Professional Documents
Culture Documents
Management
Course Notes
By
J M Marsh
This edition
November 2005
What is 4D Seismic?
“Any use of sequential seismic surveys which improves either the static or dynamic description of the subsurface, can be
described as 4D, or time-lapse, seismic.”
It is often thought that 4D, or time-lapse, seismic techniques give information of the dynamic behaviour of
the reservoir, that is a change in seismic response due to change in reservoir pressure and/or fluid content.
Although this is true, we have found in BP that sequential seismic surveys can provide information on the
reservoir static condition as well. This can be due to improved seismic data quality, or simply building up
the fold.
4D Fundamentals
The propagation of sound energy in the subsurface depends on the physical properties of all the rocks
through which it travels. These properties include mineralogy, fabric (lithology), porosity, fluid content,
pressure etc. Some of these can change if production or injection occurs causing as a result a change in the
seismic response. If we can measure the seismic response accurately enough to discern the difference, then
we may interpret what changes have occurred to cause those differences. Normally we assume that
lithology, porosity, rock physical properties etc are either constant, or can be accounted for in some way.
In common with many other remote sensing techniques in oilfield technology, we cannot measure the
properties we are most interested in directly, e.g. pressure and saturation. We have to infer them by
measuring other properties which they affect, for example seismic velocity or amplitude. However, seismic
is the best technology currently available for indicating changes away from well control.
How Is 4D Useful?
• Identifies Infill Drilling Options
• Detects well intervention opportunities
• Locates untapped reserves
• Monitors field development (surveillance)
tÜ~í=açÉë=fí=kÉÉÇ\=
• High quality seismic , ( i.e. signal > noise )
• Repeatable image
• Quick turnaround time
• Good linkage with engineering (e.g. simulation)
oÉãÉãÄÉê=>=
• 3D imaging works better some places than others
• Salt, gas “chimneys”, step dips, rugged surfaces are difficult
• 4D works better with some reservoirs than others
• Shallow, unconsolidated, clastics (“soft” rocks), high f, oil filled
So what does 4D do? By detecting changes in the reservoir image, it allows interpretation of the changes
in pressure or saturation that are taking place in the reservoir. This process usually requires data from
other sources such as well logging, production and injection measurements, and reservoir simulation to
arrive at the best overall interpretation. However, the increased certainty of how the reservoir is behaving
dynamically can reveal opportunities for further drilling, or for well interventions. For example, the
presence of bypassed oil may have been suspected, but 4D may indicate its location.
For best results 4D needs high quality repeatable data. Good repeatability means that no change will be
seen in areas where there has been no change. Poor quality, or noisy data will show change in areas where
none has taken place; it may look random or systematic depending on the origin of the noise. More will be
said about this later in the course.
Quick turnaround time is an advantage as the required results will be available sooner. The reduction in
turnaround time at all stages of the work is an important continuing aspect of the development of 4D.
Also it must not be taken for granted that practitioners of a technical discipline are equally knowledgeable
about all aspects of their subject. For example, processing geophysicists may not be fully conversant with
the intricacies of analysis. They may not be fully aware of the effects of rock physics. Neither may all
petrophysicists!
Acquisition - Experimentation
Although it is maturing, time lapse seismic is still a new technology, with commercial application growing
over the last five years. BP along with other operators carried out some early experimentation in the
1990’s. There has been an appreciation of the potential for 4D since the start of the 1980’s and a small
number of research projects have occurred since then. However it was not until the mid 1990’s that 4D
started to be used for commercial purposes. Since then its use has increased significantly. The profile of
BP’s use has reflected that of the industry as a whole.
The period of acceptance has been short, over the past 5 years, a total of 35,000 km2 of 3D seismic has
been acquired for time lapse analysis.
In BP’s case, the bulk of 4D work has been carried out in the N Sea, but ‘globalisation’ is taking place with
11 surveys for 4D analysis being acquired to date outside the N Sea by BP or its partners e.g. Pompano
(GoM), Ram Powell (GoM), Amberjack (GoM), Troika (GoM), Girassol (Angola), Chirag (Azerbaijan),
Yacheng (China), Prudhoe Bay (Alaska), Immortelle (Trinidad), Ha’py, (Egypt).
Since 1997, seismic has been used as the primary infill target selection tool. This has been achieved by the
use of lithological imaging (sand/shale discrimination), fluid imaging (oil/ brine discrimination) and fluid
change imaging (changes from oil to brine, Whitcombe, D. et al., 2002). A target is considered viable if
these seismic attributes are consistent with local well surveillance data. This process has enabled the
Whitcombe, D., Connolly, P.A., Reagan, R., L., and Redshaw, T., C., 2002, Extended elastic impedance for fluid and
lithology prediction, Geophysics, Vol 67, No 1, P 63-67.
The surveys are all surface towed except for the FARM (Foinaven Active Reservoir Management) surveys,
conducted in 1995 and 1998 which had both surface data and seabed hydrophone data taken over part of
the Foinaven field. It was this trial that gave the strong indications of 4D success that gave impetus to the
acquisition of the first repeat survey in 1999 which was carried out purely for time-lapse purposes.
West of Shetland now has the greatest number of repeat surveys on which to interpret the subsurface
dynamic performance. All new wells drilled since 1999 have been designed using the 4D results.
Costs and Benefits
Depending on the local conditions, 4D can be relatively cheap compared to other forms of subsurface
data. The West of Shetland 1999 survey cost just over $2million for 390km2 West of Shetland fields are
developed by subsea wells in around 450m water depth. Any form of well intervention is very expensive,
and a single entry to take a PLT log may cost twice as much as the entire 4D survey. A new well may cost
an order of magnitude more than the 4D.
Of course this may not always be the case, depending on the comparison of costs. For instance, on
platform wells, it may be very much cheaper to run production logs.
So What Do You Get ?
Images of an area of turbidite sand reservoir are used to illustrate the use of 4D in history matching. This
channel section has a water injector in the north, and a sub-horizontal producer in the south. The map of
pressure change ( p) produced from the simulator shown on the right is compared to the seismic
amplitude image obtained after two months, or 1.3 million stb production.
The observed areal distribution of amplitude brightening, is in accordance with that expected from the p
plot. However, it will be noticed that brightening occurs in areas both of pressure increase (in the north),
and pressure decrease (in the south). This indicates that there is a composite effect of pressure and
saturation.
