对数回归或非线性回归模型进行模型拟合,通过ForStat2.2软件进行回归拟合,得到参数估计值;最后,根据参数估计值t检验值、4个模型评价检验指标(包括调整确定系数R2a、估计值标准差SEE、平均偏差ME和平均系统误差MSE)的比较分析,选择模型参数稳定、预估精度高、最适宜的浙江省毛竹竹秆材积模型。【结果】采用似然函数法对竹秆材积模型误差结构进行分析,求得5个基础模型赤池信息量准则(AICc),计算ΔAICc大于2,模型误差项为乘积型,应采用对数回归拟合材积模型。5个竹秆材积模型参数稳定(t检验值的绝对值均大于2),各模型调整确定系数(R2a)均在0.95以上,估计值标准差(SEE)和平均系统误差(MSE)均接近于0,模型拟合效果较好,模型M2拟合预估效果最佳。分径阶进行模型拟合效果与预估精度的评价检验,5个模型在不同径阶的预估精度均较好,中等径阶时的预估精度和拟合优度最佳,而在最小径阶(4.0~5.9 cm)与最大径阶(14.0~15.9 cm)相对差一些。校正后的对数模型预估精度并无显著提高。【结论】排水法是准确测量毛竹竹秆材积的有效手段;似然函数法是进行模型误差结构分析比较与模型拟合方式选择的较好方法;引入竹高和胸高节长变量后,模型拟合优度和预估精度指标优于一元模型;考虑实践中胸高节长较竹高更易准确测量,且模型M5相较M2具有更高的拟合优度和预估精度,预估毛竹竹秆材积的最优模型为基于胸径-胸高节长的模型M5,即V=0.1912D2.1149e-6.8411/L。

【Objective】Based on the accurate measuring of the stem volume of moso bamboo(Phyllostachys edulis),the research on bamboo stem volume models was conducted to accurately estimate the stem volume and to provide a theoretical basis for the management of bamboo forest.【Method】Taking moso bamboo as the research object,based on the measured data of 216 samples of bamboo from Lin’an,Qingyuan,Wuyi,Changshan,Ninghai,Anji,Zhuji,Yuyao,Huangyan and Taishun county of Zhejiang Province,drainage method that could measure the volume of irregular shaped object was used to measure the bamboo stem volume.Firstly,based on the allometric equation and the equation proposed by Terasaki Watari,taking the diameter at breast height(D)and the internode length of bamboo at breast height(L)as the independent variables,one unary model(M1)and four binary models(M2—M5)were selected to establish the three bamboo stem volume models with all the samples data.Secondly,model error structures were analyzed with the likelihood analysis with which whether the log-transformed linear models or nonlinear weighted models were chosen for model fitting.The parameter estimations were obtained by fitting regression of the models using ForStat2.2 software.Finally,the bamboo stem volume model that fit for Zhejiang Province well and had stable parameters and higher prediction accuracy was selected according to analysis towards the results of t test for the parameter estimates and 4 validation statistics,namely the adjusted coefficient of determination(R2a),the standard error of the estimate(SEE),the mean errors(ME)and the mean systematic errors(MSE).【Result】The Akaike’s information criterion(AICc)of the 5 basic models were calculated with the likelihood analysis of model error structure.The results showed that the value ofΔAICc was larger than 2,which indicated the multiplicative error and the use of log-transformed model for model fitting.The parameters of the 5 bamboo stem volume log-models established were stable for the absolute t-value was larger than 2.All of the R2a were above 0.95,and the SEE and MSE were around 0,respectively,which reflected the well-fitting,and the model M2 fit the best.The statistics of model fitting and prediction across diameter classes were investigated as well.The results illustrated that the fitting and prediction performance across diameter classes of the 5 models was quite preferable,yet better when assuming the medium diameter classes and a bit worse as assuming the class of 4.0-5.9 cm and that of 14.0-15.9 cm.Finally,the prediction log-models with correction did not perform a better fitting effects and prediction precision.【Conclusion】Drainage method is a better method for accurate measurement of bamboo stem volume.Likelihood analysis is suitable to analyze the error structure and provide with a reliable justification of choosing the fitting models.The evaluation accuracy of binary models is higher than that of unary model as the variable H and L are introduced.Considering the easier and more accurate measurement of internodes length of bamboo at breast height(L)in practice,and the stem volume model M5 performs better than model M2,the suggested suitable model for predicting the stem volume is M5 that is based on variable D-L,i.e.V=0.1912D2.1149e-6.8411/L.