大地电磁测深法(MT)是一种基于天然源的地球物理电磁测深法，它是地球深部构造研究的最重要的方法之一。当前三维大地电磁测深法在资源勘探、地质调查和浅地表至上地幔深度地质构造研究中获得了广泛应用。传统的三维大地电磁测深法正反演算法是基于各向同性理论，且采用规则的网格进行区域离散难以适应地形和复杂的构造。然而，越来越多的大地电磁测深法数据表明地下介质的电各向异性是广泛存在的，尤其是在深部构造中。采用各向同性理论进行反演和解释会造成解释的困难和错误。只有基于各向异性理论的反演和解释才能给出合理的结果。实际勘探研究中，地形和地下构造非常复杂，传统基于规则网格的正反演方法模拟和反演复杂地形和地下构造的能力有限，难以满足实际勘探需要，限制了其实用性。当前在三维复杂地形和地质条件下，大地电磁测深法响应正演模拟、大地电磁测深法各向异性响应特征识别及大地电磁测深法各向异性数据高效反演方法和处理策略都是亟待研究和解决的问题。为了解决以上问题，本文开展了基于非结构有限元的三维大地电磁法各向异性正反演研究。 三维大地电磁测深法各向异性正演是三维各向异性反演的基础。高效的正演算法可以提高三维大地电磁测深法的计算速度和精度，是构造高效三维反演最重要的因素。在三维大地电磁测深法模拟中，为了得到高精度的数值解，需要对高度复杂的地质构造和界面进行精确建模。然而，传统的规则网格不能实现高精度地模拟任意复杂地质体和界面。最有效的解决办法就是使用非结构网格。非结构四面体网格可以灵活地构建任意的地形和地质构造，且可以灵活地实现局部优化。非结构四面体网格单元相比结构网格可以使用较少的单元拟合任意复杂的地形和地下构造。有限元法是一种适应非结构四面体网格的高效数值算法。本文采用非结构四面体单元有限元法系统地开发了各向异性介质条件下的三维大地电磁正演算法，该算法可以用于直接进行带地形的三维大地电磁测深法数据反演，克服了结构网格计算精度有限和难以拟合地形和地下复杂异常体的缺陷。 在任意各向异性数值模拟部分，本文首先分析了在水平地形条件下不同各向异性参数对MT响应的影响特征，总结了根据视电阻率识别各向异性信息的方法。然后，针对在地形存在条件下根据常规的视电阻率响应难以识别地下介质各向异性信息的问题，本文根据各向异性响应的极性特征，提出使用极坐标下的视电阻率来识别地下介质的各向异性信息的方法，并证明了其在复杂地形存在的条件下仍然具有适用性。 为了克服人工网格剖分的困难，尤其是地形和复杂构造和各向异性同时存在的地质情况，本文开发了基于加权后验误差的面向目标网格自适应算法，该方法能得到高质量的网格，从而大大提高有限元正演的计算精度。同时，本文分析了针对各向异性介质的网格自适应特征，进而研究了主轴电阻率对网格的自适应优化的影响规律，结果表明水平方向最小主轴电阻率决定了网格的自适应的疏密程度。 高内存需求和低计算效率是当前三维电磁反演实际应用中面临的主要难题。针对三维大地电磁测深法反演问题，本文开发了基于有限内存拟牛顿法(L-BFGS)的反演算法。该算法在反演迭代过程中只需要求取目标函数的梯度，避免了求取海森矩阵，可以有效地降低内存需求。本文研究了基于非结构网格和各向异性条件下的目标函数构建及其梯度的求取方法，分别实现了各向同性和各向异性条件下的三维大地电测深法反演算法。通过带地形的和实际地质背景的三维大地电磁测深反演算例表明采用非结构网格的有限元法进行反演，可以避免复杂的地形校正过程，实现直接带地形反演，且可以建立有效的先验模型。利用本文算法对美国EarthScope项目的USArray的部分长周期数据进行各向同性反演，并和基于规则网格算法的WSINV3D程序的三维反演结果进行对比，验证了本文反演算法的有效性和适用性，同时表明本文基于非结构网格的算法可以恢复更多的细节信息。 鉴于地下介质存在各向异性时，各向同性反演不能恢复真实模型的分布规律和电性特征，会造成反演解释的困难和错误。在三维大地电磁各向异性算法的基础上，本文提出了一种对三维大地电磁测深法各向异性数据进行反演解释的策略。首先，利用提出的极坐标下各向异性识别方法，定性地分析和识别地下介质各向异性特征。然后，若确定数据包含各向异性信息，采用三维各向异性算法进行反演，否则采用各向同性算法进行反演解释。其次，根据反演出的各向异性比值分布，确定各向异性构造的空间分布。最后，结合地质背景进行三维各向异性解释。通过带地形和不带地形的各向异性反演数值算例表明，本文提出的各向异性识别方法对复杂地形的各向异性数据是有效的，可以用来指导三维大地电磁测深法各向异性的识别和解释，避免了采用传统的各向同性理论容易造成的解释困难和错误；基于非结构网格和L-BFGS的反演算法，适合带地形条件下的各向异性反演；在三维大地电磁测深法各向异性反演中，P〓和P〓为可反演恢复参数，而P〓不可恢复参数。 最后，本文利用三维各向异性反演算法和提出各向异性处理策略对Coprod2大地电测深法数据进行各向异性反演解释，进一步验证了各向异性反演算法和策略的有效性。数据反演结果建立了地下深部各向异性构造的模型，解释了NACP和TOBE的位置，并发现了新的各向异性体。对USArray实测数据三维各向异性研究，证明了Cascadia俯冲带上板块运动的应力方向为各向异性电阻率的良导方向，并建立了Cascadia俯冲带的动力学模型，对研究地壳深部构造应力及运动的研究及各向异性的成因具有一定的科学意义。 本文基于非结构有限元法和有限内存拟牛顿法开展了三维大地电磁测深法各向异性正反演研究。本文实现了高效的任意各向异性介质中三维大地电磁测深法面向目标自适应有限元法正演模拟；提出了地形条件下三维大地电磁各向异性响应特征识别方法；建立了适用于实测数据的三维大地电磁测深法各向异性反演方法及处理策略。本文为三维大地电磁测深法各向异性介质正反演研究奠定了理论基础，为研究地下深部介质的电各向异性提供了技术支持。研究成果能够提高三维大地电磁测深法数据处理和解释的水平，对地球深部构造的研究具有一定的理论和实际意义。 关键词：地球物理电磁测深法，大地电磁测深法，三维正反演，非结构有限元法，面向目标自适应，有限内存拟牛顿法，各向异性介质

 As an effective geophysical tool based on natural source, magnetotelluric (MT) has been widely used in resource exploration, geological survey and the investigation of earth structures from near surface to upper mantle, especially on research of structures in the deep earth. Conventional 3D MT forward modelling and inversion assume an isotropic model and most of these methods are based on structured meshes. However, electrical anisotropy is often found in the deep earth from MT data and isotropic models can lead to misinterpretations. In such cases, one needs to use an anisotropic model for appropriate and realistic interpretation. In addition, the inversion methods with structured meshes have limited capability to handle complex geological structures and topography in realistic situation. The model accuracy is severely influenced by the quality of mesh. Currently, the study on 3D MT anisotropic forward modeling and inversion over complex terrains and geological structures has not been well studied yet. To overcome this obstacle and promote the interpretation of MT data, I developed a 3D MT forward and inversion method based on unstructured finite-element method for anisotropic media. The main content of this thesis includes 3D MT anisotropic modeling, identification of electrical anisotropy from MT data, appropriate strategies for practical anisotropic inversion and interpretation. 