【作者】 胡俊；

【导师】 朱建军； 李志伟； 丁晓利；

【作者基本信息】 中南大学 ， 测绘科学与技术， 2013， 博士

【摘要】 合成孔径雷达干涉测量(InSAR)是二十世纪六十年代发展起来一项空间大地测量与现代遥感技术,可以大范围、高精度和空间连续地监测地球表面发生的形变,为人类了解地球运动机制、防御自然及人为地质灾害等提供了重要的手段。然而,传统的单轨InSAR形变测量结果只能反映真实的三维地表形变在雷达视线方向(LOS)上的一维投影,容易引起对地表形变监测信息的误判或漏判,使得该技术的发展和应用受到了极大的阻碍。通过融合多颗SAR卫星平台获取的多个轨道数据,可以将InSAR一维形变测量拓展至三维。但是现有方法普遍缺乏严密的数据处理理论,在处理多平台、多轨道和多时域InSAR测量值时,无法充分考虑这些多源异质数据在时间尺度和测量精度上的差异,而且不能很好地顾及InSAR多源误差的影响,从而极大的限制了三维形变测量的精度和时间分辨率。现代测量平差是一种以误差理论和经典测量平差为基础的数据处理技术,并在不同层面上扩充、发展形成新理论和新方法,目前已被广泛应用于“3S”及其集成的数据处理中。InSAR三维形变测量本质上可以视为一个平差问题,因此现代测量平差理论和方法在该领域中具有良好的应用潜力。本文以InSAR三维形变估计的函数模型、随机模型和估计准则为核心研究内容。针对InSAR三维形变估计理论中的关键技术问题,借鉴和研究了滤波、插值、最小二乘、方差分量估计、卡尔曼滤波和总体最小二乘等现代测量平差方法的原理和思想,系统性地研究了基于现代测量平差的InSAR三维形变估计理论与方法,为高精度三维地表形变的获取提供了新的途径。本文的主要贡献和创新之处在于：(1)建立了基于最小二乘的InSAR三维地表形变估计的函数模型,并针对模型的观测值设计了InSAR趋势误差改正算法,显著改善了InSAR三维形变估计结果的精度。本文深入分析了从InSAR的LOS方向一维测量值求解三维地表形变的可能性和精度,系统地研究了多源异质InSAR (包括D-InSAR、Offset-Tracking和MAI)测量值与三维地表形变之间的函数关系,并通过理论推导将其扩展为适用于融合多源异质InSAR测量值和GPS测量值估计三维地表形变的函数模型。同时,为了进一步提高MAI技术获取的方位向形变测量值的精度,本文通过研究电离层误差和形变信号在空间上的不同分布规律,发展了基于方向性滤波和插值的MAI电离层趋势误差改正方法；而针对D-InSAR技术在毗邻轨道获取的形变结果的不一致问题,本文利用在毗邻轨道的公共区域提取的连接点作为附加约束,发展了基于整体同步最小二乘的多轨道D-InSAR轨道趋势误差改正模型。在此基础上,本文分别精确获取了2010年新西兰Darfield地震和2011年日本Tohoku-Oki地震的第一个同震三维地表形变场,精度可达到cm级。(2)提出了基于方差分量估计的InS(?)R三维地表形变速率估计算法,实现了InSAR测量值随机模型的后验估计,显著改善了InSAR三维形变速率估计结果的精度。在融合多源异质InSAR及GPS测量值估计三维地表形变速率时,除了需要建立精确的观测函数模型之外,还需要获取InSAR及GPS测量值的随机模型(即方差),然而目前没有一个公认的可以精确确定多源异质InSAR测量值先验方差的方法。针对这一难题,本文首先根据多源异质InSAR及GPS测量噪声的统计特性实现了多源异质InSAR和GPS测量值的最优分组；然后基于线性模型思想建立了最小二乘残差和多源异质InSAR及GPS测量值的单位权中误差之间的函数模型,通过迭代运算得到了多源异质InSAR和GPS测量值的后验方差(或权重)。此外,考虑到在实际中InSAR多余观测量较少,本文在顾及三维形变速率估计精度和计算量的前提下,通过大量实验研究得到了方差分量估计中InSAR测量值的最优配置。模拟和美国南加州地区的真实数据实验结果表明,本文方法的三维形变速率估计精度比现有方法有了明显改善。随后,将该方法应用于青藏高原冬克玛底冰川的监测中,揭示了青藏高原山岳冰川的第一个三维运动速率场,为研究青藏高原的冰水质量迁移和转换提供了一个新的视角。(3)提出了基于卡尔曼滤波的InSAR三维地表形变序列动态估计函数模型,显著提高了InSAR三维地表形变测量结果的时间分辨率和精度。传统的InSAR三维形变测量方法受限于不同平台、不同轨道SAR数据的时态差异,只能对多源异质InSAR测量值进行静态平差求解,获取地表的瞬时三维形变或三维形变速率,而无法估计三维形变序列。本文首先深入分析了融合多平台、多轨道和多时域InSAR测量值求解三维地表形变序列的局限性；然后通过联合时序InSAR测量值的成像几何建立空间联系和引入三维形变速率矢量作为时态约束,建立了三维地表形变的观测方程和状态方程,从而实现了三维形变序列的动态估计。为了提高三维形变序列的估计精度,本文引入方差扩大模型来抑制InSAR测量值中解缠误差等粗差的影响。而在实际数据处理中,利用基于最小二乘的InSAR时序测量值处理方法得到的三维低频形变和地形残差来确定滤波初始状态矢量和协助InSAR干涉相位解缠,从而进一步提高了卡尔曼滤波方程组的解算精度。模拟和美国南加州地区的真实数据实验结果表明,本文方法适用于线性或非线性形变监测,实现了三维地表形变序列的准实时估计,提高了计算效率的同时降低了数据的存储量,可以得到时间分辨率和精度明显优于现有方法的三维形变序列结果。(4)提出了基于总体最小二乘的InSAR三维地表形变估计方法,解决了传统的基于高斯-马尔可夫(GM)模型的估计准则无法考虑DEM误差影响的问题,提高了InSAR三维形变估计结果的精度。在利用InSAR技术监测地表平行位移的三维形变时,通常都借助于DEM数据提供的坡度角信息来降低对InSAR观测量的要求,这使得InSAR三维形变监测模型的系数矩阵被DEM误差所污染。传统的基于GM模型的估计方法(如最小二乘)只能考虑观测值含误差的情况,本文方法则以变量含误差(EIV)模型为核心,在满足观测值误差平方和最小的同时,使得系数矩阵误差平方和也达到最小,因此可以得到更加优化的三维地表形变估值。针对InSAR测量最为常见的配置情况,本文首先设计了两种地表平行位移的InSAR三维形变监测模型,一种是基于升降轨InSAR的LOS向测量值,另一种是基于单轨InS(?)R的LOS向和方位向测量值；然后分别研究了DEM误差(或坡度角方差)在这两种模型中的传播方式,并通过理论推导得到了相应的系数矩阵方差,进而实现了三维地表形变的总体最小二乘估计。模拟和河南大寺坡真实数据实验结果表明,总体最小二乘在估计三维地表形变时比现有方法表现更加稳健,提高了三维形变结果的估计精度。更多还原

【Abstract】 Interferometric Synthetic Aperture Radar (InSAR), a space geodesy and modern remote sensing technique developed since1960s, can monitor the surface displacements with large scale, high precision, and continuous spatial coverage, which provides essential data for the investigation of the locomotory mechanism of the earth and the prevention of the natural and anthropogenetic geologic hazards, etc. However, the displacements acquired by conventional single-track InSAR can only reflect the projection of three-dimensional (3-D) displacements in Line-Of-Sight (LOS) direction, which may induce erroneous interpretation. This greatly hampers the development and application of InSAR. By combining the SAR data acquired from multi-track and multi-sensor, the one-dimensional measurements of InSAR can be extended to3-D. However, there is generally lacks of the rigorous theory of data processing in this field. In the integration of multi-sensor, multi-track and multi-temporal InSAR measurements, the existing methods do not fully consider the discrepancies of the temporal scales and geodesic precisions between these heterogeneous observations, and can not well handle the effects of InSAR multiple inherent errors. This greatly limits the accuracies and temporal resolutions of the derived3-D displacements.Modern surveying adjustment is a geodesy data processing