文献类型：期刊文献

中文题名：基于混合效应模型和零膨胀模型方法的蒙古栎林分水平枯损模型

英文题名：Modeling Stand-Level Mortality of Mongolian Oak(Quercus mongolica)Based on Mixed Effect Model and Zero-Inflated Model Methods

作者：李春明[1] 赵丽芳[2] 李利学[3]

第一作者：李春明

机构：[1]中国林业科学研究院资源信息研究所,北京100091;[2]中国科学院遥感与数字地球研究所,北京100101;[3]河北省承德县五道河林场,承德067407

年份：2019

卷号：55

期号：11

起止页码：27-36

中文期刊名：林业科学

外文期刊名：Scientia Silvae Sinicae

收录：CSTPCD;;EI(收录号：20200808196765);Scopus;北大核心:【北大核心2017】；CSCD:【CSCD2019_2020】；

基金：国家自然科学基金面上项目“基于混合效应模型的联立方程组及概率分布模型在模拟森林生长中的方法研究”（31570625）

语种：中文

中文关键词：广义线性混合效应模型;零膨胀模型;蒙古栎;枯损;林分

外文关键词：generalized linear mixed-effects models;zero-inflated model;Mongolian oak(Quercus mongolica);mortality;stand

分类号：S757

摘要：【目的】基于混合效应模型和零膨胀模型方法构建林分水平枯损模型,为选择科学的经营措施提供理论依据。【方法】以吉林省1994年设置的295块蒙古栎固定样地为数据源,236块样地作为模拟数据,59块样地作为验证数据。构建基于林分因子、立地因子和气象因子的蒙古栎林分水平枯损模型,其基本形式包括泊松分布和负二项分布。考虑样地中存在大量零值问题,在基础模型上加入零膨胀和零改变模型。为解决模型的嵌套和纵向数据问题,在构建模型时考虑样地的随机效应,选择验证数据进行精度验证。【结果】样地断面积、株数和最暖月平均气温是枯损概率和数量最重要的影响因子;考虑样地随机效应后,可明显提高模型模拟精度;负二项分布模型因考虑数据过度离散问题,模拟精度高于泊松分布。【结论】同时考虑随机效应和零膨胀的负二项分布模型,其模拟效果最好。
【Objective】 As an important component of forest growth yield systems, it is necessary to make accurate prediction for stand mortality.【Method】 About 295 permanent sample plots were established across the natural range of Mongolian oak in the Jilin Province in 1994. All plots were measured every 5 years, and the data were measured three times. 236 plots were used as simulation data and the other 59 plots as validation data. The main objective of this study was to construct stand-level mortality model of Quercus mongolica in relation to stand factor, site factor and climate factor. The basic forms of the model include Poisson distribution model and negative binomial distribution model. Considering the existence of a large number of zero values in the sample plots, the zero-inflated and zero-altered models were added to these basic models. In order to solve the problem of nesting and longitudinal data, the random effects of sample plot were taken into account in the construction of the model. In the end, the validation data was used to verify.【Result】 The results showed that the basal area of hectare, the number per hectare and the mean warmest month temperature are the most important factors influencing the probability and quantity of mortality. The simulation precision of the model was improved obviously after considering the plot random effects. Due to the over-dispersed of the data the accuracy of the negative binomial distribution model was higher than that of the Poisson distribution.【Conclusion】 The simulation effects of the model were the best when considering the random effects and the zero-inflated negative binomial distribution model simultaneously. The validation result also supported this conclusion.