【作者】 张志东；

【导师】 臧润国； 蒋有绪；

【作者基本信息】 中国林业科学研究院 ， 生态学， 2007， 博士

【摘要】 海南岛的热带森林是我国生物多样性最为丰富的生态系统之一，同时也是遭受干扰最严重的植被类型之一。过度的森林采伐和不合理的土地利用已导致了原始林大面积减少，次生林面积显著增加。景观空间配置的变化以及景观的破碎化，对具有不同功能特征的物种组在景观中的生存、维持、生长和分布等均产生了深远的影响。我国以往对热带森林的研究多集中在群落及群落以下水平，在景观水平上的研究很少，而以功能群为基础在景观水平上的研究则还没有人涉及。本文以海南岛典型的热带天然林区—霸王岭林区为对象，应用GPS定位、公里网格群落样方调查、遥感影像解译、GIS分析和模型模拟相结合的方法在景观水平上开展了热带天然林功能群的生态学研究，主要内容及结果如下：(1)首先基于栅格数据分析了斑块类型和景观格局的变化，其次为了探讨与恢复后期林片断的毗邻程度在决定次生演替上所起的作用，进行了缓冲区分析。结果表明：在1986-1998年间，相对于其它类型外，恢复后期林是唯一变化不显著的类型，而在1998-2002年间，大多数类型均变化不显著；大多数斑块类型在1986-2002年间破碎化趋势明显，表现为斑块数、边缘密度持续增加，斑块平均大小和核心区面积不断减小；次生林分布与恢复后期林距离呈现明显的相关关系。恢复时间长的次生林在恢复后期林附近所占面积比例较高，随着与恢复后期林距离的加大，恢复时间短的次生林逐渐占据优势。(2)为了阐明在不同的时空尺度上功能群内物种丰富度和多度的变化规律，依据植被类型、干扰方式和恢复阶段划分了8个森林景观类型，同时依据生长型、演替地位和大小特征划分了7个功能群。通过研究发现功能群个数在景观水平上变化不明显，仅在低地雨林皆伐恢复前期景观缺失了主林层和超冠层乔木功能群；不同的森林景观类型都会有一组功能群占优势；低地雨林皆伐中期景观和海南松老龄林之间，低地雨林径级择伐景观和低地雨林老龄林之间，以及山地雨林各景观类型之间功能群物种丰富度和多度结构相似性水平较高；具有同样演替地位的乔木和灌木功能群，在景观类型之间具有相似的物种丰富度和多度变化格局；各功能群物种丰富度和多度在低地雨林恢复系列上存在相似的变化格局，而在山地雨林恢复系列上，则相似性较差。本研究的结论进一步证明了竞争一定居权衡和演替生态位理论的正确性而不太适用于中性理论和生态平衡假说。(3)为了探讨在热带天然林景观中不同因素对功能群分布的影响，运用冗余度分析(Redundancy analysis，RDA)分别探讨了功能群出现与否、功能群物种丰富度和多度与环境、空间和干扰因素之间的关系．运用偏冗余度分析对影响功能群分布变化的环境、干扰、空间及其交互作用等因素进行了定量分解。结果表明：利用木材密度和潜在最大高度两个指标划分出的9个功能群，能较好的区分功能群间对生境的适应差异；因子分解表明，纯环境、纯人为干扰和环境-人为干扰交互作用是影响3个功能群矩阵分布变化的主要因素，纯空间和与空间位置相耦合的因素相对较低；在诸多因子中，干扰类型、地形因子、土壤类型、砂砾含量、土层厚度以及经纬度坐标是影响功能群分布变化的主导因子；RDA排序分析表明，一般来讲，硬木多在生境条件好，远离人为干扰的立地高发生。软木则更多地出现于生境条件恶劣，人为干扰频繁的立地。而中等硬度功能群则生态幅度较宽。但除了软木灌木功能群外，其它功能群物种丰富度和多度多在土层深厚，生境条件较好的立地较高。(4)从优势功能群途径出发，建立了热带天然林景观分类体系。通过对TM影像进行低通滤波(5×5)处理，最大似然监督分类，总体精度评估和Kappa统计方法检验以及借助GIS技术进行优势功能群斑块在森林恢复阶段和地形梯度上的分布格局分析。结果表明：7个土地利用类、4个恢复阶段类和6个优势功能群类在影像上获得了较好的分离精度，总分类精度均在78％以上；先锋种功能群斑块在灌木(稀树)草地景观、恢复初期景观、平坡、半阳坡和阳坡面以及在海拔小于850 m的范围内，优势度较高。而在恢复中后期、半阴坡和阴坡以及海拔850 m以上的景观中则主要由顶极种功能群斑块所控制。功能群斑块格局在坡度梯度上敏感性不高。(5)为了探讨热带林不同演替阶段的功能群地上生物量与植被指数之间的相互关系并对其分布进行预测，在基于遥感影像和135个样地调查的基础上，分别选取了归一化差异植被指数(NDVI)、短红外湿度植被指数(MVI5)、中红外湿度植被指数(MVI7)和比值植被指数(RVI)与计算的顶极种和先锋种功能群生物量做相关分析，并利用逐步线性回归分析分别构建了基于植被指数的生物量回归模型；利用残差图对模型的有效性进行检验。结果表明MVI7和MVI5与顶极种功能群生物量关系显著，而NDVI和RVI对先锋种功能群生物量具有较好的指示作用；顶极种和先锋种功能群生物量具有高预测面积的区域分别占总面积的73.98％和88.08％，表明两个生物量模型均具有较好的模拟效果；模拟结果表明顶极种功能群生物量主要集中于研究区中部、北部和西南部区域，而先锋种功能群生物量分布并没表现出明显的规律，而是不均衡的散布于整个研究区。(6)为了探寻景观格局和物种丰富度的关系，利用最大似然监督分类方法对该区核心区5个连续的景观进行了斑块类型的划分：在每个斑块类型内计算了各功能群物种丰富度的平均数；结合主成分分析和多重回归模型进行了景观格局指数和各功能群物种丰富度的相关分析。分析结果表明在斑块类型水平上斑块类型面积比例、边界密度、总边界对比度指数和面积加权平均形状指数是影响各功能群物种丰富度分布的主要因子；相对于先锋功能群，景观破碎化对顶极功能群影响较大，表现为随着生境多样性的增加和生境形状的复杂化相对于先锋功能群，顶极功能群物种丰富度增加明显，而随着边界密度的增加和总边界对比度指数的提高，先锋功能群物种数略有增加，而顶极功能群物种数有显著减小的趋势。(7)为了探讨在热带天然林景观中关键种的潜在分布，首先在功能群框架下运用优势度指数法进行了关键种的确定；采用基于GIS的GARP生态位模型对主要关键种的地理分布进行了预测，并应用受试者工作特征分析进行了模型精度验证；应用多元线性回归分析对影响各关键种潜在分布的关键因子进行了确定。结果表明：在8个功能群中，先锋种、顶极灌木种和顶极超冠层乔木种划分出的关键种较为理想；在进行预测的8个关键种中，除了先锋主林层乔木种海南杨桐(Adinandra hainanensis)，其它3个先锋种包括毛稔(Melastoma sanguiueum)、银柴(Aporosa chinensis)和枫香(Liquidambar formosana)在研究区北部、西部以及西南部均具有较高的发生概率，而顶极种除了顶极超冠层乔木种南亚松(Pinus merkusii)外，九节(Psychotria rubra)、高脚罗伞(Ardisia quinquegona)和海南椎(Castanopsis hainanensis)具有相似的潜在分布格局，在研究区中部、东南部和南部地区具有较高的发生概率；相关分析表明极端最低温、年均温、极端最高温、年均降水量、海拔和坡向6个因子是影响研究区关键种潜在分布的关键因子；精度检验表明，GARP模型对8个关键种的潜在分布预测效果均较好。(8)为了探讨在热带天然林景观进行基于功能的潜在自然植被分布，首先采用生态位模型途径对8个功能群的潜在分布进行了模拟；其次对TM影像采用最大似然监督分类方法进行了基于功能的现状自然植被识别；然后通过对功能群发生概率图进行叠加，并在参考现状植被分布以及在功能群之间相互作用规则的制约下，构建了最终的潜在自然植被分布图。结果表明：生态位模型对8个功能群的潜在分布预测效果均较好，平均曲线下面积达到了0.81；先锋种功能群和顶极种功能群呈现出相异的分布格局，反映了两大类功能群个体生态学特性的差别和对环境的适应差异；各功能植被类型在空间分布格局和分散程度上存在着明显的差异，反映了一定的空间自相关特征。更多还原

【Abstract】 As one of the most species-diverse ecosystems in China, tropical forests of Hainan Island are increasingly affected by various disturbances. Due to extensive deforestation and unreasonable land use, the area of primary forests has been reduced and that of secondary forest has been increased markedly. Landscape fragmentation and the change of spatial configuration of patches have dramatic impacts on coexistencence, growth, maintenance, and distribution of plant functional groups in forest landscapes. The past ecological researchs conducted in tropical forest of Hainan Island were mainly at community or below community levels, ecological researches at the landscape-level and even research