Bayesian tomography-driven inversion of Bouguer gravity: application to the western Alps

The 3D western alpine lithosphere presents a complex structure and remains subject of active geophysical investigation. In this context, we propose and apply new techniques to combine seismic and gravity data. In particular, we set up a Bayesian inversion of Bouguer gravity anomaly data for the 3D distribution of density at crustal and lithospheric scales. We use the Bouguer gravity anomaly map after  Zahorec et al., (2021) across the western Alps. 

In our setup, we introduce a priori information based on existing seismic tomography models (e.g. Nouibat et al., 2022), to guide the exploration of model geometries for target density distributions. We also use flexible constraints based on known ρ-vS conversion laws (e.g. Brocher, 2005), to define a pool of candidate density models consistent with rock-physics constraints and laboratory observations.

The 3D forward gravity modeling is achieved by discretizing the target volume area into unitary voxels of constant density, accounting for surface topography as well. By pre-computing the gravity effect of each voxel, we significantly decrease the computational cost of forward modeling, thus allowing an exploration of the parameter space with a Monte Carlo sampling approach. In particular, we implement a Markov chain Monte Carlo (McMC) algorithm in a Bayesian framework.
To address the lower resolution power of gravity data, we reduce the dimensionality of the model space by describing volumetric structures with a level-set approach, based on the available seismic tomographic models. This allows to 1) incorporate a priori knowledge of the crustal structure from seismic investigations into the inversion setup and 2) define complex laterally-heterogeneous density structures with a lower number of parameters. While we allow deviations from exact ρ-vS conversion laws, the bayesian framework allows to highlight existing trade-offs among density and geometry, and to tackle the non-uniqueness that often affects gravity data inversions. Finally, this setup allows to benchmark a seismic tomographic model against gravity data while providing a new density model.

We produce a new 3D density model of the western alpine lithosphere, including the Ivrea Geophysical Body at the boundary between the European and Adriatic tectonic plates. Our setup allows us to compare the resolved density values with seismic tomography models locally and with surface geology as well, providing new constraints on subsurface rock structure and composition.