Three-dimensional Conjugate Gradients Inversion Of Magnetotelluric Impedance Tensor

We have developed a three-dimensional conjugate gradients inversion algorithm of magnetotelluric(MT) impedance tensor. During the inversion process, the 3D conjugate gradient inversion algorithm doesn't need to compute and store the Jacobian matrix but directly updates the model from the computation of the Jacobian matrix or its transpose multiplying an arbitrary vector. Requiring only one forward and four pseudo-forward modeling applications per frequency to produce the model update at each iteration, this algorithm efficiently reduces the computation of the inversion. This inversion algorithm can be used to invert 3D MT area data and 2D MT profile data. The importance of including the on-diagonal impedance tensor terms in 3D inversion is considered in the synthetic example. The results from the 3D inversion of synthetic 3D area data indicate that on-diagonal terms are not so important in the inversion when there are enough data sites on the top of the target structure; on-diagonal terms are very important in the inversion and the target structure can be better resolved only by inverting both off-diagonal and on-diagonal terms when there are few data sites on the top of the target structure. Therefore, 3D inversion should include all tensor elements if possible. The results from the 3D inversion of synthetic 2D profile data show that reasonable images for structures beneath and near a single MT profile can be obtained if all tensor elements are used in the inversion of the profile data; on-diagonal terms are as important as off-diagonal terms and the target structure can be well resolved only by inverting both off-diagonal and on-diagonal terms. Currently, most of MT data are still collected on 2D profiles. This inversion algorithm of impedance tensor can promote the practical application of 3D MT inversion. We also use this inversion algorithm to invert some real data. From the trial inversion with the synthetic and real data, the validity and practicability of this inversion algorithm is verified.