Spectral whitening based on the singular spectral analysis method

Abstract

Coherent noise is always challenging in seismic data processing. Several filtering methods have been used to improve the quality of seismic data. Singular spectral analysis (SSA), a technique based on singular value decomposition, decomposes a single trace, in time domain, into several traces with different frequency bandwidths. This idea of frequency decomposition suppresses the need of Fourier transform. Working trace to trace, this approach decomposes the seismic trace into eigentraces, from higher to lower energy, which normally correspond to frequency bands organized from lower to higher frequency content, respectively. Relying on this property, this paper proposes a new filtering technique, based on classical spectral whitening in order to balance the frequency content of seismic traces, capable of enhancing amplitude of higher frequencies while maintaining low frequency. For a better understanding of the method and a better evaluation of its results, a workflow using a Fourier-based filter has been performed in parallel. Numerical results obtained using seismic data from the Tacutu and Parnaba basins (north and northeast of Brazil, respectively) illustrated the enhancement in time resolution and lateral continuity of the reflections of stacked sections, as well as results from average amplitude spectra. The results showed that both SSA and Fourier-based methods are effective for noise attenuation and data enhancement, with SSA having the best results in terms of data resolution. As with any other filter, both methods described and utilized in this paper have their advantage and disadvantages, leading to the user to decide which is the best in each determined situation.