Integration of automatic implicit geological modelling in deterministic geophysical inversion

Abstract

Abstract. We propose and evaluate methods for the integration of automatic implicit geological modelling into the geophysical (potential field) inversion process. The objective is to enforce structural geological realism in level-set inversion, which inverts for the location of boundaries between rock units. We propose two approaches. In the first one, a geological correction term is applied at each iteration of the inversion to reduce geological inconsistencies. This is achieved by integrating an automatic implicit geological modelling scheme within the geophysical inversion process. In the second approach, we use automatic geological modelling to derive a dynamic prior model term at each iteration of the inversion to limit departures from geologically feasible outcomes. We introduce the main theoretical aspects of the inversion algorithm and perform the proof-of-concept using two synthetic studies. The analysis of results using indicators measuring geophysical, petrophysical and structural geological misfits demonstrates that our approach effectively steers inversion towards geologically consistent models and reduces the risk of geologically unrealistic outcomes. Results suggest that geological correction may be effectively applied to pre-existing models to increase their geological realism and that it can also be used to explore geophysically equivalent models.

Tomofast-x 2.0: an open-source parallel code for inversion of potential field data with topography using wavelet compression

“…Whilst the modifications described above address many of the challenges of industry and government agencies, further developments are planned, including a closer integration with level set inversions (Giraud et al, 2023a). To extend the gravity to plate-scale problems we would need to reframe the calculations with a spherical coordinate scheme.…”

“…The sensitivity kernel can also be reused when a different background field needs to be subtracted from the data, as the kernel values depend only on the model grid and data locations. Additionally, the calculated kernel can be imported into other codes to perform sensitivity analysis, null-space navigation, or level set inversions (Giraud et al, 2023a). Finally, it is possible to import the external kernel corresponding to a different geophysics or another optimisation problem (e.g.…”

Tomofast-x 2.0: an open-source parallel code for inversion of potential field data, to recover density, susceptibility and magnetisation vector, with topography and wavelet compression

“…The optimisation work described in this paper is fundamental to allowing more complete descriptions of the controls on magnetisation, including remanence, which require 3 to 9 times the computer resources of the scalar description. Whilst the modifications described above address many of the challenges of industry and government agencies, further developments are planned, including a closer integration with level set inversions (Giraud et al, 2023a). To extend the gravity to plate-scale problems we would need to reframe the calculations with a spherical coordinate scheme.…”