Magnetotelluric Data Quality Assessment And Impedance Estimation

Abstract:Magnetotelluric is a kind of geophysical method for imaging the electrical conductivity and structure of the earth by measuring time variations of orthogonal components of electric and magnetic fields at the surface of the earth. It is widely used in geological applications and the petroleum industry. It is meaningful to develop the techniques for magnetotelluric data processing, especially for the data collected in the highly industrialized area.The fundamental procedures for magnetotelluric data processing are introduced first, including pre-processing in time domain, cascade decimation, spectral estimation and least square solution for impedance estimation. The results computed by the open source code EMTF are compared with those computed by SSMT provided by Phoenix Geophysics.Natural electromagnetic field and magnetotelluric noise are introduced for short. Continuous wavelet transform and electromagnetic field polarization directions are used for describing the characteristics of magnetotelluric signal and noise. Analysis of typical data shows that signal and noise are different in these aspects. Mathematical morphology filtering with compensation is developed for time domain de-noising. The results show that the data processed by this technique are more close to the characteristics of natural electromagnetic fields.Several kinds of methods for robust impedance estimation, including M-regression estimation, bounded influence estimator and repeated median estimator, are discussed, then their robustness are tested by a series of experimental data. A collect of magnetotelluric time series is simulated and noise is extracted from the magnetotelluric data. Simulated time series with noise in different levels are processed by different kinds of robust impedance estimator. The result shows that bounded influence estimator and repeated median estimator are the most robust, they are recommended in the magnetotelluric data processing. at last, a mixed method is proposed for the real data processing.