NA

C
e a
S E
r - Ss T
uU
r D
f aY c e

A 3D finite-difference modeling study of seismic imaging

challenges in Bintuni Bay, Irian Jaya Barat
Downloaded 03/13/15 to 194.80.232.1. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

GREG J. SCHURTER, YAYAT SUPRIATNA, ANGKE NURAENI, and Supriyono, BP Indonesia

CARL J. REGONE and NURUL KABIR, BP EPTG, Houston

This material was originally presented at

the 2007 Annual Convention of the Indonesian Petroleum Association.

n this paper, we describe how we

designed and built two 3D nite
dierence (FD) models for Bintuni Bay,
Irian Jaya Barat to help us understand key
subsurface components adversely aecting
seismic data quality for clastic reservoirs
below thick and karstied carbonates.
Our modeling eorts consisted of four
main steps: creation of realistic geologic
models based on seismic interpretation;
construction of 3D velocity and density
models based on the geologic model and
well data; generation of 3D FD shot
records with dense, areal, receiver grids;
and depth migration of subsets of shots
and receivers corresponding to various
acquisition geometries.
Figure 1. Location map of Bintuni Bay, Papua, approximately 3200 km east of Jakarta.
The entire process took approximately
three months and produced a range of
synthetic seismic volumes in which the quality of the reser- construction of the 3D FD models, and analysis of depthvoir and subsurface images varied considerably. We conclud- migrated synthetic seismic volumes generated from those
ed that we were able to accurately simulate the seismic quality models. The 3D FD models were generated to address seismic
variations we see in our real seismic data.
imaging challenges for pre-Tertiary clastic reservoirs in BinAn analysis of 16 prestack depth-migrated, synthetic seis- tuni Bay, Irian Jaya, approximately 3200 km east of Jakarta
mic volumes derived from the two FD models showed that (Figure 1). The study area was loosely based on a 20 20 km
two subsurface features, depth to the top of the massive car- area of interest (AOI) for which existing data included: 3D
bonates and karsting, signicantly aect our image quality at seismic data, wells, a depositional model for the reservoirs
the target depth. Our analysis also shows that wide-azimuth of interest, and a partial seismic interpretation. Poor seismic
(WAZ) acquisition geometries capture signicantly higher image quality prevented a complete seismic interpretation of
quality precarbonate subsurface images than narrow-azimuth all intervals of interest. The seismic imaging challenges in(NAZ) acquisition geometries, even if the WAZ acquisition clude a thick karstied carbonate layer overlying pre-Tertiary
geometries are much coarser than the optimum design.
clastic reservoirs, shallow gas, and reservoir thicknesses that
The potential nancial impact of this modeling study is are both above and below tuning (Figure 2). Seismic acquisigreat because it allows us to predict the quality of our seismic tion methods range from shallow marine, to transition zone,
image using various acquisition techniques and geometries to land. Land and transition zone acquisition is generally
before committing to and acquiring seismic data. This is par- aected by sensitive environmental and harsh surface conticularly important in an area such as Bintuni Bay, where no- ditions, whereas marine acquisition is signicantly aected
toriously poor seismic image quality and severe operational by severe currents and tidal uctuations. All of these factors
environments often require application of expensive acquisi- combine to make seismic acquisition dicult and costly. In
tion techniques to achieve appraisal and development qual- addition, several hydrocarbon discoveries in Bintuni Bay are
ity seismic images. The 3D FD modeling results will help us progressing into appraisal and development. In some cases,
meet our exploration, appraisal, and development needs by accurate appraisal and ecient development planning may
allowing us to balance the cost, image quality, and expecta- require signicant improvements in subsurface imaging.
tions of future seismic operations.
Subsurface imaging challenges, operational challenges,
and increasing costs (for both wells and seismic data) necessiBackground
tate a thorough understanding of the subsurface to condentThis paper is organized into two main sections: design and ly plan eective new seismic surveys in Bintuni Bay. The sub1008

The Leading Edge

September 2009

Downloaded 03/13/15 to 194.80.232.1. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

C A S E

S T U D Y

Figure 2. Conceptual geological cross section for Bintuni Bay area. Key subsurface features for study include: thick massive limestones (Kais
and Faumai), karsting (intra-Faumai), stratigraphically complex reservoir characteristics (Paleocene channel sands), and shallow gas sands (in
Steenkool-Klasafet).

