A Bayesian approach to the tomographic problem with constraints from geodynamic modeling: Application to a synthetic subduction zone

Abstract

Geodynamic tomography, an imaging technique that incorporates constraints from geodynamics and mineral physics to restrict the potential number of candidate seismic models down to a subset consistent with geodynamic predictions, is applied to a thermal subduction model. The goal is to test the ability of geodynamic tomography to recover structures harbouring complex deformation patterns. The subduction zone is parameterised in terms of four unknown parameters namely the slab length L, thickness R, temperature T_c, and dip angle θ that define its thermal structure. A temperature-dependent viscosity is then prescribed with an activation coefficient E controlling the sensitivity. Using the full forward approach to geodynamic tomography, we generate anisotropic surface wave dispersion measurements as synthetic data. We then attempt to retrieve the five unknown parameters by inverting the synthetics corrupted with random uncorrelated noise. The final output is an ensemble of models of L, R, θ, T_c and E cast in terms of a posterior probability distribution and their uncertainty limits. Results show that the parameters are tightly constrained with the apparent existence of a single misfit minima in each of them, which implies the implicit retrieval of the complete patterns of upper mantle deformation, and correspondingly, the 21-independent coefficients defining elastic anisotropy. Each model realisation however fails to swarm around its true value. Such results are attributed to the inability of the surrogate model to accurately replicate the correct forward model for computing anisotropy due to the inherent complexity of the deformation patterns considered. Nevertheless, this proof of concept shows a self-consistent method that incorporates mantle flow modeling in a seismic inversion scheme.