However, this comparison boosts confidence that the model is ‘getting things right’ in terms of areal
distribution of pressure and saturation change. It also shows that the definition in the model is much
coarser than in reality, as shown by the seismic. Bear in mind also that the seismic response is subject to a
level of noise which creates some uncertainty as to the precise distribution of the areas of change.
Pressure (yellow line) and saturation profiles are shown in the middle figure.
Seismic amplitude is dimmed by high pressure + high Sw, and by low pressure
It is brightened by high pressure + low Sw, and by high Sg evolved from oil due to pressure reducing
below bubble point.
Net 4D Changes
This diagram summarises the 4D effects seen as a result of saturation and pressure changes. The
‘brightening’ and ‘dimming’ descriptions refer to the case, normally observed, where the over and
underburden are have a greater impedance than the reservoir.
Early Success
Use of 4D has led to the improvement of understanding of fluid movement in the reservoir, which in turn
has allowed the location of un-accessed reserves to be identified. On this basis, drilling options have been
identified, leading to the creation of value through increased production rate &/or recovery.
This diagram ranks some BP fields in terms of reflection coefficient, and thereby indicates their 4D
potential. Valuable 4D results have been obtained across the range. The diagram also shows noise
thresholds typical of the N Sea. The black line shows the threshold associated with similar processing,
where 3D surveys are processed in a similar, but not identical manner. Noise is reduced if both data sets
are processed identically. Typical levels of this parallel processing are shown by the blue line.
Harding central and south are high f(35%) and k (2-15 D) reservoirs at a depth of ~1650 mTVDss. Its
72m saturated oil column is sandwiched between primary gas cap and aquifer. 4D has been used to
monitor fluid contact movement and thereby position new wells into the thickest parts of the remaining oil
column to avoid gas and water encroachment for as long as possible.
Moving down the list, Forties was one of the first of BP’s ‘4D fields’. A combination of lithology and fluid
imaging rescued the infill drilling programme by locating un-accessed oil and thereby reducing uncertainty
in selecting infill options.
Andrew was not ranked highly as a 4D candidate, yet it has yielded excellent results, revitalising a w/o &
drilling programme. A paper in the main conference will describe the work that has been done here.
Working in the S/N region of 1, the availability of PLT logs in two horizontal producers gave the
necessary confidence to 4D interpretation.
Finally, Marnock. A gas condensate field in the Central North Sea. Using a 3D survey not acquired for
4D purposes and a pre-production 3D, useful time-lapse data have been obtained to detect pressure
change. Acquisition was not optimum for 4D as the surveys were not aligned. 4D Acquisition in 2001
The nine target fields of four surveys operated by BP and two operated by BP’s partner, Shell were
generally more challenging for seismic in terms of both reservoir structure, and rock and fluid properties.
Additionally the 2001 programme allowed testing of improved acquisition technology through the Magnus
survey.
Sources of Value: Technical Understanding
Time lapse seismic data can help both static and dynamic understanding of the reservoir. This diagram
indicates where value has been derived in the BP fields surveyed.
Changes in pressure and fluid saturation can highlight reservoir layering and faulting by building contrasts
on either side. This has been a useful aid in about half the reservoirs studied.
On the dynamic side, identification of compartments and tracking the movements of water has been the
most used features.
As more repeat surveys are acquired it is becoming easier to use the data to manage production and
injection strategy. Such management can be better achieved if surveys are acquired frequently. This keeps
the observations of field performance more closely up to date, and thereby more useful.
Conclusions
• 4-D is now an important tool in reservoir interpretation and management
• Data quality is an essential part of imaging and interpretation
• Emphasis has been on qualitative/semi-quantitative interpretation, ability to quantify is growing.
• Geoscientists, Petrophysicists and Reservoir Engineers must work closely
• Early value has been derived using reprocessing and fast-track results
Progress
• Dramatic improvements in turnaroundContinuous improvement in seismic processing
• Fully optimised acquisition
• Richer integration with production data
• …. leading to optimisation of repeat frequency
Continuous (or frequent) Monitoring
With permanently installed subsea receivers (OBC) and instrumented wells, we will be able to obtain
continuous well data and frequent repeat seismic surveys. The combination of these will provide a
powerful means of reservoir observation on which to fine tune development and production strategy.
The are several challenges incipient in this arrangement. One is how to process and interpret the seismic
data rapidly enough. Another is how to ensure that the correct amount of data is saved.
Next………
Note: The equation shown is true for normal incidence only angles (vertical travel paths) only. The ray
paths shown in the diagram are therefore misleading. Seismic amplitudes can be calculated for non-normal
incidence angles, but it is more complex.
Rock Properties
In practise this means we combine sonic and density logs to create an AI log. The AI log may be
compared directly with inverted seismic data or used to generate a synthetic seismogram for comparison
with seismic reflectivity data.
Log data QC
Warning: Logs should be carefully quality controlled before use in rock properties analysis. Non-reservoir
zones in particular may not have been edited during routine petrophysical interpretation.
Sonic and density logs are both shallow reading tools, reading only a few inches from the borehole into the
formation. As a result the sonic and density logs may measure the formation contaminated by drilling mud
filtrate. Fluid substitution (next slides) is required to correct the log measurements to values appropriate
for true formation fluids
Fluid Substitution
Fluid substitution involves the replacement of the in-situ fluids (in this case brine) with a new fluid type (in
this case hydrocarbons with irreducible brine). The density and velocity values are calculated for the new
fluid fill.
In general terms fluid substitution of the compressional velocity can be viewed as comprising:
-calculation of overall/combined hardness from sonic logs.
-estimation of the fluid hardness from fluid property data (Pressure, Temp, Sw, Salinity, API, GOR).
-subtracting the pore fluid hardness from the overall/combined hardness to give the rock frame hardness.
-estimating the new fluid hardness from fluid property data.
-adding the new fluid hardness to the rock frame hardness to give the new overall/combined hardness
-calculation of the new sonic log values from overall/combined hardness.
In practise the pore fluid contribution to the overall hardness is more complex and we usually use the
Gassmann equations.
Note: Solving the Gassmann equations requires a knowledge of both the compressional and shear velocity
data, bulk density, fluid properties, mineralogy.
K + 43 U
Vp =
ρb
where Vp = compressional velocity, K = bulk modulus, U = shear modulus. K & U are measures of how
hard a rock is.
