Bayesian Inversion of Broadband Surface Waves Dispersion Curves for Shear Velocity Structure and Anisotropy of the Crust and Upper Mantle

Abstract:

The increasing amount of broadband phase velocity dispersion measurements around the world is leading to significant improvements in shear velocity models on both regional and global scales. As the relation between surface-wave dispersion and the seismic velocity structure of the earth is nonlinear, a reliable way to perform the inversion is Monte Carlo sampling in a Bayesian framework. Considering the high sensitivity of surface waves to Vs in broad depth intervals and their low sensitivity to Vs in thin layers, there are strong trade-offs between shear speeds at neighboring depths. MC sampling provides a way to quantify non-uniqueness of the inverted shear velocity models.

We develop a Markov Chain Monte Carlo method for joint inversion of Rayleigh- and Love-wave dispersion curves that is able to yield robust radially and azimuthally anisotropic shear velocity profiles, with resolution to depths down to the transition zone. The inversion is a one step process that doesn't involve any linearization procedure or a priori bounds around a reference model. In a fixed dimensional Bayesian formulation, we chose to set the number of parameters relatively high, with a more dense parametrization in the uppermost mantle, therefore we used a Gaussian a priori distribution of the parameters in order to avoid overfitting.

We apply the MCMC algorithm to the inversion of surface-wave phase velocities accurately determined in broad period ranges in a few test regions, and present the resulting radially and azimuthally anisotropic shear velocity models.