【作者】 刘清旺；

【导师】 李增元；

【作者基本信息】 中国林业科学研究院 ， 森林经理学， 2009， 博士

【摘要】 森林是一个多资源、多功能的综合体,掌握其生长和消亡的发展规律具有重要意义。传统的森林资源地面调查方法可以获取详细的调查数据,但是周期比较长。遥感技术的出现和迅速发展,为快速、准确地获取大范围森林资源调查数据提供了一种有潜力的技术手段。激光雷达(Lighting detection and ranging, LIDAR)是一种主动遥感技术,能够精确地获取地表物体的三维特征信息。本文以揭示LIDAR探测森林冠层三维结构的机理,寻求森林参数的有效估测方法为目的,开展了以下几个方面的研究工作:(1)根据地物的后向散射特征,建立了立体散射体模型。通过对激光脉冲与森林冠层之间的相互作用关系的分析,发现已有的LIDAR方程在解释展宽的返回波形时存在局限性,因而,在此基础上建立了立体散射体模型,并将散射体划分为三种类型,即简单散射体、立体散射体和复杂散射体。简单散射体的LIDAR方程不变,立体散射体的LIDAR方程中引入了扩展卷积函数,即对已有卷积函数进行了扩展,以便解释激光脉冲在立体散射体中的后向散射特征,复杂散射体的LIDAR方程表示成另两种散射体方程的简单叠加。(2)依据不同散射体的波形特征,提出了波形特征分析方法。通过对不同散射体激光脉冲入射波形和返回波形的分析,发现激光脉冲返回能量的分布表现为不同特征;反过来,根据这些特征可以判断对应散射体的类型,提取相关的波形特征量。对于简单散射体,可以由波形特征量计算得到LIDAR到散射体之间的距离;对于立体散射体,可以得到LIDAR到散射体不同深度处的距离、散射体的深度;对于复杂散射体,可以得到LIDAR到散射体不同深度处的距离(包括到各子散射体的距离)、散射体的深度(包括子立体散射体的深度)。(3)参照不同散射体的后向散射截面,讨论了相对后向散射截面分析方法。根据不同散射体的LIDAR方程,可以得到不同散射体的后向散射截面,由于后向散射截面解算时很难获取全部的参数,因此,选择一类散射体作为参照散射体,定义了一种便于解算的相对后向散射截面。考虑到激光脉冲照射区有一定的范围,将单位照射面积上的相对后向散射截面定义为相对后向散射率,同时还定义了立体散射体和复杂散射体的等效相对后向散射截面和等效相对后向散射率。(4)基于不同散射体的波形特征量,给出了波形数据点云化方法波形特征量包括LIDAR到散射体的距离、散射体的深度等信息,结合LIDAR的位置矢量和发射脉冲方向矢量,可以得到散射体的位置。波形数据转成点云数据以后,仅保留了不同散射体的一些特征量信息,可以有效减少相关分析中不需要的冗余信息量。(5)在点云数据预处理过程中,优化了数据处理流程和相关的数据处理算法。根据激光脉冲的水平空间采样特征,提出了一种采样密度计算方法;点云数据栅格化时,根据激光脉冲采样密度,提出了一种像元尺寸的确定准则,即像元尺寸为平均点间隔的1/2,在尽可能保留采样点信息的同时,尽可能地减小信息冗余量;数字表面模型(Digital Surface Model, DSM)栅格数据内插零值像元时,提出了一种邻域内插算法,用于分析LIDAR数据存在的孔洞特征;冠层高度模型(Canopy Height Model, CHM)栅格数据平滑时,提出了一种邻域平滑算法,用于平滑树冠表面的凹入点。(6)根据CHM描述的单木树冠特征,提出了双正切角树冠边界识别算法。单木树冠特征包括树冠顶点和树冠边界等,识别树冠顶点时,采用了局部最大值搜索算法,包括固定窗口法和可变窗口法。识别树冠边界时,提出了双正切角树冠边界识别算法,包括常量法和相关法;对于连续树冠,该算法采用了等比例判别规则,按照树高之间的比例关系划分连续树冠之间的边界;该算法还采用了不相交集判别规则区分不同树冠,树冠边界矢量化采用了四方向法。(7)根据识别的单木树冠特征,找到了最优的单木参数估测方法。由单木树冠特征能够直接估测的单木参数包括树高、冠幅、枝下高等,树高采用树冠顶点位置处的高度值,根据树冠边界计算得到冠幅,计算方法包括主方向法和面积法,枝下高采用树冠边界最低点处的高度值。结果表明单木树高的估测精度最高,其次是单木冠幅,单木枝下高的估测精度最低。根据相关生长方程间接估测的单木参数包括胸径、生物量等,建立胸径的相关生长方程时,对直接估测参数和实测胸径进行回归分析,包括对参数取自然对数以后的回归分析,结果表明实测胸径自然对数与估测树高自然对数、估测冠幅自然对数的线性回归方程最优。选择现有的单木生物量的相关生长方程,由估测参数计算得到估测生物量。(8)根据LIDAR估测的单木参数,找出了最优的林分参数估测方法。由单木估测参数能够直接估测的林分参数包括林分平均高、株数密度等,研究结果表明树冠面积加权的估测林分平均高与胸高断面积加权的实测林分平均高之间的相关性最优;株数密度的估测精度受林木分布特征的影响较大,上层林木株数与下层林木株数之间的比例关系对估测株数密度的变异具有很大的贡献。根据单木估测参数间接估测的林分参数包括胸高断面积、林分生物量等,结果表明在单木参数和相关生长方程估测精度一定的情况下,株数密度对胸高断面积和林分生物量的估测精度的影响较大。总之,通过以上研究发现,高采样密度机载LIDAR能够详细描述森林冠层的三维结构特征,通过一定的数据处理流程和相关的数据处理算法,可以精确地识别单木树冠特征,并用于估测相关的单木参数和林分参数。更多还原

【Abstract】 Forests are the integrity of multi-resource and multi-function. It is very important to take the reins of developing rule of forest growth and consumption. Traditional methods of ground survey of forest resources can obtain detailed data, but require longer time. With the emerging and fast developing of remote sensing technology, it is possible to quickly and precisely acquire large area data of forest inventories. Lighting detection and ranging (LIDAR) is an active remote sensing, which can precisely acquire 3D feature information of earth objects. The main objective of this dissertation is to reveal the LIDAR detection principle of 3D structure of forest canopy and seek the effective estimation methods of forest parameters. Specially, the research topics include:(1) Established the solid scatterer model according to the back scattering feature of earth objects.It was found that the current LIDAR equation had limitation as explaining extended returned waveforms by analyzing interactive relations between laser pulses and forest canopy,. Thus, the solid scatterer model was established on the basis. The scatterers were classified as three types of simple, solid and complex. LIDAR equation of simple scatterer is kept as the same. LIDAR equation of solid scatterer extended the existing convolution function, which was defined as the extended convolution function, to explain back scattering feature of laser pulse for the solid scatterer. LIDAR equation of complex scatterer could be expressed as simple addition of two equations of the other scatterers.