Seismic Snapshots for Reservoir Monitoring

Seismic surveys acquired at different stages in the life of a reservoir can

provide time-lapse snapshots of the fluid distribution over production time.

This technique, called four-dimensional (4D) seismic reservoir monitoring, is

helping operators delineate bypassed hydrocarbons and design development

programs to optimize recovery and extend the useful life of fields.

Lars Pedersen
Statoil
Bergen, Norway

Sarah Ryan
Colin Sayers
Cambridge, England

Lars Sonneland
Helene Hafslund Veire
Stavanger, Norway

For help in preparation of this article, thanks to Olav

Holberg, Geco-Prakla, Oslo, Norway; Dominique Pajot
and Robin Walker, Geco-Prakla, Gatwick, England; and
Benoit Reymond, Geco-Prakla, Stavanger, Norway.
RST (Reservoir Saturation Tool) and TRISOR are marks of
Schlumberger. Some work described in this article was
performed as part of the Thermie project “4D Seismic,”
European Commission Contract No. OG 117/94 UK.

32 Oilfield Review
Reservoir management today is a science of
approximation when it comes to the rate
and direction of fluid-front movement. Opti-
mal management requires up-to-date infor-
mation throughout the entire reservoir vol-
ume. Access to the latest data on fluid
distribution in a reservoir, and knowledge of
how that distribution is changing with time,
allows engineers to develop cost-effective
strategies to get the most out of every field at
the lowest possible risk.
Today, in addition to static, or one-time
measurements, time-dependent answers
from various oilfield disciplines help con-
strain, refine and improve the accuracy of
reservoir models. Time-lapse logging of
fluid saturation through casing can show
which zones are contributing to production
and which are watering out or being
bypassed.1 Permanent downhole sensors
provide continual observations of pressure,
temperature and other diagnostics of reser-
voir performance.2
These measurements supply crucial infor-
mation about fluid behavior at the well
location, but fail in the vast interwell region.
One measurement technique, the 3D seis-
mic survey, has routinely been relied on to
provide interwell data. In the past, seismic
surveys were mainly interpreted for struc-
Winter 1996
tural features and stratigraphic variations
within the reservoir, but they can also be
sensitive to contrasts in fluid type. Applied
in surveys separated by periods of produc-
tion, time-lapse, or four-dimensional (4D)—
3D plus time—seismic images can map
fluid changes in a producing reservoir
(below). This article describes the technique,
the rock physics and seismic modeling
required for successful application, con-
straints in seismic acquisition, and new
interpretation methodologies that allow
changes in seismic response to be inter-
preted as changes in saturation. Application
of these techniques will be discussed in an
example from the North Sea Gullfaks field
where water is displacing oil—the ultimate
challenge in seismic monitoring.
How 4D Works
As a reservoir is exploited, pore fluid under-
goes changes in temperature, pressure and
composition. For example, enhanced oil
recovery (EOR) processes such as steam
injection increase temperature. Production
of any fluid typically lowers fluid pressure,
increasing effective pressure of overburden
on the formation rock. Gas injection and
waterflooding mainly change fluid composi-
tion and pressure. These fluid changes alter
the formation’s seismic velocity and density,
which combine to affect travel times, ampli-
tude and many other features, or attributes,
of reflected seismic waves.
When these changes are great enough, a
seismic survey acquired after years of pro-
duction—called a monitor survey in this
article—will show different attributes than
one acquired earlier, perhaps even before
production begins—the baseline survey.
With today’s computer technology, it is pos-
sible to take the difference between two sur-
veys, be that the change in amplitude, fre-
quency, phase, polarity, reflection intensity,
or of any seismic trace attribute (see “Seis-
mic Attributes,”next page).3 The key to 4D
seismic monitoring is for the change to be
sufficiently large to be seen once the differ-
ence between a baseline survey and subse-
quent monitor surveys is computed.
1. For a review: Albertin I, Darling H, Mahdavi M,
Plasek R, Cedeño I, Hemingway J, Richter P, Markley
M, Olesen J-R, Roscoe B and Zeng W: “The Many
Facets of Pulsed Neutron Cased-Hole Logging,” Oil-
field Review 8, no. 2 (Summer 1996): 28-41.
2. Baker A, Gaskell J, Jeffrey J, Thomas A, Veneruso T and
Unneland T: “Permanent Monitoring—Looking at Life-
time Reservoir Dynamics,” Oilfield Review 7, no. 4
(Winter 1995): 32-47.
3. For background on seismic attributes: Taner MT and
Sheriff RE: “Application of Amplitude, Frequency, and
Other Attributes to Stratigraphic and Hydrocarbon
Determination,” in Payton CE (ed): Seismic Stratigra-
phy—Applications to Hydrocarbon Exploration AAPG
Memoir 26. Tulsa, Oklahoma, USA: American Associ-
ation of Petroleum Geologists (1977): 301-327.
■ Seismic snapshots chronicling
fluid movement over the lifetime
of a reservoir.
33
 Seismic Attributes