Fluid Substitution
Although the overall hardness is not a simple addition of the frame and pore fluid hardness we can
generalise that:
For hard rocks (large bulk hardness) a change in the pore fluid type results in a small percentage change to
the overall/combined hardness. As a result a change in the pore fluids is unlikely to be detectable on
seismic data.
For soft rocks (small bulk hardness) a change in the pore fluid type results in a large percentage change to
the overall/combined hardness. As a result a change in the pore fluids is likely to be detectable.
Fluid Substitution
Density increases linearly with increasing the gas saturation (as density is simply the volumetric average of
the fluid and matrix densities).
Compressional velocity has a more complex response to changing saturations (resulting in part from the
increased complexity of the Gassman equations and in part because of the dependence of velocity on both
hardness (or incompressibility) and density). The first few percent gas saturation introduced into the pore
space results in a relatively large decrease in velocity (a little bit of gas goes a long way). As a result
reservoirs that have moved below (or above) the bubble point are often among the best candidates for 4D
seismic.
Temperature Changes
The effect of temperature changes on rock properties is usually small relative to changes in fluid type and/or pressure.
Exceptions are steam flooding of heavy oil reservoirs or a temp change that results in movement above below the bubble point.
In these cases the temperature change results in a large change in pore fluid properties.
The effects of large changes in pore fluid properties should be modelled as outlined in the fluid substitution section above.
Note: The effects of temperature on the rock frame are not well understood but are generally assumed to
be small. However large temperature changes may effect the reservoir stress state and hence cause a
velocity change, especially if the temperature change results in fracturing.
Effective Pressure
•Effective pressure is the difference between overburden pressure and pore fluid pressure and represents the actual stress or load
experienced by the rock frame
Similarly for a reservoir rock – it is not the absolute values of the weight of the overburden and the pore
pressure that is important – it is the difference in pressure (or effective pressure) that controls how the
rock compacts and how the rock properties change.
Pressure Changes 1
Reservoir pressure depletion = compaction
============================================= = increase in density
============================================= = increased hardness
============================================= increase in Vp
A drop in reservoir pressure results in an increase in effective pressure. The increase in effective pressure
results in compaction of the reservoir rock frame and an increase in velocity and density.
Remember:
K + 43 U
Vp =
ρb
Where
Hardness = K + 43 U
Pressure Changes 2
On the right of the plot reservoir pressures are high and effective pressures are low. On the left of the plot
reservoir pressures are lower (depleted reservoir) and effective pressures are higher resulting in increased
compaction which results in harder rock with faster velocity.
Hard rocks show less velocity variation than softer rocks for any given effective pressure change. Lab
testing of core plugs are typically used to determine how the velocity of a rock varies with effective
pressure. The velocity changes with effective pressure can be large and should by included in 4D
modelling if pressure changes are moderate-large.
Bulk density responds to changes in effective pressure in a similar manner to velocity – low porosity rocks
show less bulk density variation than high porosity rocks for any given effective pressure change. Lab
testing of core plugs to determine the porosity variations with effective pressure are often completed as
part of the standard petrophysical analysis. Bulk density variations with pressure tend to be small and are
sometimes ignored (esp in low porosity rock).
Note that if the effective pressure is increased and then decreased back to the starting conditions the
velocity and bulk density may not return to the starting values due to inelastic behaviour of the rock.
Pressure Changes 3
We have discussed the effects of pore pressure changes on the ROCK FRAME. There is also a pressure
effect on the PORE FLUID. In a depletion scenario the pore pressure is reduced which results in a
decrease in the hardness of the pore fluid. This counteracts the rock frame effect - the reduction in pore
pressure results in an increase in effective pressure and an increase in the hardness of the rock frame.
In a injection scenario the pore pressure is raised which results in a increase in the hardness of the pore
fluid. Again this counteracts the rock frame effect - the increase in pore pressure results in an decrease in
effective pressure and an decrease in the hardness of the rock frame.
The changes in pore fluid properties resulting from pressure changes are modelled as outlined in the fluid
substitution section above.
Summary
• Definitions: elastic moduli, stiffness, Poisson’s ratio, Vp, Vs
• Log analysis
• Log quality
• Fluid substitution
• Temperature effects
• Pressure effects
The effects of changing pore fluid (brine replacing oil) and pressure are shown as a function of reservoir
pressure. The fluid change (dark blue) results in an increase in AI in the reservoir of 7-10%.
The pressure effect (black) shows the AI change in the reservoir for a constant Sw. A decrease in reservoir
pressure results in an increase in AI – but an increase pressure results in a decrease in AI.
Combining the fluid and pressure effects is a simple addition. Where the reservoir pressure decreases the
fluid and pressure effects both increase AI and the combined effect is a large increase in AI. Where the
reservoir pressure increases the fluid effect remains an increase in AI but the pressure effect becomes a
decrease in AI - the combined effect is a small increase in AI.
Footnote: The fluid effect is smaller at lower reservoir pressures (higher effective stress) as compaction
has resulted in a harder, lower porosity rock.
Results of 4D modelling
The modelled density and velocity logs can be used to examine the changes in seismic amplitudes as a
result of production. In this example an injector location is modelled - the combination of a large pressure
increase and increased Sw results in a decrease in AI and a brightening of the seismic amplitudes.
In addition to the changes in seismic amplitudes there may be a change in the position of the seismic
events (as production produces changes in velocity). In this example there is a decrease in velocity over
the reservoir interval – which results in a small push down of the base reservoir reflector.
Summary of 4D changes
The effect on bulk density of changes in fluid composition and pressure are large in high porosity rocks.
Similarly the effect on velocity of changes in fluid composition and pressure are large in soft rocks.
This section of the presentation discusses how we can look for production induced changes in seismic
response on difference cross sections and maps. It will involve a discussion about seismic noise; its effects
and how to reduce them. This is done throughout the acquisition, processing and analysis workflow.
Difference sections 1
Here we see a reservoir section at two stages of production. On the left the reservoir is in an undisturbed
state. In the middle there has been some production resulting in a rise in the OWC. The colours could
represent the fluid types, or they could represent acoustic impedance. To the right of each of the columns
described is a plot of the reflectivity coefficient. At the interface between shale and oil sand there is a
negative reflection coefficient (RC)*. This indicates that the shale has a greater impedance than the oil
sand. At the fluid contact there is a kick to the right because water filled sand is harder than oil filled.
Below, shale is harder than water filled sand.