based on functional group approach were, however, rarely explored. A typical tropical forest landscape-Bawangling region of Hainan Island was selected for this study. By combining methods of GPS positioning, field investigation of systematic grid sampling plots, remote sensing imagery interpretation, GIS analyses and model, functional groups of natural tropical forests at landscape scale were studied. The main contents and results of the research are as follows:(1) Grid squares were used as the sampling units for patch type and landscape pattern change analyses. To explore the impact of proximity of late-successional forest on secondary succession, the buffer analysis was performed. The results showed that between 1986 and 1998, except late-successional forest, the proportion of all other patch types changed significantly. However, between 1998 and 2002 most of patch types didn’t changed significantly. Landscape metrics changed from 1986 to 2002. The number of patches and edge density increased, in contrast, mean patch size and core area decreased in most of patch types which indicated that fragmentation of the study region increased from 1986 to 2002. The distribution of secondary forest was significantly associated with the location of late-successional forest. The propotion of secondary forest with long recovery time was higher near the late-successional forest. However, with the increase of distance away from late-successional forest, the propotion of forest with short recovery time increased gradually.(2) To understand the spatiotemporal patterns and dynamics of species richness and abundance of functional groups in the tropical forest landscape, we classified the study area into eight landscape types based on vegetation type, disturbance manner and the time of recovery. The plant species were also aggregated into seven functional groups based on the growth form, successional status and plant size. The results showed that all functional groups, except for the emergent and canopy tree species which absent in early-successional clear cut lowland rain forest landscape, were present in all landscape types. Each landscape type had different numbers of dominant functional groups. There were similar species richness and abundance structure among functional groups between mid-successional clear cut lowland rain forest and old growth tropical coniferous forest. This similarity existed in selective logged and old-growth lowland rainforest, as well as among landscape types of montane rainforest. The functional groups with same successional status had similar patterns of species richness and abundance ratios among different landscape types. The variation patterns of functional groups along the successional stages in terms of species richness and abundance among the tropical lowland rainforest landscape types were more similar to each other than those in the tropical montane rainforest landscape types. This study provides further support for competition-colonization tradeoff and successional niche theory as opposed to models of neutrality and ecological equivalence.(3) To explore the relative influence of ecological factors on the distribution of plant functional groups in the tropical natural forest landscape, we derived functional group classification using the characteristics of species-specific wood density and potential maximum height. The three functional groups data sets were used, i.e. the presence-absence, the species richness, and abundance. Their relations to environmental, anthropogenetic disturbance, and spatial factors were analyzed with redundancy analysis (RDA). We then used a partial RDA with variation partitioning to specify which proportion of the variation in functional groups distribution pattern is explained by each of the three factors exclusively and which proportions are attributable to interactions between the factors. We found that purely environmental, purely anthropogenetic disturbance and the interaction between environmental and anthropogenetic disturbance were of