IPA03-G-004

PROCEEDINGS, INDONESIAN PETROLEUM ASSOCIATION

Twenty-Ninth Annual Convention & Exhibition, October 2003

SPECTRAL DECOMPOSITION: EXTENDING THE LIMITS OF SEISMIC RESOLUTION

FOR RESERVOIR DELINEATION

Sunu H. Praptono*
Bambang Murti**
Dylan Mair***

ABSTRACT directly in 3-D seismic data interpretation and

Accurately delineating the thickness and extent of
reservoirs is a major challenge for reliable reserves
estimates and optimal well placement.
Originally pioneered through research at BP and
Amoco, spectral decomposition is a recent seismic
imaging innovation that provides interpreters with
high-resolution reservoir detail. Detailed views of
reservoir stratigraphy and thickness are revealed by
extracting frequency-specific phase and amplitude
from the seismic data. Spectral decomposition
studies are used to improve reservoir delineation,
primarily in clastic fluvial environments but also for
carbonate environments.
This paper begins with a short review of spectral
decomposition theory, where conventional time
domain seismic data is analyzed according to its
frequency content. Interactive animation workflows
are then applied to search for thin bed information
well beyond the conventional ?/4 thickness resolution
determined from the wavelength of the dominant
frequency of the seismic data. A spectral
decomposition case study carried out in the Asia -
Pacific is reviewed to define a clastic channel system
where spectral decomposition has been found to be
effective, and to provide guidelines for the factors that
determine success when using spectral decomposition
in this region.
INTRODUCTION
Spectral decomposition is a seismic processing
technique that has only recently been adapted for use
* Kondur Petroleum S.A
** Landmark Graphics Indonesia
*** Landmark Graphics (Malaysia) Sdn Bhd
reservoir imaging workflows. This technique has the
potential to generate greater resolution of reservoir
boundaries, heterogeneities and bed thickness than
traditional broadband seismic displays and attribute
extractions.
Spectral decomposition is used to identify thin beds
through analysis of the frequency spectrum in a short
window around the time of the bed. Figure 1
demonstrates a thin bed affect. A thin bed spectral
interference pattern is imposed by the distribution of
acoustic properties within the short analysis window.
In this example, a simple homogeneous thin bed
introduces a predictable and periodic sequence of
notches into the amplitude spectrum of the composite
reflection.
Spectral decomposition can be used:
• Qualitatively, to reveal stratigraphic, structural
edges and bodies, and relative thickening or
thinning; and
• Quantitatively, to give an accurate prediction of
reservoir thickness or intra-reservoir travel time.
Interpretive processing using spectral decomposition
has been applied to: delineation of facies/stratigraphic
settings (such as flood pla in boundaries, reef
boundaries, channel sands, incised valley-fill sands,
and other thin beds); resolving the order of deposition
and migration routes; mapping detailed structural
settings involving complex fault systems (such as
reservoir compartmentalization); mapping near-
surface environmental hazards (such as expulsion
features and other near surface instabilities); and
assisting reservoir modeling (mapping fluid changes,
pressure changes and changes in 4D surveys).
The spectral decomposition interpretive processing
workflows were developed at Amoco and first
described by Partyka et al., 1999. There the authors
describe two key processing and interpretation
workflows: the “tuning cube”, and “discrete
frequency energy cubes” (also called “volume
reconnaissance”). Additional enhancements to these
workflows have since been devised, in both the
seismic processing and the subsequent interpretation.
Apache also has several patented and patent-pending
enhancements for spectral decomposition processing.
Spectral decomposition is applied to conventional
post-stack 3-D seismic data. It is generally preferable
(as for most attribute analysis work) to have the
seismic data processed for true amplitude recovery.
Ideally, band-pass filtering should not be applied and
data should be surface-consistent and migrated, with
output in either 16-bit or 32-bit format. The sample
rate should also be kept to a minimum. However,
any moderately clean 3-D dataset can be used to
investigate the potential of spectral decomposition
and evaluate whether reprocessing of the seismic data
is necessary. Ray trace modeling may be used in
more complex terrains to determine whether any
information might be found in the data.
METHODOLOGY
The spectral decomposition transforms the seismic
data into the frequency domain via a DFT (Discrete
Fourier Transform) or via a MEM (Maximum
Entropy Method Transform). The latter method,
based on a Z-transform, is potentially more stable
when analyzing very short time windows.
The transformed (phase independent) amplitude
spectra are used to delineate temporal bed thickness
variability, while the phase spectra are used to
indicate lateral geologic discontinuities. This
technique has proven to be a robust approach for
thickness estimation and fault definition.
Tuning Cube
The most common approach to characterize reservoirs
using spectral decomposition is via the zone-of-
interest “Tuning Cube.” Figure 2 indicates how the
Tuning Cube is generated. In the most typical case, a
time window around a horizon is converted into the
frequency domain; each frequency is rendered as a
map, normalized to make it comparable to other
frequencies, then stored in a single volume, or Tuning
Cube.
The tuning cube analysis depends on a reflection from
a thin bed to get a characteristic expression at a
particular frequency. This characteristic expression is
an indication of the temporal bed thickness and can be
resolved below one fourth of a wavelength.
The Tuning Cube can be viewed through animation to
reveal spatial changes in frequency that corresponds
to spatial changes in stratigraphic thickness. Spectral
decomposition reveals details that no single frequency
attribute can match.
Volume Reconnaissance
The second workflow uses “discrete frequency energy
cubes” or “Volume Reconnaissance” seismic cubes.
A short time window is applied around each sample
of the seismic volume, and a single frequency is
extracted. For efficiency, multiple volumes (each of a
different frequency) might be generated at the same
time. This provides information spatially removed
from the targeted horizon used in the Tuning Cube
workflow. This workflow can generate many seismic
data volumes (potentially one for every hertz of