technique based on the error theory and classical surveying adjustment, and has been developed in different aspects to form new theories and methods. Modern surveying adjustment has been widely used in the GPS, GIS and RS (3S) and its integrated data processing. Since the InSAR3-D displacements estimation is inherent an adjustment problem, the theories and methods of modern surveying adjustment are expected to be of great use in this field. The core research contents of the thesis are the function model, stochastic model and estimation criteria of the InSAR3-D displacements estimation. To circumvent the key difficulties in the InSAR3-D displacements estimation, the thesis studies the models and methods of modern surveying adjustment such as filtering, interpolation, least squares, variance component estimation, Kalman filtering, total least squares, etc, and thus presents a systematic research on the methodologies and models of exploiting InSAR measurements to derive3-D displacements based on the modern surveying adjustment. This provides a novel way for accurately estimating3-D surface displacements. The main contributions and innovations of the thesis are as follows:(1) The thesis constructs the function model of using InSAR to estimate3-D displacements based on the least squares (LS) adjustment, and designs the methods of InSAR ramp error correction for the observations of the function model. This greatly improves the accuracies of the InSAR3-D displacement estimations. The thesis analyses in depth the possibility and accuracy of using one-dimensional LOS measurement to estimate3-D surface displacements, and systematically studies the functional relationships between the heterogeneous InSAR (inclusing D-InSAR, Offset-Tracking and MAI) measurements and the3-D surface displacements. These are then extended into the function model of integrating heterogeneous InSAR and GPS measurements to estimate3-D displacements through theoretical derivation. In order to improve the accuracies of the azimuth displacement measurements by MAI, the thesis studies the spatial distributions of the ionospheric error and surface displacement, and develops an ionospheric ramp error correction method based on the directional filtering and interpolation. While for the discontinuity emerged between the D-InSAR measurements of adjacent tracks, the thesis exploits the connect points in the common area between the adjacent tracks as the additional constraint and thus develops a multi-track orbital ramp error correction method based on the Unified Simultaneous Least Squares (USLS) adjustment. On the basis of the developed function model and InSAR ramp error correction methods, the first3-D coseismic displacement fields of the2010Darfield, New Zealand earthquake and2011Tohoku-Oki, Japan earthquake have been mapped with an accuracy of cm level, respectively.(2) A novel InSAR3-D displacement velocity estimation model based on the variance component estimation (VCE) algorithm is proposed in the thesis. It conducts the posterior estimation of the stochastic model of the InSAR measurements. This greatly improves the accuracies of the InSAR3-D displacement velocity estimations. In order to estimate the optimal3-D displacement velocities from the LS adjustment, accurate stochastic model (i.e., variances) of InSAR measurements is also required. However, currently there is no well-developed method to estimate the accurate variances of the InSAR measurements. To slove this problem, the thesis divides the heterogeneous InSAR and GPS measurements into different groups according to their statistical natures, and then builds the function model between the LS residuals and the mean square error of unit weight of the heterogeneous InSAR and GPS measurements on the basis of the linear model. The posterior variances (or weights) of the heterogeneous InSAR and GPS measurements are finally estimated in an iterative procedure. Additionally, considering that abundant InSAR measurements cannot be acquired in practice, the thesis studies the optimal configuration of InSAR measurements in order to obtain the optimum blance between the accuracies of3-D displacement estimations and the burden of calculation. The novel method is validated with simulated data as well as real data acquired over Southern California, USA. The results show that the proposed method is superior to existing methods in terms of the accuracy. The novel method has also been applied to the Dongkemadi Glacier in Qinghai-Tibet Plateau. The first3-D movement velocity field of the alpine glacier is mapped. This provides a new perspective for the investigation of the ice mass transition and transformation of Qinghai-Tibet Plateau.