Monte Carlo Inversion of SkyTEM AEM data from Lake Thetis, Western Australia

Abstract

During 2011 Groundprobe Geophysics flew a SkyTEM airborne electromagnetic (AEM) survey in the vicinity of Lake Thetis near Cervantes in Western Australia. The survey was commissioned by the Western Australian Department of Water as part of the Mid West Groundwater Dependent Ecosystem Vulnerability Project. Groundprobe Geophysics processed and then initially inverted the dataset using the iTEM fast approximate inversion. Subsequently the data were inverted using a reversible jump Markov chain Monte Carlo (rj-McMC) 1D inversion algorithm recently developed at Geoscience Australia. Both the high- and super-low-moment data were inverted simultaneously taking into account the tilts of the transmitter-receiver frame. In the inversion of each dual-moment AEM sounding an ensemble of 300,000 models were generated in a Markov Chain, about 290,000 of which fit the data within the estimated noise envelope. The reversible jump aspect of the algorithm means that the number of layers in the 1D models varies, and the algorithm tends to favour models with the fewest number of layers that allow the data to be fitted, in essence providing a data-driven Occam's Razor. The algorithm is considerably more computationally expensive than deterministic inversions. However its advantage is that a great deal of information can be extracted from the ensemble. For example, the most likely, mean, mode and median models were all extracted and made into conductivity depth slice maps and sections. We extract the 10th and 90th percentile models and use the spread between them as a measure of model uncertainty, which we convey on maps by making uncertain areas more transparent. Another output of the algorithm is a change-point histogram which provides information on the most probable depths of the layer interfaces. By extracting the peaks from this histogram we begin to be able to automatically interpret layer boundaries to be plotted onto conductivity sections.