【作者】 徐光彩；

【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， 2013， 博士

【摘要】 森林垂直结构是陆地生态系统中重要的参数，提高遥感森林垂直结构的反演精度，对于森林资源监测、全球气候变化及其区域响应研究具有重要意义。激光雷达技术是近年来国际上发展十分迅速的主动遥感技术，在森林参数的定量测量和反演上取得了成功的应用，特别是对森林高度和垂直结构的探测能力，具有传统光学遥感数据难以比拟的优势。森林叶面积指数和生物量是森林生态系统的重要参数，其精确的估算具有十分重要的意义，本文围绕上述内容，利用机载小光斑波形激光雷达数据开展了以下几个方面的研究工作：（1）根据激光雷达波形数据特征，建立高斯分解算法和相对辐射校正模型。波形数据直接使用较为不便，需要对波形数据进行进一步处理，采用了非线性最小二乘法对波形数据进行高斯拟合处理，详细描述了波形分解的工作流程和重要的处理步骤。波形数据进行数据分类是其重要的优势之一，然而使用未经标定的数据进行分类往往存在一定的问题，因此在高斯分解的基础上，对分解结果数据基于归一化的距离和能量进行相对辐射校正处理，增强波形数据间的可比性，以期提高分类结果的准确性和精度。最后对分解结果和相对辐射定标结果进行了定量分析。（2）使用波形能量数据反演森林叶面积指数。在波形数据分解基础上，利用栅格化的数据对研究区进行分类，提取研究区的森林区域。根据基于间隙率原理的比尔朗伯定律提出了利用波形数据能量反演叶面积指数的方法。详细介绍了波形数据的分类、数据的归一化、利用波形能量数据反演LAI的方法、反演最佳尺度的确定以及森林叶面积指数的制图。结果表明，利用波形数据能够有效的反演森林的叶面积指数。（3）结合单木分割和波形特征参数进行树种识别。在波形数据进行高斯分解点云化的基础上，根据高程信息生成的DEM和DSM，并生成CHM。首先对CHM进行无效值填充，再采用结合形态学控制的分水岭算法分离CHM上的单木并提取单木的相关参数。根据单木的范围提取波形高斯分解后结果参数的统计值作为该株树的特征值，结合外业调查的数据采用SVM分类器对样地的树种进行识别和精度评价。结果显示，使用定标后的数据对研究区7个树种的识别精度达到55.07%，5个树种则达到了66.15%，均要高于未定标数据的分类精度。（4）利用波形数据反演单木生物量首先对外业获取的实测树高和冠幅与实测的胸径数据进行回归分析，建立树高和冠幅估算胸径的估测模型，结合相关生物量方程构建基于树高和冠幅的二元生物量方程。结果表明使用树高和冠幅数据能够很好的反演森林的胸径信息。在树种识别的基础上，利用单木分离提取的树高和冠幅数据结合二元生物量方程反演单木生物量。本研究工作建立了使用波形数据反演森林叶面积指数和单木生物量的完整技术流程，结果表明高密度的机载小光斑波形激光雷达能够详细的描述森林垂直结构信息，快速准确实现森林叶面积指数和单木生物量反演。更多还原

【Abstract】 The vertical structure of the forest is one of the most important parameter in terrestrialecosystem. It is of great significance for the monitoring forestry resources and global climatechange by improving its retrieval accuracy of the remote sensing. LiDAR (lighting detectionand ranging)is an rapidly developing active technology of the international remote sensing inin recent years. Especially for forest height and vertical structure detection, it has a specialadvantage comparing with traditional optical remote sensing data. Forest leaf area index andbiomass are two important parameters of the forestry ecosystem and their accurate estimateshave great significance. The main works and results are as follows:(1) Creating a Gaussian decomposition algorithm and relative radiometric calibrationmodel according to the characteristics of lidar waveform data.Full-waveform data needs for further processing because it is relatively inconvenient todirectly use. A non-linear least-squares method with the Levenberg-Marquardt algorithm wasused to fit the return waveforms by Gaussian function and Gaussian amplitude, standarddeviation and energy were extracted. Generally, different objects response to the emitted pulsediversely, which is incarnated in the waveform data. But acquired data is influenced by severalfactors, so it cannot be directly used in wide area before calibration. A relative calibrationmethod using the range between the sensor and target based on a radar equation was applied tocalibrate the amplitude and energy, and the change of transmit pulses energy was alsoconsidered in this process, which is to enhance the comparability of waveform data and toimprove the accuracy and precision of the classification results. Finally, a quantitative analysison the decomposition and relative radiometric calibration results was applied.(2) The inversion of forest LAI using the decomposition energy of full-waveform LiDARdata.Decomposed full-waveform result data was rasterized to classify the study area and theforest area was extracted. A method based on the Bill-Lambert law was proposed by using waveform data energy to esttimation Leaf Area Index. Some detailed information weredescribed including waveform data classification, data normalization,the principle of LAIestimation using full-waveform data,the best scale of LAI inversion and LAI mapping of thestudy area. The result showed that full-waveform data could effectivly estimate forest LAI.(3) Combining single tree segementation and parameters extrated from full-waveform datato identify the tree species.DEM, DSM and CHM were generated form the point groun basingon on the Gaussianwaveform decomposition. Then the morphology-controlled watershed algorithm was adoptedto separate single tree on the CHM filled with invalid value. The individual tree positions andtree crowns were acquired. The total number of return waveforms within a beam, the pulsewidth, the calibrated amplitude and energy in single tree bounds are counted as tree features todetect seven tree species by a SVM classifier. The overall classification accuracy for this studyarea was55.07%using calibrated data for seven tree species, which is5.1%higher than that ofadopting uncalibrated data. Limiting to the five main species accuracy was improved to66.15%and to conifers and broadleaved trees accuracy was85.72%using calibrated data,which are also5.75%and3.56%higher than that of using uncalibrated data. Calibration offull-waveform data is necessary for its application in tree species classification.(4) Biomass estimation of individual trees using full-waveform LiDAR data.First,regression analysis of measured tree height and crown diameter with the diameter atbreast height (DBH) was performed. The result showed that using the linear regressionequation could fit the DBH well. The collected allometric equations relating biomass could beexpressed by the tree height and crown diameter. Then combing the single tree speciesidentification result and the tree hight and crown diameter extracted from indivudual treesegementation, the biomass could be estimated with the equation.In a word, the entire workflow of forest LAI and single tree biomass inversion wasestablished in this article. The results showed that the high-density small-footprintfull-waveform LiDAR data could The results show that the high-density airborne small spot waveform lidar could get the detaile forest vertical structure information and estimate forestleaf area index and single wood biomass fast and accurately.更多还原