Seismic attributes represent a way to compress the
information contained in seismic traces. Attributes,
such as amplitude, frequency, phase, polarity and
a host of others, may be defined in different ways.1
Some are computed at one time sample—that of
the interpreted top of the reservoir, or whatever
surface is being characterized. Such attributes are
called instantaneous. Taken collectively, over
many traces, the attribute then represents some
characteristic of the reflecting surface. Display of
an instantaneous attribute is often accomplished by
color coding the display of the surface itself in two
or three dimensions (next page, left).
Other attributes may be computed over several
time samples in a trace, effectively describing a
volume. Called volume attributes, they compress a
potentially large quantity of seismic data into a
single value. For example, reflection heterogene-
ity, related to the “length” of seismic trace
between two selected time samples, says some-
thing about the homogeneity or heterogeneity of
the internal reflector pattern in the volume of
reservoir between the two time samples.2
A third type, a surface attribute, can be com-
puted on a defined area of the interpreted surface.
Examples are the apparent dip and dip azimuth of
the surface.
Different attributes can be sensitive to different
reservoir properties. Instantaneous phase, for
example, highlights continuity of reflectors.
Instantaneous frequency has been reported as a
light-hydrocarbon indicator. Acoustic impedance
contrast has been used for porosity mapping.3 Vol-
ume reflection heterogeneity is sometimes a
lithology indicator.4 Dip and dip azimuth are pow-
erful fault and fracture indicators.5 Finding the
right attribute, or combination of attributes, that
will show sensitivity to the reservoir properties of
interest, is the job of the interpreter.
Interpreting 4D seismic data for fluid contact
changes in Gullfaks field requires mapping the top
of the reservoir, pinpointing areas where different
caprocks overlie the reservoir and identifying the
type of fluid below. This process relies on a com-
bination of instantaneous and volume attributes
(below). Instantaneous attributes show sensitivity
to the top surface of the reservoir and volume
attributes describe the fluid content.

1. For the classic introduction to seismic attributes:

Taner and Sheriff, reference 3, main text. Instantaneous Amplitude Amplitude Deviation
2. Sønneland L and Barkved O: “Use of Seismic
Attributes in Reservoir Characterization,” in Buller AT,
Berg E, Hjelmeland O, Kleppe J, Torsæter O and
Aasen JO (eds): North Sea Oil and Gas Reservoirs—II.
London, England: Graham & Trotman (1990): 125-
128.
3. Alam A, Matsumoto S, Hurst C and Caragounis P:
“Qualitative Porosity Prediction from Seismic
Attributes,” presented at the 65th Society of
Exploration Geophysicists International Exposition
and Annual Meeting, Houston, Texas, USA, October
8-13, 1995, paper IN2.2.
4. Risch DL, Donaldson BE and Taylor CK: “3D Seismic
Sequence Stratigraphy of Lowstand Deposits,” pre-
sented at the SEG Summer Research Workshop on 3-
D Seismology: Integrated Comprehension of Large
Data Volumes, Rancho Mirage, California, USA, Polynomial Factor Polynomial Factor
August 1-6, 1993.
5. Heggland R: “Detection of Ancient Morphology and
Potential Hydrocarbon Traps Using 3-D Seismic Data
and Attribute Analysis,” presented at the 65th Society
of Exploration Geophysicists International Exposition
and Annual Meeting, Houston, Texas, USA, October
8-13, 1995, paper IN2.3.

■Seismic attributes of the top reservoir surface. The instantaneous attributes describe the top surface of the
reservoir, and volume attributes are sensitive to fluid content. The instantaneous amplitude (top left) shows posi-
tive amplitudes in red grading to negative amplitudes in blue. Instantaneous amplitude deviation (top right)
enhances definition of regions where amplitude changes from positive (white) to negative (blue). The two
attributes plotted in the bottom figures are related to principal components of a polynomial that approximates the
trace shape.

34 Oilfield Review
 1.1 10
Water

Normalized compressional velocity

2000 Oil

Compressional velocity, m/sec

9
1.0 Gas

Acoustic impedance
0% oil
100% brine 8 Saturation
0.9 1600 change 4%
7 Pressure
50% oil
0.8 50% brine change 12%
6
1200
0.7 100% oil
Seismic Data Cube 5
0% brine Initial Producer
stress stress
0.6 800 4
0 50 100 150 200 0 20 40 60 80 100 0 10 20 30 40 50
Temperature, deg C Gas saturation, % Effective stress, MPa
x y
■ Effect of gas and high temperature on seismic velocity. Increasing temperature
decreases velocity in oil-filled rocks (left). Introduction of gas also decreases velocity dra-
Time

matically (center). A decrease in fluid pressure has the opposite effect (right), increasing
acoustic impedance—the product of velocity times density.

