Near-Surface Bayesian First-Arrival Tomography with Uncertainty using Natural Neighbor Interpolation

Abstract

Summary Various velocity parameterizations are used in Bayesian first-arrival tomography. We conduct a short review of the existing approaches and suggest the natural neighbor interpolation as a viable alternative. This parameterization possesses numerous useful properties. It provides naturally smooth models, which is particularly suitable for a refraction setting. It does not need any specific treatment at model boundaries, and, finally, does not need any additional parameters apart from velocities defined on a set of nodes. We compare this parameterization with a more conventional linear barycentric approach on a synthetic near-surface seismic dataset. The comparison shows that natural neighborbased tomography results in a more accurate estimation of seismic velocity inside the near-surface low-velocity anomaly and provides a lower estimate of velocity uncertainty in the whole model.