Ultimately, a seismic image serves interpreters as a means of building a geological model of the subsurface. Automatic edge detection schemes can help to produce images that emphasize critical geological features such as faults and river channels.
Showing some striking coherency images of complex geological structures, Bahorich and Farmer 1995 brought the coherency attribute to the attention of the geophysical community at large. The initial method scanned dips within a local image volume, estimated the dominant planar contribution and subtracted it. Gersztenkorn Gersztenkorn and Marfurt (1996) improved the original method by using the eigenvector representation of a local image volume to remove the coherent contribution. Luo et al. 1996 utilize two alternative shemes to yield similar coherency images. Luo's first scheme is based on simple, normalized trace differences while his second scheme calculates the spatial derivative of the traces' instantaneous phase.
In a previous article Schwab et al. (1996), I used Prediction Error (PE) filtering to remove the coherent component of a local seismic image volume. PE filters are data adaptive and tend to whiten the spectrum of their input data Claerbout (1994). In another article in this report Schwab (1997), I discuss a set of PE filters that is designed to remove the plane layer component of a local data volume. Unfortunately, this set of PE filters removes almost all the coherent information and leaves only a white output image that is useless for interpretation.
In this paper, I describe my attempt to control the spectrum of the output of the PE filtering more carefully. In particular, I will attempt to remove horizontally coherent events while preserving most of the temporal coherency. The resulting images are rather disappointing. In general, pre-whitening in the time domain allows us to shape the temporal image spectrum. However, targets, such as fault reflections and salt dome flanks, were spatially predicted and removed along with the locally planar layered volumes. The final coherency images are too white to be useful.
In the first section, I will discuss how the subsequent application of two PE filtering steps permits us to shape the output spectrum. In the second section, I apply PE filtering to a few field data examples and discuss the individual results. PE filtering fails to produce the desired coherency images. In the conclusion, I will discuss my difficulties in finding an optimal parameter choice.