If we subtract one column from the other, we will only see the differences. Here we have subtracted the
left hand (or ‘pre’) column from the middle (or ‘post’) column. It is differenced this way so that the swept
zone has the properties of water invaded reservoir.
This is perhaps the most logical way of differencing, as a reduction in impedance relative to initial
conditions will appear as a reduction in the difference.
Difference sections 2
• Black colours indicate an increase of impedance
In a real example shown here, the same differencing convention has been used (ie diff = post – pre). The
black colour indicates an increase in impedance due to the replacement of oil by water.
Is the 4D signal > background noise ?
• The magnitude of the 4D signal ‘S’ will depend on:
o the rock and fluid properties
o the size of the production effect (change in pressure and/or saturation)
> The higher the S/N, the more quantitative the 4D data, and the better risk reduction it provides
In reality, seismic data with a signal to noise ratio of 1 may be very informative. This is because the signal
will be coherent and related to the reservoir structure or geology, whereas noise is likely to have a greater
Improving Acquisition
With time, the technical requirements of seismic acquisition have become more demanding in order to
improve repeatability. Important steps have been to repeat survey azimuth, source postion, feather.
1 NRMS
• Compares the size of the difference data to the original
• Value range = 0 (perfect repeatability) to 1.41 (random noise) to 2 (anti-correlated)
• NRMS is sensitive to phase and time shifts
2 Cross coherency
• Measures the correlation between the two vintage
• Values in the range 0 (no correlation) to 1 (perfect)
• Cross-coherency is not sensitive to phase and time shifts
• Cross-coherency is influenced by the amount of coherent energy (‘geology’ & multiples) in the input volumes
A 4D effect will generally only be detectable if the strength of its signal exceeds that of the noise. There
are a number of ways of quantifying noise and three measures are noted above.
RMSA − RMSB
NRMS =
RMS ( A + B)
2
Cross-coherency is influenced by the amount of coherent energy (‘geology’ & multiples) in the input
volumes e.g. reduction in multiple energy on input volumes will result in worse coherency leading to an
erroneous assumption of a decrease in repeatability.
Noise thresholds 1
The green vertical band shows the size of background noise level obtained in the Forties Field by applying
‘similar processing’ to the seismic data. In Forties, the 4D signal just exceeded this noise level.
Noise thresholds 2
The green vertical band now shows the size of background noise level obtained by ‘parallel processing’.
This opens up the 4D possibility of using 4D data an many more fields.
Seismic acquisition
• Make the acquisition as similar as possible
• Acquisition alignment is the key thing to get right
By monitoring the repeatibility throughout the processing sequence any mistakes can be spotted quickly.
Extended elastic impedance for fluid and lithology prediction. Whitcombe, D N., Connolly, P A., Raegan, R L. and
Redshaw, T C. Geophysics, 67 No 1, 63-67
Coloured Inversion
Seismic inversion is regarded by the industry as a specialist activity.
o typically takes several weeks to carry out
o skills generally not in Asset teams
Coloured Inversion* was developed in BP as ‘80% solution’:
o very quick – e.g. overnight
o surprisingly accurate – appears to be more robust than alternatives
o now available as add-on to Landmark software
*
Fast-track ‘coloured’ inversion. Lancaster, S. and Whitcombe, D N. SEG Conference, Calgary, August 2000.
Using AVO
Production:
o As Sw increases, Acoustic Impedance increases
We can think of the interpretation and integration of time lapse seismic data as a continuum ranging from
the qualitative to the quantitative. In this diagram the continuum is discretised into four divisions.
1) Direct Inference from 4D Data: In many cases, useful conclusions about the dynamic behaviour of a
reservoir may be drawn from examination of the seismic data alone, given that there is an adequate
understanding of where production and injection has taken place. Imaging of fluid fill may indicate where
the hydrocarbon saturation has changed, and where it has not, giving a good idea of where remaining oil
may occur.
2) Visual Comparison with Simulation: Taking things a step further, comparison of seismic conclusions
with the predictions of a simulator may strengthen confidence in interpretation, or indicate where there are
problems. A solution can be sought, and used to improve the performance of the simulator – or the
interpretation of the seismic!
3) 4D seismic synthetics: With a knowledge of the rock properties, we can calculate the changes in
impedance resulting from changes in reservoir pressure and saturation. The history matched simulator can
provide values of pressure and saturation for each cell, and so we can calculate impedance for each cell.
We can therefore compare calculated values with real values and use any mismatch to inform our
interpretation of the reservoir.
4) Derivation of P and S from seismic: The ultimate aim is to quantify pressure and saturation directly
from the seismic. This would give us numerical data in a 3D sense to use in calibrating the reservoir
simulator.
In this example, the hydrocarbon zone is clearly visible as a bright. The oil-water contact is clearly
identifiable. There was a concern that the northern part of the gas cap may be isolated. The 4D data was
of direct use in determining whether this was the case.
The change in seismic response showed less change than expected from the reservoir model, but broadly
agreed with the changes in reservoir pressure which had been modelled in the simulator. The lack of
change in a small segment to the north east showed a potential new target if the interpreted hydrocarbons
were isolated from the rest by a barrier.
Results: Surprise!
A well was drilled into the new target. The depletion was low, but the quality of the rock was much poorer
than expected. This shows that interpretation of null signals has to be made very carefully. In this case,
the favoured interpretation was one of good and undepleted reservoir, however the difference in seismic
response was due to the lack of change in the pressure, and told nothing about the reservoir quality.
• 4D successful in helping the well stay away from water; reservoir pressure 1400 psi higher than existing wells,
but…
• Much poorer reservoir presence & quality than expected
• Need to consider what a “lack of 4D signal” might represent:
o Good quality reservoir that has not been drained?
o Lower quality reservoir?
o Low N/G reservoir?
Integration with engineering: visualising production data
In this map of the Forties Field, we are co-visualising fluid saturation change with production volume. In
this way we can see together where water has replaced oil, and where oil has been produced.
Overlay Visualisation
One of the major problems in analysing 4D is the capability of computer software to allow easy
comparison of different data sets. The lower part of the slide is an overlay combination of the three upper
images which from left to right are: seismic amplitude, seismic coherence and a simulator grid.
The combination allows us to co-visualise the effects of pressure and/or saturation change (derived from
the amplitude), possible flow disruptions, or baffles (indicated by the seismic coherency) and the simulator
grid on which cell to cell transmissibility is displayed. Seeing these features together makes it easier to test
if the model barriers are inserted correctly, and to change them if not.
