Professional Documents
Culture Documents
Abstract
Seismic reservoir characterization of unconventional gas shales is challenging due to their
heterogeneity and anisotropy. Rock properties of unconventional gas shales such as porosity,
pore-shape distribution, and composition are important for interpreting seismic data amplitude
variations in order to locate optimal drilling locations. The presented seismic reservoir
characterization procedure applied a grid-search algorithm to estimate the composition, pore-
shape distribution, and porosity at the seismic scale from the seismically inverted impedances
and a rock-physics model, using the Haynesville Shale as a case study. All the proposed rock
properties affected the seismic velocities, and the combined effects of these rock properties on
the seismic amplitude were investigated simultaneously. The P- and S-impedances correlated
negatively with porosity, and the VP/VS correlated positively with clay fraction and negatively
with the pore-shape distribution and quartz fraction. The reliability of these estimated rock
properties at the seismic scale was verified through comparisons between two sets of elastic
properties: one coming from inverted impedances, which were obtained from simultaneous
inversion of prestack seismic data, and one derived from these estimated rock properties. The
differences between the two sets of elastic properties were less than a few percent, verifying
the feasibility of the presented seismic reservoir characterization.
Keywords: reservoir characterization, rock physics, shale gas, seismic impedance, modeling
1
Currently at Shell, New Orleans, LA, USA.
(a) (b)
Figure 1. (a) Haynesville Shale (gray shaded area) is at the boundary of Texas and Louisiana. Light orange areas are basins, and yellow
areas are structural highs. The extent of the Sabine Uplift is marked by the dashed area. This figure was modified from Hammes et al
(a)
(b)
Figure 2. (a) A map view showing the negative loop with maximum amplitude in the Haynesville Shale. The white block shows the seismic
survey for this study, and the two vertical wells are marked by the black star (Well A) and white star (Well B). The white horizontal ticks show
the crossline from 2001 to 2367, and white vertical ticks show the inline from 1001 to 1341. The thick white line marks the seismic profile
(crossline 2184) shown in (b). The black line marks the seismic profile shown in the seismic profile visualization example. The colors in the
white square indicate the maximum negative seismic amplitudes near the base of the Haynesville Shale. (b) Poststack P-wave seismic data. Blue
colors represent positive reflections, and red colors represent negative reflections. The seismic profile is marked by the thick white line in (a), and
the profile is oriented roughly along the east–west direction. Well A is in the middle of the seismic profile, at inline 1166, with the top and bottom
of the Haynesville Shale marked by blue arrows. The green line shows the horizon near the bottom of the Haynesville Shale. The correlation
coefficient of the seismic well tie was 0.79. The artificial two-way reflection time is used to maintain confidentiality of the data location.
limestone, and dolomite) and found that the Marcellus for- and closure pressure, which are critical parameters for the
mation in their study area was highly heterogeneous elasti- fracture simulation process. Guo et al (2013) c onstructed a
cally. Sena et al (2011) developed a workflow that integrated rock-physics workflow in which the clay and kerogen particle
analysis of azimuthally varying prestack seismic data with orientations were defined, and they applied this workflow to
well data to estimate geomechanical properties of shales and the Barnett Shale and evaluated porosity, lithology and brit-
to identify sweet spots. They estimated the ratio of differ tleness index. They found that Poisson’s ratio and Young’s
ential horizontal stresses, the pressure to initiate fractures, modulus along the well path decreased as clay content
221
J. Geophys. Eng. 13 (2016) 220 M Jiang and K T Spikes
2360
2370
2380
Artificial Depth m
2390
2400
2410
2420
2430
(a) (b)
Figure 3. (a) Well log curves from the study well. Left to right are logs of gamma ray, P-wave velocity, S-wave velocity, and density. The
shaded area is the Haynesville Shale. Depth is artificial due to the data confidentiality. (b) A zoomed-in view of the well logs within the
Haynesville Shale from Well A. At an artificial depth of about 2418 m, velocity and density decrease.
222
J. Geophys. Eng. 13 (2016) 220 M Jiang and K T Spikes
Figure 6. Prestack seismic inversion analysis at the Well A location. The first four panels show P-impedance, S-impedance, density, and
VP/VS. The blue curves indicate smoothed well logs, black curves indicate initial models, and red curves indicate seismic inverted results.
