【作者】 王艳霞；

【导师】 吴承祯；

【作者基本信息】 福建农林大学 ， 森林经理学， 2010， 博士

【摘要】 随着全球工业的不断进步发展,大气温室气体含量不断上升,全球气候变暖,越来越引起人们的关注。森林生态系统作为陆地生态系统的主要组成部分,其固碳和吸碳作用日趋显著。森林生态系统碳储量包括土壤有机碳储量、植被有机碳储量和凋落物层有机碳储量,其现有碳储量是评价森林生态系统吸收CO2功能的主要尺度。杉木(Cunninghamia lanceolata)、马尾松(Pinus massoniana)、毛竹(Phyllostachys hterocycla)和桉树(Eucalyptus grandis)都是中国南方重要的森林资源,经济用途广泛,是福建省分布最广的人工林生态系统。本文采用野外调研与室内试验相结合的方法,研究了福建杉木、马尾松、毛竹和桉树人工林生态系统碳密度及其空间分布特征,并用最小二乘—支持向量机法分别对各人工林土壤有机碳含量和有机碳密度进行了预测估计,分析了土壤有机碳密度与林龄、坡向、坡位的线性和非线性回归关系;运用多元线性回归模型预测不同人工林乔木层各器官生物量,在此基础上计算乔木层各器官的有机碳含量和有机碳密度;对杉木、马尾松和桉树人工林生态系统中,土壤各层有机碳密度与乔木层的干、枝、叶和根,林下植被的草本层和灌木层以及凋落物层进行了典型相关分析研究,深入分析土壤碳密度与地上部分碳密度的关系;最后,运用刀切法、投影寻踪等方法对乔木层碳密度与胸径、树高的关系进行模拟。主要研究结果如下:1、土壤有机碳含量和有机碳密度的空间分布特征:有机碳含量和有机碳密度均垂直方向上表现为土壤表层有机碳含量随着土层深度的增加而逐渐降低,其中表层到第二层的降低速度最快;水平方向上,从大到小依次为毛竹>桉树>杉木>马尾松,下坡>中坡>上坡、阳坡>阴坡;杉木林和毛竹林土壤有机碳含量和有机碳密度表现出幼龄林>中龄林>成熟林的变化特征;而马尾松林地、桉树林地在不同发育阶段土壤有机碳密度趋势则相反,表现为幼龄林<中龄林<成熟林的变化特征。但因其受土壤碳含量、土壤容重、土层厚度的影响,又有一定变异性。2、应用多元线性回归模型对杉木、马尾松、毛竹和桉树人工林生态系统中土壤有机碳密度与环境因子坡向和坡度以及林龄的关系,进行了多元回归分析。结果表明,林龄对杉木土壤剖面有机碳密度的影响比较显著,而对马尾松、毛竹和桉树3个林分土壤剖面有机碳密度的影响不显著,坡向和坡位对4个林分土壤剖面有机碳密度的影响均较小。但因多元线性回归的前提是假设土壤有机碳密度与各因子之间的线性关系。3、进一步应用最小二乘支持向量机法对各人工林土壤碳密度进行预测,结果表明,对建阳和永安杉木人工林土壤表层有机碳密度预测的平均误差分别为0.077%和0.116%;对建阳和永安杉木人工林土壤剖面有机碳密度预测的平均误差为0.098%和0.082%;对建阳和永安马尾松人工林土壤表层有机碳密度预测的平均误差分别为0.082%和0.081%;对建阳和永安马尾松人工林土壤剖面有机碳密度预测的平均误差为2.559%和0.049%;对建阳和永安毛竹人工林土壤表层有机碳密度预测的平均误差分别为2.709%和3.928%;对建阳和永安毛竹人工林土壤剖面有机碳密度预测的平均误差为2.895%和2.474%;对桉树人工林土壤表层有机碳密度预测的平均误差为0.066%;对桉树人工林土壤剖面有机碳密度预测的平均误差为0.049%。说明LS-SVM模型对福建4种人工林土壤有机碳密度值的预测效果良好。4、森林生态系统中各部分有机碳密度,特别是土壤有机碳密度受诸多因子影响,而且各因子之间可能存在线性或非线性的关系,且有主要因素和次要因素。本文采用尝试用建立扩展离散灰色序列模型,来预测土壤有机碳密度值,并确定影响土壤有机碳密度的主要变量。结果表明,杉木、马尾松和桉树人工林的林龄与土壤表层有机碳密度和土壤剖面有机碳密度之间的复相关系数均为最大值,因此确定林龄为杉木林、马尾松林和桉树林的主变量,其余因子为辅助变量。运用扩展灰色离散序列高阶动态预测模型对杉木、马尾松和桉树人工林表层土壤有机碳密度和土壤剖面有机碳密度进行模拟,结果表明相关系数R均大于0.7832,且预测值与实测值误差较小,说明扩展灰色离散序列高阶动态土壤有机碳密度预报模型对福建不同人工林土壤有机碳密度的预测效果良好。5、选取福建建阳、永安地区的杉木、马尾松和桉树人工林表层土壤有机碳密度和土壤剖面有机碳密度为因变量,林龄(A)、坡位(P)、坡向(S)、林分密度(D)、坡度(G)、土壤容重(V)等为自变量,利用实数编码自适应约束优化遗传算法进行参数搜索,建立投影寻踪回归预测模型。结果表明,该模型在3个人工林土壤表层和剖面有机碳的预测中,其预测值平均误差区间为5.73%~10.56%,标准误差区间为7.25%~15.24%,相关系数介于0.886~0.940之间,说明改进的投影寻踪模型用于土壤有机碳密度预测,能达到理想效果,方法简单有效,模型适用性及实用性较强。6、生物量模型的改进:乔木层各器官的生物量采用相对生长量法进行模拟,即利用各器官的生物量与D2H(其中D为胸径、H为树高)的关系进行模拟;本文在模拟W = a( D2H)b的基础上,对各器官生物量的改进模型W = a?DbHc进行优化模拟,结果表明模拟效果理想,其结果可以用来计算乔木层各器官有机碳密度。7、不同森林类型乔木层有机碳密度存在差异,其中杉木林乔木层有机碳密度最高,为15.145 kg/m2,其次为马尾松林乔木层有机碳密度,为13.723 kg/m2,最小是桉树林乔木层有机碳密度,为5.662 kg/m2,表现出杉木林>马尾松林>毛竹林>桉树林的趋势。从各林分类型乔木层各器官有机碳密度来看,均表现出干>根>枝>叶的趋势。就树干而言,有机碳密度从大到小依次为:马尾松>杉木>桉树>毛竹;就树根而言,有机碳密度从大到小依次为:杉木>马尾松>毛竹>桉树;就树枝而言,有机碳密度从大到小依次为:杉木>马尾松>毛竹>桉树;就树叶而言,有机碳含量从大到小依次为:杉木>马尾松>桉树>毛竹。8、不同林分的林下植被有机碳密度以马尾松最大,为0.263 kg/m2,其次为桉树和杉木,毛竹林下植被的碳密度最少,只有0.031 kg/m2。林下植被各层中,灌木层和凋落物层均以马尾松林的碳密度最大,分别为0.133 kg/m2和0.107 kg/m2,桉树和杉木其次,毛竹最少;草本层的有机碳含量则以杉木林最大,桉树和马尾松次之,毛竹最少。这进一步说明本研究中的几种人工林中,毛竹的林分结构比较单一,林分碳储量受人为干扰严重;马尾松和杉木人工林的林分结构相对比较丰富,林下植物种类多样,生物多样性较高。9、4种人工林生态系统有机碳贮量各不相同,杉木人工林生态系统的有机碳储量最大,达到28.125 kg/m2;马尾松人工林生态系统的有机碳储量次之,为27.779 kg/m2;毛竹人工林生态系统的有机碳储量第三,为22.884 kg/m2;而桉树人工林生态系统的有机碳储量最小,为22.381 kg/m2。10、森林生态系统是一个有机整体,各部分的有机碳含量除受环境因子的影响外又有一定得自相关性。本文将土壤亚层即h1、h2、h3和h4层的有机密度以及0-100cm的剖面有机碳密度一起设为第一组的x1、x2、x3、x4、x5典型变量,乔木层的干、根、枝、叶、草本层、灌木层和凋落层分别为第二组的y1、y2、y3、y4、y5、y6、y7典型变量,采用典型相关分析讨论地下部分碳密度对地上部分碳密度的影响,深入分析土壤碳密度与地上部分碳密度的关系。结果表明,土壤碳有机碳密度对u1的相对作用大小依土壤深度依次减小,即:第h1层( 0~20㎝)>第h2层(20~40㎝)>第h3层(40~60㎝)>第h4层(60~100㎝),其中第h1层是作用最大,第h4层的作用很小。