surface is seismically complex, limiting the usefulness of less

expensive traditional data acquisition techniques. Application
of advanced or new seismic technologies in the area will come
with increased cost and logistical complexity. Therefore it is
critical to understand the physics of the subsurface in order to
understand the limits that physics imposes on our ability to
produce an accurate subsurface image. We have used 3D FD
models to investigate those limits and avoid spending money
and time on costly seismic surveys that may provide little new
information, or may even lead to costly drilling failures. Here
we only focus on the application of FD modeling to help
understand the subsurface imaging challenges, as the computational particulars of FD modeling are beyond the scope
of this paper.
Historically, conventional seismic survey design relies on
the assumption that uniform midpoint coverage will lead to
uniform illumination in the subsurface. However, when dealing with complex subsurface structures, this is no longer valid
because reection points are not situated at midpoint locations. In the case of Bintuni Bay, large velocity contrasts, rugose surfaces, karsting, and thick carbonates cause signicant
wave distortion.
The rst FD model was based on 33 interpreted and inferred horizons, eight wells, and a 1997 3D seismic data set.
Seven horizons were interpretable from the seismic data, and
26 were only interpretable from the well data. The inferred

horizons were the basis of geobodies (karst zones, shallow gas,

and channel model features) that were created primarily from
well data and analogues.
Once the framework model was completed, 3D velocity
and density models were derived from well logs to build the
nal input to the FD shot-record computations. Generation
of realistic 3D shot records using FD acoustic modeling is
very compute intensive, and so choosing parameters and a
manner of acquisition simulation were very critical to conducting ecient and eective modeling.
Shots were extracted from the computed 3D FD model
based on a set of 3D acquisition geometries. The geometries
fell into two main types: narrow azimuth (NAZ) and wide
azimuth (WAZ). Each set of extracted shots was then prestack
depth migrated using the exact velocity model. These depthmigrated synthetic seismic volumes formed the basis for validating the integrity of the 3D FD model and drawing nal
conclusions regarding the key subsurface features controlling
image quality and the ability of acquisition geometries to accommodate those features. Lastly, a second FD model was
created by decreasing the depth to top of carbonates, and the
same process was followed to generate a second set of synthetic seismic volumes.
The following discussion provides details of each of the
above model building steps, and shows how an FD model
representing a complex subsurface imaging problem can be
September 2009

The Leading Edge

1009

Downloaded 03/13/15 to 194.80.232.1. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

C A S E

S T U D Y

Figure 3. (a) Shallow gas accumulations are based on high-amplitude seismic reectors in the Steenkool-Klasfet. (b) Shallow gas model was
included in the subsurface geological model.

Figure 4. (a) Coherency slice at Faumai zone from ocean-bottom cable (OBC) seismic data, karst feature represented in black. (b) Inferred
horizons representing karstied zone.

completed in a relatively short time. The resulting synthetic

seismic volumes indicate that the complexity of the FD model was sucient to represent key subsurface features which
detrimentally impact image quality, and provide key information as to which acquisition types are best suited to address
the imaging challenge.
Modeling methodology
There were three main steps in the FD modeling work ow:
building a subsurface model, generating synthetic seismic
data by using an FD acoustic method, and prestack depth
migrating (PSDM) shots extracted based on specic acqui1010

The Leading Edge

September 2009

sition geometries. Assessment of the imaging potential for

each acquisition geometry is discussed later in this study.
The rst step was to build a 3D subsurface model that
represented our complex geological features as accurately as
possible. Some geologic intervals were interpretable from
seismic data (framework horizons), and some were not (inferred horizons). The key horizons used to frame the model
came from an existing seismic interpretation project on the
Wiriagar Deep Field. Seven framework horizons were used,
representing the top and base of Kais limestone, base of Faumai limestone (or top of Eocene clastics), top and base of the
Paleocene, base Cretaceous, and top Ayot limestone. Faults

Downloaded 03/13/15 to 194.80.232.1. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

C A S E

S T U D Y

Figure 5. Channel model in Paleocene interval: (a) mudprone, (b) upper member, (c) middle member, and (d) lower member.

