考虑电流型畸变的大地电磁三维反演

大地电磁测深(MT)的观测数据易受到由近地表小尺度非均匀体或地形起伏引起的电流型畸变干扰,消除或压制这种干扰对获取可靠的深部电性结构至关重要.当区域结构为二维时,电流型畸变可采用张量分解等方法予以消除或压制.当区域结构为三维时,畸变问题更加复杂和严重,传统张量分解方法往往效果不佳或无效,严重地制约了MT三维反演技术的实用性.对此,本文提出一种考虑电流型畸变的MT三维反演算法,将完整的电流型畸变参数引入到目标函数,并采用非线性共轭梯度法与电阻率参数同时反演,从而达到压制畸变的目的.该算法有两个关键点:一是通过分析实测数据所遭受畸变的分布特征,在目标函数中对其进行有效约束;二是在迭代过程中,通过自适应地调整双正则化因子保障算法的稳定和效率.理论模型测试结果显示,常规三维反演算法不能合理解释数据中的畸变成分,而只能通过引入虚假异常体强制地拟合受畸变数据,从而造成电阻率模型严重失真.与之相比,本文算法能够在反演中自动求解各测点所受到的畸变,获得更接近真实的电阻率模型.

Three-dimensional inversion of magnetotelluric resistivity model with galvanic distortion

Galvanic distortion, which is caused by shallow small-scale inhomogeneities or topography, has long been recognized as an obstacle in magnetotelluric (MT) sounding studies. Removal of these distortions from the field-observed MT data is essential to obtain a reliable model of the subsurface electrical structure. In two-dimensional (2-D) situations, a number of impedance decomposition methods have been widely used for distortion removal. For data from three-dimensional (3-D) structures, however, the effects of galvanic distortion become far more complex and cannot be solved by these traditional methods. This paper proposes a method that employs the Nonlinear Conjugate Gradient algorithm (NLCG) to solve both the regional resistivity structures and the parameters of full galvanic distortion during iterations. The statistical distribution of galvanic distortion is firstly estimated by using the phase tensor method and suggests that the off-diagonal elements of the distortion tensor are quasi-symmetrically distributed around zero, and the diagonal elements are centralized around one. Using the prior information contained in the field data, we build a new objective function that takes the galvanic effects into consideration and adds an extra constraint on the distortion parameters. Furthermore, to ensure stability and efficiency of the algorithm, an adaptive regularized strategy is used to determine the trade-off factors between the data misfit term and the roughness terms of the resistivity model and galvanic distortion respectively. The synthetic model study suggests that conventional inversion approach cannot reasonably explain the distorted part of the synthetic data and will inevitably bring erroneous structures to the resistivity model. In contrast, our new inversion algorithm can reliably retrieve both the regional resistivity structures and galvanic distortion, irrespective of whether galvanic distortion is present or not.