Documente Academic
Documente Profesional
Documente Cultură
C
(Lake 1989). In low-temperature oxidation, the oil viscosity
is lowered and the oil is converted into alcohols, ketones and
aldehydes (no phase change from liquid to gas). In high-
temperature oxidation, the combustion proceeds entirely to
carbon dioxide or monoxide, this involves a phase change
from liquid to gas.
The heavy oil behaves as solid at high frequencies and
lowtemperatures (laboratory condition), due to non-negligible
shear modulus. This solid-like properties of the heavy oil
violates Gassmanns uid substitution (Das and Batzle 2008).
The velocity of heavy oil is a function of temperature, API
gravity (density), viscosity and seismic frequency. However,
at low (seismic) frequencies and high temperatures, heavy oil
does not support shear wave propagation and have liquid-like
properties. The liquid point is the temperature at which the
shear rigidity vanishes. This point depends on the API and
wave frequency. When the reservoir has a temperature above
the liquid point, heavy oil properties are similar to that of light
oil. In this case, the oil bulk modulus shows a linear trend with
increasing temperature and depends mainly on API gravity
(Han et al 2008). The expected behaviour of the liquid point as
a function of oil density (API) and the frequency measurement
is given by Han et al (2008). For the Balol eld, with
17
H M Zadeh et al
Table 1. Initial properties used for uid replacement modelling (FRM). Reproduced from Kumar and Mohan (2004).
Fluid Matrix
Component Brine Oil Gas Quartz Clay Rock matrix Fluid mixture
Density (g cc
1
) 1 0.92 0.007 2.56 2.58 2.56 0.93
Bulk modulus (GPa) 2.48 1.79 0.021 38 20.9 36.5 1.95
Shear modulus (GPa) 0 0 0 44 6.9 34.7 0
Saturation (%) 30 70 0 95 5 72 28
Figure 2. P-wave velocity (top) and density (bottom) changes
versus gas saturation.
15.50
API gravity oil, the reservoir temperature (72
C) is
higher than liquid point (lower than 5
C in seismic frequency
range) and the shear modulus of the oil is negligible. Thus,
for this case Gassmanns uid substitution equation remains
valid.
The rock properties for Gassmanns uid substitution are
summarized in table 1 and are reproduced from Kumar and
Mohan (2004). They calculate the bulk modulus and density
of the rock matrix from assumed mineral composition (i.e.
95% quartz and 5% clay). The uid properties are calculated
from available PVT data and rock physics equations (Batzle
and Wang 1992, Vasquez and Beggs 1980).
The changes in P-wave velocity and density versus gas
saturation are shown in gure 2. The maximum drop in
P-wave velocity is 8% which corresponds to a 15% increase in
gas saturation and a 1.5% decrease in density. This represents
the maximum change in P-wave velocity (corresponding to
the minimum in gure 2). However, if the reservoir is fully
saturated with gas (100%gas case), the drop in P-wave velocity
is 3.6% which corresponds to a 12% decrease in density. This
represents the maximum change in density (see gure 2).
Auid replacement modelling (FRM) using available well
log and eld seismic data was done. The initial physical
properties of the matrix and uid used for the FRM are shown
in table 1 for the initial case. In this study we use a porosity
of 28% and assume it is constant after in situ combustion.
There is no S-wave log available in our study. Therefore,
we use Castagnas mud-rock line to generate the S-wave log.