v1与地上部分各层次有机碳密度(yi)的原始数据相关关系如下:v1与地上部分树干、树根有机碳密度(yi)的原始数据存在明显的正相关。11、根据福建省永安和建阳杉木、马尾松、毛竹和桉树人工林64株调查材料,按刀切法原理对人工林乔木层有机碳密度的进行估算。经计算,人工林乔木层有机碳密度的估计量为12.740 kg/m2,且估计精度为93.239%。因此,利用刀切法估算人工林乔木层有机碳含量的结果可靠。12、基于投影寻踪的乔木层有机碳密度回归模型:投影回归是用于分析和处理非正态、非线性数据的一种新方法。由于建立多元回归模型的前提是:影响因子与预报因子之间确切存在模型的假定相关关系,而各预测因子之间的相关关系并不是一致的线性或非线性,而是存在多种相关形式,因此采用一致线性或非线性形式建立的回归模型不能真实的反映回归关系。本文引入加权的思想,在改进单纯形算法的基础上建立了乔木层有机碳密度模型,取得了较满意的效果,12个训练样本拟合值的总体拟合精度较高,在建阳杉木人工林乔木层有机碳密度预测模型中,预测和实测值相比较,平均误差为3.163%,标准误差为2.096%,相关系数达0.923,精度较高,预测效果良好。更多还原

【Abstract】 With the continuous progress of global industry, the content of greenhouse gases in atmosphere is rising continuously, and global warming is more and more cause the attention of people. Forest ecosystems are the main component of terrestrial ecosystems, their function of carbon fixation and carbon sinks become more significant. Carbon storage in forest ecosystems including organic carbon storage in soil, organic carbon storage in vegetation and organic carbon storage in litter layer, is the major source of evaluating forest ecosystem’s carbon dioxide absorption function. Cunninghamia lanceolata, Pinus massoniana, Phyllostachys edulis and Eucalyptus grandis are important forest resources in south of China with wide economic use, and they are the most widespread plantation ecosystems in Fujian province. This thesis use the method of combining field investigation and laboratory testing to study the plantation ecosystem’s carbon density and spatial distribution of Cunninghamia lanceolata, Pinus massoniana, Phyllostachys edulis and Eucalyptus grandis in Fujian province. Based on the data, the least squares -support vector machine method is used to separately predict and estimate the plantations’organic carbon content and organic carbon density in soil and analyze the linear and nonlinear regression between organic carbon density in soil, and stand age, slope exposure, slope position. Therefore this thesis apply the multivariable linear regression model to predict the biomass of different plantation’s various organs in tree layer and calculate the organic carbon content and organic carbon density of various organs in tree layer. In plantation ecosystems of Cunninghamia lanceolata, Pinus massoniana and Eucalyptus grandis, the canonical analysis research is carried out to analyze the relationships between each soil layer’s organic carbon density and tree layer’s trunk, branch, leaf and root, the herb layer, shrub layer and litter layer of undergrowth vegetation. Furthermore, the relationship between soil carbon density and aboveground carbon density is studied. Finally, the jackknife method, projection pursuit method, etc. are used to simulate the relationship between carbon density in the tree layer and DBH, tree height. The major research results are as follows:1 The characteristic of soil organic carbon content and carbon density’s spatial distribution: organic carbon content and carbon density are all represented as the surface soil organic carbon content‘s decreasing along with soil depth’s increasing in the vertical direction, in which the fastest decline is from the surface layer to second layer. In the horizontal direction, the order is followed