The fifth panel shows wavelets extracted from the partial stacked angle gathers at small, middle, and large angle ranges, the synthetic
seismic traces, partial stacked angle gathers from data, and the difference between the two. The black box indicates the Haynesville Shale,
to which the rock-physics modeling and grid searching was applied.
Haynesville description and data set and silisiclastic shelves (Hammes et al 2011). The shale and
carbonates were deposited during a worldwide transgressive
Description event, in a quiet-water environment, below the wave base
(Hammes et al 2011). High temperatures and overpressure
The Haynesville Shale of the upper Jurassic is an organic- were indicated in the burial and thermal history (Ewing 2001,
rich mudrock located in the East Texas basin of east Texas Becker et al 2010, Nunn 2011). The overpressure results in
and northwestern Louisiana (figure 1(a)). Stratigraphically, a high initial production rate, as well as a dramatic decline
it sits between the Cotton Valley Group and the Smackover rate (Hammes et al 2011). The Haynesville Shale is 3000
Limestone (figure 1(b)). The Haynesville Shale was depos- m to 4000 m in depth. Average permeability is <0.001 mD,
ited in a restricted basin, surrounded by carbonate platform and porosity is 3–14% (Wang and Hammes 2010). The main
223
J. Geophys. Eng. 13 (2016) 220 M Jiang and K T Spikes
Seismic data
224
J. Geophys. Eng. 13 (2016) 220 M Jiang and K T Spikes
and pores with a variety of aspect ratios and volumes. Rock-physics modeling
Chapman’s (2003) model provided a vertical transversely
isotropic (VTI) medium by introducing aligned micro- In the rock-physics modeling, the effective shale was built
fractures. The rock-physics model was calibrated using based on different mineral and pore phases. The mineral
the measured well log data. Before inversion, the seismic phases included quartz, calcite, clay, kerogen and pyrite. The
data was processed to increase the S/N. The inverted P- and pore phases included non-spherical/spherical pores, non-
S-impedances were then input into grid searching steps spherical cracks and aligned fractures. The detailed descrip-
to estimate rock properties at the seismic scale. The grid- tion and calibration of the rock-physics model is described in
search algorithm is described later in the paper. Finally, the Jiang and Spikes (2013a).
elastic properties derived from the estimated rock proper- The rock-physics model transformed prior distribu-
ties were compared to the ones inverted from the prestack tions of mineralogy, porosity and pore shape to elastic
seismic data to verify the reliability of the estimated rock properties. The prior composition included quartz, calcite,
properties. pyrite, kerogen and clay. Based on core analysis and XRD
225
J. Geophys. Eng. 13 (2016) 220 M Jiang and K T Spikes
measurements, pyrite comprised only a small portion of the Prestack seismic inversion
composition, and it was fixed at 2% in the rock-physics mod-
Simultaneous inversion (Hampson et al 2005) of prestack seismic
eling. For the other compositions (quartz, calcite, kerogen
and clay), 400 scenarios (case 1 to case 400) were intro- data was applied to obtain P-impedance, S-impedance and den-
duced, and each scenario had a composition assemblage sity. Three assumptions are central to this algorithm (Hampson
with certain percentages of each mineral. The 400 scenarios et al 2005): a linearized approximation for reflectivity, the angle-
provided a wide enough composition range while main- dependent reflectivity can be calculated by Aki–Richards equa-
taining low computational cost. In the 400 scenarios, the tions (Aki and Richards 2002), and the natural logarithms of
quartz and calcite content were negatively correlated with P-impedance, S-impedance and density are linearly related.
the kerogen and clay content. From case 1 to case 400, the The angle gathers at the well location (figure 5(a)) have
quartz and calcite percentages decreased linearly, and the incidence angles from about 4° to 50°. Overall, these angle
kerogen and clay percentages increased linearly. Therefore, gathers have a relatively low S/N. In particular, the traces
the higher the case number, the more compliant the rock at large incidence angles (between about 30° to 50°) were
is; the lower the case number, the stiffer the rock is. The noisier than the traces at smaller angles. A seismic-well-tie
prior porosity was from 0 to slightly more than 0.4, and the was done based on poststack seismic data that has much
prior pore aspect ratio was from 0 to 0.12. The ranges of higher S/N than the prestack seismic data. Supergather anal-
prior porosity and pore aspect ratio were large enough to ysis and partial stacking were applied to increase the S/N.