Castagnas relation is a statistical relationship between P- and
S-wave velocities (Castagna et al 1985). We assume a constant
Poissons ratio (0.1) for dry rock and estimate the dry rock
bulk modulus from Gregorys method (Gregory 1977). In
Gregorys method, the dry bulk modulus is obtained by solving
the following quadratic equation:
y
2
(s 1) + y
_
s
_
k
s
k
f
1
_
s +
M
k
s
_
_
s
M
k
s
__
k
s
k
f
1
_
= 0 (1)
where
y = 1
k
dry
k
s
(2)
s =
3(1 )
1 +
(3)
and M is given by
M = v
2
p
(4)
where , k
s
, k
f
are porosity, rock matrix bulk modulus and
uid mixture bulk modulus, respectively. In these equations,
Poissons ratio and dry rock bulk modulus are noted by and
k
dry
, respectively. In equation (4), and v
p
are density and
P-wave velocity of the saturated rock, respectively. The
density of the saturated rock can be calculated by
= (1 )
s
+
f
(5)
where
s
and
f
are solid and uid part densities of saturated
rock, respectively. The assumption behind this method is from
the fact that for most dry rocks and unconsolidated sands,
Poissons ratio is about 0.1 and is independent of pressure as
long as the effective pressure is non-zero. As the effective
pressure approaches zero, it is shown by Duffaut and Landr
(2007) that the Poissons ratio will change. However, if
the shear-wave velocity is available, Poissons ratio can be
calculated directly.
The moduli of the mineral mixture can be predicted using
HashinShtrikman bounds (Hashin and Shtrikman 1963).
We use these bounds due to the fact that they give the
narrowest possible range of moduli, without specifying the
geometry of the constituents (Avseth et al 2005). Since
the Gassmann model assumes a homogenous mineral modulus,
it is useful to represent this mixed mineralogy with an average
mineral modulus, equal to the upper and lower bound average.
Therefore, we use an average of lower and upper Hashin
Shtrikman bounds as frame bulk and shear modulus. We
use the Reuss lower bound (Mavko et al 1998) to estimate
the effective elastic modulus of the uid mixture, assuming
homogeneous saturation of uid components.
18
Seismic monitoring of the in situ combustion process in a heavy oil eld
Figure 3. Synthetic modelling result of different gas saturation scenarios. From left to right: resistivity log, natural gamma ray log,
synthetic seismic data (ve repeated traces for 0, 15 and 100% gas saturation), P-wave velocity log and density log. The horizontal lines
show the approximate top (955 ms) and base (975 ms) of the reservoir. The time axes are calculated from depth conversion using the
baseline velocity model.
Figure 4. Comparison of real and synthetic data for a realistic changes in P-wave velocity (4%) and density (0.75%). From left to right:
natural gamma ray log, acoustic impedance log, synthetic seismic data and real seismic data. The approximate top (955 ms) and base
(975 ms) of the reservoir, for baseline, are indicated in logs. The time axes are based on the baseline velocity model.
The reservoir pressure change is negligible according to
the eld history (Kumar and Mohan 2004). This is due to
strong aquifer support. Hence, we keep the pressure constant
before and after combustion to 10.2 MPa.
Nur and Simmons (1969) and Devilbiss et al (1979) show
that the temperature-related changes in the velocity are largely
due to the effect of pore uids while the skeleton properties
remain approximately constant. We, therefore, consider the
effect of temperature in terms of change in uid phase, i.e.
from heavy oil to vapour.
Synthetic, zero offset traces corresponding to 0, 15 and
100% gas saturation are shown in gure 3. These gas
saturation values were chosen judiciously, as they correspond
to pre-combustion (0%), maximum v
p
change (15%, see
gure 2) and post-combustion (100%) cases. The well logs
are from a production well (B-183) that was logged before
the in situ combustion process started (baseline case). The
result from FRM shows an amplitude decrease at the top and
an amplitude increase at the base of the reservoir for 15%
and 100% gas saturations compared to 0.0% gas saturation
case. Since there are no log data available for the post-
combustion case, we estimated the realistic change in P-wave
and density within the bounds obtained from FRM which
satises the monitor seismic data. The estimated change in
19
H M Zadeh et al
Figure 5. Schematic geological models for base (left) and monitor (right) cases. The changed area (in monitor case) has 4% and 0.75%
decreased in P-wave velocity and density, respectively.