If temperature and pressure in the reservoir Introduction of gas into liquid-filled rock
are known, the effects on seismic properties or an increase in temperature of hydrocar-
expected from a change in fluid properties bon-filled rock both cause a decrease in
can be estimated from laboratory experi- seismic velocity (above). Introduction of gas
Amplitude Attribute (high in middle) ments on core samples. Most of the changes decreases velocity substantially by making
in seismic behavior come from fluid effects the fluid mixture compressible. The effect of
on the formation’s seismic velocity rather increasing temperature makes hydrocarbons
than on its density. Laboratory experiments less viscous, reducing overall rigidity and
on fluid-filled rock show how temperature, therefore reducing seismic velocity. Both
y
x pressure and fluid content can either effects are most prominent at low overbur-
decrease or increase seismic velocity.4 den stress, such as in shallow, unconsoli-
Frequency Attribute (low in middle) dated sands. A dramatic increase in velocity
4. Murphy W, Reischer A and Hsu K: “Modulus Decom-
position of Compressional and Shear Velocities in
can occur with decrease in fluid pressure, as
Sand Bodies,” Geophysics 58 (February 1993): 227- typically occurs during oil production. The
239. decrease in fluid pressure increases effective
Clark VA: “The Effect of Oil Under In-Situ Conditions stress on the reservoir rock, stiffening the
y on the Seismic Properties of Rocks,” Geophysics 57,
x (July 1992): 894-901. matrix and increasing velocity.
Tosaya C, Nur A, Vo-Thanh D and Da Prat G: “Labora- Replacing oil with water gives rise to a
tory Seismic Methods for Remote Monitoring of Ther- moderate increase in seismic velocity. This
■ Computing attributes on a time surface in a 3D mal EOR,” SPE Reservoir Engineering 2 (May 1987):
seismic volume. The seismic character of each trace 235-242. increase reaches about 10% in clean, high-
is analyzed at the selected time (top) and assigned a Wang Z and Nur A: “Wave Velocities in Hydrocar- porosity sandstones, and can be even greater
value. For example, the amplitude of each trace on a bon-Saturated Rocks: Experimental Results,” Geo- in unconsolidated sands. Above 30% poros-
physics 55 (June 1990): 723-733.
surface can be mapped on a separate plot (middle). ity, seismic velocities of oil-filled and water-
Watts GFT, Jizba D, Gawith DE and Gutteridge P:
In this example, higher amplitudes near the center of “Reservoir Monitoring of the Magnus Field Through filled rocks become distinguishable, at least
the 3D seismic volume plot as higher values in the 4D Time-Lapse Seismic Analysis,” Petroleum Geo- in theory and in the laboratory (next page).
center of the 2D amplitude plot. Other attribute sur- science 2, no. 4 (November 1996): 361-372. Most successes in early tests of seismic
faces are computed in the same way (bottom). 5. Greaves RJ and Fulp TJ: “Three-Dimensional Seismic
monitoring occurred where the dramatic
Monitoring of an Enhanced Oil Recovery Process,”
Geophysics 52 (September 1987): 1175-1187. effects of temperature increase and gas
He W, Anderson RN, Xu L, Boulanger A, Meadow B introduction gave clear results. In the 1980s,
and Neal R: “4D Seismic Monitoring Grows as Pro- the first reported time-lapse surveys were
duction Tool,” Oil & Gas Journal 24, no. 21 (May 20,
1996): 41-44, 46. made on combustion and steamflood pro-
6. Ariffin T, Solomon G, Ujang S, Bée M, Jenkins S, Cor- jects, where temperature effects are easiest
bett C, Dorn G, Withers R, Özdemir H and Pearse C: to see.5 Today, in some reservoirs, seismic
“Seismic Tools for Reservoir Management,” Oilfield
Review 7, no. 4 (Winter 1995): 4-17. monitoring of EOR progress continues to be
7. Johnstad SE, Uden RC and Dunlop KNB: “Seismic a cost-effective reservoir management tool.6
Reservoir Monitoring Over the Oseberg Field,” First Successes have also been reported in
Break 11 (May 1993): 177-185.
time-lapse seismic monitoring of gas-oil
Bossert A, Blanche J-P, Capelle P, Marrauld J and
Torheim E: “Seismic Monitoring on the Frigg Gasfield contacts. Norsk Hydro began seismic moni-
(Norway) Using AVO Attributes and Inversion,” pre- toring of the gas-oil contact in the giant,
sented at the 55th Meeting and Technical Exhibition of North Sea, Oseberg field in 1991, and Elf
the European Association of Exploration Geophysi-
cists, Stavanger, Norway, June 7-11, 1993. Aquitaine performed a similar project in the
nearby Frigg field.7

