Wavelet transforms, neural networks and migration applied to magnetotellurics
In the magnetotelluric (MT) method, naturally occurring electromagnetic fields are used to study the electrical structures of the Earth. The impedance estimates are dependent on the natural sources and conventional Fourier transform techniques are not particularly suitable. The main focus in this dissertation is to automatically detect and process the magnetotelluric signals and enhance the signal-to-noise ratios of the recordings with the wavelet transform method. It is shown that a MT transient can be identified and its signal-to-noise ratio can be enhanced by taking advantage of the time-scale localization properties of these natural sources. When a complex Morlet wavelet is used, it is proved that the impedance can be directly calculated in the wavelet transform domain. Wavelet coherency and apparent wavelet impedance ratio are defined. The MT signals can be easily detected on the wavelet transform plane with the definition of wavelet coherency and the apparent impedance ratio. Synthetic and real examples are presented in the thesis. The MT c-response test for one dimensionality has been extended to data with errors. Two scalar indices are defined so that we can ascertain whether or not a structure is one-dimensional. This knowledge is important prior to any attempt at inversion or migration. For quickly imaging the near surface structure, the regularized Hopfield neural network inversion scheme and a fast imaging technique with EM migration are presented. The massively parallel processing of the Hopfield neural networks makes them suitable for hardware implementation; therefore, using the artificial neural networks has the potential of greatly speeding up MT inversion even in field operations. The EM migration technique can also effectively construct the conductivity discontinuities (i.e. reflection surfaces). Both methods are fast and can be easily implemented on a personal computer.