contain most combinations expected in this interval. Jiang The number of common depth point (CDP) gathers added
and Spikes (2013a) showed that at the well log scale, the in the super gathers was 25, which increased the S/N (part
porosity and composition estimations matched the observa- icularly for the large angle gathers between about 30° to
tions quite well. For example, for a well (outside the seismic 50°), and preserved the main features shown in the original
survey) that is about 3 km away from the vertical pilot hole, angle gathers (figure 5(b)). Partial stacking further enhanced
the average modeled porosity was 4.6% compared to density the S/N (figure 5(c)). Three angle ranges were selected:
porosity of 4.2%. Meanwhile, the average modeled quartz, 5°–15° (near), 15°–25° (middle), and 25°–35° (far). Angles
calcite, kerogen and clay were 31.3%, 23.3%, 4.7% and from 35°–50° were excluded due to relatively low S/N and
38.7% compared to XRD results of 31.6%, 25.8%, 5.3% nonexistence below about 700 ms. For each of the three angle
and 35.3%, respectively. ranges, the middle angle value was assigned to the angle
226
J. Geophys. Eng. 13 (2016) 220 M Jiang and K T Spikes
stack, e.g. the angle stack from angle range 5°–15° had an standard deviation were calculated for each rock property at
angle value of 10°. In the simultaneous inversion, these three each sample. The standard deviation shows the uncertainty of
angle stacks (10°, 20°, and 30°) were treated as prestack angle the estimations laterally and vertically.
gathers.
Comparison of the estimated elastic properties
Grid search The spatially continuous rock properties were simultaneously
A grid searching method was used to estimate the rock prop- estimated by inputting the prestack seismic inverted elastic
erties and the associated uncertainties at each angle gather properties into the rock-physics modeling and grid searching.
location. In the grid searching, P- and S-impedances from the For each of the rock properties, both the estimation and uncer-
rock-physics models were compared with the ones inverted tainty (standard deviation) were calculated. The reliability of
from the seismic data. The modeled impedances that satisfied these rock properties was verified further through the com-
the criteria (L1 norm) parisons between the elastic properties derived from them and
the ones inverted from the prestack seismic data. Based on
Obj_P = IP model − IP inverted ⩽ 0.2 the estimated rock properties at the seismic scale, the rock-
(1) physics model generated a set of elastic properties. Those
Obj_S = IS model − IS inverted ⩽ 0.1 elastic properties should be close to the ones from the prestack
seismic inversion if the estimated rock properties are reliable.
provided the estimated rock properties. The value 0.2 and
0.1 for P- and S-impedances were chosen to make their
errors 2–2.5%. This limit was chosen to account for physi- Results
cally reasonable possibilities and to ensure the computational
Prestack seismic inversion
efficiency.
For each trace at each time sample, the grid searching Analysis of the prestack inversion results at the well loca-
provided a large number of estimated rock properties based tion is shown in figure 6. In the first four panels, the well log
on the objective function (equation (1)). The mean value and P- and S-impedances, density, and VP/VS are plotted in blue,
227
J. Geophys. Eng. 13 (2016) 220 M Jiang and K T Spikes
the initial models are plotted in black, and the inverted results shows that the rock-physics models explained the seismic
are plotted in red. To utilize the poststack data that had higher inverted P- and S-impedances simultaneously. In figure 7, the
S/N and generate a proper initial model, the poststack inverted crossplot of S-impedance versus P-impedance is colored by
P-impedance was chosen as the initial P-impedance model. The the prior porosity (figure 7(a)) and prior composition (figure
initial density and VP/VS models were derived from log data. The 7(b)). The black points are from the well logs, gray points are
Haynesville Shale is marked by the black box. Due to the large from the smoothed well logs, and magenta points are from
and abrupt impedance contrast at the base of the Haynesville the seismic inversion at the well location. The three sets of
(about 40% increase in about 10 m), it is difficult to obtain an P- and S-impedances show similar trends. The model trends
accurate inversion result at the bottom of the Haynesville Shale. explained all the data points quite well.