3D MT anisotropic modeling is the basis for inversions with anisotropic models. An efficient forward algorithm can improve the calculation speed and accuracy of 3D MT modeling, and is one of the most important factors in constructing 3D MT anisotropic inversion. In MT forward modelling, it is important that the geological structures and interfaces in the numerical domains can be modeled as accurately as possible, which can not be efficiently achieved by the conventional structured grids. Unstructured tetrahedral grids enable representing arbitrary structures more accurately with fewer cells compared to regular structured grids, and they allow efficient local refinement. Fortunately, the finite-element method naturally supports unstructured grids. In this thesis, unstructured tetrahedral grids with finite-element method are used for the 3D forward modeling of MT data, which allows for the direct inversion of MT data with topography. It can overcome the defects of regular structured grids such as high computational cost for improving accuracy and difficulty of inverting with topography. In MT numerical experiment section, under the condition of anisotropic media, the influence of different anisotropic parameters on MT responses is analyzed. According the effect of anisotropy to MT apparent resistivity, I have summarized how to identify anisotropy without topography. However, it doesn't work when obvious topography occurs. To solve the problem, I proposed a technique that use the apparent resistivity in polar coordinates system to analyze the anisotropic features, which proved to be a useful way to identify the anisotropy on condition of topography. When one uses artificially refined grids, the accuracy of forward modeling is severely influenced by the quality of the grids, especially for complex geology and topography. In this thesis, I have developed a goal-oriented adaptive unstructured finite-element method for accurate and efficient 3D MT anisotropic modeling, rendering the hard work of artificially refined grids unnecessary. In addition, the anisotropic effect on the adaptive meshes are analyzed, which shows the fact that the grids refinements depends largely on the most conductive components in the horizontal directions. High memory cost and low computational efficiency have limited the practical application of traditional 3D MT inversion. In this thesis, I have used the limited-memory quasi-Newton (L-BFGS) method to perform 3D MT isotropic and anisotropic inversions. This avoids explicitly calculating the Hessian matrices, and greatly reduces the memory requirements. In the L-BFGS method only the objective function and its gradient need to be calculated. I have constructed the objective functional and its gradient with the unstructured grids for isotropic and anisotropic media, respectively. The isotropic inversion experiments show that my inversion method with unstructured grids can achieve good inversion result without carrying out