Winter 1996 35
■ Distinguishing gas, 5 Acquisition Concerns
oil and water. Com-
pressional velocities
of gas-, oil- and
water-filled clean 4
sandstones are simi-
Normalized Vp
lar at low porosities,
but are different
3
enough at high
porosities to allow
fluid identification.
Symbols represent
2
laboratory velocity
measurements, while
curves are theoreti-
cal predictions.
1
(From Murphy et al, 0 10
reference 4.)
Until recently, the technique was consid-
ered unproven for monitoring movement of
an oil-water contact (OWC). However, Sta-
toil petrophysicists working on sonic logs
from the Gullfaks field uncovered inconsis-
tencies in log response that turned into
favorable conditions for 4D seismic moni-
toring. Openhole logs from three wells
drilled through the original OWC, com-
pared to logs from wells drilled into water-
flushed areas, showed consistently higher
velocities in the water-filled zones than in
those zones filled with oil. Sonic logs run in
a steel-cased observation well were found
to be unrepeatable from year to year.
Periodic logging with the RST Reservoir
Saturation Tool in the same well showed
that the OWC was rising 13 m [42 ft] per
year. Analysis showed that sonic logs above
the OWC were repeatable from one year to
the next, while below the OWC, sonic
velocities increased where water had dis-
placed oil. If borehole sonic waves could
detect saturation changes at the well scale,
perhaps seismic waves could do the same
across the whole field.
Predicting Success
Seismic response to a change in fluid prop-
erties at a reflector can be predicted through
forward modeling, if elastic properties of the
rock and fluids are known. Relationships
published by Gassmann, and later reworded
by many authors, can be used to predict
density and seismic velocity through what is
called fluid substitution—knowing the prop-
erties of a rock filled with the original fluid,
as well as the properties of the new fluid,
allows computation of the properties of the
newly filled rock.8
36
Water
Oil
Gas
20 30 40
Porosity, %
Fluid substitution requires knowledge of the
density, porosity, and bulk and shear moduli
of the rock frame, the bulk modulus of the
grains making up the rock, and the density
and bulk modulus of the two fluids, all at the
pressure and temperature conditions of the
reservoir. Rock grain properties are usually
measured in the laboratory, while the rock
frame and fluid properties may be measured
in the lab or inferred from borehole measure-
ments of compressional and shear velocities,
porosity and density (next page, top).
The computed compressional velocity can
be used to model the anticipated change in
seismic response at the top of the reservoir,
in this case, the top of the Tarbert sand (next
page, bottom). When the rock is filled with
oil, a normal-incidence reflection has a very
slight swing to the right. When rock is filled
with water, the reflection shows a greater
swing, and to the left. The difference in
modeled responses to oil-filled and water-
filled reservoir becomes more pronounced
when the effects of seismic trace stacking
are taken into account.
This seismic modeling has assumed that
the Tarbert sand is completely filled with
either oil or water, but intermediate stages of
saturation can also be modeled. If the OWC
is an abrupt change in saturation that occurs
over just a few feet or meters—as was prob-
ably the case when the reservoir was at
equilibrium before any fluids were pro-
duced—the contact may also be a seismic
reflector. After years of production, the satu-
ration change may occur over a wide transi-
tion zone, and may or may not appear as a
discontinuity to seismic waves.
Predicting that the “before” and “after” pic-
tures will look different is just the first step
in setting the stage for 4D seismic snap-
shots. Seismic modeling usually assumes
that survey parameters in the baseline and
monitor surveys are identical. Survey
parameters include receiver positions,
source positions, source signature, and any
directivity or coupling effects associated
with the environment. In the past, this has
limited most 4D experiments to land, where
source positions can at least be marked and
receivers permanently implanted and revis-
ited for monitoring surveys.
In the marine environment, permanent
sensors have been used for decades, but
usually for earthquake and other seismicity
detection. Only in recent feasibility and
pilot tests have ocean bottom cables been
permanently installed over reservoirs for
repeat seismic surveying. In what may be
the biggest 4D seismic monitoring develop-
ment so far, BP and Shell have installed per-
manent cables in the seabed overlying the
Foinaven field in the deep waters—
480 m [1575 ft]—of the North Sea west of
Shetlands.9 A baseline survey was acquired
in 1995, and first oil will be produced in
1997. Monitor surveys are planned every
year. In this deepwater environment, opera-
tors hope that seismic imaging of bypassed
hydrocarbons will help optimize future
development plans.
Permanent sensors for marine seismic
monitoring are still in the experimental
stage. And though they will likely ease some
problems associated with imperfectly
repeated experiments, they cannot promise
the same weather, currents or other tempo-
ral conditions that affect all marine surveys.
This does not mean that 4D marine moni-
toring is infeasible. While some geophysi-
cists press the case for exact repeatability,
others say a high degree of repeatability is
desirable but not necessary. Most agree it is
essential to eliminate as many physical vari-
ables as possible between surveys.10
8. Gassmann F: “Uber die Elastizät Poröser Medien,”
Vierteljahrsschr. Naturforsch. Ges. Zürich 96 (1951):
1-23.
See also Murphy et al, reference 4 and Wang and
Nur, reference 4.
9. Kristiansen P and Currie MT: “Seismic Imaging Capa-
bilities Optimize Reservoir Management,” Petroleum
Engineer International 67 (December 1995): 22-23,
25.
10. Von Flatern R: “Adding Time Makes Seismic Data a
Production Tool,” Petroleum Engineer International
67 (December 1995): 17-18, 20-21.
Oilfield Review
 ■ Fluid substitution.
Density, Oil-Filled Velocity, Oil-Filled Replacing the oil
g/cm3 m/sec (red) in the pore