were not explicitly built into the model; however, horizons

were interpolated but not smoothed, and so major faults were
identiable as kinks in the horizons and subsequent velocity and density models. The inferred horizons were designed
to represent geological features that were inferred from the
seismic data (shallow gas), or were interpreted from the well
analyses and geological modeling (karstied limestone in
Faumai, and complex Paleocene turbidite channels and sheet
sands). All were important to include in the model because
they were either related to seismic image quality, drilling hazards, or were prospective reservoir targets. The overall goal of
the subsurface model was to incorporate all of these features
into the model such that some regions of the model would
be inuenced by only one feature, other regions would be inuenced by a combination of features, and still other regions
would be inuenced by all of the features.
Creating the inferred horizons was tricky. The shallow gas
distribution was generated from an amplitude extraction of
1012

The Leading Edge

September 2009

3D seismic data over the Wiriagar Deep Field. The inferred

shallow gas accumulations were based on high-amplitude
seismic reectors in the Steenkool-Klasafet section in real 3D
seismic data. The thickness of the shallow gas intervals was
correlated to shallow gas encountered in well data, which averaged 912 m thick, and widths of 14 km for these intervals
were taken directly from the real seismic amplitude events.
Tops and bases for the shallow gas intervals were drawn in our
workstation project, and the areal distribution of the gas was
varied such that some areas of the model would not contain
shallow gas features (Figure 3).
The karst interval at the top of the Faumai was built to
vary karst thickness, areal density, and karst connectivity.
These parameters were modeled after geologic descriptions of
karsting (Yose et al., 2004) and a karst imaging study undertaken by BP at Vorwata Field using coherency volumes from
ocean-bottom cable (OBC) seismic data. The karst interval
was generated by duplicating the top Kais framework horizon

C A S E

S T U D Y

Downloaded 03/13/15 to 194.80.232.1. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Figure 6. Time slice of migrated synthetic volume (a) and cross section
in Y direction resulting from narrow-azimuth acquisition geometry
(b). Red arrow points to area of interest.

Figure 7. Time slice of migrated synthetic volume (a) and cross section
in Y direction resulting from wide-azimuth acquisition geometry (b).
Red arrow points to area of interest, better signal-to-noise ratio and
good reector continuity.

to form a top and base of karst. The base of karst was then
shifted in time to the bottom of a seismic volume below the
acoustic basement, snapping it to the nearest high-amplitude
reector (just noise), which served to randomize the reector.
The base was then shifted back up to a position beneath the
top of karst, at the appropriate time interval for the karsting. Both horizons were then cut spatially into pieces using a
polygon delete tool (Figure 4). The dimension of karst varies
between 0200 m thick and 100500 m wide. The lateral occurrence ranged from very prevalent to sparse.
The Paleocene consisted of four intervals: Mudprone,
Upper Member, Middle Member, and Lower Member (Figure 5). Each interval was modeled to contain a large number
of sands, varying in reservoir quality and distribution (discontinuous-conned channels system through unconned
sheet sands). The channel orientations were north-south 45
based on well data. Lithostratigraphy zonation referenced to
biostratigraphy provided the basis for gross thickness of sand/

shale packages within each member. BP benchmark analogues

constrained the geometry of sand bodies (width ranging from
0.5 to 5 km). The thicknesses of modeled channels varied
from 6 to 20 m in the Mudprone, Upper and Lower members, whereas a thicker stacked-channel body of 45 m was
modeled in the Middle Member. All sand bodies started with
top and base horizons copied from the nearest framework horizon. Pairs of top and base horizons were then cut spatially to
form channel geometries using a polygon delete tool.
The completed subsurface time model was then converted
to depth using a combination of interval velocities from a
2005 Kirchho prestack time migration (PSTM) processing ow and checkshot data. We used well-log data to get
constant interval velocity and density for each layer in the
model. For QC purposes, all layers in the model were loaded
in depth to form a 3D volume, with which we could visually
ensure that all horizons and intervals were in the right position according to our geologic model. This was particularly
September 2009

The Leading Edge

1013

Downloaded 03/13/15 to 194.80.232.1. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

