Interpretation of subsurface geology is greatly enhanced by 3-D seismic data, and this accounts for its ubiquity in today’s search for hydrocarbons. Seismic interpretation has two fundamental disciplines at its core: seismic geomorphology and seismic stratigraphy. Plan views allow the interpreter to apply principles of geomorphology, based on analogies with modern sedimentary systems, to interpret depositional environments and even predict facies distributions. On the other hand, section views require stratigraphic interpretation and give insight into stratal architecture and the temporal development of the depositional system. Both approaches give insight into the geological processes that formed the hydrocarbon play (eg Posamentier 2003).

This paper explores some of the techniques available today to take this insight to deeper levels than ever. The widespread availability of powerful computers and high performance graphics have opened new doors to the seismic interpreter, allowing him or her to test and tweak parameters to fine-tune an attribute, and to make unique and compelling displays to help them tell their story. Many of these techniques are especially potent if time is short: a quick, simple interpretation of a couple of easily-interpreted horizons is enough input to guide a rich assemblage of sophisticated algorithms through the dataset, revealing hidden patterns and subtle features in a matter of minutes. If more time is available, techniques such as 2-D seismic modelling can shed light on difficult interpretation problems and give m o re confidence that he or she is indeed interpreting geological features and not geophysical artefacts. We present examples of both of these approaches in this paper.

Fig. 01
Figure 1. Dip and strike seismic lines across the sand ridge feature that forms the reservoir interval. This subtle feature is discernible on the map on the right (outlined) and has a distinctive seismic expression that reflects its complex internal character and reveals a challenging reservoir characterization problem. A carbonate platform that existed before deposition of the reservoir sediments is a dominant feature of the map. The top of the key reservoir zone is shown as a green seismic surface. The zone of interest is enveloped by the blue and orange seismic surfaces which were used for calculation of interval attributes and isochores. On the map, red is shallow, blue is deep; on the seismic sections, black is a peak indicating a positive acoustic impedance contrast. [The images are from PowerView.]

The example seismic dataset, shown in Figure 1, is a migrated full-stack, filtered, 16-bit floating-point amplitude volume from a producing oilfield in western Canada. We interpreted three horizons with sparse seed lines and an autopicking tool; all three horizons were finished in under an hour. The horizons were interpolated and smoothed and are used for building a velocity model for depth conversions, and for guiding the attribute extraction algorithms along structure. The reservoir interval is interpreted as a tidal sandridge complex deposited during a lowstand and transgression in the early part of the Mississippian. Offshore shales are unconformably overlain by heterolithic lower- and then sandy upper-shoreface sandbodies, which are subsequently partially reworked and redeposited as tidal sand-ridges during a period of transgression. These sands are then conformably overlain by the ensuing highstand shales. The sand ridges are around 20–40 km long, 3–7 km wide, and 20–30 m thick, and these are the subject of our investigation.

Seismic Stratigraphy


Seismic volumes are complex volumes of reflections, interference and tuning effects. It can be difficult to know what to expect a given sandbody or stratigraphic geometry to look like in the seismic. Seismic modelling can give the interpreter this insight, and allow him or her to try multiple iterations in order to best match a known response. Stratigraphic interpretations can then be made in areas where there is little or no well control.

A workflow for seismic modelling to illuminate stratigraphic relationships is illustrated in Figures 2–6.