Flow Barrier Breakdown
In this example, we are history matching the performance of a producing well. The bottom hole pressure
and the GOR are shown as the blue crosses on the graphs on the left of the slide. Values calculated by the
simulator are shown as the purple lines. In the BHP plot at the top left, the match between measured and
calculated data appears to be good up to about half way across the figure, then the match deteriorates.
Bottom hole pressures are measured by downhole gauges, and are therefore reliable. GOR is derived from
allocated well production, and so is less reliable, as correct allocation depends on the frequent well testing.
A reason for the mismatch in the right hand portion of the plot had to be found. Here the measured
pressure exceeded that predicted by the simulator. At the same time, the estimated GOR was much less
than that predicted by the simulator. Comparison of the two seismic sections indicates a possible
explanation for this. Prior to production the main sand appeared to be disconnected between the injector
and the producer. By the time the repeat survey was acquired, the effects of injection were visible, eg
increased amplitude in the oil around the injector, dimming where the injection water is located. The
coherence of the reservoir sand appears to have improved. On coherency maps of this horizon, the same
phenomenon appears. The channel boundary appears to have weakened. If this indicates an increase in
the transmissibility of the channel margin, then modelling this change should result in a better history
match. This was found to be the case.
4D Interpretation and Integration 3
4D seismic synthetics: With a knowledge of the rock properties, we can calculate the changes in
impedance resulting from changes in reservoir pressure and saturation. The history matched simulator can
provide values of pressure and saturation for each cell, and so we can calculate impedance for each cell.
We can therefore compare calculated values with real values and use any mismatch to inform our
interpretation of the reservoir.
Co-visualisation
In interpretation, it may be necessary to co-visualise simulation results with seismic data. This can be
facilitated by using a software program such as STP. Written by ARC CLS, this is designed to facilitate the
transfer of seismic like data (i.e. regular in X, Y, Z) between different vendor packages. STP currently
supports Landmark, GeoQuest, RMS, and also generic SEGY format files. This means that data from any
of these systems (or formats) can be passed into any other. STP uses 3rd party application toolkits to read
and write to the respective databases, so that the end user is shielded from changes in format in the 3rd
party systems.
STP obviates the need for intermediate SEGY-like files, and means that the transfer is accomplished in one
step, rather than an output and then input step.
Simulation models rarely use the same geometric grid pattern as the seismic. Resampling reservoir models
onto the seismic grid enables the co-visualization of simulation data in the seismic environment, which is
useful when interpreting 4D signals.
STP additionally enables various data manipulations to be carried out as it transfers data. For example it
can simultaneously read two data sets from SeisWorks (two different seismic vintages perhaps), carry out a
mathematical operation (differencing for example) and then write the data (back into SeisWorks, or to
RMS for example).
Sim2Seis: SE Forties
These figures show the AI change based on simulator predictions of pressure and saturation for the same
two periods as shown earlier.
Quantitative Integration
The main part of quantitative integration of 4D into reservoir management is to use it to improve our
reservoir simulations, and thereby our ability to predict reservoir performance. There are many challenges
in doing this: the scales and resolution of seismic and simulation grids are usually very different; repeat
seismic surveys are typically few in number, and at discrete times in a field’s history.
With more frequent seismic surveys we will be in danger of not being able to process and interpret the data
fast enough, leading to a certain amount of data indigestion!
Quantification Goals
So the goal of quantification is to make our models more representative of the subsurface ‘truth’, and
thereby better at prediction. As the problem is a function of many variables, there are multiple models that
can satisfy the performance history. Additional data gives additional opportunity to discriminate between
these multiple models.
To eliminate unlikely models, we need to concentrate on what is significant and important, and be able to
do it quickly. A complex reservoir model constructed in the ‘conventional’ way can take a year or more to
produce.
Yet good predictions yield value by enabling correct technical and commercial decisions to be made. The
value in acquiring the data to do this results in a reduction of risk.
A means of checking the model is to use it to calculate the seismic properties of its components. This can
be done as previously described, based on a knowledge of the rock properties and the prevailing conditions
of fluid saturation and pressure.
Section Comparison
These two cross sections illustrate planes from a model where this process has been carried out. The
upper section is a coloured inversion of full stack data and therefore represents the acoustic impedance at
this location. Below it is an equivalent section of synthetic impedance, calculated from the reservoir
geological model and re-sampled on to the seismic grid. The good comparison gives comfort that the
geological model is representative of the geology in this area.
Synthetic impedances are shown only for the immediate region of the reservoir units. This is because the
calculation is derived from the simulator which is only populated with reservoir units. The overburden and
under burden are not included. This can be an important consideration if it is important to consider the
geomechanical state of the surrounding rocks.
Section Comparison
The result of this process for the model shown before is added to the cross section figure. The coarse
granularity resulting from the coarser grid can be seen, but the overall match is satisfactory.
On the assumption that the model building and upscaling checks have been done as described, this
diagram illustrates the essence of the ensuing traditional subsurface workflow that leads to a history
matched simulator. Typically this process takes several months and in spite of upscaling leads to a high
definition model which takes hours to run. Modifications to the structure and geological model are
difficult to do, and attempts at the quantification of uncertainty is limited to a sampling of the effect of the
uncertainty range made by sensitivity studies where individual parameters are adjusted in turn.
Bear in mind that a detailed and slow model is not always appropriate. If the business decision is to define
a pipeline capacity, but the reservoir is compartmentalised, the fluids are complex, seismic data are poor,
and there are few appraisal wells, why attempt to model in detail, or attempt to add detail, when it may be
unimportant to the decision?
BP’s Top Down Reservoir Modelling (TDRM) philosophy holds that the model complexity should not be
more than absolutely necessary to enable the required decision to be made. In a recent study, an ‘industry
standard’ fine scale model which ran in 10 hours was replaced by a simpler model which ran in 10 minutes
and gave similar results. Given the relatively small amount of data, and large uncertainty, this approach
was justifiable. With short run times, many runs can be made to explore uncertainty, and TDRM, uses a
case management system to generate several hundred model runs (~10*number of variables) based on the
uncertainties defined and quantified by the integrated subsurface team. The history match is optimised
using a genetic algorithm which identifies the quality of fit using an objective function, and generates new
model runs based on higher quality fit parent cases.
The history match worktime can be reduced by an order of magnitude. In the case quoted, 310 hours of
effort with the fine scale model, was replicated and enhanced with less than 10 hours using TDRM.