C A S E

S T U D Y

helpful in complicated areas of thin Paleocene channels that

sometimes inadvertently overlapped other horizons.
The next step in the modeling process consisted of generating thousands of 3D FD shot records. Required inputs
were the velocity and density volumes that were created in
the rst step. The dimensions of these models were 20 km in
the x direction, 20 km in the y direction, and 4.5 km in the
z direction with grid sizes of 18.5 m in the x and y directions
and 10 m in z. With these grid sizes and model velocities, a
maximum frequency of 30 Hz could be generated. For each
shot we recorded a 10 10-km square patch of receivers centered on the shot with a 25-m group interval in both the x
and y directions. The shot grid was 18 18 km (1 km inside
the edge of the model on all sides) and the shot interval was
150 m in both the x and y directions. Thus, there were a total
of 14,641 shots for each of the two models built. The shot
generation process took 15 days for each model with computer resources consisting of 720 dual-processor 3.4 GHz
Xeon-based computers.
The appropriate subsets of data were extracted from the
shot records for each acquisition geometry. A waveeld common-shot depth migration method was used with the exact
velocity model to migrate the synthetic shot gathers. The nal
result for this study was the generation of migrated synthetic
seismic volumes representing six acquisition geometries.
Modeling results
Modeling the NAZ geometries was straightforward. Two
subsets of sources and receivers were extracted with the following geometry: 12 3-km cables, and then 12 5-km cables,
spaced 50 m apart with a 25-m group interval and a sail line
separation of 300 m oriented in the y direction with a source
interval of 150 m. No multiple attenuation was applied before migration. Figure 6 shows the time slice at target level
and cross section in the y direction, and red arrows show the
reection from the Jurassic reservoir.
Modeling WAZ geometry required special treatment. We
chose to simulate OBC acquisition with dense patch shooting. Because the number of receiver locations in such an
experiment is much smaller than the number of shots, reciprocity was used to swap sources and receivers. Five geometry
congurations were extracted to provide a wide range of cost
versus data quality options. The receiver spacing was varied as
follows: (a) 150-m receiver line interval and 150-m receiver
interval, (b) 300-m receiver line interval and 150-m receiver
interval, (c) 450-m receiver line interval and 150-m receiver
interval, and (d) 600-m receiver line interval and 150-m receiver interval. All of these geometries had a 25-m shot interval in both the x and y directions and a shot grid extending 5000 m on both sides of each receiver line. The resulting
PSDM is shown in Figure 7 with no multiple attenuation
applied before migration. The reectors in the zone of interest
(indicated by red arrows) in the WAZ data show signicantly
more continuity and a better signal-to-noise ratio than in the
NAZ data.

1014

The Leading Edge

September 2009

Figure 8. Comparison of inline 1486 from both NAZ geometry

volumes, showing that the quality of both images is similar.

Figure 9. Inline 1486 from the densest WAZ acquisition geometry

volume. Image quality is considerably improved over the same line
from the NAZ volumes.

Downloaded 03/13/15 to 194.80.232.1. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

C A S E

S T U D Y

Figure 10. A
typical inline
comparison from the
NAZ and WAZ150
volumes (Model 1),
running north-south
from an area of
dense karsting and
shallow gas presence
in the north, to a
relatively karst- and
shallow gas-free area
in the south. The
image beneath the
karsting on both
lines is signicantly
degraded compared
to the areas without
karsting. However,
many subkarst
reections on the
WAZ150 image
retain continuity
and are easily
picked.

Figure 11. A typical

inline comparison
from the NAZ and
WAZ150 volumes
(Model 2). Here the
top of the carbonate
(Kais) was moved
shallower, and the
interval between
the Paleocene
and the Permian
was expanded
proportionally.
The degradation
of the entire NAZ
subsurface image
is obvious when
compared to the same
line from Model
1 (Figure 7). The
degradation of the
WAZ subsurface
image is signicantly
less than that of the
NAZ image.

