ABSTRACT
The aim of this research is to compare the results of three 3D inversion methods applied to greenfield airborne electromagnetic (AEM) data. The first two methods of inversion are the CDI3D and Maxwell codes, each parameterising the 3D conductor with a thin-plate approximation; inversion of which is overdetermined with AEM data. The third inversion method is an underdetermined 3D voxellated inversion using the finite volume method and regularisation based on minimising differences from a starting model. Deterministic, iterative, non-linear inversion requires a starting model, which in greenfield exploration is not available a priori using geological knowledge. We investigated three different starting models, each generated from fitting the data. These were: (1) a 1D background from CDI3D, (2) the approximate CDI3D solution including a discrete target, and (3) a stitched 1D starting model often called a conductivity–depth image (CDI). The CDI starting model is similar to the expected output of stitched 1D inversion codes. We investigated a volume of earth surrounding the IR-2 conductor at the Forrestania, Western Australia test site, located under five lines of VTEM Max data. The conductor response is seen in the central three of the five selected lines. Both plate approximations fit easily with aspects of the observed data, and their interpreted conductor location is consistent with the intersection of sulphides in the one drill hole on site. The voxellated 3D inversion consumed significant computer resources compared with approximate solutions. The result of the voxellated inversion was unsatisfactory when the stitched CDI was used as a starting model. We believe that this was because the stitched CDI shows an unreasonably deep, small conductor at the position of the true conductor. Because this deep conductor produces an undetectable 3D modelling response, it persists through the inversion process. We infer that all stitched 1D solutions, even those derived from inversion with or without lateral constraints, are likely to produce poor starting models for 3D inversion of finite conductors.
Acknowledgments
We thank Geotech and Southern Geoscience Consultants for access to the VTEM and drill hole data. Program CDI3D was developed with sponsorship by Spectrem Air. We thank anonymous reviewers for their constructive suggestions. The authors declare no conflicts of interest.
Disclosure statement
No potential conflict of interest was reported by the authors.