One-Dimensional Magnetotelluric Inversion Techniques

摘要

The general nonlinear regression problem associated with the inversion of magnetotelluric data to a one-dimensional, horizontally homogeneous earth model is investigated. An automatic program that will invert most magnetotelluric data encountered by the author to a significant one-dimensional model without any interaction on the part of the computer operator is presented. A comparison is made of the Gauss-Newton method, a modified Newton-Raphson method, and the Nelder and Mead flexible simplex method for fitting data with a model in a least square sense. It is concluded that the modified Newton0Raphson method with a gradient starter is generally the most successful algorithm for inverting magnetotelluric data and also for use in the automatic program. Another study is made to determine what range of frequencies of the surface data are altered most by perturbations in the model at various depths. From this study and other work an algorithm is presented that generates a first guess model for a magnetotelluric inversion program. In addition, this perturbation technique is useful in determining what frequency range of data is necessary to define a particular geological profile. Many program and model insensitivities that occur during the inversion process are categorized and discussed, and corrections for each type of insensitive program termination are suggested for use in the automatic program. Several examples of the automatic program inversions of theoretical data with and without noise and actual field data are presented and discussed. (Author)