Gleaning meaningful information from seismic attributes

Abstract

Seismic attributes form an integral part of most of today’s interpretation projects. Attributes enhance subtle features in the seismic data that may otherwise be overlooked or require a great deal of time to map. The quality of attribute displays is directly proportional to the quality of the input seismic data. Ideally, all amplitude, phase, and travel time distortion effects due to near-surface and overburden heterogeneities as well as those introduced by acquisition and processing should be optimally handled, even if they cannot be totally eliminated. In practice, even with careful acquisition, processing, and imaging, our data will still exhibit a certain level of noise. We show how structure-oriented filtering can eliminate random noise, with the principal component filter providing better results than the more familiar mean and median filters. The acquisition footprint is a form of coherent, rather than random, noise and requires a different filtering approach, ideally in the prestack domain prior to stacking. Finally, we show that different implementations of a given attribute can make a difference, with energy ratio coherence providing more robust images than semblance. We illustrate these findings through application to a suite of examples from Alberta, Canada.