A Bayesian Approach to the Inversion of Airborne Electromagnetic Data for a Multidimensional Earth

Resistivity distributions for the subsurface based on airborne electromagnetic data are commonly derived using deterministic inversion methods. It is well known that this inverse problem is inherently non-unique; if one model can be found, it is likely that there exist alternative models that fit the data equally well, particularly once noise on the data is taken into account. Probabilistic approaches, like the one introduced in this work, allow exploration of the posterior distribution which represents the distribution of models that are in agreement with the data and the prior information. This work uses multipoint geostatistical models to represent the prior information, and a Markov Chain Monte Carlo technique to sample the unknown posterior distribution. The airborne electromagnetic data are predicted by employing a 2.5D forward solver, so that lateral changes in structure along the flight path are taken into account. The inversion aims at determining the probability of individual lithologies to be present. We use synthetic examples to demonstrate how the method recovers well-known facts of airborne electromagnetic imaging, for example the reduced resolution if targets are located under conductive regolith cover. Application of the method to field data collected over the Harmony Ni-S deposit in Western Australia shows that our sets of samples of the posterior distribution provides a more complete picture of solution space when compared to deterministic inversion results. Such probabilistic images of the subsurface can ultimately be beneficial for the mitigation of exploration risk, because of their quantification of uncertainties.