【作者】 李文梅；

【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】 随着遥感技术在林业应用领域的日趋发展，合成孔径雷达（Synthetic Aperture Radar，SAR）以其独特的优势在森林资源分类和生物参数提取研究中取得了许多重要的研究成果。近年来，极化SAR（Polarimetric Synthetic Aperture Radar，PolSAR）、干涉SAR（Interferometric Synthetic Aperture Radar，InSAR）、极化干涉SAR（PolarimetricInterferometric Synthetic Aperture Radar，Pol-InSAR）和层析SAR（Tomography SyntheticAperture Radar，TomoSAR）已经成为森林散射机制分离、林下地形提取、树高估测、森林地上生物量（Above Ground Biomass，AGB）估测和后向散射功率垂直分布信息提取等研究的重要手段。但目前对森林资源信息定量提取的研究主要集中在基于后向散射强度信息、InSAR相干系数和Pol-InSAR的森林平均树高、单位面积蓄积量和地上生物量的反演，而充分利用基于TomoSAR提取的森林垂直方向雷达反射率垂直分布信息的研究较少。而森林垂直方向雷达反射率垂直分布信息是雷达对森林内部各种散射体垂直分布的响应，有助于理解复杂垂直结构森林的雷达散射机理，并使森林三维结构信息反演成为可能，也有利于提高森林平均树高、蓄积量、AGB等森林结构参数的估测精度。TomoSAR是提取森林垂直结构信息的主要技术，主要包含极化相干层析（PolarizationCoherence Tomography，PCT），多基线InSAR层析（Multi-baseline InSAR Tomography，MBInSAR Tomo）和多基线Pol-InSAR层析（Multi-baseline Pol-InSAR Tomography，MBPol-InSAR Tomo）三种主要的层析方法。为此，本文开展了极化干涉SAR层析估测森林垂直结构参数的方法研究，探讨、分析三种层析技术提取森林垂直结构信息的有效性，主要内容如下：（1）单基线PCT提取森林垂直结构信息基于德国Traunstein研究区E-SAR2003年航飞试验获取的L-波段单基线Pol-InSAR数据和地面测量数据，提出了一种基于对象的PCT技术来估测森林覆盖区AGB的新方法。在罗环敏等研究的基础上，进一步完善了森林AGB反演模型，应用Pol-InSAR分割与面向对象分割技术将森林覆盖区划分为均质多边形，进而将基于实测林分建立估测模型推广到实测林分之外的森林覆盖区。从垂直结构剖面提取出的参数对森林AGB估测最为重要的是林分层析测量高，林分层析测量高对应于冠层相对反射率最大的高度，能够在一定程度上表征森林AGB。研究发现，林分尺度垂直结构剖面与森林AGB相关，这一研究结果与Luo(2011)和Cloude(2009)观测结果一致，并且应用剖面参数与地面实测林分AGB通过后向逐步回归分析建立的估测模型的决定系数R2为0.883，均方根误差（RootMean Square Error，RMSE）为39.98tons/hm2，相对RMSE（Relative RMSE，RRMSE）为13.15%。研究结果表明，虽然PCT提取的林分层析测量高比采用经典三阶段反演方法提取树高精度较低，但森林AGB提取精度较高。PCT剖面提取出的参数总体上能够描述垂直结构剖面的几何特征，能够用于建立森林AGB反演模型。对垂直结构剖面进行参数化，应用多元分析方法建立这些参数与实测生物量之间的关系是可行的，并能够推广到林分边界周围区域。值得注意的是，该方法估测的森林AGB没有出现饱和现象，即使生物量达到500tons/hm2，相关关系依然成立。（2） MBInSAR Tomo提取森林垂直结构信息应用瑞典Raminstorp研究区E-SAR于2007年3月-5月采用重轨飞行模式获取的机载L-，P-波段BioSAR多基线InSAR数据，提取了后向散射功率垂直分布信息，分析了后向散射功率估测森林AGB的可行性，以及基线数量、时间基线和波长对森林垂直结构信息提取的影响。研究发现，HH极化MBInSAR Tomo估测树高与Lidar H80相比R2为0.65，RMSE为2.35m，相关系数为0.80。HV极化MBInSAR Tomo估测树高与Lidar H80相比R2为0.55，RMSE为3.27m，相关系数为0.74。VV极化MBInSAR Tomo估测树高与Lidar H80相比R2为0.34，RMSE为5.13m，相关系数为0.58。研究结果表明，Capon方法在高噪声背景条件下能够准确地获取目标信号信息，但噪声抑制能力稍差，适于提取森林垂直结构及冠层高度信息；基线数量、时间基线和波长均对森林垂直结构信息提取具有一定的影响：对于P-波段而言，3条航线即可提取森林垂直结构信息，但基线数量越多所能提取的森林垂直结构信息越丰富，并且31天的时间基线对P-波段而言影响不大，所得后向散射功率分布较为分散，较短时间基线所得后向散射功率集中分布于近地表；L-波段所得后向散射功率谱随机性较强，可在一定程度上表征垂直结构变化，但地表及冠层识别能力较差，P-波段则更稳定，能够有效识别地表、冠层边界，有利于森林垂直结构信息综合提取；P-波段HH极化MBInSAR Tomo技术提取的树高精度能够在一定程度上满足林业应用需求；P-波段HH/HV/VV极化MBInSAR Tomo提取的某些特定高度处的后向散射功率对针叶林区森林AGB的提取贡献有限。（3） MBPol-InSAR Tomo提取森林垂直结构信息同样，应用瑞典Raminstorp研究区BioSAR2007机载L-，P-波段多基线Pol-InSAR数据，采用MBPol-InSAR Tomo技术提取后向散射功率垂直分布信息和极化角，估测树高，分析应用后向散射功率估测森林AGB的可行性。其中，极化角为Cloude-Pottier极化SAR分解中代表散射机制的物理参数。同时，分析了基线数量、时间基线和波长对MBPol-InSAR Tomo提取森林垂直结构信息的影响。研究发现，MBPol-InSAR Tomo估测的树高与Lidar H80相比，R2为0.53，R为0.73，RMSE为4.08m，MBPol-InSAR Tomo提取的某些特定高度处的后向散射功率与实测森林AGB没有明显的相关性。研究结果表明，与MBInSAR Tomo类似，基线数量、时间基线和波长均对MBPol-InSAR Tomo提取森林垂直结构信息具有一定的影响。P-波段MBPol-InSAR Tomo技术提取的树高精度与MBInSAR Tomo提取的树高相比没有明显提高，即极化信息的引入对树高估测没有明显作用，同时，极化信息的引入对针叶林区森林AGB的估测基本没有贡献。极化角是MBPol-InSAR Tomo与MBInSAR Tomo相比新增的反演量，能够表征后向散射机制在垂直方向的变化规律。更多还原