as Phyllostachys edulis>Eucalyptus grandis>Cunninghamia lanceolata>Pinus massoniana, downgrade>mesoslope>upslope, sunnyslope>dark-slope. Cunninghamia lanceolata forest and Phyllostachys edulis forest’s soil organic carbon content and organic carbon density represent the mutative characteristic of young forest>middle-aged forest>mature forest, but Pinus massoniana forest, Eucalyptus grandis forest at different developmental stages, the trend of soil organic carbon density is opposite, represent young forest<middle-aged forest <mature forest. However, due to the effect of soil carbon content, soil bulk density, thickness of soil, the trend showed certain variability.2 The multivariable linear regression model was used to carry out multiple regression analysis for the relationship between soil organic carbon density and environmental factors exposure, slope, and forest age in the plantation ecosystems of Cunninghamia lanceolata, Pinus massoniana, Phyllostachys edulis and Eucalyptus grandis. The results showed that forest age’s effect on organic carbon density of soil profile is more significant and emerges positive correlation, while the slope aspect and slope position’s effect on organic carbon density of soil profile is relatively small and negatively correlated. But the premise of multiple linear regressions is the hypothetical linear relationship between soil organic carbon density and each factor. 3 The least square support vector machine method is used to predict the plantation soil carbon density estimation. This thesis adopt (0,1) the normalization method for data processing and the Gaussian radial function as the kernel function to estimate the numerical value of surface soil organic carbon density and the profile organic carbon density. The prediction result is good, but the operation process is generally "black-box operation".4 Among of various parts of the organic carbon density in forest ecosystems, especially in soil organic carbon density, which affected by many factors and each factor may exist between the linear or nonlinear relationship, and there were the main factors and secondary factors. In this paper, it used an extended sequence of discrete gray model to predict soil organic carbon density, and determined the impact of soil organic carbon density of the main variables. The results showed that, in fir, pine and eucalyptus forest plantation, the complex correlation coefficient value of age was the maximum, which related to soil organic carbon density and surface soil organic carbon density, thus the age of this forest fir, pine and eucalyptus forest was the main variables, the remaining factors were auxiliary variables. The correlation coefficient R were greater than 0.7832 when using gray high-end discrete sequence generalized dynamic model to simulated the fir, pine and eucalyptus plantations soil organic carbon density and soil organic carbon density. And the predicted values were similar with measured values, which indicated that the gray discrete series high order prediction model was fit to soil organic carbon density of soil organic carbon density in Fujian forests.5 Surface soil organic carbon density and soil organic carbon density of Fir, pine and eucalyptus