(3) A novel InSAR3-D time series displacements estimation function model based on the Kalman filtering (KF) algorithm is proposed. It improves the temporal resolution and accuracies of InSAR3-D time series displacements. The conventional methods are limited by the temporal discrepancies between different sensors and different tracks SAR data, and can only resolve the3-D instantaneous displacements or3-D displacement velocities through the integration of the heterogeneous InSAR measurements. There is no method for InSAR 3-D time series displacement estimations. The thesis first analyzes in depth the limitation of using multi-sensor, multi-track and multi-temporal InSAR measurements to estimate3-D displacement evolutions, and then builds the observation and state functions for the3-D time series displacements estimation by exploiting the spatial-temporal correlations of the acquired InSAR measurements as constraints. They can be used to derive the dynamical estimations of3-D displacement. In order to improve the accuracies of the3-D time series displacement estimations, a variance inflation model is applied to suppress the gross errors such as unwrapping errors. In the real data processing, the3-D low-pass displacements and topographric residuals estimated by the InSAR time series analysising method based on LS are used to determine the filtering initials and to assist the InSAR phase unwrapping. This can further improve the estimating accuracy of the Kalman filtering. In the simulated and real data experiment carried out over Southern California, the novel method is proved that can estimate the3-D displacement evolutions for linear or nolinear ground movements in quasi-real-time, and reduce the computational burden and data storage. It also reveals that the novel method can obtain better3-D time series displacements in terms of temporal resolutions and accuracies than existing methods.(4) A novel InSAR3-D displacement estimation method based on the Total Least Squares (TLS) adjustment is proposed. It can fix the problem induced by the DEM error in conventional adjustment based on the Gauss-Markov (GM) model and thus improve the accuracies of the InSAR3-D displacement estimations. In the investigation of the surface-parallel ground displacements with InSAR, the DEM-derived slopes are needed to relax the requirement for multi-sensor and multi-track InSAR measurements. The error of the derived slope however will propagate into the coefficient matrix of the InSAR3-D displacement observation model. Conventional adjustment (e.g., LS) based on the GM model can only handle the errors in observations. Based on the Errors-In-Variables (EIV) model, the proposed method can minimize the summation of square errors of the observation errors and the coefficient matrix as well, and therefore can yield more accurate3-D displacement estimations. According to the commonly used configurations of InSAR measurements, the thesis designs two InSAR3-D models for surface-parallel displacements, one of which is based on the ascending and descending InSAR LOS measurements and the other based on the single-track InSAR LOS and azimuth measurements. The thesis then studies the propagation of the DEM errors (or the variances of slope angles) in the two models, respectively, and thus yields the variances of the coefficient matrix by theoretical derivation. The TLS estimations of the3-D surface displacements are then obtained. The results from the simulated and real data experiment carried out over Dasi Slope in Henan province, China show that the TLS adjustment is more robust in3-D displacement estimations than existing methods, and can improve the accuracies of the final results.更多还原