In TDRM, a scalable solution is developed that begins with basic analytical calculations and moves to
material balance, coarse model, FFM and detailed full physics model. TDRM focuses on downscaling
rather than upscaling. A FFM with detailed geological and displacement descriptions is still required for
deep technical analysis, but this is built later once the "best solution" has been supported by the TDRM
results rather than on the basis of an initial set of assumptions.
In the workflow shown in the figure, the simulation model generated from the geological model has the
reuired complexity to solve the current problem, but no more. It is used as a basis for generating many
different simulation runs. The uncertain properties are identified, e.g. permeability, transmissibility barriers
and a range for each is determined. These ranges are used by the genetic algorithm (GA) optimiser to
populate the simulation runs. The simulation results are then compared to the matching data, e.g.
production rates, bottom hole pressures, watercut and a numeric match quality determined using an
objective function. The objective is to minimise the mismatch. In doing this, it is likely that several
models will have the same match quality, in spite of being quite different in terms of properties.
To show the improvement in fit another way, in the lower diagram, each axis represents a possible
uncertainty, such as porosity, permeability. Dark areas are the locations of alternative solutions, and each
simulation result is represented by a dot. The genetic algorithm begins with a random population of
The objective function is a weighted RMS difference between the observed and simulated data for as many
wells and properties as required. Lower values of the match quality (MQ) indicate better matches.
Although not included in the representation of the equation in the figure, the individual components of the
MQ calculation can be weighted in respect of the confidence placed in them. Thus measured bottom hole
pressures may be weighted higher than back-allocated GOR data. The weighting factors used are likely to
be a result of their measurement accuracy, and an interpretation of their importance based on experience.
It should be noted that this does allow an opportunity for bias in the results.
Using the same two-property diagram as has been shown earlier, we see here the possible solution spaces
for well data, and separately for seismic data. If we superimpose these we see that the match space
supported by both data items is considerably reduced. As a result of using time lapse data, we have
therefore delineated possible match spaces much better.
The use of 4D seismic data in model matching assumes that there is a reasonable basis for comparison of
seismic data with engineering data. Geophysics and reservoir engineering observe and use different features
of the reservoir. Seismic acquisition measures two-way travel time, amplitude and incident angle (offset).
Reservoir simulation models reservoir pressure and fluid saturations. Rock properties link seismic and
engineering properties. Changes in P and S will result in changes in seismic impedance, and therefore in
the measured properties of travel time and amplitude.
With knowledge of the rock properties, we can estimate seismic properties from simulator pressure and
saturation. As we usually have limited rock property data, this calculation carries uncertainty. Likewise we
can convert seismic data to impedance by an inversion. With more uncertainty we can attempt to
differentiate and possibly quantify changes in pressure and saturation.
The selection of the best attribute to compare simulator and seismic depends on the reservoir case, and is
usually a compromise. We have one set of seismic data for each survey and with reasonable confidence we
can calculate impedance, either acoustic, or elastic if we have AVO data. We may have many hundreds of
simulator derived data sets, and we can calculate impedance from each, but there will be a computing
overhead to be borne.
Here, additional steps have been added to represent the addition of 4D seismic to the history matching
process. Each model solution is converted to a synthetic seismic volume via the petroelastic model. The
result is compared with the comparable seismic attribute, e.g. acoustic impedance. Further discussion of
this follows in a later slide. Other phenomena can be compared too. For example in fields that
demonstrate measurable changes in fluid contacts, then these could be compared.
The objective function now has to include well data components and seismic components resulting in a
‘combined’ objective function.
If only well data components are included in the objective function, they can be weighted to recognise
their accuracy and importance to the history match. The seismic components also have to be weighted in
the same manner. The appropriate weightings are determined by trial and error.
To be most useful, the match quality achieved using the well data should not correlate with that obtained
using the 4D seismic. Ie the two data types should provide information about different features of the
reservoir, but both of these should be of relevance to the reservoir performance, and therefore the
observed production history.
In this example, two seismic attributes have been used as the seismic components to a history match.
These are time shift (TS) and the sum of negative amplitudes (SNA*). The time shift which results from a
change in the seismic velocity, in this case due to reservoir compaction, has been used to monitor longer
term reservoir changes (> 1 year) and the SNA has been used to monitor short term changes (< 1 year).
The observed values of time shift and SNA are shown after upscaling from the seismic grid dimensions to
those of the simulator. Next to the observed data is shown synthetic data based on a history match using
only well data to match to. There is a reasonable match to the TS data, but a poor match to SNA. If
seismic data are used as well, the match is much improved.
* Changes in SNA correlate with the changes in acoustic impedance so can be used to represent the
impedance changes.
Predictions
A reservoir simulator is only of value if it can be used to predict future reservoir performance. Good
history matches may be expected to predict better than poor matches. In this example, the best 24
matches from the previous example were used to predict incremental recovery arising from two infill wells
drilled in the study area. In the left hand graph they are the best 24 where only well data have been used.
In the right hand graph, seismic has also been used. Cases with identical MQ results were removed.
Seismic data is typically on a lateral grid scale of metres, whereas the simulator is typically ona grid scale of
tens of metres. Vertically the simulator will probably resolve to a few metres, whereas seismic to tens of
metres. Seismic is collected on an orthogonal grid, simulators now typically use a variable shaped grid to
respect geological structure and faulting.
Non reservoir rock will usually be omitted from simulators, whereas it is important acoustically, and must
be accounted for in acoustic modelling.
Finally, geophysicists commonly work using time as the vertical axis. Reservoir engineers use depth.
Horizontal Scale
The different scales of seismic and simulation data have also been discussed in the section concerned with
model building. It is also important when comparing seismic data with simulation data in the objective
function.
The lateral, or x,y scale of most models is 100m or more, whereas seismic data lateral resolution
approaches the acquisition bin size, today typically 12.5m.
There is no extra information to be gained by portraying the simulator at higher resolution, so we are
compelled to make comparison at the simulator scale.
There are a number of ways of making the comparison between the seismic and the overlain simulator grid
cell. Compare with a centre point location; an average over the area covered by the grid cell etc.
Vertical Scale
Simulator layer thickness will often be only a few metres. Several model layers may be used to represent
the sands shown in the upper thick unit of this reservoir. Seismic resolution is limited by physics, and is
typically tens of metres. Seismic is incapable of differentiating changes occurring in reservoir layers below
its resolution. Model data has to be averaged, or upscaled in order to compare with seismic response.