frequency). The Tuning Cube analysis is usually
applied first to determine key frequencies worthy of
further investigation.
Subsurface “Satellite” Images
Other workflows are possible using data that has been
transformed into the frequency domain. One example
is the use of a satellite imaging technique to display
multiple frequency maps simultaneously. For
instance, any three frequencies (or horizon slices)
may be displayed together, us ing red, green and blue
(RGB) colour bars for each. This approach is similar
to how remote sensing uses sub-bands of frequencies
to map interference at the earth's surface. Just like
remote sensing, it is very important to dynamically
observe the response of the reservoir to different
frequency bands (Laughlin et al., 2002).
CASE STUDY
Several published case studies focus on delineating
clastic channel systems, such as Partyka et al., 1999.
Partyka captures detailed information about the
relative thickness and spatial relationship of channel
systems in the Gulf of Mexico, determined using
analysis of the spectral decomposition Tuning Cube.
Channels have also been studied using seismic data
acquired onshore (Peyton et al., 1998). Carbonates
are less prevalent in the spectral decomposition
literature. Davis (2002) used the technique in
association with other attributes to delineate subtle
features and change of features in carbonate facies of
New Mexico. Companies using spectral
decomposition have observed significant detail from
these images at great depth. Interpretation and
integration with well data and models is critical to
understanding how these details map to stratigraphy
(Laughlin et al., 2002).
The authors have reviewed many spectral
decomposition projects carried out throughout
Southeast Asia, China, India and Australia. Due to
the sensitive nature of these projects, most of these
case studies are anonymous. Some images are
provided for a clastic prospect; further discussion
below captures specific benefits and limitations
encountered by spectral decomposition studies of
other Asia -Pacific clastic and carbonate
environments.
Clastic Case Study
We begin with the results of a clastic play involving a
single channel identified in onshore seismic data.
This channel was first identified using conventional
amplitude attributes. Tuning Cube analysis provided
sharper images of the channel (Figure 3). In addition,
the edges of the channel were found to be much
thicker than the middle. The Tuning Cube was also
able to identify thinning beds (increasing frequency)
stepping away from the channel. This supports
interpreting the narrow floodplain as a paleo-low.
Volume Reconnaissance analysis resolved the
channel into a stacked series of channels on this
narrow floodplain (see Figure 5). This result was
confirmed by recent drilling results (Figure 6). The
previous Tuning Cube analysis window was too broad
to separate the individual channels, resulting in a map
that shows the channels thickening toward the edges.
The paleo-low appears to ensure the channels are
neatly stacked on top of each other. The thicker
edges of this feature can be interpreted as sands
deposited preferentially along the edges of the
floodplain (Figure 4). It may be possible to map
individual channels within this sequence by carefully
interpreting animations of the different volume
reconnaissance (single -frequency) data.
The presence of stacked channels in this sequence has
been confirmed through drilling (see Figure 6).
Further study of this prospect using spectral
decomposition would likely involve phase volume
animation to detect smaller faults, and kurtosis maps,
to capture the spatial distribution of stacked channels.
CONCLUSIONS
Inevitably, the question is: "how are these images
better than conventional attribute analysis?”
Conventional attributes will often capture similar
images if the tuning frequency of the feature is similar
to the peak frequency of the data. However, with
spectral decomposition you can see thinner (and often
thicker) features. You might have begun focusing on
a particular feature because you'd already seen it.
Tuning Cube analysis may identify finer details;
Volume Reconnaissance may help to identify other
similar features that were not immediately obvious.
You can separate features based on thickness.
Stacked channels are an obvious example. Detailed
mapping within a carbonate structure or simply
contouring the thickness of a single feature is
possible.
Don't forget to work with a phase Tuning Cube. The
phase cube shows boundaries, like faults, reef edges,
and the gradual thickening of stratigraphy. There
might even be potential for revealing larger fracture
features.
Most importantly: you know what you're looking at
with spectral decomposition, which separates it from
many other seismic attributes. It is either thin beds (if
they exhibit amplitude tuning), or thin bed boundaries
(in phase). Understanding the theory of spectral
decomposition will enable the interpreter to
understand the serendipitous details that may provide
more information than simple channel thickness.
Understanding is also essential for devising creative
workflows that eke more detail out of challenging
terrains.
ACKNOWLEDGEMENTS
The author thanks to Kondur Petroleum SA
management and BP-MIGAS for their support and
permission to release the data, also to Rusdinadar
Sigit, Richard Leaver, Supriadi, Shiwen Ding,
Qingrong Zhu and Jonathan Zwaan of Landmark
Graphics. Thanks especially to the explorationists of
oil and gas companies throughout Asia -Pacific who
have applied spectral decomposition to improve the
seismic resolution of their prospects.
REFERENCES
Davis, D.A., Marfurt, K., Sullivan, C., Evans, I.,
2002. Seismic Stratigraphy, Reservoir
Characterization, and 3-D Visualization of Seismic
Attributes of Carbonate Facies and Karst in the San
Andres Formation, Vacuum Field, Lea County, New
Mexico, 2002 Annual Meeting, AAPG, Technical
Program.
Laughlin, K., Garossino, P., Partyka, G., 2002.
Spectral Decomposition for Seismic Stratigraphic
Patterns, Geophysical Corner, AAPG Explorer, May
2002.
Partyka, G.A., Gridley, J.M., and Lopez, J. 1999.
Interpretational Applications of Spectral
Decomposition in Reservoir Characterization, The
Leading Edge, vol. 18, No. 3, p. 353-360.
Peyton, L., Bottjer, R., Partyka, G., 1998.
Interpretation of Incised Valleys Using New 3D
Seismic Techniques: A Case History Using Spectral
Decomposition and Coherency, The Leading Edge,
vol. 17, no. 9, p. 1294-1298.
Figure 1 - The spectral interference pattern is imposed by the distribution of acoustic properties within the
short analysis window (Partyka et al., 1999).

