GCThe Use and Abuse of Seismic Attributes*
Hans E. Sheline1
Search and Discovery Article #40143 (2005)
Posted March 1, 2005
*Adapted from the Geophysical Corner column in AAPG Explorer, January, 2005, entitled “Don’t Abuse Seismic Attributes” and prepared by the author. Appreciation is expressed to the author, to Alistar R. Brown, editor of Geophysical Corner, and to Larry Nation, AAPG Communications Director, for their support of this online version.
1VeriNova, Sugar Land, Texas ([email protected])
Seismic interpretation is a cornerstone of our industry, as interpretation success has grown increasingly dependent on ever-newer combinations of seismic attributes (SAs). Attributes are simply defined as information extracted or computed from seismic data. What combinations work best depend on reservoir characteristics, the available data, and, most importantly, human expertise.
Seismic attributes are not magic, but the explosion of 3-D seismic at the end of the 20th century resulted in dramatic increases in the types, combinations, and uses of SAs (Figure 1). We now have available multi-trace, prestack, horizon, wavelet, and 4-D attributes, in addition to those derived from shear wave volumes. These allowed for significant improvements in estimates of reservoir properties from seismic (RPFS).
Figure Captions and Table Caption
Table 1 defines terms used in seismic attribute analysis today. The explosion of potentially useful attributes requires seismic interpreters to keep current and to have the most effective, efficient “fit for purpose” work flows. One important aspect of these work flows is “starting with the end in mind.” For example, if amplitudes or wide azimuths are critical, the seismic acquisition and processing for those SAs should be optimized.
Unfortunately, the potential for abusing seismic attributes also has increased. One common abuse of seismic attributes often occurs “because it’s there.” Interpreters today have access to many SAs on their workstations but often have very little time to understand these attributes properly, model them, and correctly correlate them with ground truth and the principles of physics.
Be wary of pretty SAs that are not well understood. This can damage your credibility while tarnishing the true potential of SAs. Don’t expect your workstation to pop out the solution. Be wary of “black box” answers. Instead, commit the resources to correlate, model, and understand your SAs and what they can and cannot do.
Workstations now make it very easy to generate, for example, the third derivative of the instantaneous phase or the second derivative of instantaneous frequency. Even if this SA correlates with ground truth somehow, will you understand it or trust its significance?
Another abused shortcut often sounds like: “Just give me the one attribute that solves my problem.” In some unusual areas, interpreters have been able to succeed using only a single attribute interpretation. However, I have not yet found an area where a single attribute provides the optimum answer.
Note in Figure 2 how none of the four attributes alone shows the sand channel very well – but when they are combined, the result is both quantitatively and areally more accurate than any individual attribute. This example shows the dramatic potential of SAs for lithology prediction. We add value by using experience to improve estimates of reservoir properties from seismic (RPFS), reducing risk and helping to quantify uncertainty.
Therefore, avoid grabbing the first attribute or attributes that seem to work. Instead, develop a robust, efficient work flow that quickly considers many of the most promising attributes and objectively correlates them with seismically scaled and corrected ground truth. Then model these attributes to understand and optimally guide the SA combination to estimate the reservoir properties best and quantify the uncertainty of those estimates.
There is also a real danger of using too many SAs to “over-fit” the data. With unlimited attributes – and therefore unlimited degrees of freedom – statistical accidents will occur. The critical step is testing for significance – for example, by blindly dropping one well or zone at a time. The number of attributes ideal for reservoir property estimation typically varies from two to four, depending on the area, data, and objective.
Despite the pitfalls in seismic attribute analysis (SAA), there are many successful examples of predicting RPFS. For example, consider Figure 2: It is often important and valuable to define the 3-D extent of a channel or sand body. Figure 2 shows an example of combining four 3-D attribute volumes along with the appropriate well information to predict lithology.
Once you have optimized your SAA workflow, it can dramatically improve property and risk estimates. Robust work flows have been developed on data sets around the world, in clastic and carbonate environments, onshore and offshore. The accuracy of estimates varies with location, data quality, and objectives.
The speed and accuracy of reservoir modeling and simulation have also been improved using RPFS estimates and associated uncertainty cubes.