Fig. 02
Figure 2. This seismic section shows wells that penetrate a stratigraphic reservoir. On either side of the well bore are displayed a sonic curve and the corresponding synthetic. The geology has been tied to the seismic aligning synthetic character with seismic signatures. The 2 wells on the right clearly show matching seismic and synthetic reflections, however the well on the left does not, even though it seems to penetrate the reservoir (in circles). [Image from SeisVision, synthetics fro m LogM.]
Fig. 03
Figure 3. This is a stratigraphic geological cross section of the same 3 wells we see in the seismic section from Figure 2. The sonic curves are displayed in the tracks and between the wells is an interpolated colour fill based on the sonic curve values highlighting the slowness of the data in microseconds per metre. Reds are low velocity and blues higher velocity. The low velocity sand body shown in this figure tapers towards the left well where we had no synthetic reflection in Figure 2. [Image from GESXplorer Cross Section.]
Fig. 04
Figure 4. To get a clear understanding of the correlation between the geological stratigraphy and the seismic, modeling can be a key element. The middle image (Step 2) in Figure 4 is a seismic trace model generated from the cross section in Figure 3. Again this model shows a leftward tapering of the same sand body identified in Figure 3. By cropping a representation of this model and overlaying it on the seismic in the zone of interest (Step 3) you can visually validate the correlation between the geology and the seismic; thus giving you an invaluable confidence in understanding the stratigraphic seismic signatures that are present. [Images from SeisVision, model created in LogM. ]
Fig. 05
Figure 5. Well number 1 (the left-most well in Figure 2) still leaves a lot of uncertainty regarding where the last synthetic reflector sits on the upper part of the reservoir, leaving little indication of where the sandbody may lie in the seismic. Is this because our sonic curve is incorrect, because we are on the edge of the reservoir? What if we increased the sand body thickness in the well—what synthetic response would that give us? Here we have taken well number 1 and copied it 4 more times in a cross section, displaying the sonic curve and the synthetic for each well. In the middle well we have added about 20 metres below the top of the sandbody, having started at 5 metres first, and then increased to 10 m, 15 m and finally 20 m until we got the synthetic response expected which indicates a significant positive reflection. We can now take this cross section and model it to generate an interpolated seismic trace model and compare it to the real seismic. [Image from LogM Stratigraphic Model Builder. ]
Fig. 06
Figure 6. Step 2 is the result of the seismic trace model generated from the cross section in Figure 5. We can see the sand wedge the model has created right where we have added 20 metres of sand in the zone of interest from the middle well. By cropping a representation of this model, capturing the sand wedge and overlaying it on the seismic in the zone of interest (step 3) you can align the correlation polygon by matching the strong upper reflectors from the model with the strong reflectors on the seismic in the top part of the reservoir. As a result we can see that the sand wedge matches the intermittent seismic anomalies reflecting the sand body of the reservoir. This workflow scenario, using ‘what if’ approaches, validates the stratigraphic relationships of the geology with the seismic anomalies of a potential target. These results are invaluable in helping an interpreter feel confident of his or her interpretation. [Images from SeisVision, model created in LogM.]

Volume seismic attributes

Attributes drawn from a group of seismic traces in a 3-D volume are capable of representing complex three-dimensional reflection characteristics and relationships. For example, it is possible to generate a volume representing the amount of dip or degree of parallelism displayed by the reflectors in the seismic. These are calculated directly fro m the seismic traces, with no interpretive input (though the quality of the results are improved by the use of a three-dimensional velocity model, which normally would incorporate some surface interpretations to guide the velocity fields). Other such attributes quantify aspects of the reflectors such as continuity, hummockiness, divergence and azimuth. It is even possible to combine and reprocess these attributes to produce new ones, such as shaded relief images (Barnes 2003; see example in Figure 7a). These attributes all represent aspects of apparent palaeotopography, which can aid in geological interpretation. This is especially true when the results are combined with other images. Two effective ways of achieving this are semi-transparent overlays (Figure 8) and colour blending (Figure 9). As with all 3-D seismic datasets, using these visualization techniques in a 3-D space is more powerful still.

