A three-dimensional magnetotelluric regularized inversion using minimum support gradient

Speaker: Luolei Zhang
Three-dimensional (3-D) magnetotelluric (MT) inversion has progressed fast in recent a few decades. Most of these inversion methods above are classified as a regularized inversion with smoothness constraint. These inversions give smooth solution, and are not suitable for clearly imaging geo-electrical interfaces. In this study we introduce a new stabilizer to solve this problem. Portniaguine and Zhdanov (1999) proposed the focusing geophysical inversion images by using minimum gradient support functional and used it in gravity and magnetic inversion. Zhdanov (2008) also applied it to invert gravity and electromagnetic data. Here we apply the same functional in 3-D MT inversion. In the forward calculation of the inversion process, the subsurface resistivity structure is divided by cubes. The conductivity in each cube is assumed uniform. Through changing the cube’s volume, the resolution and accuracy of inversion can be ensured by their trade-off. The model parameter is defined as log of conductivity normalized by initial conductivity. Modified Iterative Dissipative Method (MIDM) which was proposed by Singer (1995) and Avdeev et al. (2000) is used for forward calculation, which allows us to avoid calculation of large-scale linear equations. GPBi-CG is used to get the solution in modified Neumann series, and the efficiency is increased. The quasi-Newton method is used to optimize the objective functional. This approach is a kind of Newton method with simplified calculation of the Hessian matrix by using BFGS update (Koyama, 2002). We tested the performance of our code by using a synthetic model which consists of an anomalous body in a uniform half space, and by comparing with the results obtained by other smooth inversion code. >From the comparison, we can find the geo-electrical interfaces are imaged more clearly and accurately by our code, while the conductivity of the anomalous body is estimated more or less similar to the result of smoothing inversion. The minimum data misfit value is also smaller than that of smoothing inversion. Comparing with smoothing stabilizing functional, the use of minimum gradient support functional for local geophysical study allows us to determine clear geo-electrical interfaces.