Depth, m
1.95 2.95 2177 7622 Porosity
Gamma Ray space with water
Density, Water-Filled Velocity, Water-Filled (blue) increases the
API g/cm3 m/sec %
0 150 1.95 2.95 2177 7622 0 100 density and com-
pressional velocity
Top Tarbert in the high-porosity
1880 Tarbert sand.

1900

1920

1940

1960

1980

Gamma Oil-Filled Synthetic Water-Filled Synthetic

Ray Seismic
Real Seismic Section
Trace at Well
50 API 125 NIP Stack NIP Stack
1866

1875

Top
Tarbert

1900
Time, msec

1925

1950

■ Modeling seismic response to different fluids using log input. Normal-incidence compressional (NIP) synthetic
seismograms computed for oil-filled Tarbert sand (track 2) and water-filled sand (track 5) show different charac-
teristics at the top reservoir reflection. Real seismic data traces, however, are the result of stacking traces from
many angles of incidence. The synthetic oil- and water-filled equivalents to the real stacking process are in
tracks 3 and 6, respectively. The seismic section near the well is shown in track 4. The seismic trace at the well
location is in track 7, repeated several times for ease of comparison with the synthetics.

Winter 1996 37
 For some time, conventional towed marine
3D has involved careful positioning of in-
sea elements, with differential global posi-
tioning system antenna and acoustic ranging
routinely providing receiver streamer posi- Gullfaks
tioning accuracy to within a couple of
meters.11 Imminent technological improve-
ments that help correct for tidal changes will
improve overall data quality and accuracy. Area
Similarly, improvements in source monitor- Gullfaks of detail
ing possible with the TRISOR source control South
system should lead to the elimination of
NORWAY
most of the variability in individual shot Bergen
characteristics, or signatures. With this vari-
ability minimized, processing should be
able to give “matched” data sets, in which Stavanger
differences between surveys are removed.
To ensure that the data from the two sur- N O R T H
veys resemble repeat acquisitions, the S E A
repeatability must be quantified. This Aberdeen
requires a diagnostic tool to verify that the UK
repeat accuracy requirements are met. In
N DENMARK
addition, there must be a compensation pro-
cedure to fix data that deviate from ideal
conditions. These diagnostic and compensa- ■ Gullfaks field, Norwegian North Sea.
tion tools may be used during acquisition or
processing, depending on the problem.
An example of repeatability diagnosis and
compensation may be found in a study of 0
Dipping seafloor
source depth during marine seismic acquisi-
tion. The depth of the source in the water
can be monitored with the TRISOR source 500

control system to within one meter [3.3 ft].

There may be, without intention or knowl-
1000
edge, a perturbation of the depth, say to 2 m
[6.6 ft] deeper than specified, starting at
Depth, m

some shot point. The effect on the recorded

1500
waveforms may be a slightly changed ampli-
tude, which, when processed, could be
interpreted as a change in seismic reflection Tarbert formation
2000
characteristics at the target. Diagnosing the
origin of the amplitude change as a source
depth shift can be accomplished by moni-
2500 400% vertical
toring the source signature. Continuity of exaggeration
signal is expected from shot to shot if the
0 2000 4000 6000 8000 10,000
source depth is constant. When a perturba-
Distance, m
tion of the standard signal is detected, the
problem may be fixed during acquisition. If ■ Effects of Gullfaks structural complexity on seismic data quality
source monitoring is not done in real time, and modeling. The steep change in seabottom depth over the
the archive of acceptable shot signals may field, strong reflectors overlying the target and a highly faulted
be used to create a match filter, forcing the target combine to make imaging and characterization of Gull-
faks reservoir properties difficult.
perturbed source signal to conform to those
for the correct depth.
Monitoring the quality of the seismic sig-
nal and other essential measurements, such
as navigation accuracy, during acquisition,
allows identification of those perturbations
that can be accepted, those that can be
compensated for, and those that require a
new experiment.