Fig. 07
Figure 7. Multi-panel seismic attribute analysis. Commonly, the interpreter generates a large number of seismic attributes, most of which are displayed as maps. These can be cumbersome and time-consuming to compare, but views like this one can make the job easier. The maps can be zoomed and panned together or individually, and each has its own colour-bar. a) Shaded relief attribute: a 3-D view generated directly from the seismic, without any interpretation; b) an isochore map, the result of subtracting one horizon from another and converting the result to thickness with interval velocity calculations—produced in a single step from the original time horizons; c) a timeslice through the amplitude volume; d) a waveform classification map showing seismic facies tracts—each colour re p resents a discrete class of waveform shapes based on the result of a standard K-means test; e) a red-green-blue blended image showing the amplitude responses at three different frequencies in a single view; f) a bedding thickness prediction map generated from the first spectral peak frequency attribute; g) a vertical section through the seismic amplitude volume showing the interval of interest along the green Top Reservoir horizon. The blue and orange horizons were used for calculating interval attributes. [The tools used are PostStack Shaded Relief, PA L Waveform Classifier, SpecDecomp, PowerCalculator, and the image is from PowerView. ]
Fig. 08
Figure 8. Red-green-blue blended image showing three seismic volumes simultaneously. Red represents the Hummockiness attribute, green is Parallelism, and blue is Dip. The resulting colours represent the varying reflection patterns in the seismic. They could assist in solving correlation problems, or qualitatively compare seismic facies. [The volumes are PostStack reflection pattern attributes, and the image is from PowerView. ]
Fig. 09
Figure 9. Comparison of two seismic volumes. A black-white seismic amplitude image is shown, with a seismic attribute volume overlain. The attribute is Dip and is displayed with the scale shown to the right of the image. No interpreted horizons are provided to the algorithm, but a three-dimensional velocity model is used. In this display, the Dip volume is semi-transparent, allowing the interpreter to see both volumes at the same time. This can assist with resolving correlation issues and with qualitative stratigraphic analysis. [Dip is a PostStack reflection pattern attribute, and the image is from PowerView. ]

Waveform Classification

An important component of seismic stratigraphy is recognizing systematic variations in the character of a seismic reflector or set of reflectors. Such variations might reflect changes in stratal geometry or lithologic stacking patterns. This can be a powerful way of elucidating subsurface stratigraphy qualitatively and, when coupled with seismic modelling, quantitatively (eg Williamson 2003). An important feature of waveform classification is that wavelets are compared by shape, not by amplitude. This can mean that waveform classification is less affected by tuning effects and porosity variation, for example, than other types of seismic attribute.

There are two fundamental approaches to waveform classification: supervised and unsupervised. Supervised approaches require the interpreter to select a time-window (usually centred on a seismic horizon) and provide any number of reference waveforms (as an inline/crossline location). These are often chosen to be representative of distinct lithologies at nearby wells. For example, one might provide the location of an especially thickly-developed reservoir interval, a thinner reservoir interval, more heterolithic reservoir, a shaled-out reservoir section, and an eroded, unconformable section. These waveforms represent 5 discrete classes. Every trace in the seismic volume is then classified according to which of the reference waveforms it most closely resembles; there is also a ‘zeroth’ or reject class for traces which resemble none of the references. The result is a map of wavelet classes in which each class is represented by an integer.

Fig. 10
Figure 10. Waveform classification analysis for a 40 ms window around the Top Reservoir surface using a standard K-means testing algorithm for six classes. (a) shows a horizon-guided waveform classification map. Each colour represents a discrete class of trace shapes, based on a statistical classification of the analysis window in every trace. Like colours represent like wavelets. The sand ridge complex is clearly differentiated from other areas of the map, with some internal differentiation also visible. (b) again shows the classification map, but semi-transparent and overlain on an attribute re p resenting the confidence with which the waveforms were classified, and corresponds to the probability of the given trace being in the assigned class. Darker areas were classified with lower confidence. (c) shows the same classification map, again semi-transparent, but this time overlain on a timeslice from the amplitude volume. Views like this can help the interpreter distinguish seismic facies variations from other effects and data quality variation. [The tool used is PAL Waveform Classifier, and the images are from PowerView. ]

The second approach, supervised classification, automatically calculates a reference wavelet from the data before assigning classes. This is achieved by an iterative statistical analysis of a subset of the data, the results of which are then applied to the entire dataset. As before, the result is a map of waveform classes, such as those shown in Figure 10, with each class being re p resented by an integer. The classes are organized according to relative similarity, so class 1 more closely resembles class 2 than it does class 5. This approach is especially useful if there are no wells in the area of interest, or if the interpretation is in the early stages and no firm correlations with wells have yet been established.