The density match between inversion and well log data is not as Using the seismic reservoir characterization procedure, the
good as impedances because (1) the density log itself fluctuates rock properties at the well location were estimated (figure 8).
due to the rugose borehole environment; (2) the seismic traces At each time sample, we obtained multiple estimates of pore
at large angles have low S/N. The inverted VP/VS has some misfit aspect ratio, porosity and composition assemblages based on
with the borehole measurements but fits the log measurements the objective function in the grid searching, and then randomly
within the Haynesville Shale fairly closely. The synthetic angle selected 100 estimates. The probability of each estimate was
stacks (red seismograms) are similar to the seismic data (black obtained (background color in figures 8(a) and (b)). The mean
seismograms). The correlation coefficient is about 0.90, with value of the estimates (blue curves in figures 8(a) and (b)) and
relatively small error (red seismograms on right). The similarity the value with the highest probability of the estimates (magenta
between the synthetic seismic traces and observed seismic data curves in figures 8(a) and (b)) are similar, while the mean
indicates that the inversion results are acceptable. value curve is smoother. There is an increase of both porosity
and pore aspect ratio below 720 and 740 ms. The estimated
porosity at the top part of the Haynesville Shale is similar to
Calibration of the procedure at the well location
the density porosity log (red curve, figure 8(b)), particularly to
The seismic inverted P- and S-impedances were then input the smoothed density porosity (black curve, figure 8(b)). The
into the rock-physics modeling. The modeling result (figure 7) mismatch at the bottom part of the Haynesville Shale could
228
J. Geophys. Eng. 13 (2016) 220 M Jiang and K T Spikes
be related to the challenge to obtain accurate seismic inver- rock properties. These two sets of P- and S-impedances are
sion results due to the very large and abrupt impedance con- very similar, indicating that the estimated rock properties are
trast (about 40% increase in about 10 m) from the Haynesville reliable.
Shale to the Smackover Limestone. The average value of the
estimated porosity indicated by the blue curve is 5.4% within
Seismic profile visualization
the Haynesville Shale, and the average density porosity from
the well log is 6.1%. The composition assemblage (figure The seismic reservoir characterization procedure is a trace-by-
8(c)) is from the mean value of the estimated composition at trace operation, and it provides both seismic profile visualiza-
each time. Each time corresponds to various percentages of tion and horizon-based visualization. Figure 9 shows a seismic
the minerals. Calculations of the vertically averaged miner- profile representation of the inverted P-impedance (figure
alogy is 30.8% for quartz, 22.8% for calcite, 2% for pyrite, 9(a)), S-impedance (figure 9(b)) and VP/VS (figure 9(c)).
4.8% for kerogen and 39.6% for clay. These numbers are close The location of the seismic profile is marked by the black
to the ones from core measurements from a well that is about line in figure 2(a), with Well A (inline 1166) in the middle
3 km away, which are 31.6%, 25.8%, 2%, 5.3%, and 35.3%, of the seismic profile. The hot colors indicate large values,
respectively. and cold colors indicate small values. Spatial variability is
Additional steps were applied to verify the estimated rock clearly present in each section. Vertically, each inverted para
properties. Based on the estimated rock properties, the cali- meter shows a decrease at the base of the Haynesville Shale
brated rock-physics model provided a set of elastic properties between about 720 ms and 740 ms. Laterally, the color inten-
(P- and S-impedances). Those derived elastic properties were sity changes among different inlines.
then compared with the ones from the prestack seismic inver- The estimated porosity (figure 10(a)) varies vertically and
sion. The black set of impedances (figure 8(d)) represents the laterally. Vertically, the porosity is larger between about 720 ms
ones from seismic inversion, and the blue set of impedances and 740 ms relative to the other time ranges. This relatively
(figure 8(d)) represents the ones derived from the estimated larger porosity is consistent with the low P- and S-impedances
229
J. Geophys. Eng. 13 (2016) 220 M Jiang and K T Spikes
observed from the inversion result at the same time range. The estimates of composition include fractions of clay, quartz,
The porosity is smaller on the top and very bottom of the calcite, and kerogen. Clay fraction estimation (figure 11(a))
Haynesville Shale, which is consistent with the large inverted and quartz fraction estimation (figure 11(b)) are shown as
P-impedances at the top and very bottom of the Haynesville examples. Similar to the pore-shape distribution, the clay
Shale. Laterally, the color intensity of the porosity estima- and quartz fraction distributions appear to be correlated with
tion varies at different locations. For example, between about VP/VS. More clay corresponds to larger VP/VS, and less clay
720 ms and 740 ms, the porosity has large values between corresponds to smaller VP/VS. More quartz corresponds to
inline 1130 and 1140, at inline 1150, 1170, and 1200, and smaller VP/VS, and less quartz corresponds to larger VP/VS.