38 Oilfield Review
 ■ Difficult field development. Complex faulting of the Gullfaks reservoir structure presents a challenge to
reservoir management.
The following 4D seismic monitoring
example shows how the combination of
high-quality marine acquisition and innova-
tive interpretation techniques has helped
Statoil optimize development of the Gull-
faks field (previous page, top).
Monitoring Gullfaks Production
Under the current production plan, begun
in 1986, the Gullfaks field in the Norwegian
North Sea will produce 50% of its 480 mil-
lion m3 [3020 million bbl] original oil in
place. About half the remaining oil could be
produced with infill wells, but locating
those wells optimally and reducing costs
continues to be a challenge.
Development of Gullfaks has been ham-
pered by the complexity of the faulted reser-
voir structure (above). New seismic data and
results from development wells have led to
improved knowledge and refinement in
fault mapping. Carefully planned data
acquisition and flexible drilling programs
Winter 1996
resulted in a sound understanding of reser-
voir properties early in the development.
Reservoir simulation has been used exten-
sively to continually evaluate production and
development options for the field. But even
with detailed reservoir models, engineers
don’t know where a fluid front is until it
arrives at a well. By tracking fluid-contrast
fronts before they get to wells, reservoir engi-
neers are able to take action to avoid poten-
tial problems. With 4D seismic monitoring,
the project team—comprising geophysicists,
geologists, petrophysicists and reservoir engi-
neers—will be able to map reservoir
drainage and optimize future production.
Several sets of seismic data have been
acquired over the Gullfaks field. Following
the initial exploration interpretation based
on 2D lines spaced on a 2-km [1.2-mile]
grid, a first 3D survey was acquired in 1979
and a second in 1985. Despite the marked
improvement in the quality of the second
survey, and careful reprocessing in 1992,
the data are still very complex due to vari-
able reflectivity and the highly faulted target
zone (previous page, bottom).12
A pilot reservoir volume was chosen to test
the ability of 4D seismic monitoring to map
changes in the oil-water distribution. The
selected area exhibited reservoir conditions
known to be favorable for seismic monitor-
ing: high porosity—averaging 34%—and a
reliable seismic interpretation.
The three production platforms created an
acquisition challenge for the 1995 monitor
survey: seismic vessels had to navigate to
avoid the platforms, leaving a gap in the
1995 3D volume. The missing volume was
filled in by undershooting—a source vessel
shoots to a separate receiver vessel posi-
11. Beckett C, Brooks T, Parker G, Bjoroy R, Pajot D, Tay-
lor P, Deitz D, Flaten T, Jaarvik LJ, Jack I, Nunn K,
Strudley A and Walker A: “Reducing 3D Seismic
Turnaround,” Oilfield Review 7, no. 1 (January
1995): 23-37.
12. Petterson O, Storli A, Ljosland E and Massie I: “The
Gullfaks Field: Geology and Reservoir Develop-
ment,” in Buller AT, Berg E, Hjelmeland O, Kleppe J,
Torsæter O and Aasen JO (eds): North Sea Oil and
Gas Reservoirs—II. London, England: Graham &
Trotman (1990): 67-90.
39
1985 Attribute Map 1995 Attribute Map (uncalibrated)
2

Apparent polarity
Low
coverage
0

Oil
-2 Gas
Water
Calibration Function 1995 Attribute Map (calibrated ) Nonreservoir

-4
-1 0 1 2
Factor

Low coverage
1

Amplitude
■ Compensating for coverage differences between baseline and monitor surveys. The
1985 Gullfaks survey, acquired before oil production, was able to achieve full coverage
of the top of the reservoir (top left). For the 1995 survey, seismic vessels had to undershoot 0
three platforms, leaving gaps in the 1995 3D volume (top right). If the overburden is not
affected by reservoir production, then comparison of the two surveys allows computation
of a calibration function (bottom left), which, when applied to the 1995 data boosts the
amplitudes to those that would have been recorded had the platforms not been present -1
(bottom right).

-2
-1 0 1
1985 Baseline Survey Factor

■ Attribute space plots showing bad and

good correlation between reservoir proper-
ties and attributes. Reservoir property val-
ues from well data are color-coded dots
with oil in red, gas in yellow, water in blue
and nonreservoir in green. Values are plot-
ted as points in the space defined by the
“apparent polarity” attribute on the verti-
cal axis and on the horizontal axis an
attribute related to a principal component
of a polynomial function fit to the seismic
trace (top). This pair of attributes does not
show good correlation with reservoir prop-
1995 Monitor Survey
erties: members of the nonreservoir class
are mixed with those of the oil and water
classes. A similar plot made with trace
amplitude shows good correlation, and
better separation of classes (bottom).

■ Tracking the top of the Tarbert. The 1985 baseline survey (top) was interpreted for the
reservoir top and all associated faults. The 1995 monitor survey (bottom) was interpreted
for the same feature.