Analysis
Our analysis of the 3D FD models consisted of three primary steps: determine which form of the data would provide the
most information about dierences in the seismic volumes;
determine which seismic volumes in fact contained dierent
information than others; and then analyze the dierences between the volumes. The exact position, velocity, and density
of each geologic interval were known since we worked from
a model, so extracting amplitude and continuity information
at key horizons was trivial. However, evaluating 20 20-km
3D seismic volumes representing six acquisition geometries
1016

The Leading Edge

September 2009

and two FD models, each containing 33 horizons, presented

a major data management challenge. A complete map view
analysis of each reector in every seismic volume would require generation and evaluation of more than 100 maps, even
if we had looked at only one attribute. These maps would be
in addition to the inline and crossline comparisons. Initially,
we addressed this data abundance challenge by working with
only one seismic volume. We then ran a complete analysis of
each geologic interval in that volume, as well as of multiple
inlines and crosslines, to determine if a particular attribute
was best for evaluating dierences in subsurface details in

Downloaded 03/13/15 to 194.80.232.1. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

C A S E

S T U D Y
Figure 12.
Comparison of
Model 1 WAZ
acquisition
geometries using
inline 1486.
Dierences are
much more subtle
than the dierences
between the NAZ
and WAZ.

Figure 13.
Comparison of
Model 2 WAZ
acquisition
geometries using
inline 1486. The
degradation due
to the shallower
top of carbonate is
clear, but variation
in image quality
between the WAZ
volumes is not
enormous at the
target intervals.

the models, and if it was necessary to analyze every volume.

For the rst step of the analysis, we used the seismic volume generated using a WAZ 150-m receiver line and receiver
interval spacing, and 25-m carpet shooting (which will be
henceforth referred to as WAZ150). This volume represented
what we consider the theoretically best seismic acquisition
that could be achieved, although in reality it was unrealizable
due to operational, time, and cost considerations. We analyzed all horizons and intervals to evaluate how the reection
strength and continuity of the pre-Tertiary (Jurassic and Paleocene) reectors were impacted by the presence of shallow
gas and karst in the overburden, and by decreasing the depth

to top of carbonates (Model 2).

First, we extracted rms amplitudes on the geologic intervals, with particular attention paid to the karst, Paleocene,
and Jurassic intervals. In addition, we ran coherency on the
equivalent intervals. Next, we applied a bluing operator to
the seismic volume, and repeated the same rms amplitude
extractions and coherency calculations. Lastly, we inverted
the reectivity volume to relative impedance using a derived
colored inversion operator, and again ran the rms and coherency computations on the inverted volume. Visual inspection
of the key geologic intervals from all these tests showed some
variation in details of geologic features, but overall the same
September 2009

The Leading Edge

1017

Downloaded 03/13/15 to 194.80.232.1. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

C A S E

S T U D Y

Figure 14. Comparison of the reection strength of the Jurassic reservoir, which is modeled as a continuous sheet sand of constant (~100 m)
thickness. High and continuous reection strength gives some indication that the surface is well illuminated. This appears to indicate that the
WAZ acquisitions are capturing signicantly more reected energy from the pre-Tertiary section.

features are interpretable on all tests. Consequently, and in

order to help reduce the volume of data input to the analysis, we made the decision to use the reectivity volumes and
rms amplitude extractions from those volumes for horizon
and inline/crossline data comparisons. The original synthetic
seismic reectivity volumes generated from the FD modeling
provide the basis for all additional illustrations and analyses
presented in this paper. We also compared various amplitude
extractions, and determined that rms amplitude was a sucient horizon-based attribute for evaluating lateral resolution
of the key geologic intervals.
The second step of the analysis was to evaluate each of the
six synthetic seismic volumes for dierences in imaging the precarbonate reectors. We evaluated dierences by visual comparison of the seismic data in the inline direction, the crossline
direction, and using rms amplitude extractions on the karst,
Paleocene, and Jurassic intervals. We used the density model
built for the FD model to ground truth the images, for both
inline and crossline comparisons, and the horizon-based comparisons. The rst result of this step of the analysis was elimination of the NAZ volume modeled on a 5-km cable length.
Part of this decision was based on the fact that the images from
1018

The Leading Edge

September 2009

both NAZ volumes were similar (Figure 8), and dropping one
volume would reduce the volume of data to be reviewed. But
more importantly, a large part of the decision was because the
WAZ volumes all provided signicantly higher quality images
than either NAZ volume (Figure 9). Therefore, we decided to
focus the initial analysis on the dierences between one of the
NAZ volumes and the WAZ volumes.
Results
The geologic framework underpinning the FD model was
created so as to maximize the amount of information we
could derive about the interaction of subsurface elements.
Consequently, some areas of the model contained shallow
gas intervals, some contained karst, some contained subseismic resolution Paleocene channels, and some areas contained
all three. Figure 10 illustrates a typical inline comparison
from the NAZ and WAZ150 volumes (Model 1), running
north-south from an area of dense karsting and shallow gas
presence in the north, to an area relatively free of karst and
shallow gas in the south. The Top Permian represented the
base of our imaging zone of interest, and the Top Paleocene
represented the top of our imaging zone of interest. On both