Figure 6. Seismic section of unmigrated synthetic data, base (top), monitor (middle) and difference (bottom). The data are stacked. The
diffraction curves due to sharp velocity change around the fault are visible in all sections. The top (red line) and base of the reservoir are
located around 955 ms and 990 ms, respectively.
v
p
and density is around 4.0% and 0.75%, respectively.
A synthetic trace of such changes is shown in gure 4. The
synthetic result is compared with real traces which are post-
stack traces from baseline survey (B) and monitor (M) surveys
at injector well B-153 (situated approximately 635 m south
of production well B-183). A good correspondence between
the synthetic and the real trace is observed. The normalized
RMS (NRMS) amplitude change in this case is around 30%
for a 50 ms time window including both top and base of the
reservoir.
4. Synthetic data study
Amajor challenge for the Balol 4Dseismic is that the expected
4D changes are close to a major fault. The typical vertical
displacement of the major fault in this area is reported to be
around 300400 m. We study the effect of fault to 4D analysis
by a synthetic 2D elastic nite difference modelling. Two
sets of synthetic data are generated. The source and receiver
separation are 25 mand 12.5 m, respectively, with a maximum
offset of 2500 m. The source and receiver depths are 6 m and
8 m, respectively. The models for baseline and monitor cases
are shown in gure 5. For simplicity, we add a water layer to
the model to avoid ground roll noise with the expense of adding
multiples. In the monitor case, we introduce an anomaly
close to the fault with 50 m extension. The anomaly has a
4% decrease in P-wave velocity and a 0.75% decrease in the
density in comparison to the base model, which was estimated
from the rock physics model. A stacked seismic section
for baseline, monitor and difference is shown in gure 6.
The diffraction curves are evident in the difference section.
20
Seismic monitoring of the in situ combustion process in a heavy oil eld
Figure 7. The corresponding base (top), monitor (middle) and difference (bottom) synthetic seismic sections after post-stack time
migration. The multiples are present. The top (red line) and base of the reservoir are located around 955 ms and 990 ms, respectively.
Figure 8. Amplitude analysis for the synthetic data in a 50 ms time window (9501000 ms), including both top and base reservoir primaries.
This suggests that migration may improve the 4D image.
The results after post-stack migration are shown in gure 7.
The 4D signal is clear and interpretable in this case. Note
that the time shift is not observable due to small extension
of the 4D anomaly that limits the resolution. Therefore,
amplitude analysis helps to detect such a small anomaly. The
amplitude analysis of the baseline, monitor and the difference
is shown in gure 8. The RMS amplitude analysis of the
difference gives the most pronounce 4D effect in this case.
Based on this analysis, we chose to focus on amplitude
analysis of the real data as time shifts are expected to be
less signicant. Since the injector wells are close to the
fault in the eld, we think that migrated eld data are a
better candidate to monitor the re front, despite the fact
that the migration may introduce artefacts. Additionally,
we perform a resolution test to nd the minimum detectable
extension of anomaly for the eld case. According to this
test, the rst Fresnel zone is 180190 m. Hence, the expected
anomalies for the Monitor-1 survey (100 diameter) might be
beyond the noise level, while for the Monitor-2 survey (200
m diameter) might be detectable. However, we emphasize
that the migration collapses the Fresnel zone to a much
smaller area.
21
H M Zadeh et al
Figure 9. The location of the sources for baseline (red) and monitor line (blue).
5. Time-lapse 3D seismic acquisition
The main goal of the time-lapse 3D seismic surveys was
to monitor changes in the reservoir condition due to in situ
combustion and track the in situ combustion front movement.