Seismic Geomorphology

Spectral decomposition

Ordinary 3-D seismic data typically has a 60–80 Hz bandwidth, so it contains energy reflected from the subsurface at a wide range of frequencies, all of which are compounded in a typical seismic volume. In certain circumstances, especially subtle stratigraphic plays, it may be helpful to see the amplitude (or phase) of reflections at particular frequencies. These tuned amplitudes may tell the interpreter something about the physical spacing of the acoustic impedance contrasts in the image—in other words the bedding thicknesses (eg Partyka et al., 1999). There are also examples of mapping the attenuation of high-frequency signal by hydrocarbons, allowing their direct detection (eg Castagna et al., 2003). Spectral decomposition analysis allows the interpreter to quantify amplitude variation with frequency, and thereby gain insight into the distribution of different stratigraphic packages and/or hydrocarbons.

There are three key spectral decomposition workflows. The first is to process a discrete window a round a very smooth seismic horizon interpretation, transforming the amplitude or phase data into the frequency domain (in other words, a new volume results, with frequency represented by the z-direction; a typical volume might contain 100 slices, representing amplitude or phase at 1–100 Hz). Such a volume is called a ‘tuning cube’. There are two helpful ways of visualizing this dataset. The first is simply to animate through the volume, exactly as one would animate through timeslices. This yields images similar to those in Figure 11 a – 11c. The second visualization technique is to blend three timeslice images at a time, using the techniques illustrated in Figures 11d and 11e. Such blended images can provide an attractive and data-rich snapshot of very subtle reservoir characteristics. This workflow should help establish which features of the reservoir interval are brought out by which frequencies.

Fig. 11 a,b,c Fig. 11 d,e
Figure 11. Spectral decomposition analysis using blended map images. The analysis was performed using the discrete Fourier transform method on a 60 ms window centre d on the Top Reservoir horizon, with a Gaussian taper (required for eliminating spurious high-frequency responses from the edges of the window). (a)–(c) show amplitude at 30 Hz, 40 Hz and 50 Hz respectively. A bright response is represented by white. (d) is a red-green-blue (RGB) blended image comprising of the individual images above displayed in red, green and blue respectively. The colours are additively combined to produce the full-spectrum image shown. For example, yellow hues indicate that the re d (30 Hz) map and the green (40 Hz) map have coincident high amplitude responses in that area. Likewise, white indicates that all three frequencies are responding.(e) is a hue-saturation-brightness (HSB) blended image. The same three amplitude slices are re p resented by hue (colour), saturation (colour intensity), and brightness (or colour value) respectively. The images are then combined to give the final map; a bright, saturated red colour indicates that all three frequencies are ‘bright’ in that area. The interesting thing about these images is that while they re p resent the same datasets, they illustrate quite different characters of the reservoir interval. For example, the RGB blended image seems to give a more definite sense of where the sweet-spot might be for the reservoir, whereas the HSB-blended image resolves the ‘Swiss-cheese’ seismic texture very well. [The tool used is SpecDecomp, and the images are from PowerView. ]

The second workflow is to leave the data in the time domain, but to process a series of time windows (a ‘running window’). Typically, a flattened seismic volume is used, with the same guiding horizon as a datum. The result is a flattened volume which is tuned to a specific frequency or set of frequencies, such as those established as useful or revealing in the tuning cube workflow. The reason for using a flattened volume is to give insight into the geological evolution of the interval. Animating through a tuned volume shows the same reservoir subtleties revealed in the first workflow, but now with the added dimension of geological time.

