Using SVD filters for velocity analysis and ground-roll attenuation

Abstract

This study investigates the adaptive filtering approach based on the Singular Value Decomposition (SVD) method to improve velocity analysis and ground-roll attenuation. The SVD filtering is an adaptive multichannel filtering method where each filtered seismic trace keeps a degree of coherence with the immediate neighboring traces. Before applying the adaptive filtering, in order to flatten the primary reflections the seismogram is corrected using the Normal Move Out (NMO) method. The SVD filtering helps to strengthen the spatial coherence of reflectors. It works as multichannel and can be applied by selecting a set of seismic traces taken from around the target trace. Thus traces from different shots can be represented by a five-point areal operator, which we call five-point cross operator. In this paper we run this operator along the coverage map of the seismic survey. At each operator position, the filtered trace (center of the operator) is obtained by taking the first or adding the first eigenimages. Thereby we enhance the coherence corresponding to the primary reflections in detriment of the remaining events (ground-roll, multiples, and other non-correlated events) remained in the other eigenimages. The method was tested on a seismic line of the Tacutu Basin, Brazil. The obtained results show the velocity spectra with better definition, as well as better post-stacked section exhibiting better continuity of seismic reflections and lower noise, compared with the raw processing results (without SVD filtering).