One of the simplest ways to compare is to compare maps of both the seismic and synthetic attributes.
This reduces the 3D comparison to a 2D one which has the advantage of avoiding depth conversion issues
where the seismic data is in the time domain, and the simulator in depth.
The following three figures illustrate cases of a well model history match.
In the first the well data (gas and water production rate) are matched satisfactorily, but the seismic data
match is poor
In the second, the seismic is matched well, but the well data is not matched.
In the final case, both data types are matched reasonably well.
Summary
• Value achieved over entire qualitative – quantitative range
• Co-visualisation is a useful qualitative integration method
• Synthetic seismic allows
• QC of model building
• Quantitative comparison with field data
• 4D seismic provides independent model matching data
• AI commonly used for matching, but other attributes also
• Prediction range is reduced
• Be aware of differences of scale
What makes this story different is that in the Andrew Field, the 4D seismic can be semi-quantitatively
matched to two PLT (Production Logging Tool) runs acquired at the same time as the 4D survey.
Within the industry, this is fairly unique, and it has provided a powerful way of checking the 4D response
and calibrating it.
Summary
Andrew Background
4D Results- Match to PLT/ RST logs
o A02 and A03 Wells
o Structural Context
4D Calibration
Imaging Infills and Interventions (3 I’s)
o A08 Example
Where we are now
The presentation starts with some background information on the the Andrew field and why a 4D survey
was needed. Then some results from the 4D and the match to the PLT are shown. Lastly it is shown that
4D can be used in a directly calibrated fashion to help identify both infill and intervention candidates.
Andrew Background: 2001
The Andrew field was discovered in the late 1970’s within a clear Palaeocene anticline, 100 miles west of
Aberdeen. Andrew has a significant gas column (~100m) compared with its 59m oil column and is
underlain by an active aquifer.
A platform was installed and a series of 12 horizontal wells were drilled into the oil column from 1996-
1999 pressure supported from the rising water and gas cap expansion.
Full Field Reservoir Simulation
This shows the prediction of the remaining hydrocarbon thickness from two reservoir model scenarios at
the start of 2002. The map on the left shows the A15 location, chosen not solely for the size of the pool,
but its robustness across a range of simulation runs
In particular, the effects of faults is very marked in generating a target “A16” to the east of the field. At the
time, faulting was not considered to be a major factor in reservoir dynamics. The faults are at the limit of
seismic resolution, with the greatest throw being about 20m, in a 85% net:gross sandy system.
In fact, faulting and shale barriers emerged as major influences on fluid flow, and the key to identification
of infill and intervention candidates.
4D Surveys
This field was not the most obvious place to acquire 4D as the rocks are moderately hard and the reservoir
is at a depth of 2500mtvdss. The oil column is approximately a seismic wavelength and there is
considerable noise created by the multiple diffractions off the overburden.
It was decided to acquire a 4D survey across Andrew and its satellite Cyrus repeating the area that had
been covered by 3D baseline surveys in 1991 and 1992.
The data were fast track processed so that the 4D could be used to influence drilling and intervention
decisions within 6 months of acquisition. Based on the success of the 2001 survey a low cost, speculative
survey re-shoot was acquired in 2002.
Although BP has tested a large number of North Sea fields, Andrew-Cyrus remained at the bottom of the
list of tested fields to date.
This meant that careful processing was required to give the best chance of seeing a 4D signal
Andrew 4D Method 1
The 4D workflow starts with the acquisition of the repeat 3D survey. Over Andrew, the baseline was
acquired in 1991, and two repeats have been acquired, in 2001 and 2002.
Data acquired in the horizontal wells can be extremely useful for the interpretation of the observed 4D
effects. In Andrew we have production logs in two wells. These have been used to interpret where fluid
flow is occurring, and to determine the changes in saturation that have occurred.
The 3D data has been used to define the structure of the reservoir, and repeat data have helped to
reinforce the interpretation. Knowledge of the structure of the field provides a context in which to
interpret the observed saturation changes, and also to identify where no change has taken place. In this
way, potential infill targets can be identified.
Andrew 4D Method 2
This slide reiterates the message of the last one, but with pictures.
In the bottom left hand corner, PLTs taken in three different years show how water has advanced along
the length of the horizontal section. In the top right hand corner the fluid contacts and the horizontal well
are shown in relation to the reservoir structure. The position of the perforations are also shown, along
with a gamma ray log. The 4D acoustic impedance difference cross section indicates where water and gas
have moved in the reservoir. We will look at this in more detail later.
Water movement results in acoustic hardening and consequent reduction in seismic amplitude, here shown
by the blue colour. An increase in gas saturation causes the opposite effect, and results in the red colour.
So we can see that water has moved up to the producer and gas has moved down. The arrows signify the
direction of movement which is influenced by the stratigraphy. The stratigraphy is interpreted from the
composite log which also shows the water saturation near to the wellbore. This corresponds well with that
expected from the 4D observations. Note the difference in scale here – the logs measure saturation close
to the well bore, the seismic is detecting the changes from the surface, ie hundreds of metres away.
A02: 4D Difference 2001-1991
The water has advanced to the well, and in fact over the well for most of its length. At the heel the
advance is slightly less, but the gas seems to have moved down a bit more.
Calibrated 1992 Hydrocarbon Height map: Field STOIIP 286mmbl, Actual 315mmbl
It is possible to calibrate the pre-production seismic data to the hydrocarbon height seen in the appraisal
wells.
This resulting map shows the predicted oil distribution prior to any production with a 60m oil column in
red, across much of the field and thinning out to zero (the blues and purples) on the edges.
When combined with porosity and net-to-gross, this gives a predicted STOIIP of 286mmbl, which is
within 10% of the volumetric figure of 315mmbl.
Most of the field looks well swept, but some parts appear to have remaining oil which has potential for
recovery by infill wells. The area around wells A12-A14 looks reasonably undrained. These are two recent
wells, water cut is low and this oil is being produced from them.
The heel section of A08 formed the first intervention target, and there seems to be potential for a new well
in the SE and one to the West, between A06 and A04. Well A16 came on at 7mbd. It is still doing well
with water cut limited to 30%.Thus, this map can be used to find targets and assess their likely size.