comparable importance for controlling the changes of functional groups distribution. However, purely spatial influence and the interaction between spatial and other factors were relatively less important. Furthermore, we found that anthropogenetic disturbance types, topographical factors, soil types, gravel content, soil depth and X, Y coordinates all had unique impacts on functional groups distribution. Generally, hardwood functional groups were of high presence at the well-conditions, off-disturbance sites. But for softwood, the situation was the opposite. The distributional range of moderate density wood was wider than the other two groups. Except softwood shrub, the others having high species richness and abundance occurred at the well-conditioned sites, especially where soil depth was deeper.(4) A functional groups based approach was tried in this paper to classify landscape types in the species rich forest ecosystems. In this process, the TM imagery was treated by 5×5 low-pass mean filter, and the identification was made by the supervised maximum likelihood classification technique combining with field investigation data. The accuracy of classification was assessed by the classification error matrix and Kappa statistics. Finally, the distribution patterns of patches composed of different dominant functional groups and their relationships with successional stage and topography were analyzed by GIS. The results showed that seven land-use classes, four successional stage classes and six predominant functional group classes were classified from Landsat TM data with a reasonable degree (an average of up to 78 percent of overall classification accuracy) of class separability. Patches dominated by the pioneer were mainly distributed in the shrub lands, early recovery stage stand, flat area, sunny slopes and elevation ranges of＜850 m. While the patches dominated by the climax species had more occupancies in the middle and late recovery stages, shaded slopes, and elevation ranges of＞850 m. The slope gradient had no significant influences on the distribution patterns of patch types dominated by different functional groups.(5) To evaluate relationships between biomass and vegetation indices and to determine the spatial distribution of biomass for functional groups of different successional status in the tropical forest, we used the modelling procedure based on vegetation index and 135 sampling plots distributed over the study area. Biomass of different functional groups was obtained by regression analysis. Four vegetation indices (normalized difference vegetation index (NDV1), moisture vegetation index using Landsat’s 5 (MVI5), moisture vegetation index using Landsat’s band 7 (MVI7) and ratio vegetation index (RVI)) were correlated with measurements of aboveground biomass (climax and pioneer species biomass). Models describing the relationships between biomass and vegetation indices using stepwise linear regression analysis were also developed. Three biomass components maps were produced using the developed models. Residual maps were used to test the validity of the models. The results showed that MVI7 and MVI5 were the most important vegetation indices that are effective determinants of biomass for the climax species functional group, whereas NDVI and RVI seem to be good indices of biomass for pioneer species functional group. The strongly predictive percent areas for climax and pioneer species biomass model were 73.98 and 88.08 respectively. Simulated biomass of climax species was mainly continually distributed in the center, north and southwest parts of study area. However, simulated biomass of pioneer species was scattered over the study area.