3D seismic surface multiple attenuation: algorithms and analysis

Abstract

The aim of seismic exploration is to provide a comprehensive description of subsurface geologic structure in terms of its reflectivity function at the boundaries between geological units. Seismic multiples are coherent noise that obscure primary events and considerably degrade the quality of seismic images in the target zones. In spite of the fact that many methods have been designed to suppress multiples, only a limited success has been achieved. I have developed two different approaches to address the problem of seismic multiples. The first approach attempts to suppress multiples in terms of decomposition of the measured seismic wavefields into its upgoing and downgoing waves. The separation process is accomplished by using some statistical characteristics of the data in the plane-wave p domain. The ratio of these two components yields the true reflectivity function free of multiples. Although encouraging results are obtained in the separation process, instability occurs during the wavefield division step. As a result, the effectiveness of this approach is limited. I have also investigated seismic multiples for 3D geology and proposed a new methodology in which 3D multiples are predicted and attenuated successfully. The departure of the predicted multiple arrival times from the observed multiple arrival times explains why demultiple algorithms that assume two-dimensional multiple reflections often fail. In this approach, I employed 3D ray tracing to predict the arrival times of the primary and its multiples in individual shot gathers generated from a three-dimensional reflector. A non-linear optimization method, called Very Fast Simulated Annealing (VFSA) is used to determine geometry of the subsurface reflector in 3D. This is achieved by applying a ray traced normal moveout (NMO) correction to seismic reflections with respect to the zero offset time. Based on the optimized NMO-corrected shot gathers, the autoconvolution of the seismic trace is employed to predict the multiple reflections, which are then scaled and subtracted from the original data. The application of this technique to real data demonstrates that the new method successfully suppresses many surface multiples, and is able to recover several deep primary events. This algorithm is robust and computationally very efficient.