Probabilistic inversion of audio-frequency magnetotelluric data and application to cover thickness estimation for mineral exploration in Australia

Abstract

We have applied a Bayesian inference algorithm and released open-source code for the 1D inversion of audio-frequency magnetotelluric data. The algorithm uses a trans-dimensional Markov chain Monte Carlo technique to solve for a probabilistic resistivity-depth model. The inversion employs multiple Markov chains to generate an ensemble of millions of resistivity models that adequately fit the data given the assigned noise levels. The trans-dimensional aspect of the inversion means that the number of layers in the resistivity model is solved for rather than being predetermined. The inversion scheme favours a parsimonious solution and the acceptance criterion ratio is theoretically derived such that the Markov chain will eventually converge to an ensemble that is a good approximation of the posterior probability density (PPD). Once the ensemble of models is generated, its statistics are analysed to assess the PPD and to quantify model uncertainties. This approach gives a thorough exploration of model space and a more robust estimation of uncertainty than deterministic methods allow.We demonstrate the application of the method to cover thickness estimation for a number of regional drilling programs. Comparison with borehole results demonstrates that the method is capable of identifying major stratigraphic units with resistivity contrasts. Our results have assisted with drill site targeting and have helped to reduce the uncertainty and risk associated with intersecting targeted stratigraphic units in covered terrains. Interpretation of the audio-frequency magnetotelluric data has improved our understanding of the distribution and geometries of sedimentary basins. From an exploration perspective, mapping sedimentary basins and covered near-surface geological features supports the effective search for mineral deposits in greenfield areas.