【作者】 王建军；

【导师】 刘琪璟； 孟京辉；

【作者基本信息】 北京林业大学 ， 森林经理学， 2021， 博士

【摘要】 内蒙古大兴安岭林区是我国四大重点国有林区中国有林地最集中、面积最大、生态地位最重要、纬度最高的国家森林生态功能区。兴安落叶松是林区主要的优势树种,如何科学的经营兴安落叶松林,充分发挥大兴安岭林区的生态功能是林区的主要任务。森林生长模型是描述林木或林分生长规律的一组方程式,它能够了解森林生态系统的结构与功能,预测林分未来的发展动态,为科学经营森林打下理论基础。本研究基于内蒙古大兴安岭重点国有林区第6次(2003年)、第7次(2008年)、第8次(2013年)和第9次(2018年)四期国家森林资源连续清查固定复测样地,以内蒙古地区的兴安落叶松林为研究对象,进行如下研究。一是基于142块兴安落叶松人工林样地,采用Reineke自然稀疏方程,研建兴安落叶松人工林的密度控制图,基于密度控制图模拟相同密度不同抚育措施条件下,林分达到目标直径的林龄、优势高、蓄积量以及总收获量等林分参数,进而确定最优经营方案。二是基于2013和2018年两期一类清查数据,共计667块兴安落叶松天然林样地,建立分树种的可变概率矩阵模型和固定概率矩阵模型,应用构建的模型,预测未来五年林分的进界数量和近百年林分参数。三是基于2003到2018年四期一类清查数据,共计626块天然林样地,采用线性混合效应方法构建主要树种的断面积生长模型。此外,还引入了自相关矩阵和异方差函数来描述样地内的自相关和异方差性。最后,根据赤池信息准则(AIC)、贝叶斯信息准则(BIC)、对数似然值(Log Lik)以及似然比检验(LRT)确定最终的模型。本研究主要得到以下结论:(1)基于142块兴安落叶松人工林,筛选出完满立木度样地,采用Reduced Major Axis回归,构建Reineke自疏方程。根据兴安落叶松的初始密度,构建了兴安落叶松人工林最优密度上限方程、最优密度下限方程和林分郁闭线。基于样地数据,拟合了兴安落叶松的优势高模型、蓄积模型以及碳储量模型作为密度控制图的辅助方程。采用收获预估模型模拟了一定密度下不同措施林分总收获量和林分平均收获量,发现合理经营抚育措施能提高林分的收获量。(2)基于一类清查数据的667块兴安落叶松天然林样地,将林分的树种分为桦木、软阔、以栎类为主的硬阔、兴安落叶松四类,构建了四个树种的生长、进界和枯死的转移概率模型,模型具有较好的解释性和预测性。通过构建的模型对林分的状况进行了短期预测和长期预测,短期预测的结果与理论值基本无差异,长期预测符合理论值。此外,采用固定参数平均的方法所建立的转移矩阵模型在进行实际林分结构预测时容易产生比实际值偏高的预测结果,而采用考虑林分状态的可变参数转移矩阵模型法则符合理论值。(3)基于一类清查数据的626块兴安落叶松天然林样地,将林分的树种分为白桦、栎类和兴安落叶松,分别构建了三类树种的胸高断面积生长模型。引入样地随机效应的混合模型AIC、BIC显著降低,Log Lik显著增大。考虑不同随机参数组合的拟合结果表明三个树种的混合效应模型均优于传统模型。考虑到础模型存在异方差性,在消除异方差时,指数函数、幂函数和常数加幂函数均能在一定程度上缩小残差的变化范围。模型检验中,考虑样地效应的混合效应模型大大提高了预估的精度,混合效应模型的模拟精度显著优于传统模型。与基于最小二乘法的基础模型相比,考虑层次结构的混合效应模型显著地改善了模型的表现,所构建的模型有一定的生物学意义和统计可靠性。更多还原

