You are on page 1of 15

Marine and Petroleum Geology 25 (2008) 845–859

Contents lists available at ScienceDirect

Marine and Petroleum Geology


journal homepage: www.elsevier.com/locate/marpetgeo

Exploration for gas hydrates in the deepwater, northern


Gulf of Mexico: Part II. Model validation by drilling
Jianchun Dai*, Niranjan Banik, Diana Gillespie, Nader Dutta
Schlumberger, 10001 Richmond Avenue, Houston, TX 77042, USA

a r t i c l e i n f o a b s t r a c t

Article history: This study examines the accuracy of the predictions of gas hydrate saturations made based on five-step
Received 31 January 2007 analysis of 3D seismic data prior to 2005 drilling, logging, conventional coring, and pressure core
Received in revised form 3 September 2007 sampling through the gas hydrate stability zone at two focus sites in the northern Gulf of Mexico. These
Accepted 15 February 2008
predictions are detailed in Part I (2008). Here we conduct a detailed analysis of the gas hydrate saturation
using both resistivity and P-wave velocity log data and analyze the pre-drilling predictions, which were
Keywords: made almost exclusively on the basis of seismic data, with no local logging control. Well log measure-
Gas hydrate
ments, core data analysis, and pressure core-degas experiments all indicated general agreement with the
BSR
Inversion
pre-cruise analysis regarding the location and approximate concentration of gas hydrates in the sedi-
Seismic detection ments. We find that seismic predictions are generally consistent with log-based estimates after upscaling
Calibration to seismic frequencies. We recalibrated the pre-drill model based on the new field data so that a refined
Gulf of Mexico version of the model could be used for future work.
Published by Elsevier Ltd.

1. Introduction reliable due to borehole conditions. Lack of quality shallow-log data


in a given hydrate-exploration basin poses additional constraints
In recent years, the detection and delineation of gas hydrates on the reliability of gas hydrate saturation predictions based on
have drawn significant attention from the scientific community a seismic approach.
worldwide due to its potential as an alternative energy source, In 2005, the DOE–Chevron Joint Industry Project (JIP) drilling
cause for drilling hazards, and being an agent for global climate and coring program completed seven holes for the study of gas
changes. Reflection seismic technology, being the principal method hydrates in the northern GoM. One objective of the program was
in hydrocarbon exploration, has been extensively used for hydrate to collect sediment cores and obtain well logs in areas where there
detection (e.g., Collett et al., 1999a; Dai et al., 2004; Diaconescu was substantial seismic evidence of the occurrence of gas hydrates
et al., 2006; Ecker et al., 1998; Kvenvolden and Barnard, 1983; and thus aid in the verification of the exploration model. The JIP
Shipley et al., 1979; Xu et al., 2004). In Part I, we presented results of targeted Keathley Canyon (KC) 151 and Atwater Valley (AT) 13 and
gas hydrates detection and characterization based on a seismic 14, both at water depths of about 1300 m. These sites are shown in
approach termed the five-step workflow (Dai et al., 2004; Xu et al., Fig. 1 of Part I (Dai et al., 2008). The Atwater Valley site has
2004). a prominent gas hydrate mound in a high fluid flux area (Ellis
Inherent in seismic prediction are uncertainties. In the case of et al., 2008; Wood et al., 2008), and Keathley Canyon site had
gas hydrate exploration in the Gulf of Mexico (GoM), the sources of a regional bottom simulating reflector (BSR; Hutchinson et al.,
uncertainty are mainly due to a general lack of ‘‘ground-truth’’ in- 2008). The lithologies in both locations consisted of clays and silts
formation about the formation and accumulation of gas hydrates. with small amount of sands. In both locations, the inferred in situ
Since the prediction is model based, the validation of a particular gas hydrate concentrations were low and highly variable.
model and calibration of model parameters through drilling are Five holes were drilled in AT13 and 14, and two holes were
important first steps to improve the accuracy of seismic prediction. drilled in close proximity to each other in KC151 (holes KC151-2 and
Additionally, the properties of near-seafloor sediments vary con- KC151-3). KC151-2 was drilled first and Logging-While-Drilling
siderably, and well log data in such shallow sections are often not (LWD) and Resistivity-at-the-Bit (RAB) were run in this well during
drilling. Following completion of KC151-2, coring started at KC151-
3. wireline dipole sonic/gamma ray (GR)/general purpose in-
* Corresponding author. clinometer Tools (GPIT) were successfully run after the completion
E-mail address: jdai@houston.oilfield.slb.com (J. Dai). of the KC151-3 hole. Two possible substantial hydrate zones (Fig. 1)

0264-8172/$ – see front matter Published by Elsevier Ltd.