3D Seismic Volume

Interpreted 3D
Seismic Volume

Zone of Interest Sub-

volume

Zone-of-Interest Tuning
Cube (cross-section)

Frequency Slices through

Tuning Cube (plan view)

Figure 2 - Generating the Tuning Cube (Partyka et al., 1999).

Figure 3 - This display shows the peak frequency map resulting from spectral decomposition of frequencies
from 10Hz to 50Hz. The channel tunes around 20 to 25 Hz. Closer inspection showed the sands to
be thicker towards the edges of the broad channel. A line of section, A-A’ along the channel, is
shown in Figure 5.

Figure 4 - Diagram showing one possible interpretation of the stacked channels. The thin, stacked channels
have accumulated in a consistent paleo-low. Surrounding sediments produce a thin bed response
that indicates thinning away from the paleo-low. Thicker sands on the edges of the paleo-low may
result from differential deposition of sand on the point bars of more sinuous meandering streams.
 A A’

Figure 5 - Seismic cross-section (top) through the stacked channel sequence, along the line of section shown in
Figure 3. Individual stacked channels can be isolated using volume reconnaissance to capture sands
that tune at a specific frequency, such as 26 Hz (middle) and 20 Hz (bottom), with channel
responses shown in white. The “low frequency” appearance of these tuned events results from the
sliding time window of the Fourier transform – the channel is positioned at the peak of the event.
Animating through these images helps separate the stacked channels.
Figure 6 - Well correla tion indicates presence of the “smart” stacked channel. Adjacent well indicates different
facies over short lateral distance.