Volumetric Attributes and Curvature

Interpreters should distinguish between attributes which are properties of a horizon and attributes
which are extracted from a volume of seismic data using a window. An attribute along a horizon
obviously requires a structurally interpreted horizon, at least a preliminary one, and thus will be
used after a significant amount of tracking has been completed. Much of this chapter has been
addressing this type of attribute, and the uses have centered on verifying and enhancing the
interpretation in progress.

Volumetric attributes, which operate directly on the data themselves, can be used earlier in the
interpretation before a horizon exists. Coherence, which has been discussed at length in this
chapter, is the best example of this type of attribute, and is used primarily for detecting faults. Other
attributes can now also be extracted from the volume and the most significant of these is volumetric
curvature. Chopra and Marfurt (2007) tell us that there are many ways of calculating volumetric
curvature but that the most important are most-positive curvature and most-negative curvature.
These can also be extracted from the curvature volume along a horizon surface to give a horizon
slice in curvature.

Curvature has emerged as an attribute with many applications. It can be used in conjunction with
dip, azimuth, or residual to cross-validate the presence of a subtle fault. It can be used to create a
display emphasizing the relief on a horizon surface (Figure 8-58). For a brittle reservoir like a
carbonate encased in shale, fracturing will probably be most pronounced where the layer is most
bent. Thus a high value of curvature can be used as circumstantial evidence for fracturing in
carbonate reservoirs. Figure 8-59 shows two types of volumetric curvature extracted along a
horizon. Most-positive curvature shows lineations in red, indicating principally channel levees.
Most-negative curvature shows lineations in blue, indicating principally channel thalwegs.

Amplitude Variation with Offset

Amplitude Variation with Offset for the identification of gas is discussed briefly in Chapter 5. AVO
has become a very popular subject and is covered by Castagna and Backus (1993). However, the
amount of AVO work performed in 3-D is extremely small, and Castagna and Backus barely mention
the subject. Furthermore, the 3-D AVO projects which have been performed show little incremental
benefit over the 3-D analysis of the post-stack amplitudes. The example presented in Figures 8-60
through 8-63 is one exception.

Figure 8-60 is the horizon slice displaying conventional stacked amplitude for a gas reservoir in the
deepwater Gulf of Mexico. Figures 8-61 and 8-62 show horizon slices on the same horizon from
individually migrated data volumes incorporating respectively near-offset traces only and far-offset
traces only. The two are similar and both show separation into eastern and western lobes. The
differences between the two do not at first appear significant.

Figure 8-63 is the arithmetic difference between the amplitudes of Figures 8-61 and 8-62; in other
words Figure 8-63 is a display of the AVO attribute far-near amplitude difference. A channel is now
clearly visible in yellow, red and green, which indicates increasing amplitude with offset. This
channel, meandering across parts of both Eastern and western lobes, is not visible on any normal
amplitude display and is interpreted as the sand-filled channel facies of a slope fan. Furthermore, it
probably indicates the areas of highest porosity and permeability within the reservoir.

Use of Multiple Attributes

This chapter has so far focused on the attributes available and their individual uses. Approaches are
now in place for using multiple attributes together in the derivation of a reservoir property map.
Attributes are normally extracted from the relevant level in the seismic data as horizon attributes
and/or windowed attributes over at least one half-period. They are then selected on the basis of
geophysical and petrophysical reasoning; that is, we use attributes which appear reasonable. Each
attribute may be crossplotted against the reservoir property of interest using multiple wells and the
one that correlates best is selected. The other attributes are then tested in turn to find how much
of the remaining variance in the relationship they explain. The statistics then help with the selection
on the basis of their contribution to variance reduction. The resultant attributes, one or several
together, are then used in geostatistical cokriging to interpolate the reservoir property being
mapped between wells.

There are many dangers in statistics, and much has recently been written about these dangers as
applied to reservoir characterization. Hirsche et al. (1997 and 1998) have been particularly vocal
with remarks including: Neglecting geology and geophysics reduces geostatistics to a purely
statistical process that may give false confidence in spurious results. Schuelke and Quirein (1998)
offer a similar caution: The use of statistical methods without the foundation of a physical basis for
the correlation between the seismic attribute(s) and the rock property is very risky. Hart (2002)
makes the same point: There must be a known or suspected link between the attribute and the log-
based physical properties to be imaged. Hart and Balch (1999) remind us that the probability of
obtaining a statistically significant but spurious correlation between an attribute and a log-derived
property is proportional to the number of attributes tested and inversely proportional to the
number of Wells used in the calibration. Kalkomey (1997) discusses this further and calculates
alarmingly high probabilities.