(2) Developed an analysis method of waveform feature in term of the waveform features of different scatterers.On the basis of analyzing transmitted and returned waveforms of laser pulses of different scatterers, I discovered that the distributions of returned energy had different properties. In the other words, it was easy to judge the scatterer types and extract the relative value of waveform features according to these properties. For simple scatterer, the distance between LIDAR and scatterer could be computed from value of waveform feature. For solid scatterer, distances between LIDAR and different position of scatterer could be obtained; also, including the depth of scatterer. For complex scatterer, distances between LIDAR and different position of scatterer (including sub-scatterers) could be obtained; depth of scatterer (including solid sub-scatterers) could be also obtained.(3) Proposed an analysis method of relative back scattering cross section refer to the back scattering cross section of different scatterersThe back scattering cross section of different scatterers could be deduced from the corresponding LIDAR equations. It was difficulty to obtain all the parameters as resolving the back scattering cross section. Such that, the paper defined the relative back scattering cross section that was easy to be resolved by selecting a kind of scatterer as the reference scatterer. Considering the illumination area extent of laser pulse, relative back scattering ratio was defined as relative back scattering cross section of per unit area. In addition, equivalent relative back scattering cross section and equivalent relative back scattering ratio of solid and complex scatterer were defined.(4) Presented a transformation method from waveform to point cloud in tern of the feature values of waveforms of different scatterersThe feature values of waveforms include the distances between LIDAR and scatterers, the depths of scatters et al. The location of scatters could be calculated in combined with the location vector of LIDAR and direction vector of transmitted pulse. The cloud point data that was transformed from waveform data only contained the feature value of different scatterers, which could effectively reduce redundancy information that was not needed for the relative analysis.(5) Optimized the flow and relative algorithms of data processing in the preprocessing of cloud point dataAccording to sampling feature of horizontal space of laser pulse, the paper proposed a method of calculating sample density. According to sample density of laser pulse, proposed a judgment rule of pixel size, which was 1/2 of average point space, for rasterizing cloud point data. This would keep sample point information while reduce redundancy information in the whole way. In order to interpolate the zero pixels of digital surface model (DSM) raster data, a neighbor interpolation algorithm were developed, which used for feature analysis of the hole in LIDAR data. In order to smooth canopy height model (CHM) raster data, an effective algorithm was also developed for smoothing concave point of crown surface.