Radially anisotropic 3-D S-wave model using multi-mode surface waves: Comparisons of linearized and nonlinear Bayesian approaches

Abstract:

Structural mapping for shear wave structure in the crust and upper mantle have been based mostly on linearized inversions using surface-wave dispersion data. Such linearized inversions involve intrinsic limitations, such as the influence of subjective a priori constraints (e.g., initial model, damping, model parameterization) and the possibility of entrapment into local minima. Recently, a fully nonlinear method based on the trans-dimensional hierarchical Bayesian (TDB) inversion has become a popular approach for reconstructing vertical shear velocity profiles, jointly inverting receiver functions and surface wave dispersion data. Such an approach enables us to better constrain both the absolute shear velocity and internal discontinuities. However, the application of such a joint inversion is limited to a location where a broad-band seismic station exists.

In this study, we employ the fully nonlinear TDB inversion to reconstruct localized vertical shear velocity profiles using multi-mode dispersion maps of Love and Rayleigh waves in Australia, which forms a 3-D radially anisotropic S wave model in the upper mantle. In the TDB approach, little a priori information on model parameters is necessary, and entrapment into local minima can be avoided. Also, we can quantify model uncertainties since the final model is obtained as probability density functions. Since we do not impose any linearization in our TDB inversion, it requires a significant computation time compared with a linearized approach. The TDB model is then compared with an earlier 3-D model derived from the conventional linearized method of iterative least-squares inversion using the same dispersion data set.

The TDB approach allows us to better resolve vertical changes in the shear wave velocity and radial anisotropy without strong a priori constraint. While large-scale features of these two models are consistent, the TDB model tends to exhibit more enhanced strength of heterogeneity and well-resolved spatial distribution of radial anisotropy. A better fit to multi-mode dispersion data can be achieved in the TDB model owing to a better estimate of data errors from the hierarchical Bayesian approach.