small values at 1140 and 1180. In fact, the porosity affects The relationship between clay and VP/VS is consistent with the
both P- and S-impedances. Generally, large porosity corre- fact that clay has higher VP/VS than the other minerals in the
sponds to low P- and S-impedances and vice versa. The rela- Haynesville Shale. Similarly, the relationship between quartz
tive standard deviation (figure 10(a)) is 20% to 25%. and VP/VS is consistent with the fact that quartz has lower
The pore-shape distribution is represented by the pore VP/VS than the other minerals in the Haynesville Shale. The
aspect ratio. Similar to porosity, the pore-shape distribution relative standard deviations for the composition estimation are
also varies vertically and laterally (figure 10(b)). These varia- very small (figures 11(a) and (b)).
tions correlated negatively with the VP/VS (figure 9(c)): higher
pore aspect ratios correspond to lower VP/VS, and vice versa.
Horizon-based visualization
In particular, between about 720 ms and 740 ms, the hot spots
in the pore-shape distribution between inline 1130 and 1140, A horizon close to the base of the Haynesville Shale (black
at inline 1150, 1170, and 1200 correspond to the cold spots dashed curve in figure 9) is shown in terms of prestack
in the VP/VS (figure 9(c)), respectively. Similar to the porosity seismic inverted P-impedance (figure 12(a)), S-impedance
estimation, the relative standard deviations (figure 10(b)) of (figure 12(b)) and VP/VS (figure 12(c)). Well A is in the middle
the pore-shape distribution is about 20%. of the view (inline 1166, crossline 2184). In each panel, the
230
J. Geophys. Eng. 13 (2016) 220 M Jiang and K T Spikes
hot colors represent large values, and cold colors represent similarities between the two sets of elastic properties indicate
small values. P- and S-impedances show similar spatial varia- that the estimated rock properties are reliable.
tions. In the north area, P- and S-impedances are relatively
small, and in the southeast area, P- and S-impedances are rela-
tively large. The VP/VS also shows spatial variation, whereas Discussion
the variation pattern is different from P- and S-impedances. In
general, the VP/VS in the north and west areas are higher than The rock properties were estimated at the seismic scale, and
the ones in the south and east areas. they were correlated with the seismic impedance and VP/VS
Figure 13 shows the estimates of porosity, pore shape variations. Some correlations are explicit, such as the negative
(pore aspect ratio) distribution, quartz fractions and clay frac- correlation between porosity and P- and S-impedances, the
tions. These rock properties are all correlated with the P- and negative correlation between quartz fraction and VP/VS, the
S-impedances and VP/VS. The porosity estimation (figure 13(a)) positive correlation between clay fraction and VP/VS, the posi-
is relatively higher at the north area and lower at the south- tive correlation between pore shape and quartz fraction, and
east areas. These spatial variation patterns in porosity are the negative correlation between pore shape and clay fraction.
negatively correlated with the ones in P- and S-impedances. Some correlations are implicit, e.g. the relationship between
Similarly, the pore shape (pore aspect ratio) distribution and composition and impedances, and the relationship between
quartz fraction is negatively correlated with VP/VS, and the the pore shape and impedances. The rock property estima-
clay fraction is positively correlated with the VP/VS. tion result shows a low clay fraction and high quartz frac-
The elastic properties were then computed from the esti- tion between about 720 ms and 740 ms (figure 11), whereas
mated rock properties (figure 14) using the rock physics P- and S-impedances are relatively low in the same time range
model. Those elastic properties are very similar to the (figure 9). In horizon visualization, the rock property esti-
inverted elastic properties in figure 12. Residuals between mation result shows low clay fraction and high quartz frac-
the rock-property-derived elastic properties and the seismic- tion toward the east of Well A (figure 13), whereas P- and
inverted elastic properties are all close to zero (figure 15). The S-impedances are low in the same areas (figure 12). These
231
J. Geophys. Eng. 13 (2016) 220 M Jiang and K T Spikes
relationships seem contrary to the normal case in which the Obj_P and Obj_S were too small, then the constraint would
stiffer quartz matrix has higher impedance and the softer clay be too harsh. On the other hand, if the maximum values for
matrix has lower impedance. From microstructure images Obj_P and Obj_S were too large, then there would be almost
(Chalmers et al 2012), the majority of the porosity in the no constraint, and the estimated rock properties would be
Haynesville Shale sits in between stiffer mineral grains (i.e. quite random. The objective function used in this study was
quartz and calcite) instead of clay platelets. Accordingly, the L1 norm, and other types of objective functions might pro-
the estimated porosity is higher where the quartz fraction is vide different estimations of the rock properties.