(6) Discovered a double tangent crown edge recognition (DTCER) algorithm adapt to the crown features of individual tree described by CHM.The crown features of individual tree include crown top, edge et al. The local maximum search algorithm with fixed or variable window was used to recognize crown tops; the DTCER algorithm with both constant and relative running mode was developed to recognize crown edges. In case of continuous crowns, the DTCER algorithm introduced judgment rule of equal proportion, which was that the boundaries between continuous crowns were divided by proportion of corresponding tree heights. In addition, the DTCER algorithm adopted judgment rule of disjoint sets to partition different crowns. The crown edge vectorization made use of one four directions algorithm.(7) Found the optimal estimation methods of parameters of individual tree according to the recognized crown features of individual tree.The parameters that could be directly estimated from the recognized crown feature of individual tree included the tree height, crown diameter, and crown base height (CBH) et al. The tree height was estimated from the detected height value at the crown top position. The crown diameter was calculated from crown edge, for which main directions algorithm and area algorithm have been developed. The CBH was extracted from the height value at the lowest position of crown edge. The result shows that the estimation accuracy of tree heights of individual trees is the highest, the lower is that of crown diameters of individual trees, and the lowest is the crown base heights (CBHs) of individual trees. The indirect parameters of individual tree were consisted of diameter at breast height (DBH), and biomass et al., which were estimated by the allometric growth equation. In order to establish the allometric growth equation, regression analysis was performed between directly estimated parameters and field-measured diameters at breast height (DBHs), including regression analysis after natural logarithm operation of parameters. The result shows that the linear regression equation is the most optimal between the natural logarithm operation of field-measured DBHs and the natural logarithm operation of estimated tree heights and estimated crown diameters. The estimated biomass could be calculated from estimated parameters by already existed allometric growth equation of biomass.(8) Sought the effective estimation methods of stand parameters through the estimated parameters of individual tree.The direct parameters of stand included average height of stand, stem density et al., which were directly estimated from the estimated parameters of individual tree. The result shows that the most optimal relation is that between estimated average heights of stand weighted by crown areas and field-measured average heights of stand weighted by DBHs. The estimated accuracy of stem density is strongly influenced by the distribution feature of trees. The proportion of stem numbers between overstory and understory would contribute to variation of estimated stem density. The indirect parameters of stand included basal areas, stand biomass et al., which were directly estimated from the estimated parameters of individual tree. The result shows that the estimated accuracies of basal areas and stand biomass are easily influenced by stem density in case that the estimated accuracies of individual tree parameters and allometric growth equation are given.In a word, it concludes that the airborne LIDAR with high sample density could describe 3D structure feature of forest canopy in detail; the crown feature of individual trees could be precisely recognized by definite flow and relevant algorithm of LIDAR data processing, which can also be used for estimating relative parameters estimation of individual trees and forest stands.更多还原