In summary, for a geostatistical reservoir study we would like a large number of wells but we should
not use a large number of attributes. Thus about three, or up to a maximum of five, attributes should
be selected primarily on the basis of geophysical and petrophysical reasoning. Little reliance should
be placed on correlation coefficients. The selected attributes are then submitted to the multiple
regression analysis and cokriging.

Spectral Decomposition

Spectral decomposition (SD) is a technique that breaks down seismic signal into narrow frequency
sub-bands. When these sub-bands are examined in a spatial context (i.e., plan view of a 3-D survey)
they reveal interference that is occurring across the available bandwidth of signal. This approach is
analogous to remote sensing techniques commonly used to image the surface of the earth. Whereas
the remote sensing techniques rely on substantially higher electromagnetic frequencies to map
interference at the surface of the earth, spectral decomposition makes use of much lower seismic
frequencies to image the reflective nature of the subsurface rock mass. Such decomposition
provides greater resolution and detection of the layer-stacking heterogeneity, boundaries, and
thickness variability than are possible with traditional broadband seismic attributes.

The interference we observe in seismic data is controlled by the interaction of a band-limited signal
with local distribution of impedance contrasts (Figure 8-64). This interaction causes geologic
features to tune in at some frequencies and tune out at other frequencies (Figure 8-65). Finding
frequencies at which the geologic features stand out (i.e., either tune in or tune out) from the
background amplitude is the key to the successful application of spectral decomposition.

A reflective layer in the subsurface can be thought of as a boxcar in time. A boxcar in one domain
transforms to a sinc function in the other domain. For example, a steep-sided boxcar-shaped filter
in the frequency domain transforms to a sinc waveform in the time domain. A layer/boxcar in the
time domain transforms to a sinc function in the frequency domain. Thus a single layer in time
becomes a sinc function in frequency, and multiple layers form a summation of sincs. The
mathematical expression of the layer-to-sinc transform allows examination of the relationship
between layer-stacking combinations and the associated spectral signatures.

Interpretation using only full-bandwidth amplitude means disregarding the sub-band interference
information that is available. This is much like the loss of information that occurs when stacking AVO
gathers that exhibit significant amplitude changes with offset. The goal of processing in preparation
for SD work is to maintain as broad a signal bandwidth as possible. If possible, utilize AVO processing
to preferentially image rock or fluid properties prior to SD analysis. This broad bandwidth can then
be examined as a series of narrow bands to characterize a zone of interest with significantly more
fidelity than is obtained from a combined full-bandwidth response.

(1) Reconnaissance tuning cube (Figure 8-66):

Transform a short window around the zone of interest from time to frequency. Perform spectral
balancing to suppress the wavelet overprint (Figure 8-67). Display the resulting (x,y,frequency)
volume in plan view as a series of frequency slicesone slice per frequency sub-band.

(2) Running window spectral decomposition:

Select a few key frequency bands that reveal heterogeneity of interest. Generate isofrequency
running window spectral decomposition volumes for the key frequencies. Perform spectral
balancing to suppress the wavelet overprint. Animate up and down in time to identify temporal
positions that best image the zone of interest.

(3) Targeted tuning cube:

Transform a short window around the zone of interest from time to frequency, using targeted
parameters. Perform spectral balancing to suppress the wavelet overprint. Display the resulting
(x,y,frequency) volume in plan view as a series of frequency slicesone slice per frequency sub-
band.
 Significance of Spectral Balancing

A raw tuning cube contains three main components:

layer interference (the geologic heterogeneity part of most interest to the interpreter)

noise, and

the seismic wavelet

Frequency slices of amplitude in narrow frequency bands above and below the dominant frequency
of the seismic wavelet will appear significantly weaker in amplitude than slices extracted at
frequency bands near the dominant frequency. To remove this effect, spectral balancing is applied
(Figure 8-67) by determining the median amplitude in each frequency slice and scaling to a target
amplitude value. Spectral balancing suppresses the seismic wavelet overprint and reveals the
underlying geologic heterogeneity while not destroying the spatial variations in interference. What
is left behind is the reflective subsurface amplitude heterogeneity and noise. It provides a view of
the zone-of-interest reflective energy that is no longer biased towards just the dominant frequency
of the wavelet. In other words, the spectral balancing pro-cess allows every frequency band to be
exposed to the same degree, although higherfrequency slices will tend to contain progressively
more noise because the signal-tonoise ratio normally decreases with increasing frequency.