【Abstract】 With the development of forest remote sensing technology, Synthetic Aperture Radar(SAR) obtains a lot of important research results in forest resource classification and biologicalparameters extraction with its unique advantage. For the past few years, Polarimetric SyntheticAperture Readar (PolSAR), Interferometric Synthetic Aperture Radar (InSAR), PolarimetricInterferometric Synthetic Aperture Radar (Pol-InSAR) and Tomographic Synthetic ApertureRadar (TomSAR) have been the important means for forest scattering mechnicham separation,understory topography extraction, forest height estimation, forest above ground biomass (AGB)estimation and backscattering power profile extraction and so on. However, forest researchmainly concentrates on forest height, volume and forest AGB estimation based onbackscattering coefficients, InSAR coherence and Pol-InSAR, there is few studies on verticaldistribution of radar backscattered power using TomoSAR. Vertical distribution of radarbackscattered power in forest is the response of vertical distribution of scatterers, and it ishelpful for understanding radar scattering mechanism of complex forest vertical structrue. Itmake possible to inverse forest3-dimension(3D) information, and it helps to improve theestimation accuracy of forest height, volume, forest AGB and other forest structure parameters.TomSAR is the major technique for forest vertical structure information extraction, it containsthree tomographic approach: Polarization Coherence Tomogragphy (PCT), Multi-baselineInSAR Tomography (MBInSAR Tomo) and Multi-baseline Pol-InSAR Tomography(MBPol-InSAR Tomo). Hence, we used these three tomographic SAR techniques to extractvertical distributions of backscattering power and analyze the applications of these information.There are three main research contents:(1) Forest vertical structure information extraction using single baseline PCTWe proposed a new method for forest AGB estimaion using object-based PCT techniqueby applying L-band single baseline Pol-InSAR data, and the data was acquired by E-SAR in Traunstein test site in Germany in2003. Based on the study of Luo (Luo, Chen et al.2011), wecontinued to improve the forest AGB estimation model, applied Pol-InSAR separation andobject-orientated segmentation techniques to make the forest coverage areas divide into severalhomogenous polygons, and extended the forest AGB estimation model to the polygons.Tomography canopy height (TomH) is the most important parameter for forest AGB estimaion,and all the parameters were extracted from vertical structure profile. TomH corresponds to thelargest relative reflectivity in canopy, and it co uld represent forest AGB in some degree. It isdiscovered that vertical structure profile in forest stand level relates with forest AGB or volume,and this result is agreed with that of Luo(2011) and Cloude(2009). Backward step-wiseregression method was used to establish estmation model using profile parameters and in-situforest AGB in forest stand scale. The coefficient of determination R2is0.883, Root MeanSquare Error (RMSE) is39.98tons/hm2, and relative RMSE (RRMSE) is13.15%.The results indicate that TomH extracted from profile produced by PCT has less accuracythan forest height estimated using classical three-stage inversion approach, but it has higherprecision in forest AGB estimation. Parameters extracted from vertical profiles could describethe overall geometric characteristics of vertical structure profile and some of them were used tobuild forest AGB inversion model. It is feasible to parametrize vertical structure profile andused these parameters to estimate forest AGB by multivariant analysis method and the modelcould be extended to other regions outside of the measured forest stands. It is worth noting thatestimated forest AGB does not saturate using the approach we proposed. The estimation modelstill valid even the biomass up to500tons/hm2.