| Popis: |
Properties of subsurface structure determine principal effects of local ground motion. Attributes of primary importance, i.e. S- and P-wave velocity profiles, can be inferred from Rayleigh and Love wave dispersion and ellipticity curves as retrieved from the single-station and array measurements. However, the measured data are subject to uncertainty and the solution exhibits significant inherent non-uniqueness as different velocity models provide a similar fit to observed data. This highlights the importance of rigorous treatment of uncertainty. Standard non-linear optimization inversion techniques chasing velocity models that provide a minimal data misfit; then a set of tested models might not be representative in terms of solution uncertainty. This can be overcome by an inversion formulation in the Bayesian (probabilistic) framework, where the uncertainty of the measured data is rigorously propagated to results. This research is focused on the development and application of a novel Bayesian inversion method intended for local subsurface characterization. Our inversion method is formulated in the transdimensional Bayesian parameter space, where 1-D velocity models may have a varying number of layers. The number of layers is treated as the model complexity that is governed by the data-driven law of parsimony. This parameter space is explored by a reversible-jump Markov chain Monte Carlo algorithm that produces an ensemble of models representative in terms of solution uncertainty. To suppress a possible dependency on an initial model, we use multiple parallel Markov chains with independent and random initial models (parallel tempering technique). We introduce also a multizonal formulation of the prior, allowing to include additional information to the inversion (from geological profiles, etc.). In this contribution, we present a validation of the inversion method using a synthetic test and application to a selected Swiss site. Our method is suitable for sites with low-velocity zones and it provides reliable estimates of solution uncertainty. |
