Bayesian Inversion of Transient Airborne EM Data from the McMurdo Dry Valleys, Antarctica

Inversion is a standard tool for evaluating geophysical measurements, yet most geophysical inverse problems are highly non-unique, so assessing the set of models that adequately explain the data is paramount. Yet the canonical method for inverting geophysical data - gradients and regularization - produces only a single “optimal” model with little to no information about uncertainty.

Bayesian inverse methods, such as Markov chain Monte Carlo, provides an ensemble of models, all of which fit the data adequately. The ensemble provides a wealth of statistical information, in lieu of just a single model.

We develop a trans-dimensional Markov chain Monte Carlo algorithm (McMC) for inverting transient AEM data for 1D Earth models and apply it to a unique AEM dataset collected over Taylor Valley, Antarctica. Prior work on this dataset using standard inversion techniques found evidence for conductive channels in the sediments beneath the valley. We confirm the presence of these channels and estimate quantitatively the range of conductivities compatible with the data. We translate this uncertainty into a probable range of values for porosity and pore fluid conductivity. In addition, we present our method as a novel way to obtain accurate estimates of the depth of investigation (DOI).