Selecting velocity models using Bayesian Information Criterion

ABSTRACT We present a strategy for selecting the values of model parameters by comparing walkaway vertical seismic profiling data with a multilayered model in the context of Bayesian information criterion. We consider P‐wave traveltimes and assume elliptical polar velocity dependence. A model with different propagation speeds, depending on the angle of propagation, can be a good approximation for a medium composed of thin layers. While elliptical anisotropy in a one‐layer model yields good results, an efficient tool for multilayer modelling would provide improved inversion results. To obtain the proper set of velocity values for specific parameterizations, we require two steps of optimization. In the first step, we find the signal trajectory; in the second step, we obtain parameter values by minimizing the misfit between the model and the data. By comparing models and data, we choose the best model in the sense of the Bayesian information criterion.