A Bayesian approach to elastic full-waveform inversion: application to two synthetic near surface models

Imaging of the first metres of the subsurface with seismic methods constitutes a key challenge for several applications. In this context, the analysis of Rayleigh waves can reveal information about the S-wave velocity structure in the first metres of the subsurface. The waves recorded can be inverted using several techniques, of which the most widely used is the multichannel analysis of surface waves, where dispersion curves are picked on the velocity-frequency spectrum. A full-waveform inversion of surface waves has been implemented, offering the possibility to exploit the complete information content of the recorded seismograms. This method has only recently been tested with elastic approximation on synthetic data, as the application in near-surface scenarios is very challenging due to the high nonlinearity of the problem and the considerable computational costs. This paper presents a gradient-based Markov chain Monte Carlo elastic full-waveform inversion method, where posterior sampling is accelerated by compressing data and model spaces through the discrete cosine transform and, also, by defining a proposal that is a local, Gaussian approximation of the target posterior probability density. The applicability of the approach is demonstrated by performing two synthetic inversion tests on two different near-surface models: a two-layered model with lateral velocity variations, and a four-layered model with velocity inversions.