Interpretation Recommendations

(1) Dont limit the interpretation to one frequency band. Animate through a tuning cube after
spectral balancing. Make sure to animate from the lowest frequency all the way through the usable
frequency range. Geologic insight comes from the variations in amplitude from frequency to
frequency (Figure 8-68). The first notch, marked in orange, is initially seen in the thickest part of the
channel. This notch moves to thinner portions of the channel towards the outside with increasing
frequency. Note all the channels visible on the horizon slice of Figure 8-69, and then note the
different appearance of most of the channels on the frequency slices of Figures 8-70 and 8-71.

(2) Tuning out is just as important as tuning in. Dont limit the interpretation to features that are
just tuning in (i.e., higher amplitude than the background). Subsurface features will tune in at certain
frequencies and tune out at other frequencies. Examine the heterogeneity that is exposed in both
the tuning-in and tuning-out cases.

(3) Consider red-green-blue blending of amplitude from different frequencies. Figure 8-72 combines
three frequency slices, two of which are the slices shown in Figures 8-70 and 8-71. Although not as
complete a picture of the zone-of-interest interference as a full tuning cube, RGB blending provides
an image that is better than limiting the interpretation to a single band or to a single full-bandwidth
attribute. Note, in Figure 8-72, that the centers of the channels are predominantly blue, indicating
increased sedimentary thickness. Figures 8-73 and 8-74 both combine frequency slices from three
frequencies in the same way and demonstrate a channel system at two different levels. Figure 8-75
combines three levels of the channel system for one frequency. The progression red to green to
blue takes the observation from younger to older sediments and thus deeper into the valley fill.
(4) Interference is a first-order effect in seismic data. Attenuation does occur, but so does
interference. Consider interference before jumping to the conclusion that a frequency anomaly is
caused by attenuation. Consider, for example, a blocky, lowimpedance, brine-filled reservoir layer
embedded in a higher-impedance material. If we replace the brine with lower-impedance
hydrocarbon, the time thickness and amplitude will both increase. For temporal thicknesses in the
range shown in Figure 8-65 and frequencies less than 50 Hz, increasing the temporal thickness will
cause a shift in the dominant frequency to a lower value. The higher-frequency content has not
been attenuated, it is simply being tuned by the temporal thickness that is present.

(5) Strong amplitude events dominate the interference pattern. The spectral interference pattern is
dominated by the strongest amplitude events in the analysis window, as demonstrated in Figure 8-
76. If additional length does not include significant additionalreflection energy, lengthening the
analysis window will not change the overall interference pattern.

(6) Dont forget about the phase component. Transforming data from time to frequency domain
produces both amplitude and phase spectra. The phase is very useful for imaging lateral
discontinuities, both structural and stratigraphic, because phase instability exposes edges. Note the
detailed information, especially the faults, in the frequency slice in phase shown in Figure 8-77.

(7) Tuning cube animations provide insight regarding stratigraphic and structural edges, as well as
relative temporal thickness. Amplitude contours move from thicks toward thins as you proceed from
low to high frequency. Subtle discontinuities and reflection heterogeneity are often best imaged at
higher frequencies, as long as a reasonable signal-to-noise ratio exists at those higher frequencies.

(8) Low-frequency components often provide better imaging in poor data reas because the highest
signal-to-noise ratio often exists at lower frequencies. Those lower-frequency slices are sometimes
able to provide better images of the subsurface.

(9) Layer-stacking patterns exhibit characteristic spectral signatures. Changes in the reflective
interfaces sampled by the analysis window show as changes in interference pattern. For the case of
a single layer of reflective material in an analysis window, the period of the notching in the
amplitude spectrum is inversely proportional to the temporal thickness of the layer. Changing the
layering by shifting any one layer boundary within the layer stack alters the frequency domain
interference profile in a characteristic manner.

Resolution Benefits of Spectral Decomposition

Traditional approaches to seismic resolution (separability and visibility) relate resolution limits to
the seismic wavelet shape and its associated tuning frequency and tuning thickness (Chapters 1 and
6). With spectral decomposition the wavelet shape is removed during the spectral balancing step.
Resolution is then no longer controlled by the wavelet. It is instead dependent on the available signal
bandwidth (higher frequencies give increased resolution), signal-to-noise at each frequency (higher
signal-tonoise ratio yields better imaging), complexity and gross thickness of the target (simpler
layer-stacking patterns require less bandwidth to be detected), and the strength of the impedance
contrasts (the higher the contrast, the stronger the reflected energy, and the higher the signal-to-
noise).
 Expresiones de gratitud