(6) To identify and characterize the relationships between landscape patterns of different habitat types and species richness of functional groups in the natural tropical forest region, we obtained patch types from a Thematic Mapper imagery classification using a supervised maximum likelihood classification in five adjacent sub-regions in the study region. The mean number of species of different functional groups was calculated for each patch type in each of the five sub-regions. The relationships between landscape patterns and the mean number of species of different functional groups were evaluated with multiple linear regression and principle component analysis. The results showed that significant relations existed between four landscape metrics and the mean number of species of different functional groups at the patch type level. The four landscape metrics included percentage of landscape (PLAND), edge density (PD), area-weighted mean shape index (SHAPE_AM) and total edge contrast index (TECI). The nature of these relations changed according to the functional groups considered, indicating that different functional groups reacted to the landscape pattern differently. The influences of landscape fragmentation appeared to be stronger for climax than for pioneer functional groups. With the increase of habitat diversity and more irregular shape of patch types, the species richness of climax functional groups increased more than that of the pioneer functional groups. While with the increase of PD and TECI, the species richness of pioneer functional groups didn’t change significantly, on the contrary, the species richness of the climax functional groups decreased significantly.(7) Keystone species were identified within the context of functional groups using Dominance Index (DI). The GARP was used to estimate the keystone species’ potential distribution and then the Receiver Operating Characteristics was used to evaluate the predictive performance. On this basis, by applying multiple linear regression analysis, we identified themajor factors determining the potential distributions of keystone species. The results showed that identification of keystone species within the pioneer, climax shrub and emergent tree functional groups was easier than within the climax subcanopy and climax canopy tree functional groups. Among the eight keystone species, pioneer species such as Melastoma sanguiueum, Aporosa chinensis and Liquidambar formosana (but except Adinandra hainanensis) had high occurrence probabilities in the north, west and southwest areas of study area. However, climax species such as Psychotria rubra, Ardisia quinquegona and Castanopsis hainanensis (but except Pinus merkusii) have high occurrence probabilities in the central, southeast and south regions of the study region. Minimum and maximum temperature, mean annual temperature and precipitation, aspect and altitude were the key factors determining the potential distributions of keystone species. Evaluation of model’s performance indicated excellent predictive abilities of the model for predicting distributions of eight keystone species.(8) To reconstruct the functional groups based potential natural vegetation (PNV) in the tropical forest landscape, a series of procedures were conducted in this study. Firstly, we used the GIS-based GARP model to estimate the potential distributions of the eight functional groups. Secondly, we identified functional groups based realized natural vegetation (RNV) by using a supervised maximum likelihood classification technique. Thirdly, through overlaying occurrence probability surfaces of the functional groups while considering the distribution of RNV and interactions among the different functional groups, we derived the final functional groups based PNV. The results showed that the well model predictive performances for the functional groups were obtained, with all area under curve (AUC) up to 0.81. The distribution patterns of pioneer and climax functional groups were significantly different which reflected the differences of autoecological functional traits and reactions to environmental changes between the two functional group categories. The differences in spatial distribution patterns and in dispersal extents among functional groups based vegetation types indicated that the spatial autocorrelation existed in compositions of the different vegetation types.更多还原