40 Oilfield Review
tioned on the other side of the platform— Amplitude ■ Creating attribute-
but the acquisition geometry, and therefore space plots. In this
the raypaths and seismic energy reflected at example, the goal
is to classify the

Amplitude
the target, were different from the rest of the attributes in the
survey (previous page, top left). Polarity areas marked by
Researchers are testing ways of compen- the triangle, square
sating the undershot data for the difference and circle. Each
shape contains
in the amount of energy reaching the target.
many data points,
One method that appears to give good Polarity plotted as points
results is to assume that across the survey with that shape on
area the overburden does not change with the corresponding
time. This translates into the constraint that graph. When only
two attributes (top)
the total energy contained in the seismic Amplitude
are examined—
trace from the seabottom to the top of the amplitude and
reservoir be constant from one survey to the polarity—the result-
next. A calibration function, or match filter, ing 2D plot fails to

can be applied to the low-energy 1995 data
to fulfill this constraint.
Interpretation Innovations
The goal in interpreting seismic monitoring
data is to extend reservoir knowledge at
well locations into the interwell volume—to
predict reservoir properties where there are
no wells. The Gullfaks well data used for
this purpose consist of basic rock and fluid
information—whether the reservoir sand is
present, and if so, the type of fluids present.
To extend this information away from the
well into the reservoir, first the reservoir vol-
ume, especially the top surface, must be
accurately interpreted from the 3D seismic
data. In the case of Gullfaks, this surface is
the top of the high-porosity Tarbert sand,
and the interpretation includes all faults that
intersect it (previous page, bottom left). The
top picked in 1985 is a positive-polarity
reflection, and the same feature was inter-
preted in the 1995 data. Automatic horizon-
picking software is used to achieve the
required accuracy and avoid inconsistencies
introduced through manual interpretation.
The second step is to characterize the seis-
mic data, at or surrounding the well loca-
tion, that correlate with the information on
reservoir properties at the well. The seismic
information is captured in attributes. For the
Gullfaks interpretation, at least 20 different
instantaneous and volume attributes were
computed for the Tarbert sand.
13. Two unsupervised classification methods were
tested—competitive learning, sometimes called neu-
ral network classification, and K-mean, or nearest
neighbor classification. The supervised methods
tested were Bayesian and Fisher. In the case of the
Gullfaks data, neural network classification was
used, since the assumptions of the other methods
were violated. For more on the methods: Johnson RA
and Wichern DW: Applied Multivariate Statistical
Analysis. Englewood Cliffs, New Jersey, USA: Pren-
tice Hall, 1992.
Polarity
Phase
ase
Ph
Next, the attributes are examined for cor-
relation with a number of classes of reser-
voir properties. Four classes were identified
in the Gullfaks pilot area: 1) oil-filled Tarbert
sand; 2) water-filled sand; 3) gas-filled sand;
and 4) nonreservoir. Correlation of a class
with a set of attributes is determined by the
closeness with which members of the class
are located in a multidimensional cluster
plot with the attributes as axes (previous
page, right). The multidimensional space
thus created is called attribute space. When
members of a class lie close together, or the
cluster is tight, they correlate well with that
collection of attributes. The class members
have a distribution in attribute space that
can be quantified statistically.
In addition to mutual proximity, the mem-
bers of a class must also be well separated
from members of other classes. Several
combinations of attributes must be tested to
obtain a set of attributes that are sensitive to
the class properties and able to distinguish
the classes from one other (above).
To find this correlation between well data
and seismic attributes, only seismic data
near wells have been used. In a manner of
speaking, the seismic data have now been
“trained” to recognize desired reservoir
properties, perhaps in the way a person may
learn to relate sounds or letters to words.
Amplitude
distinguish squares
from triangles.
When three
attributes—ampli-
tude, polarity and
phase—are used
Polarity (bottom), all three
shapes can be dis-
tinguished because
the phase attribute
now distinguishes
squares from trian-
gles. On real seis-
mic data, any num-
ber of attributes
may be examined.
Attributes may be thought of as the letters
needed for the spelling of words—the alpha-
bet of a language, and so can be used to dis-
criminate between different words with dif-
ferent spellings or sounds. When the right
combination of letters yields the desired
words, the next step is to hunt through the
entire volume of sound for other groups of
sounds that resemble the words. For the
Gullfaks seismic interpretation, when the
optimum set of attributes has been found,
that next step is classification of the attribute
surfaces according to the four classes of
reservoir properties.
Several methods in multivariate statistical
analysis have been tested to achieve classifi-
cation. Some techniques, labelled “unsuper-
vised” classification, do not require training
with log data, but simply look for correla-
tions within the seismic data. The number of
classes to find must be specified. “Super-
vised” classifications use log constraints in
an initial classification and make assump-
tions about the statistical distributions of
class members in attribute space. Classifica-
tion results from different methods will be
similar if the statistical assumptions made by
the methods are not violated.13

Winter 1996 41
1985 Fluid Distribution 1995 Fluid Distribution

Oil Gas Water Nonreservoir

■ Fluid distributions before and after ten years of oil production. The best classification, obtained by analysis of 14 attributes, shows how
fluid distributions have changed between the 1985 survey (left) and the 1995 survey (right). Overall, oil (red) has been replaced by
water (blue) in the lower, western sections of each fault block. This indicates a relatively smooth sweep. In some areas, however, such
as in the northern part of the middle fault block, oil and gas (yellow) appear to be in a separate compartment that has not been tapped.