We thank the SEG and the SEG Distinguished Lecture Program for permission to include material
from the SEG Distinguished Lecture on Spectral Decomposition; BP for their permission to publish
this material as part of the 2005 SEG Distinguished Lecture; teams at BP and Amoco for collaboration
and support that led to the development of Spectral Decomposition; Chuck Webb and Craig Cooper
for recognizing the value of this technology early and for providing the positive feedback and
support that enabled its development, and OpenGeoSolutions Inc. for providing a framework for
continued research and development of spectral decomposition and related technology

Fig. 8-58. Curvature on a horizon surface, emphasizing drape over pinnacle reefs.

Fig. 8-59. Most-positive curvature extracted along a horizon (above) and most-negative curvature
extracted along the same horizon (below), showing lineations associated with channels. (Courtesy
Arcis Corporation.)

Fig. 8-61. Horizon slice in amplitude from near offsets only for a gas reservoir in the deepwater Gulf
of Mexico. Note the separation of the reservoir into eastern and western lobes. (Courtesy Amoco
Production Company.)

Fig. 8-60. Horizon slice in conventional stacked amplitude for a gas reservoir in the deepwater Gulf
of Mexico. (Courtesy Amoco Production Company.)

Fig. 8-62. Horizon slice in amplitude from far offsets only, for comparison with Figure 8-61. (Courtesy
Amoco Production Company.)

Fig. 8-63. 3-D AVO horizon slice in far-near amplitude difference; that is, the subtraction of Figure
8-61 amplitudes from Figure 8-62 amplitudes. Reds, yellows and greens indicate amplitude
increasing with offset. The pattern in these colors indicates a channel not visible in normal amplitude
and probably containing the highest porosities and permeabilities in the fan. (Courtesy Amoco
Production Company.)

Fig. 8-64. Amplitude spectrum of a thin-bed reflection. From Partyka et al. (1999). Used by
permission.

Fig. 8-65. A simple wedge model illustrates the interference that occurs as a function of temporal
thickness and frequency. From Partyka et al. (1999). Used by permission.

Fig. 8-66. The tuning cube is a frequency domain representation of a target zone within the seismic
data. From Partyka et al. (1999). Used by permission.

Fig. 8-67. Spectral balancing removes the seismic wavelet shape in a way that does not destroy the
spatial variations in interference. From Partyka et al. (1999). Used by permission.
Fig. 8-68. A tuning cube view of a simple channel cross-section. Animating through the tuning cube
from low to high frequency causes amplitude contours to move from thick to thin. From Partyka
(2005). Used by permission

Fig. 8-69. Horizon slice showing full-bandwidth amplitude along a Pleistocene horizon from the Gulf
of Mexico. From Partyka et al. (1999). Used by permission.

Fig. 8-70. Frequency slice showing amplitude in a narrow frequency band around 16 Hz, using the
same horizon as Figure 8-69. From Partyka et al. (1999). Used by permission.

Fig. 8-71. Frequency slice showing amplitude in a narrow frequency band around 26 Hz, using the
same horizon as Figure 8-69. From Partyka et al. (1999). Used by permission

Fig. 8-72. Frequency slice showing amplitude in narrow frequency bands around 36 Hz as red, 26 Hz
as green, and 16 Hz as blue, using the same horizon as Figure 8-69. Such Red-Green-Blue (RGB)
blending provides a way to simultaneously view three frequency components. From Partyka (2005).
Used by permission.

Fig. 8-73. Frequency slice from offshore Africa showing amplitude in narrow frequency bands
around 60 Hz as red, 50 Hz as green, and 40 Hz as blue, at 80 ms above a picked horizon. From
Partyka (2005). Used by permission.

Fig. 8-74. Frequency slice showing amplitude in narrow frequency bands around 60 Hz as red, 50 Hz
as green, and 40 Hz as blue, from 40 ms deeper than the level shown in Figure 8-73. From Partyka
(2005). Used by permission.

Fig. 8-75. Frequency slice showing amplitude in a narrow frequency band around 60 Hz at 80 ms
above the picked horizon as red, at 40 ms above the picked horizon as green, and at picked horizon
as blue. In this case, red, green, and blue allow you to look progressively deeper into the valley fill.
From Partyka (2005). Used by permission

Fig. 8-76. Strong amplitude events dominate the spectrum. Whether we use a shorter analysis
window or longer analysis window, the strong reflection dominates the interference pattern. From
Partyka (2005). Used by permission.