plantations in Jianyang, Yon’an area, Fujian province, were selected as dependent variable, age (A), slope position (P), slope (S), Lin Density (D), slope (G), soil bulk density (V) as the independent variables. Based on which, it established an projection pursuit regression model using adaptive real-coded genetic algorithm constrained optimization parameter search. The results showed that the model in three plantation soil surface and profile prediction of organic carbon, the average error range of the predicted value of 5.73% ~ 10.56%, standard error interval of 7.25% ~ 15.24%, the correlation coefficient between 0.886 ~ 0.940, which indicated that the improved projection pursuit model for prediction of soil organic carbon density, could achieve the desired results, and the model was simple and effective, applicable and practicable strongly.6 The biomass of each organ in tree layer is simulated with relative growth method which is based on the relation between the biomass of each organ and D2H (D is diameter at breast height, H is the tree height). In this thesis, based on the simulation of W = a( D2H)b, optimized simulation is used for the improved model W = a?DbHc of various organs’biomass. The results showed that the simulation results are satisfactory and can be used to calculate the organic carbon density of tree layer in different organs.7 The organic carbon density of different forest types in tree layer is different. Among them, tree layer of Cunninghamia lanceolata forest’s organic carbon density is the highest of 15.145 kg/m2, followed by the tree layer of Pinus massoniana forest’s organic carbon density of 13.723 kg/m2, the smallest is the tree layer of Phyllostachys edulis forest’s organic carbon density of 5.662 kg/m2. This showed the trend of Cunninghamia lanceolata forest> Pinus massoniana forest> Phyllostachys edulis forest. The organic carbon density of various organs in different kinds of forest’s tree layer all showed the trend of trunk> root> branch>leaf. In detail, on the part of trunks, the big to small order of organic carbon density is followed as: Pinus massoniana > Cunninghamia lanceolata > Eucalyptus grandis > Phyllostachys edulis. On the part of roots, the big to small order of organic carbon density is followed as: Cunninghamia lanceolata > Pinus massoniana > Phyllostachys edulis > Eucalyptus grandis. On the part of branches, the big to small order of organic carbon density is followed as: Cunninghamia lanceolata > Pinus massoniana > Phyllostachys edulis > Eucalyptus grandis. On the part of leaves, the big to small order of organic carbon density is followed as: Cunninghamia lanceolata > Pinus massoniana > Eucalyptus grandis > Phyllostachys edulis.8 The largest organic carbon density of understory vegetation in different forest is Pinus massoniana of 0.263 kg/m2, followed by Eucalyptus grandis, Cunninghamia lanceolata, Phyllostachys edulis. The carbon density of the vegetation under Phyllostachys edulis forest is the least of only 0.031 kg/m2. Underground vegetation in each layer, the carbon density of pinus massoniana in shrub layer and litter layer is the largest of 0.133 kg/m2 and 0.107 kg/m2, followed by Eucalyptus grandis and Cunninghamia lanceolata, Phyllostachys