The third workflow gives a valuable quantitative estimation of bedding thickness in the analysis window, without the need for accurate seismic interpretation or phase concerns. Figure 12 shows an example of such a thickness estimation map, derived from the first spectral peak frequency attribute. This attribute is simply the frequency of the first peak in the amplitude spectrum . This is an important number, because it is exactly half the value of the spacing of the interference ‘notches’ in the spectrum. The notches are, in turn, spaced in inverse relation to bedding thickness in the analysis window. See Partyka et al. (1999) for a full explanation of the physics of this analysis. A simple horizon calculation can therefore transform the first spectral peak frequency attribute into time thickness, or, given the interval velocity of the zone, vertical thickness. Since this is an independent measure of bedding thickness, this workflow is potentially an important risk reduction tool for the explorationist. The approach is especially powerful when coupled with synthetic modelling and/or crossplotting the predicted thicknesses with actual vertical thicknesses as measured in offset wells.

Fig. 12
Figure 12. Thickness prediction from spectral decomposition analysis. The discrete Fourier transform is guided through the dataset by the Top Reservoir horizon which has had a very aggressive smoothing filter applied to it. As such, the results are almost completely independent of interpretation. Because of this factors, spectral decomposition can be an important risk reduction tool in exploration. The first spectral peak frequency of the data is a by-product of the decomposition process, and, since it is inversely proportional to time thickness, can easily be transformed into a bedding thickness prediction map. This map clearly shows the sand ridge running in a NE-SW orientation. The apparent decrease in bedding thickness across the reservoir is explained by the discrete nature of the depositional sandbody, in which the reservoir zone is typically 20–25 m thick, compared to the thickly-bedded shales of the ‘swale’ regions, which contain few significant acoustic impedance contrasts. [The tools used are SpecDecomp and PowerCalculator, and the image is from PowerView. ]

These workflows, well-proven by several years of application in the industry, can give valuable insight into the internal organization of the reservoir interval. We have found the images and animations particularly effective communication tools, particularly in plays with strong stratigraphic components such as channel sands and deep marine gravity deposits.

Instantaneous seismic attributes

There is an almost bewildering number of instantaneous seismic attributes, each of which describes some aspect of a single trace at a single time or time-window. Examples include simple attributes such as phase or reflection strength, and more sophisticated interval statistics such as number of zero-crossings or total time thickness above a given amplitude. Maps of these attributes can often be interpreted geologically, with some features drawn out by particular examples. The ability to view many maps simultaneously, with a range of different colourbars and zoom-levels, can be a significant aid to interpretation. An example is shown in Figure 7, in which six maps are compared side-by-side. This allows the interpreter to quickly spot correlations or discrepancies between attributes which may have stratigraphic explanations. Opacity overlays and blended images may also help with interpretation, as previously illustrated in Figures 7–11.

Fig. 13
Figure 13. Volume visualization and interpretation give new insight into the spatial arrangement of stratigraphic features and broaden the interpreter’s perspective. This makes it possible to interpret seismic or plan well trajectories interactively and dynamically, with input from the geophysicist, geologist and engineer, each of whom can see their own data in the same space. This allows them to significantly reduce risk and make complex decisions with much more confidence. Images like the one shown are replacing the venerable contour map as the fundamental decision tool. See Meyer et al. (2001) for more examples. [The tools used are GeoProbe and Wellbore Planner. Data courtesy of Seitel Inc.]

Another technique for multi-attribute analysis, and for correlating seismic attributes with reservoir properties or with each other, is crossplotting. An in-depth review of this technique is beyond the scope of this paper, but many examples can be found in the literature. Establishing a relationship with a reservoir property is especially useful, since it allows the interpreter to model and predict the property directly from the seismic, with a known and quantifiable level of certainty. This makes the analysis of exploration risk considerably easier.

The Power of Volume Interpretation

While this paper is focused on two dimensional views of the seismic data, volume visualization and interpretation is now well-established as an interpretation discipline. A 3-D seismic volume has not been truly interpreted until it has been volume interpreted (Figure 13). Despite this, it remains curiously ‘niche’ or somehow ‘advanced’ in the Canadian geoscience community (but that’s another story!). Two basic aspects of volume interpretation make it an important tool for the stratigraphic investigator: rapid volume interpretation and surface visualization.