topography correction. The long period MT data from the USArray component of EarthScope project in the United States has been used to test my inversion algorithm. The results were compared with the published results by the WSINV3D code. The comparison was very positive that verified the validity and applicability of my algorithm. In addition, my inversion recovers more detail about the underground structures compared with the published result with structured mesh, which shows the effectiveness of the unstructured mesh for 3D inversion. The results of isotropic inversion for anisotropic data are far beyond the original models, ignoring anisotropic may leads to misinterpretation. Only the anisotropic model can give a reasonable interpretation. Following the 3D MT isotropic inversion algorithm, I have developed a triaxial anisotropic inversion. After that I have proposed an effective strategy for anisotropic inversion for 3D MT data. Firstly, I use the anisotropy identification method I proposed in polar coordinates system to analyze the MT data, which can identify the anisotropic features if anisotropic structures do exist Secondly, if the data contains anisotropic information, then I do 3D inversions using anisotropic inversion algorithm. Thirdly, the spatial distribution of the anisotropic structure is determined according to the of the distribution anisotropy ratio. Finally, a reasonable 3D anisotropic interpretation in combination with the geological background is carried out I also verify the effectiveness of my strategy and algorithm via two numerical examples with anisotropic bodies embedded in a flat or topography half space. From the numerical experiments I can draw the conclusions. The unstructured finite-element method with L-BFGS algorithm are effective for 3D MT triaxial anisotropic inversion with topography. In triaxial anisotropic media,〓 and ρ〓 are resolvable, ρ〓 can't be resolvable due to nature of plane wave propagating vertically. In condition of complex geologies with anisotropy, to better recover the real underground electrical structures, one should use anisotropic inversion for MT data. The validity of the anisotropic inversion strategy and algorithm are verified by the Coprod2 and USArray MT data. The inversion of Coprod2 data provides an anisotropic interpretation of the NACP and TOBE anomalies together with a new revealed anomaly. With the inversion of USArray data, a dynamic model of the Cascadia subduction zone is established, which has scientific significance for studying the tectonic stress and motion of the deep earth and the cause of anisotropy. The results of field data inversion can provide a reference to the study of anisotropy in the deep earth. The algorithms addressed in this thesis lay a foundation for 3D MT anisotropic forward modling and inversions, which provides basis for studying the electrical anisotropy of deep earth. In this thesis, I have developed a goal-oriented adaptive finite-element method for 3D MT anisotropic modeling with topography, a noval identification technique of electrical anisotropy from MT data, and an effective strategy for anisotropic inversion for 3D MT data. The achievements have important theoretical and practical value in promoting the level of 3D MT data interpretation. Keywords: EM geophysic, Magnetotelluric, 3D forward and inversion, Unstructured finite-element method, Goal-oriented adaptive method, Anisotropic media, Limited memory qusi-Newtun method