基于零膨胀模型和栅栏模型的赣南杉木林林分枯损模型  

Stand-Level Mortality Model of Cunninghamia lanceolata Forest in Southern Jiangxi Based on Zero-Inflated Model and Hurdle Model

作　　者：刘军 潘萍 欧阳勋志[1,2] 臧颢 郭杨[2] 游景晖 LIU Jun;PAN Ping;OUYANG Xunzhi;ZANG Hao;GUO Yang;YOU Jinghui(Key Laboratory of National Forestry and Grassland Administration for the Protection and Restoration of Forest Ecosystem in Poyang Lake Basin,Nanchang 330045,China;College of Forestry,Jiangxi Agricultural University,Nanchang 330045,China)

机构地区：[1]鄱阳湖流域森林生态系统保护与修复国家林业和草原局重点实验室,江西南昌330045 [2]江西农业大学林学院,江西南昌330045

出　　处：《江西农业大学学报》2022年第6期1428-1437,共10页Acta Agriculturae Universitatis Jiangxiensis

基　　金：国家自然科学基金项目(31760207)。

摘　　要：【目的】分析林分的枯损情况,构建林分水平枯损模型,探讨树木死亡的影响因素,为科学经营和有效管理提供参考依据。【方法】以江西省赣州市2009年森林资源二类调查中地类为纯林的1973块杉木样地为数据源,将样地按4∶1的比例随机划分成模拟数据(1579块)和验证数据(394块),分别用于模型的构建和验证。选取基于泊松分布和负二项分布形式的零膨胀模型和栅栏模型等4种模型,以林分枯损株数为因变量,以林分因子、立地因子和气候因子等17个环境因子为自变量,构建杉木林林分水平枯损模型。利用方差膨胀因子(VIF)排除各环境因子间相关性较大的因子;采用赤池信息准则(AIC)、贝叶斯信息准则(BIC)和-2倍对数似然函数值(-2logL)3种评价指标分析比较各模型之间的拟合效果,选用平均绝对误差(MAE)和均方根误差(RMSE)等比较其预测效果,筛选出最优林分水平枯损模型。【结果】(1)除最暖月平均温度、最冷月平均温度和无霜期天数外,其余14个环境因子间均不存在多重共线性(VIF<10)。(2)在模型的零部分,各参数估计值均在0.001水平上显著,海拔、林龄和株数密度是影响杉木林林分枯损的重要因子;在模型的离散部分,各参数估计值均在0.01水平上显著,且海拔和株数密度的参数估计值均为正值,林龄均为负值,说明林分枯损株数随海拔的升高和株数密度的增大而增加,随林龄的增大而减少。(3)通过比较模型评价指标,零膨胀负二项模型的AIC值(4841.73)、BIC值(4890.01)和-2logL值(4823.73)均小于其他模型,且其MAE值(39.4293)和RMSE值(116.5089)也均小于其他模型,表明零膨胀负二项模型的拟合效果和预测效果最好,其次为栅栏负二项模型、零膨胀泊松模型,栅栏泊松模型最差。【结论】基于模型比较结果,零膨胀负二项模型用于构建赣南杉木林林分水平枯损模型最好。[Objective]This study aims to analyze the situation of stand mortality,construct the stand-level mortality model,and discuss the influencing factors of tree mortality thus providing a reference basis for scientific management and effective management.[Method]1973 sample plots of Cunninghamia lanceolata with pure forest in the forest resources inventory of Ganzhou City,Jiangxi Province in 2009 were selected as the data source.These sample plots were randomly divided into simulation data(1579 plots)and validation data(394plots)according to the ratio of 4∶1 for model construction and verification,respectively.Four models such as the zero-inflated model and hurdle model based on Poisson distribution and negative binomial distribution forms were selected.The number of mortality trees in the stand was taken as the dependent variable,and a total of 17environmental factors including stand factors,site factors and climatic factors were taken as independent variables to construct the stand-level mortality model of Cunninghamia lanceolata forest.The variance inflation factor(VIF)was used to exclude the factors with large correlation between environmental factors;the three model evaluation indexes of Akaike information criterion(AIC),Bayesian information criterion(BIC)and-2log-likelihood function value(-2logL)were used to analyze and compare the fitting effect between each model,and the mean absolute error(MAE)and root mean square error(RMSE)were selected to compare the prediction effect,in order to select the optimal stand-level mortality model.[Result](1)Except the mean temperature of the warmest month,the mean temperature of the coldest month and the number of frost-free days,there was no multicollinearity among the other 14 environmental factors(VIF<10).(2)In the zero part of the model,the estimated values of each parameter were significant at the 0.001 level,indicating that altitude,stand age and density of trees were important factors affecting the stand-level mortality of Cunninghamia lanceolata forest;In the count

关 键 词：林分枯损 杉木林 零膨胀模型 栅栏模型 

分 类 号：S757[农业科学—森林经理学]