Inversion of controlled-source audio-frequency magnetotelluric data

This dissertation addresses an inversion algorithm to recover the Earth's electrical structure from controlled-source audio-frequency magnetotelluric data. The algorithm is based on the concepts of rapid relaxation inversion (RRI) of magnetotelluric data. The inversion uses the same technique to compute sensitivities as RRI and these approximate sensitivities are validated by comparison with exact sensitivities. The comparison shows that the approximate sensitivities have similar depth variations although different magnitudes as the exact sensitivities when transmitter-receiver offsets are greater than one skin depth in the Earth. A relationship between rapid relaxation inversion and the standard inversion is established through the sensitivity matrix. It is shown that RRI computes the sensitivities of data to the cells directly below the observation site by using approximate analytic formulas similar to the 1-D Frechet derivative, and approximates the sensitivities of data to the rest of cells to be zero. The combination of sensitivity comparison and successful inversion imply that the accuracy of the sensitivity matrix need not be very high for an iterative inversion.; The 2.5-D finite-element forward modeling method of Unsworth et al. (1993) is used in this inversion algorithm and extended to include a magnetic source. Thus the 2.5-D finite-element forward modeling method can model any source exciting a 2-D Earth. In principle, this inversion algorithm can invert CSAMT data from an electric or a magnetic source. The algorithm is tested on synthetic and field data from a electric dipole, a horizontal magnetic dipole and a finite-length horizontal electric dipole, and gives promising results.