Adaptive primary-multiple separation using 3D curvelet transform

Abstract

In this paper, we propose a method to enhance the separation of primaries and multiples by utilizing the ultra-sparseness property of the 3D curvelet transform. By extending our earlier work on the 2D method, our current 3D primary-multiple separation method takes into account the coherence between neighbouring gathers, and extends the Bayesian Probability Maximization (BPM) based separation mechanism into the 3D curvelet domain. The primaries and multiples are differentiated by utilizing the traces of neighbouring gathers in an additional dimension; this further promotes their separation compared to the 2D curvelet domain method. Moreover, this 3D curvelet domain separation method produces robust results regardless of the ordering of data as long as they are organized in a volumetric manner. Additionally, we have also introduced a 3D spatiotemporal constraint for handling the deviation from linearity or planarity of the seismic events. We demonstrate the improvement of the 3D curvelet domain primary-multiple separation method on synthetic and field data examples, by comparing the results with those produced by existing separation methods.