A multi-scale approach for seismic tomography

The goal of seismic tomography is to derive a velocity model. In this paper, we describe a multi-scale approach which iteratively derives a velocity model. In the early stage of the iterative process, we apply large smoothing constraints to derive the large scale features of the velocity model. As the data misfit reduces, we gradually reduce the smoothing constraints, adding finer details to the derived model. The key component of our approach is effective control of the smoothness of a derived model. To achieve this, we have developed a new implementation of smoothing constraints based on stochastic inversion. For a discrete and uniformly gridded model, it can be shown that our new implementation based on the stochastic inversion is equivalent to the conventional implementation based on the regularization method