Speaker: Jeremie Giraud
Date: Thursday 15th of May 2025, 11:00 am.
Abstract:
We present and apply a pseudo trans-dimensional inversion method for 3D geometrical gravity inversion, in which the number of rock units, their geometry, and their density can vary during sampling. The method builds on a multiple level set framework and uses a birth-death process to insert or remove rock units from an existing model. Interface geometries are perturbed using random fields, and densities are sampled from distributions informed by prior geological knowledge. Sampling is performed using a non-reversible Metropolis-Hastings algorithm designed to efficiently explore complex model spaces while ensuring a parsimonious solution.
The method is applied to gravity data from the prospective Boulia region (Queensland, Australia) to image rocks beneath sedimentary cover. In this field case, an implicit geological model—constructed from the interpretation of 2D seismic lines, borehole data, and geological rules—is used to define prior geological constraints on the inversion. To aid interpretation, a workflow combining dimensionality reduction and clustering is applied to the ensemble of sampled models, allowing identification of families of geologically plausible solutions. Preliminary results suggest that up to two dense rock units, not initially identified by the geological model, may be needed to explain the observed data. Overall, our analysis of results suggests the ability of the method to infer the presence of previously unrecognized geological features, such as buried intrusions or facies variations, and indicates its potential as a tool to support exploration.