Final results of classification show how the 1985 Saturation Distribution

fluid distribution has changed in the Tarbert
sand over the course of nine years (above).
The fluid distribution changes inferred from
4D seismic monitoring are in agreement
with the expected drainage in the area. Sim-
ulated fluid saturations for 1985 and 1995
were extracted from the drainage simulation
models, which have been history matched
to the well data (right).
These simulated saturation distributions
show the same general features as the seis-
mic interpretations. Oil has been drained
from the westernmost fault blocks, and from
the western edges of the two central fault
blocks. A fault in the northern portion of the
center block appears, however, to be isolat-
ing oil and gas from the southern part of the
fault block, where the oil sweep appears to
be effective. 1995 Saturation Distribution
Based on the positive results of the 4D
seismic monitoring study, two major deci-
sions have been taken toward modifying
Gullfaks development. First, a new 3D sur-
vey was acquired to cover the rest of the
field in 1996, together with simultaneous
logging of pressure and saturation distribu-
tion in key wells. The new 3D survey also
covers satellite fields to the south and will ■ Fluid saturation
be the baseline survey for future monitoring distributions from
in those fields. Second, a new extended- reservoir simula-
reach well is being drilled from the C Plat- tion. Simulations of
the 1985 (top) and
form in the eastern part of the field, where 1995 (bottom) satu-
production from zones immediately below rations show sweep
has declined. The trajectory of the well is patterns similar to
designed to tap multiple compartments those imaged with
the 4D seismic
identified by the 4D survey as containing monitoring. Red
bypassed oil. indicates high oil
saturation.
0 100
Oil saturation, %
42 Oilfield Review

4D Seismic Data
Petrophysical Modeling
■ Central role of the common earth model in the many disciplines contributing to 4D seismic monitoring.
Now that Statoil has confirmed the feasi-
bility of 4D monitoring in the prime condi-
tions of the Tarbert formation in the Gullfaks
field, other more challenging applications
await. Deeper reservoir layers with lower
porosity might also be candidates for moni-
toring, pushing the technique to its limits.
4D Monitoring Will Change with Time
Many operators and service companies are
talking about and gaining experience with
4D seismic monitoring. The technique is
expected to decrease uncertainty in reser-
voir models and reduce risk in drilling new
wells. With faster and improved-quality 3D
seismic data acquisition, the investments in
a 4D seismic monitoring survey have
become low enough to be offset by the
expected increase in production associated
with a better understood reservoir. But 4D
seismic monitoring is still in its infancy: 3D
seismic technology took ten years to
become established. More field studies and
further developments are needed to prove
the value of additional knowledge brought
by 4D monitoring.
Winter 1996
Well Logs
Common Earth Model
Classification System
Further work is required in many areas to
bring this extraordinary technique into ordi-
nary practice. Repeatability of acquisition
may never be perfectly realized, but the lim-
its and tolerances of acquisition differences
should be better appreciated. There will
probably not be a single acquisition scheme
that will have universal application for 4D
monitoring, but rather a range of solutions
to cover a spectrum of different reservoirs.
Some of the failures of early attempts at
4D monitoring have been attributed to the
“incidental” nature of time-lapse surveys—
most 3D surveys are acquired without any
intention of comparing the results to a later
monitor survey. With 4D monitoring on the
horizon, there is great interest in making the
extra effort to ensure that baseline surveys
have all the information imaginable to
extend their value as long as possible.
More research and experimentation with
permanent sensors, both in ocean-bottom
cables and in boreholes, will advance
knowledge of hardware limitations and
bring down the costs associated with their
occasional use today.
Reservoir Simulations
Seismic Modeling
Work remains to be done on understand-
ing the relationships between seismic
attributes and rock and fluid properties. Cur-
rently, finding the right attributes for a
desired reservoir property is a time-consum-
ing interpretation project requiring an
expert. Forward modeling may help predict
which attributes are useful for a given fluid
change or rock type, and might allow some
automation of the interpretation process.
Central to integrating the efforts of geolo-
gists, geophysicists, petrophysicists and reser-
voir engineers is the model—a common
earth model—that can be tapped and refined
at every stage of reservoir management
(above). Continued refinements to 4D seis-
mic monitoring will be driven by the project
teams charged with optimizing recovery
from existing fields. These teams require
methods that allow them to improve the
location and timing of development drilling,
to not only understand but also control reser-
voir behavior. With that as a goal, seismic
monitoring may one day become the most
powerful of reservoir management tools.
—LS
43