【Abstract】 The Greater Khingan Mountains Forest Region in Inner Mongolia is a national forest ecology function zone with the most concentrated forest,the largest area,the most important ecological status,and the highest latitude among the four key state-owned forest areas in China.Larix gmelini is the main dominant tree species in the forest area.How to manage the Larix gmelini forest scientifically and give full play to the ecology functions of the forest area in the Greater Khingan Mountains Forest Region is the main aim of the forest area.The forest growth and yield model is a set of equations that describe forest trees or stands growth dynamics.It can help understand the structure and function of forest ecosystem,predict the development of forest stands,and lay a theoretical foundation for scientific forest management.In this study,based on the data collected from sample plots from the 6th(2003),7th(2008),8th(2013)and 9th(2018)Chinese National Forest Inventory in Inner Mongolia Autonomous Region,north China,the Larix gmelini forest is researched.One is based on 142 sample plots of Larix gmelini plantation,using Reineke natural sparse equation to develop a density control chart.Based on the density control chart,the parameters of stands which reach the target diameter such as age,dominant height,volume and total yield under the conditions of the same density and different thinning measures are simulated and then the optimal management plan is determined.The second is to develop the variable transition model and the fixed parameter transition matrix model of the tree species using data collected from 667 natural forest sample plots of Larix gmelini from the 2013 and 2018 Chinese National Forest Inventory in Inner Mongolia Autonomous Region,north China.Then,applying the developed models to predict the number of advanced forest stands in the next five years and stand parameters in nearly one hundred years.Third,the linear mixed-effects individual-tree growth models for main tree species are developed using data collected from 626 natural forest sample plots from the 2013 and 2018 Chinese National Forest Inventory in Inner Mongolia Autonomous Region,north China.In addition,the variance functions and autocorrelation structures are applied to describe within-plot heteroscedasticity and autocorrelation.Finally,the optimal mixed-effects model is determined based on the Akaike information criterion(AIC),Bayesian information criterion(BIC),log-likelihood(Loglik)and the likelihood ratio test(LRT).The main conclusions of this study are as follows:Based on 142 sample plots of Larix gmelini plantations,sample plots with fully stocking percent were selected.Using reduced major axis regression,the renike self-thinning equation was constructed.According to the initial density of Larix gmelini,the upper limit equation of optimal density,the lower limit equation and the closed canopy line were constructed.Based on the data,the dominant height model,volume model and carbon storage model were fitted as the auxiliary equations of the density control chart.The total yield and average yield of stand under different densities and different measures were simulated by using the yield prediction model.It was found that reasonable management and thinning measures could improve the yield of stand.Based on the 667 natural forest sample plots of Larix gmelinii,the tree species are divided into four types: birch,soft broad-leaved species,hard broad-leaved species,and Larix gmelinii,and the growth model,recruitmen model and mortality model were developed.The developed models have good explanatory and predictive ability.The short-term and long-term prediction of the forest stand were made through the developed model.The results of the short-term prediction are basically the same as the theoretical value,and the long-term prediction was in line with to the theoretical value.In addition,the transition matrix model established by using the fixed-parameter average method is likely to produce higher prediction results than the actual value when predicting the actual forest stand structure,while the variable parameter transition matrix model method that takes into account the forest stand status is consistent with the theoretical value.Based on the 626 natural forest sample plots of Larix gmelinii,the tree species are divided into three types: birch,oaks,and Larix gmelinii,and the basal area increment models were developed.Using sample plot as a random effect,the AIC and BIC reduced,and the Log Lik increased.It is found that the results of different combinations of random effects while considering all independent variables and the intercept in the basic model are better than the basic model for the three tree species.The general positive-definite matrix was selected for the optimal random effects variance–covariance structures.Considering the heteroscedasticity of the basic model,exponential function,power function and constant plus power function can all correct the heteroscedasticity of the model to a certain extent.In model prediction and evaluation,the mixed-effects model that considers sample plot as a random effect outperformed the basic model,which significantly improved the accuracy of prediction.Compared with the basic model developed using least square method,the mixed-effect model considering the hierarchical structure significantly improved the performance of the model,and the developed model has certain biological significance and statistical reliability.更多还原