To achieve this goal, P-wave time-lapse seismic data were
acquired at regular intervals, keeping the same acquisition
parameters. The baseline data were acquired during
OctoberNovember 2003, before the in situ combustion
process started. After the base data acquisition, four
injection wells were put on in situ combustion successively
from north to south. Wells B-147, B-162, B-145 and
B-153 were ignited respectively on November 2003, (19)
December 2003, (29) December 2003 and January 2004 and
continuous air injection is going on in all these wells to
sustain the re. Subsequent to the baseline survey, two
monitor surveys (Monitor-1 and Monitor-2) were carried
out at the interval of 12 months under similar climate
conditions with the same survey parameters (Kumar et al
2006). The acquisition consists of 12 receiver lines with
72 channels in each line. Group and inline shot intervals are
20 m with a minimum near offset of 20 m. The data are
recorded for 5 s with a 2 ms time sampling interval using
the end-on shooting method. Shot and receiver locations for
baseline and monitor data are shown in gures 9 and 10,
respectively. The fold map of the data sets after matching
is shown in gure 11. Note that the injectors B-162 and B-152
are in the low fold area.
6. 3D processing
Consistent 3D processing ows are applied to both baseline
and monitor data. Since the data are processed by ONGC, we
have limited knowledge on specic details related to the 3D
and 4D processing. The available data processing details are
given in the appendix. The data are contaminated by ground
roll noise. FK-ltering, in view of amplitude preservation,
is avoided for attenuation of this noise. Instead, surface
noise attenuation is performed. Despite applying the noise
attenuation process, the high amplitude of ground rolls in short
offset traces could not be fully eliminated and caused noisy
stack in shallower zones. Surface noise attenuation, consistent
amplitude balancing and deconvolution are the processes that
are in favour of 4D. However, approximately 10% of traces
are relinquished in the noise editing process that can weaken
4D processing.
Since the injectors are close to the major fault, as shown
in gure 1, it is expected that the time migrated data are more
reliable than the unmigrated due to scattering and the shadow
effect of fault on the anomalies. This is conrmed by the
study of both synthetic data and processing of unmigrated
eld data. Therefore, we use time migrated data for the time-
lapse seismic analysis in this case. The reservoir engineering
estimate of the thermal front movement was reported to be
50 m/year, thus we focused our 4D study on baseline and
second monitor data which has time interval of 2 years and
expected to bear the signature of thermal front movement
by 100 m.
22
Seismic monitoring of the in situ combustion process in a heavy oil eld
Figure 10. Receiver positions for baseline (red) and monitor line (blue).
7. 4D processing
In order to look for injection-related time-lapse changes and
non-processing artefacts, both data sets are treated the same
way, when possible. The pre-stack time migrated data sets are
fully stacked using the same velocity eld.
Despite the best effort to acquire and process the data
sets in the same way, systematic differences are observed.
To remove such a difference, a process that we call pre-
4D is applied to the monitor data. The idea of the pre-4D
process is to minimize the difference in the overburden (static
window). This can be achieved by removing the time and
wavelet differences in common trace pairs within the static
window. In this study, the static window is assumed to be
from 600 ms to 750 ms, since there is a strong reector
in this window. No changes are expected above 600 ms;
hence, we do not include it in static window. The window
below 750 ms is not contributed in the static window, since
there might be temperature changes right above the reservoir.
We do understand that the in situ combustion process affects
the overburden and underburden; however, the calculation of
heat losses to overburden and underburden shows negligible
effects in this case (Lake 1989, Farouq Ali 1966). All pre-
4D processes are designed in the static window and applied
to the whole volume. Therefore, the pre-4D process corrects
time invariant changes. To quantify the repeatability issues,
we use the normalized root mean square (NRMS) concept. In
this method, the percentage-normalized RMS difference of the
two traces (say a
t
b
t
) from two different surveys within a
given window t
1
t
2
is computed using the formula (Kragh
and Christie 2002)
NRMS = 200
RMS(a b)
RMS(a) + RMS(b)
(6)
where NRMS is measured in per cent, and the RMS operator
is dened as
RMS(a
t
) =
_
t
2
t
1
(a
t
)
2
_
N (7)
where N is the number of samples in the time interval t
1
t
2
.
The value of NRMS is not limited to the range 0100%.
If both the traces contain random noise, the NRMS is 141%
(