A08 4D Reperforation
The opportunity to perforate the heel section of this well was seen on the 4D and confirmed by an RST
log. The interpretation indicated that if perforated, the heel section would produce dry oil. This is in fact
the deepest part of the well, the producing horizontal section is slightly higher. The workover was carried
out in August of 2002. Its effect can be seen in the plot of water cut and GOR. The oil recovered to date
has more than paid for the 4D work which created the option.
Alsos, T., Tøndel, R., Aanvik, F. & Solheim, O.A. Quantifying Rise In Gas Water Contact From Time-
Lapse Seismic On The Sleipner Øst Field. 65th Conference & Exhibition, EAGE, Stavanger, 2003. Expanded
abstract A08.
Askim, O.J. Seismic Forward Modeling In A Chalk Reservoir With Permanent Monitoring. 65th Conference
& Exhibition, EAGE, Stavanger, 2003. Expanded abstract A16.
Barkved, O. K., Buer, K., Halleland, K.B., Kjelstadli, R., Kleppan, T., & Kristiansen, T. 4d Seismic
Response Of Primary Production And Waste Injection At The Valhall Field. 65th Conference & Exhibition,
EAGE, Stavanger, 2003. Expanded abstract A22.
Bogan, C., Johnson, D., Litvak, M. & Stauber, D. Building Reservoir Models Based On 4D Seismic &
Well Data In Gulf Of Mexico Oil Fields. SPE 84370
Clifford, P.J., Trythall, R., Parr, R.S., Moulds, T.P., Cook, T., Allan, P.M. & Sutcliffe, P. 'Integration of 4D
Seismic Data into the Management of Oil Reservoirs with Horizontal Wells between Fluid Contacts. SPE
83956, 2003.
Dubucq, D., Lefeuvre, F. & Bertini, F. Deep offshore seismic monitoring: the Girassol field, a West Africa
textbook example. 73rd Ann. Internat. Mtg., Soc. Expl. Geophys., Dallas. 2003, Expanded abstract RCT1.3
Furre, A-K., Munkvold, F.R. & Nordby, L.H. Improving Reservoir Understanding Using Time-Lapse
Seismic At The Heidrun Field. 65th Conference & Exhibition, EAGE, Stavanger, 2003. Expanded abstract A20.
Guderian, K., Kleemeyer, M., Kjeldstad, A., Pettersson, S.E. & Rehling, J. Draugen Field – Successful
Reservoir Management Using 4d Seismic. 65th Conference & Exhibition, EAGE, Stavanger, 2003. Expanded
abstract A01.
Hague, P., Staples, R., Weisenborn, T. & Ashton, P. 4d Seismic For Oil Rim Monitoring. 65th Conference
& Exhibition, EAGE, Stavanger, 2003. Expanded abstract A03.
Hall, S.A. & MacBeth, C., Stammeijer, J. & Omerod, M. Time-lapse seismic analysis of pressure depletion
in the Southern Gas Basin. 73rd Ann. Internat. Mtg., Soc. Expl. Geophys., Dallas. 2003, Expanded abstract
RCT2.4
Hatchell, P.J., van den Beukel, A., Molenaar, M.M., Maron, K.P., Kenter, C.J., Stammeijer, J.G.F., van der
Velde, J.J. & Sayers, C.M. Whole earth 4D: reservoir monitoring geomechanics. 73rd Ann. Internat. Mtg.,
Soc. Expl. Geophys., Dallas. 2003, Expanded abstract RCT1.1
Herawati, I. and Davis, T.L. The Use of Time-Lapse P-wave Impedance Inversion to Monitor CO2 Flood
at Weyburn Field, Saskatchewan, Canada. 73rd Ann. Internat. Mtg., Soc. Expl. Geophys., Dallas. 2003,
Expanded abstract RCT1.6
Johnston, D.E., Eastwood, J.E., Shyeh, J.J., Vauthrin, R., Khan, M., & Stanley, L.R. Using legacy data in
an integrated time-lapse study: Lena Field, Gulf of Mexico. 2000. The Leading Edge NV pp. 294-302
Johnston, D.H. & Gouveia, W.P., Solberg, A. & Lauritzen, M. Integration of Time-Lapse Seismic and
Production Logging Data: Jotun Field, Norway. 73rd Ann. Internat. Mtg., Soc. Expl. Geophys., Dallas. 2003,
Expanded abstract RCT1.5
Kloosterman, H.J., Kelly, R.S., Stammeijer, J., Hartung, M., van Waarde, J. & Chajecki, C. Successful
application of time-lapse seismic data in Shell Expro's Gannet Fields, Central North Sea, UKCS. Petroleum
Geoscience, Vol 9 2003, pp. 25-34.
Kovacic, L. & Poggialiomi, E. Integrated time lapse reservoir monitoring and characterisation of the
Cervia Field - a case study. Petroleum Geoscience, Vol 9 2003, pp. 43-52.
Lancaster, S. and Whitcombe, D N. Fast-track ‘coloured’ inversion. Soc. Expl. Geophys,, Calgary, 2000.
Expanded abstracts. pp1572-1575.
Landrø, M., 2001, Discrimination between pressure and fluid saturation changes from time lapse seismic
data. Geophysics, Soc. of Expl. Geophys., SSI pp. 836-844.
Lefeuvre, F., Medina, S., Charrier, P., L’Houtellier, R. & Dubucq, D. Improved Reservoir Understanding
through, Rapid and Effective 4D: Girassol field, Angola, West Africa. 73rd Ann. Internat. Mtg., Soc. Expl.
Geophys., Dallas. 2003, Expanded abstract RCT1.2
Lumley, D., Meadows, M., Cole, S. & Adams, D. Estimation of reservoir pressure and saturations by
crossplot inversion of 4D seismic attributes. 73rd Ann. Internat. Mtg., Soc. Expl. Geophys., Dallas. 2003,
Expanded abstract RCT6.7
Lygren, M. et al. A method for performing history matching of reservoir flow models using 4D seismic
data. Petroleum Geoscience, Vol 9 2003, pp. 85-90.
MacBeth, C., Stammeijer, J. & Ormerod, M. A petroelastic-based feasibility study of monitoring pressure
depletion in a UKCS gas reservoir. 73rd Ann. Internat. Mtg., Soc. Expl. Geophys., Dallas. 2003, Expanded
abstract RCT4.1
Marsh, J.M., Whitcombe, D.N., Raikes, S.A., Parr, R.S. and Nash, T. BP's Increasing Systematic Use Of
Time-Lapse Seismic Technology. Petroleum Geoscience, Vol 9 2003, pp. 7-13.