edulis is the least. And, the largest organic carbon content in herb layer is Cunninghamia lanceolata forest, followed by Eucalyptus grandis, Pinus massoniana and Phyllostachys edulis. Hence, in this study, the forest structure of Phyllostachys edulis is simple and seriously disturbed by human. Pinus massoniana and Cunninghamia lanceolata plantations are relatively rich in forest structure, understory plant species is diversity and the biodiversity is high.9 The carbon storage of four kinds of plantation’s ecosystem is different, The organic carbon storage of Cunninghamia lanceolata plantation’s ecosystem is the largest of 28.125 kg/m2, Pinus massoniana plantation is the second of 27.779 kg/m2. Phyllostachys edulis is the third of 22.884 kg/m2, and Eucalyptus grandis plantation is the smallest of 22.381 kg/m2.10 Forest ecosystem is an organic whole, each part of the organic carbon content not only influenced by environmental factors, but also there must be self-relevant. This thesis makes the sub-layer organic density of the soil h1, h2, h3 and h4 and 0-100cm’s profile organic carbon density as the typical variable of first group x2, x3, x4, x5, and the tree layer of trunk, root, branch, leaf, herb layer, shrub layer and litter layer are the second group as y1, y2, y3, y4, y5, y6, y7. Canonical correlation analysis was carried out to discuss the effect of underground carbon density to aboveground carbon density and analyze the relationship between soil carbon density and aboveground carbon density. The results showed that the extent of relative effect of soil organic carbon density to u1 will successively reduce as the soil depth decline, and the order is the h1 layer (0 ~ 20 cm) >the h2 layer (20 ~ 40 cm) > the h3 layer (40 ~ 60 cm) >the h4 layer (60 ~ 100 cm) in which the role of the h1layer is the largest, the role of the h4 layer is little. The correlativity between v1 and the raw data of each layer’s organic carbon density on the aboveground part is as follows: a significant positive correlation is existed between v1 and aerial parts of tree trunks, roots organic carbon density (yi).11 Based on the 64 survey materials of Cunninghamia lanceolata, Pinus massoniana, Phyllostachys edulis and Eucalyptus grandis in Yongan and Jianyang County, Fujian province the jackknife method was used to estimate the organic carbon density in tree layer of plantations. According to calculate results, the organic carbon density’s estimated amount is 12.740 kg/m2 in plantation tree layer, and the estimated accuracy is 93.239%. Therefore, the result of estimating the organic carbon content of plantation tree layer is reliable by using the method of jackknife.12 Because the premise of establishing multiple regression models is that there is exactly exist the model’s assumed correlation between the impact factor and predicting factor and the correlation between predictors is not consistent linear or nonlinear and show several related forms, projection pursuit regression model of soil organic carbon density is adaptable for simulating. This thesis introduced the idea of weighting and established the soil organic carbon density model in the constrained based constringent and optimized genetic algorithm.更多还原