(2) Forest vertical structure information extraction by MBInSAR TomoThe airborne L-and P-band multi-baseline InSAR data was applied to extract verticaldistribution of backscattering power, and the data was acquired by E-SAR in Raminstorp testtest in Swenden from March to May in2007BioSAR campaign with repeat track mode. Forestheight was estimated using empirical method, the feasibility of forest AGB estimation usingbackscattering power was analyzed, and the impacts of baseline numbers, temporal baseline and wavelength on forest vertical structure information extraction were analyzed. It isdiscovered that R2is0.65,0.55,0.34, RMSE is2.35m,3.27m,5.13m and correlationcoeddicient (R) is0.80,0.74,0.58between the estimated forest height using HH, HV andpolarimetric MBInSAR Tomo and Lidar H80, respectively.The results show that Capon method could accurately obtain the target singal in high noiseenvironment, but is has less noise immunity, and it is suitable for forest vertical structure andcanopy height information extraction. Baseline numbers, temporal baseline and wavelength hassome influences on forest vertical structure extraction, respectively. For the P-band data, morebaselines extracte more forest vertical structure information and31days temporal baseline haslittle effect on P-band data and the distribution of backscattered power scattered, meanwhilebackscatterd power obtained using shorter temporal baseline concentrates on ground. Thebackscattering power spectrum extracted using L-band data has strong randomness, it coulddescribe the variant of vertical structure but could not effectively identify the surface andcanopy, and that extracted from P-band data is more stable, it could distinguish surface andcanopy features, which aids to forest vertical structure extraction. The precision of forest heightestimated using P-band HH polarimetric MBInSAR Tomo is able to meet the demand of forestapplication in some degree. However, extracted backscattered power at certain heighte ofP-band HH/HV/VV MBInSAR Tomo contributes little to forest AGB estimation in coniferousforest.(3) Forest vertical structure information extraction by MBPol-InSAR TomoThe data is the same with that of (2) besides it contains polarimetric information, that ismean multi-baseline Pol-InSAR data was used. MBPol-InSAR Tomo technique was applied toextract vertical distribution of backscattering power, polarization angle, estimate forestheight and analyze the feasibility of backscattering power for forest AGB estimation.Meanwhile, the effects of baseline number, temporal baseline and wavelength on forest verticalstructure information extraction using MBPol-InSAR Tomo were analyzed. It is discoveredthat, R2is0.53, R is0.73, and RMSE is4.08m between the esimated forest height using MBPol-InSAR Tomo and Lidar H80. But the backscattering power at certain height contributeslittle to forest AGB estimation.The results show that similar with MBInSAR Tomo, baseline number, temporal baselineand wavelength have effects on forest vertical structure information extraction usingMBPol-InSAR Tomo technique. The precision of forest height estimated using MBPol-InSARTomo is not better than that from MBInSAR Tomo. Meanwhile, the introduction ofpolarization information basically has no contribution on forest AGB estimation in coniferousforest. Polarization angle is the new inverted parameters for MBPol-InSAR Tomocompared with MBInSAR Tomo, it could describe vertical variations of backscatteringmechanism in forest.更多还原