3D parametric hybrid inversion of time-domain airborne electromagnetic data

Abstract

We have developed a method to invert time-domain airborne electromagnetic (AEM) data using a parametric level-set approach combined with a conventional voxel-based technique to form a parametric hybrid inversion. The approach was designed for situations in which a voxel-based inversion alone may struggle. Such an example is where a distinct anomaly is present with sharp boundaries, and there is a large contrast between a low-resistivity target and a high-resistivity background. The first step of the proposed hybrid method used our novel parametric inversion to recover a best-fitting skewed Gaussian ellipsoid that represented the target of interest. Subsequently, the parametric result was set as an initial and reference model for the second stage, where smooth features with smaller resistivity contrasts were introduced into the model through a conventional voxel-based approach. The approach was tested with synthetic and field data. In the synthetic case, we recovered the size and dip of a conductive, thin, dipping plate with better accuracy compared with a voxel-based inversion. In the field example, we inverted AEM data over the Caber volcanogenic massive sulfide deposit. Based on information from past drilling, our results improve upon previous parametric plate inversions of the deposit itself, while additionally imaging the conductive cover over the deposit. These findings showcased how our parametric hybrid method can improve the accuracy of time-domain AEM inversions for thin dipping targets with large resistivity contrasts compared with the background. Preliminary results presented at SEG 2014 in a talk entitled: "Recovering a thin dipping conductor with 3D electromagnetic inversion over the Caber deposit." Manuscript

Large Scale 3D Airborne Electromagnetic Inversion - Recent Technical Improvements

“…For such systems as SkyTEM (Sørensen and Auken, 2004) that have a dual-pulse, this modular approach makes performing a joint inversion of both pulses of data a simplistic task. It is also trivial to perform a parametric inversion or a hybrid parametric inversion (McMillan et al, 2015) in both the time and frequency domain as it just requires an additional layer of parametrization and derivatives using the same forward operators and inversion framework as the voxel inversions. A visual example of the parametric and hybrid parametric is displayed in Figure 3 …”

Open-source software for simulations and inversions of airborne electromagnetic data

“…First, we see from the data-misfit curve in Figure 6c that this has been a challenging problem for the voxel inversion; between iterations 5 and 14, little progress is being made in reducing the misfit. This may be related to some of the challenges observed by McMillan et al (2015), where the circular nature of the sensitivity of an airborne EM survey tends to promote "ringing" artefacts in the inversion. The authors overcame this problem by introducing a parametric inversion.…”

A framework for simulation and inversion in electromagnetics

“…In the case of a parametric inversion, the inversion model is defined on a domain that is independent of the forward modelling mesh and the mapping takes the parametric representation and defines a physical property on the forward modelling mesh (e.g. a gaussian ellipsoid defined geometrically) (Li et al, 2010;Pidlisecky et al, 2011;McMillan et al, 2015b;Kang et al, 2015). Maps can be composed, for instance, a layered, 1D log conductivity model defined only in the subsurface may be mapped to a 2D cylindrical Mesh, as shown in Figure 5.…”