Downloaded 03/13/15 to 194.80.232.1. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

C A S E

S T U D Y

Figure 15. Comparison of a line taken from both the Vorwata Field WAZ and NAZ seismic data sets. The top reservoir (Jurassic) annotated on
the gure is the key horizon for this eld. The improvement in reservoir reector quality in the WAZ seismic data was sucient in the Vorwata
case to help the interpreter directly pick the top reservoir for the rst time in the elds history.

seismic lines, the image beneath the karsting was signicantly degraded compared to the areas without karsting. The
many subkarst reectors on the WAZ150 image, however,
still retained continuity, and were easily picked. The presence
of shallow gas near and on the line in the north appeared to
have little impact on image degradation relative to the impact
of the karst. Figure 11 shows data from the same north-south
line, but from Model 2. Here the top of the carbonates (Kais)
was moved shallower, and the interval between the Paleocene
and the Permian has been expanded proportionally. The degradation of the entire NAZ subsurface image was obvious
when compared to the same line from Model 1. The degradation of the WAZ subsurface image was also signicant
as compared to Model 1, especially at the Paleocene level;
however, overall the image degradation was signicantly less
than that of the NAZ image. The image quality showed a
signicant dependence on the depth to top of carbonates,
as well as the presence of karsting. It also highlighted the
impact of WAZ acquisition.
The next step in the analysis was to compare the range of
WAZ acquisition geometries, because in both of the previous data comparisons (Figures 10 and 11), the WAZ image
was superior to the NAZ image, regardless of the subsurface
element causing image degradation. Although the image obtained by the WAZ150 was clearly superior to the NAZ geometries, it was not realistically achievable in the eld due to
operational and nancial constraints. Analysis of the sparse
versions of the WAZ150 provided a more useful insight into
what might be achievable in the eld with new seismic acquisition. Figure 12 shows a comparison of Model 1 WAZ
acquisition geometries for the same inline used in previous
gures. There was a dierence in image quality between the
1020

The Leading Edge

September 2009

volumes, but the dierences were much more subtle than the
dierences between the NAZ and WAZ. Figure 13 shows
the same line again from the same WAZ acquisition geometries, but from Model 2. The degradation due to the shallower top of carbonates was clear, but again, the variation in
image quality between the WAZ volumes was not enormous
at the target intervals. The dierence between the WAZ150
and the sparsest WAZ survey (150 600) equated to a fourfold increase in line spacing. The corresponding fold decrease
was approximately the same amount. This result suggested
that the sparse version of the WAZ acquisition did not negate
the large increase in image quality the WAZ acquisitions gain
over the NAZ acquisitions.
Comparisons shown thus far were based on inlines and
crosslines. We made additional horizon-based comparisons,
as previously stated, especially to examine spatial details captured by the various acquisition geometries. We used horizon-based rms amplitude extraction comparisons to focus on
two primary intervals into which signicant stratigraphic elements were built in the subsurface model (Faumai/karsting,
Paleocene/channeling). The rst interval analyzed was the
karst near the top of the Faumai carbonate. The karst bodies
were in fact well imaged by both the NAZ and WAZ volumes. The next interval analyzed was the Paleocene interval,
which was modeled to contain a variety of subseismic resolution turbidite-like channels. Although the channels were
clearly seen on the extraction from the velocity model, we
were disappointed to nd that most channel features in the
model were scarcely imaged in the WAZ data when they were
subkarst, and they were almost completely absent in the NAZ
data. Another informative comparison we found was of the
reection strength of the Jurassic reservoir, which was mod-

Downloaded 03/13/15 to 194.80.232.1. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