higher (and clay fraction is lower), and vice versa. The higher The accuracy of the seismic inversion is very important. If
porosity generates the lower P- and S-impedances. Similarly, the elastic properties were not inverted correctly, then the esti-
the above relationship between composition and porosity mated rock properties would also be inaccurate, and it would
helps to interpret the relationship between pore-shape distri- be difficult to correlate the rock properties with the seismic
bution and impedances. The explicit and implicit relationships attributes. For our case, the impedance contrast at the bottom is
between rock and elastic properties imply the need to investi- very large and abrupt (about 40% increase in about 10 m), due
gate simultaneously the rock properties when interpreting the to the formation change from shale to limestone. Therefore, it
seismic attribute variations. These interpretations also rely on is difficult to obtain an accurate inversion result at the bottom
the prior distributions or the correlations between them and of the Haynesville Shale. In addition, the seismic-inversion
the weight that has been put on P- and S-impedances during algorithm is isotropic, whereas the rock-physics modeling
the rock-physics modeling. is anisotropic. On the one hand, it is difficult to perform
In the seismic reservoir characterization workflow pre- anisotropic seismic inversion with a lack of anisotropy meas-
sented here, there are three major parts: rock-physics mod- urements at both the log scale and laboratory scale. On the other
eling, seismic inversion, and value estimation from grid hand, theoretically, the anisotropic features would mainly be
searching. Each of these affects the accuracy of characteri- related to the far offset signals in the seismic data, whereas in
zation. The rock-physics model accounted for the anisotropy. our case, the data quality is poor in the far offsets (figure 5).
The grid searching results depended on the objective function Therefore, isotropic seismic inversion was performed in this
(equation (1)). On the one hand, if the maximum values for study. Anisotropic seismic inversion could be performed if
232
J. Geophys. Eng. 13 (2016) 220 M Jiang and K T Spikes
there were anisotropy measurements, and the seismic data Chalmers G R, Bustin R M and Power I M 2012 Characterization
quality at far offsets was better. of gas shale pore systems by porosimetry, pycnometry, surface
area, and field emission scanning electron microscopy/
transmission electron microscopy image analyses: examples
Conclusion from the Barnett, Woodford, Haynesville, Marcellus, and Doig
units AAPG Bull. 96 1099–119
Chapman M 2003 Frequency-dependent anisotropy due to meso-
In this study, a seismic reservoir characterization proce- scale fractures in the presence of equant porosity Geophys.
dure was developed. The rock properties were estimated at Prospect. 51 369–79
the seismic scale through rock-physics modeling, prestack Curtis J 2002 Fractured shale-gas systems AAPG Bull. 86 1921–38
seismic inversion, and grid searching. Different distribu- Curtis M E, Ambrose R J, Sondergeld C H and Rai C S 2010
tions of the rock properties helped to interpret the seismic Structural characterization of gas shales on the micro- and
nano-scales Proc. of the Canadian Unconventional Resources
amplitude variations. All these rock properties affected the and International Petroleum Conference, ‘Society of Petroleum
seismic velocities, and their effects on the seismic ampl Engineers’ SPE Paper 137693
itude were investigated simultaneously. Generally, the P- and Ewing T E 2001 Review of Late Jurassic depositional and potential
S-impedances are negatively related to porosity, and VP/VS is hydrocarbon plays, northern Gulf of Mexico Basin Gulf Coast
positively related to clay fraction distribution, and negatively Assoc. Geol. Soc. Trans. 51 85–96
233