Figure Captions

Noise Suppression

Suppression of Random Noise

Mean filters and median filters are commonly used during interpretation to suppress random noise. One valuable application is the use of dip-steered mean or median filters, which enhance laterally continuous events by reducing random noise yet does not suppress details consistent with geologic structure. Such a filter spans a defined number of data samples (N) within an aperture that follows local dip and azimuth and replaces the amplitude of the central sample with the median value of all N amplitudes spanned by the filter. Median filters can be applied iteratively, reducing random noise at each successive iteration, but they do not significantly increase the high-frequency geological components of the surface along which they operate.

Dip-steered mean filters work well on prestack data in which discontinuities appear as smooth diffractions, but they tend to smear faults and stratigraphic edges on migrated data. Dip-steered median mean filters work somewhat better, but they too can smear faults. Structure-oriented filters operate parallel to reflectors and do no filtering or smoothing perpendicular to a reflector.

Suppression of Acquisition Footprint

An acquisition footprint is defined as any amplitude or phase anomaly observed in seismic data that correlates to surface data-acquisition geometry rather than to subsurface geology. Spatially periodic changes in stacking fold, source-receiver azimuths and source-receiver offsets cause spatial periodicity in enhanced seismic signal and in noise rejection. Most seismic attributes react to these periodic changes in seismic data quality and create artifacts that mimic the source-receiver geometry.

One of the simplest methods for suppressing data-acquisition footprints is to apply kx-ky filters on seismic amplitude time slices. We show an example of this type of noise suppression on Figures 1a and 1b, where much of the vertical striping seen on the amplitude data exhibited on Figure 1a is removed on Figure 1b. Attributes calculated from seismic data that have no acquisition footprint do not display acquisition geometry artifacts and provide more accurate geologic interpretation. As an example, notice vertical receiver-line imprints that appear on the time slice of the most-positive curvature attribute on Figure 1c are not seen on the equivalent time slice in Figure 1d after the data-acquisition footprint is filtered from the amplitude data.

Frequency Enhancement

Thin-bed spectral inversion is a process that removes time-variant wavelets from seismic data and extracts reflectivity to image bed thicknesses far below seismic resolution. In addition to enhanced images of thin reservoirs, these frequency-enhanced inverse images are useful for mapping subtle onlaps and offlaps, thereby facilitating the mapping of parasequences and the direction of sediment transport.

In addition to viewing spectrally broadened seismic data in the form of reflectivity, data also can be filtered to any desired frequency bandwidth that allows useful information to be better seen for interpretational purposes.

Depending on the quality of data being interpreted, as well as access to the methods discussed here, data need to be preconditioned to optimize noise removal (whether the noise removal involves random noise or unwanted acquisition footprints) and to achieve optimal frequency-enhancement before attributes are computed. Once such data preconditioning is done, attribute computation then yields attribute maps devoid of artifacts and allows a more accurate geologic interpretation.

To illustrate the importance of data preconditioning, Figure 2 shows stratal slices from coherence volumes run on (a) input data, (b) input data with inverse Q filtering, (c) spectrally whitened input data, and (d) input data transformed to filtered thin-bed reflectivity inversion.

Notice these coherence slices show increased resolution in this a-b-c-d order of data preconditioning, with the highest lateral resolution seen for coherence computed from filtered thin-bed reflectivity inversion.

We emphasize that computation of attributes is not a process that involves pressing some buttons on a workstation, but requires careful examination of input seismic data in terms of signal-to-noise ratio, noise contamination, and frequency content.

Conclusions

In our studies, we find that:

• Attributes calculated from seismic data that have a high signal-to-noise ratio and that are processed for acquisition footprint suppression exhibit geological features clearly without any masking.
• Enhancement in the frequency content of data volumes occurs in the order shown on Figure 2.

Some of these data-conditioning methods may not be available to an interpreter; we hope these examples assist in decisions about how seismic interpretation software and workstation capabilities may need to be adjusted to improve data interpretations.

Acknowledgment

We wish to thank Arcis Corporation for permission to present these results.

Return to top.