The ability to instantaneously track an amplitude anomaly or apparent pinchout, and see the results in three dimensions alongside key reference data such as well logs and fault interpretations transforms the slow, methodical, line-by-line interpretation task into a dynamic and much more creative process. Hunches can be instantly verified, or disproven, and leads can be rapidly generated and investigated. More powerful still is the ability to bring multiple seismic attributes to bear simultaneously on a problem. Some great examples of applying these techniques to stratigraphic problems are shown in Meyer et al. (2001) and Alexander et al. (2001).

Surface visualization is an important tool for seismic geomorphological studies (for example, see Posamentier 2003). Viewing angle, lighting direction, colour, texture, and vertical exaggeration can be instantly adjusted to bring out the particular features of a seismic horizon. Misties and other anomalies can be seen and corrected. Attributes can be overlain on structural surfaces and visual correlations made. The interpreter can learn more in minutes of playing in 3-D than in days of looking at maps.


A very rich set of tools is available to the seismic interpreter for gaining insight into stratigraphy. Which tools one chooses to apply depend on the play type being investigated, the amount of time available for analysis, the stage in the interpretation process, the quality and nature of the available data, and the preferences , experience and intuition of the individual. There is no formula for the ‘right’ attribute for a given situation. In fact, our own approach is to build as many models, look at as many attributes, and try as many parameters as time allows. Time saved by automatic surface interpretation and volume interpretation can be put to good use by applying the available tools, letting the data itself offer up the answers. This seems like a better use of time, and technology, than the frustrating process of line-by-line interpretation of difficult surfaces, the results of which are always risky and uncertain.



We would like to acknowledge Landmark Graphics for permission to publish this paper. PowerView, PowerCalculator, SpecDecomp, PostStack, GeoProbe, Wellbore Planner, SeisVision, LogM and GESXplorer are trademarks of Landmark Graphics.


About the Author(s)

Eric Trouillot is manager of business development for GeoGraphix at Landmark Graphics Canada. He received a D.U.T. in Civil Engineering in France where he practiced field engineering before moving to Calgary in 1988. Since then he has entered the high tech world of software development in the oil and gas industry, accumulating over 16 years of experience between Digi-Rule, GMA, GeoGraphix and Landmark.

Matt Hall gained his B.Sc. degree in geology from the University of Durham, UK, in 1993 and his Ph.D. in sedimentary geology from the University of Manchester, UK, in 1997. He joined Statoil as a geologist in Stavanger, Norway, working on the Tertiary of the North Sea. In 2000 he moved to Canada and joined Landmark Graphics, where he is now responsible for business development in prospect generation systems.


Alexander, C, et al. (2001). The Plutonio discovery, Block 18, Angola—A3-D visualization and multiattribute approach to exploration success. The Leading Edge, December 2001.

Barnes, A (2003). Shaded relief seismic attribute. Geophysics 68 (4), July-August 2003, p 1281–1285.

Castagna, J, S Sun & R Siegfried (2003). Instantaneous spectral analysis: Detection of low-frequency shadows associated with hydrocarbons. The Leading Edge, February 2003.

Meyer, D, et al. (2000). Use of seismic attributes in 3-D geovolume interpretation. The Leading Edge, December 2001.

Partyka, G, J Gridley & J Lopez (1999). Interpretational applications of spectral decomposition in reservoir characterization. The Leading Edge, March 1999.

Posamentier, H (2003). Integration of seismic geomorphology and seismic stratigraphy: principles and applications. CSEG/CSPG Annual Convention Abstracts Volume, June 2003.

Williamson, A, et al. (2003). Quantitative interpretation of neural network seismic facies—Oriente Basin, Ecuador. CSEG/CSPG Annual Convention Abstracts Volume, June 2003.

Wood, L, et al. (2000). Seismic attribute and sequence stratigraphic integration methods for resolving reservoir geometry in San Jorge Basin, Argentina. The Leading Edge, September 2000.


Join the Conversation

Interested in starting, or contributing to a conversation about an article or issue of the RECORDER? Join our CSEG LinkedIn Group.

Share This Article