Active source refraction tomography using a reversible jump Markov Chain Monte Carlo approach.

Abstract:

We present an approach to recover subsurface velocity structure and its associated uncertainties from controlled-source seismic data. Here, we cast the ray-based traveltime tomography problem in a Bayesian framework and estimate the posterior probability density (PPD) in model parameters using the Reversible Jump Markov Chain Monte Carlo (rj-MCMC) approach. The number of model parameters is treated as a variable, like the P-wave velocity information. We use an adaptive cloud of nuclei points and Voronoi cells to represent our 2D velocity model. In our rj-MCMC, we use the birth-death-move approach to add, remove or move the velocity nuclei in a given model and accept or reject this model using the Metropolis-Hastings-Green criterion. We compute ray paths and their corresponding travel times for all source-receiver pairs in all sampled models using a shortest-path method. We then use all accepted models after burn-in phase to calculate various statistical measures and estimate the uncertainty in our tomography derived model. We apply our approach to existing 2D active source seismic data acquired at mid-ocean ridges and convergent margins.We show that our global optimization approach leads to higher-resolution subsurface velocity structure as compared to the gradient-based approach, particularly in regions of sharp velocity gradient. We further report that this methodology allows for the quantification of uncertainty in subsurface velocities, from which related physical properties can subsequently be derived (e.g. porosity).