C A S E

S T U D Y

eled as a continuous sheet sand of constant (~100 m) thickness (Figure 14). High and continuous reection strength
gave some indication that the surface was well illuminated.
The extractions on the top of the Jurassic reservoir from the
WAZ volumes showed similar reection strength, which was
generally good over roughly half of the extraction surface. The
extraction from the NAZ volume showed signicantly more
discontinuous and overall lower reection strength than any
of the WAZ extractions. This suggested that the WAZ acquisitions were capturing signicantly more reected energy
from the pre-Tertiary section.
To this point, the analysis of synthetic seismic volumes
generated from both NAZ and WAZ acquisition geometries
indicated that WAZ acquisition geometries were providing
superior subsurface imaging of the 3D subsurface model
from which they were derived. However, there remained the
question of how accurately the 3D FD model represents the
true subsurface being investigated. Fortunately for this study,
overlapping sparse OBC WAZ seismic and conventional
towed-streamer NAZ seismic have been acquired over Vorwata Field in Bintuni Bay. Also, drilling and seismic analysis
have shown that signicant karsting exists over Vorwata Field
near the top of the Faumai carbonate. Figure 15 shows a comparison of a line taken from both the Vorwata Field WAZ and
NAZ seismic data sets. The top reservoir (Jurassic) annotated
on the gure was the key horizon for this eld. The improvement in reservoir reector quality in the WAZ seismic data
was sucient in the Vorwata case to help the interpreter directly pick the top reservoir for the rst time in the elds
history. The uplift in pre-Tertiary image quality in the WAZ
data appears to agree with the preliminary conclusions of this
FD modeling study.
Conclusions
The goals of this study were to compare WAZ and NAZ acquisition and to identify key subsurface features impacting
image quality. Previous seismic surveys were used to decide
on a shot patch width for the FD model. Due to time and
budget constraints, we did not conduct patch-width tests.
These tests would be required to optimize the patch size and
the shot grid, in order to economize the acquisition geometry
for execution in the eld.
The 3D FD modeling results have afforded us signicant insight into the impact of key subsurface features on image
quality based on generalized Bintuni Bay
subsurface geology. The key features affecting seismic image quality are karsting
and depth to the top of the massive carbonates. The imaging impact of shallow
gas appears to be secondary to those of
both karst and depth to top of carbonates. For this 3D FD model, WAZ acquisition geometries resulted in pre-Tertiary images that were superior to those
of the NAZ acquisition geometries. A
four-fold increase in receiver line spacing

of the densest WAZ geometry modeled resulted in relatively

small changes in overall pre-Tertiary image quality, which was
still signicantly better than the NAZ seismic images. Evaluation of real NAZ and WAZ seismic data sets acquired over
Vorwata Field displayed the expected step change in WAZ
image quality predicted by this 3D FD modeling analysis.
This result suggested that the subsurface model underpinning
our FD model was a reasonable representation of the Bintuni Bay subsurface. Lessons learned from this study can and
should be extrapolated to areas with similar subsurface geology and structure, with special attention given to density of
karsting and depth to top of carbonates, and should be used
to design future seismic acquisitions.
The seismic quality variations in the migrated synthetic
volumes appeared to reect image variations seen in real
seismic data. The amount of detail built into the subsurface
model was reected in some but not all of the synthetic volumes, suggesting that our choice of acquisition geometries
was broad enough to highlight dierences between NAZ and
WAZ data. However, the geobodies in our subsurface model were based on our best estimate of geobody dimensions
(thickness, width). In hindsight, for a particular geobody, it
would have been more useful to create a range of geobodies that represented our uncertainty regarding what the true
dimensions might be. For example, many of the Paleocene
channel geobodies were not imaged in most of the synthetic
volumes. We would have a better understanding of the minimum channel dimension that could be imaged for each acquisition geometry if we had input a larger range of channel
widths and thicknesses.
Suggested reading. A modeling approach to wide-azimuth
design for subsalt imaging by Regone (TLE, 2006). New
frontier in 3D seismic characterization of carbonate reservoir:
Example from a supergiant eld in Abu Dhabi by Yose et. al.
(SPE Exhibition and Conference, 2004).
Acknowledgments: The authors thank BP Indonesia, Tangguh
Joint Venture Partners, Indonesia Petroleum Association, and the
Government of Indonesia for permission to publish this study. The
IPA is acknowleged for permission to publish this paper.
Corresponding author: schurtgj@bp.com

September 2009

The Leading Edge

1021