Refine
Has Fulltext
- no (3)
Document Type
- Article (3) (remove)
Language
- English (3) (remove)
Is part of the Bibliography
- yes (3)
Keywords
Institute
We have developed a 1D laterally constrained inversion of surface-wave dispersion curves based on the minimum gradient support regularization, which allows solutions with tunable sharpness in the vertical and horizontal directions. The forward modeling consists of a finite-elements approach incorporated in a flexible nonparametric gradient-based inversion scheme, which has already demonstrated good stability and convergence capabilities when tested on other kinds of data. Our deterministic inversion procedure is performed in the shear-wave velocity log space as we noticed that the associated Jacobian indicates a reduced model dependency, and this, in turn, decreases the risks of local nonconvexity. We show several synthetics and one field example to demonstrate the effectiveness and the applicability of the proposed approach.
Frequency-domain electromagnetic (FDEM) data are commonly inverted to characterize subsurface geoelectrical properties using smoothness constraints in 1D inversion schemes assuming a layered medium.
Smoothness constraints are suitable for imaging gradual transitions of subsurface geoelectrical properties caused, for example, by varying sand, clay, or fluid content. However, such inversion approaches are limited in characterizing sharp interfaces. Alternative regularizations based on the minimum gradient support (MGS) stabilizers can, instead, be used to promote results with different levels of smoothness/sharpness selected by simply acting on the so-called focusing parameter.
The MGS regularization has been implemented for different kinds of geophysical data inversion strategies. However, concerning FDEM data, the MGS regularization has only been implemented for vertically constrained inversion (VCI) approaches but not for laterally constrained inversion (LCI) approaches.
We present a novel LCI approach for FDEM data using the MGS regularization for the vertical and lateral direction. Using synthetic and field data examples, we demonstrate that our approach can efficiently and automatically provide a set of model solutions characterized by different levels of sharpness and variable lateral consistencies.
In terms of data misfit, the obtained set of solutions contains equivalent models allowing us also to investigate the non-uniqueness of FDEM data inversion.
Given the range of geological conditions under which airborne EM surveys are conducted, there is an expectation that the 2D and 3D methods used to extract models that are geologically meaningful would be favoured over ID inversion and transforms. We do after all deal with an Earth that constantly undergoes, faulting, intrusions, and erosive processes that yield a subsurface morphology, which is, for most parts, dissimilar to a horizontal layered earth.
We analyse data from a survey collected in the Musgrave province, South Australia. It is of particular interest since it has been used for mineral prospecting and for a regional hydro-geological assessment. The survey comprises abrupt lateral variations, more-subtle lateral continuous sedimentary sequences and filled palaeovalleys. As consequence, we deal with several geophysical targets of contrasting conductivities, varying geometries and at different depths. We invert the observations by using several algorithms characterised by the different dimensionality of the forward operator.
Inversion of airborne EM data is known to be an ill-posed problem. We can generate a variety of models that numerically adequately fit the measured data, which makes the solution non-unique. The application of different deterministic inversion codes or transforms to the same dataset can give dissimilar results, as shown in this paper. This ambiguity suggests the choice of processes and algorithms used to interpret AEM data cannot be resolved as a matter of personal choice and preference.
The degree to which models generated by a ID algorithm replicate/or not measured data, can be an indicator of the data's dimensionality, which perse does not imply that data that can be fitted with a 1D model cannot be multidimensional. On the other hand, it is crucial that codes that can generate 2D and 3D models do reproduce the measured data in order for them to be considered as a plausible solution. In the absence of ancillary information, it could be argued that the simplest model with the simplest physics might be preferred.