doi:10.1016/j.marpetgeo.2008.02.005
846 J. Dai et al. / Marine and Petroleum Geology 25 (2008) 845–859

were penetrated in these holes at depths between 200 and 300 volume of shale (VCL) is estimated using a simple transform of the
meters below seafloor (mbsf). The depth interval corresponds to GR log. It shows a trend of increasing shale content with depth. It
two-way time interval between 2020 ms and 2120 ms from the must be noted, however, that the GR log contains a major baseline
mean sea level. shift at 125 mbsf (Fig.1), and we lack good quality coring information
In Part I we presented a detailed pre-drill analysis and pre- to define the GR trend for pure sands. Therefore, the accuracy of VCL
diction results for the drilling areas based on the five-step seismic estimation may be questionable.
approach (Dai et al., 2004; Xu et al., 2004). In this paper, we eval- Two anomalously elevated resistivity zones are indicated by the
uate hydrate concentrations based on drilling results in the KC151 resistivity logs in Fig. 1: one between 220 and 240 mbsf and the
gas hydrate wells using resistivity from KC151-2 and dipole sonic other between 260 and 300 mbsf. The Vp log also shows high
data from KC151-3. We compare these results to predictions de- velocities at the upper zone, but the velocities are not high and,
tailed in Part I (Dai et al., 2008) and derived from prestack wave- almost constant with depth in the lower zone. These two anoma-
form inversion (PSWI) (Mallick, 1995, 1999; Dutta, 2002) on lous zones may be indicative of appreciable hydrate concentration.
existing 3D seismic data. We then present an updated seismic Detailed discussions of these potential gas hydrate zones will be
model based on the current drilling data. We also present results given in later sections of this paper.
using a fast, yet robust simultaneous prestack seismic inversion The Vs log does not show any coherency with Vp or other logs.
technology (Rasmussen et al., 2004). A review of the up and down logging runs of acoustic and shear
sonic velocities shows high levels of consistency in Vp, but noisy
2. KC151 log evaluation and well ties to seismic and poor consistency in Vs (Fig. 2). We used the up-going Vp after
measurements minor editing. Because of the poor consistency of the shear logs,
we considered the Vs log to be unsuitable and therefore did not
As shown in Fig. 1, the KC151 suite of logs (combination of KC151- use this log for gas hydrate estimation in the current work. The
2 and KC151-3) includes gamma ray (GR) and photoelectric ab- density and the neutron porosity logs appear to be reliable below
sorption factor (PEF); shallow, medium and deep resistivity averages approximately100 mbsf. They also extend to depths below the
(BSAV, BMAV, BDAV); differential caliper average (DCAV); P-wave BSR. At depths shallower than 100 mbsf, these logs are of poor
velocity (Vp); S-wave velocity (Vs); bulk density (RHOB); neutron quality. This is evident from the caliper data through this shallow
porosity (NPHI); and nuclear magnetic resonance (NMR). The section.

Fig. 1. Combined log suite from KC151-2 and 3, including gamma ray (GR), photoelectric absorption factor (PEF), shallow, medium and deep resistivity averages (BSAV, BMAV,
BDAV), differential caliper average (DCAV), P-wave velocity (Vp); S-wave velocity (Vs), bulk density (RHOB), neutron porosity (NPHI), and nuclear magnetic resonance (NMR). The
volume of shale (VCL) is estimated using a simple transform of the GR log. The BSR is marked by the black line.
J. Dai et al. / Marine and Petroleum Geology 25 (2008) 845–859 847

where the Vp is not available, Gardner’s equation (Gardner et al.,


1974) was used to estimate Vp for these zones. The application of
Gardner-type velocity transformation may be questionable for the
unconsolidated rocks. Also, the quality of the shallower density is
poor due to bad hole conditions as indicated by the enlarged caliper
(Figs. 3 and 4). Nonetheless, this does not affect the well-to-seismic
tie in general since we focus on the interval where Vp observations
are available.
We used an 18 Hz central frequency Ricker wavelet to create
the synthetic seismic traces. This wavelet was consistent with the
wavelet extracted from the seismic data near the well location. The
tie between the synthetic and the seismic data at the well location
was reasonably good in the zone of interest (1900 ms–2220 ms
two-way travel time). The poor tie for the shallower section is due
to a poor quality density log and the absence of Vp and Vs logs (Figs.
3 and 4). The characteristic peak at the top of the high P-wave
velocity zone at 2020 ms as shown in Figs. 3 and 4 greatly facili-
tated matching the events in synthetic and seismic data. The arti-
fact in the synthetic data arising from the absence of good-quality
Fig. 2. KC151-3 up and down logging runs of acoustic and shear sonic velocities. well data at and near the water bottom was addressed in later
analyses through the use of PSWI (see below), where we demon-
strate 90% goodness of fit between the PSWI synthetic gather and
Well-to-seismic ties provide a basis for ascertaining hydrate the real gather at the well. This also underscores the value of
events on seismic data and for the calibration of a hydrate- pseudo-well data derived from prestack inversion for hydrate
estimation model. We created a well-based, zero-offset synthetic detection and quantification.
seismogram and compared it with the reprocessed poststack seis- Fig. 5 shows the overlay of Vp and resistivity curves on the
mic trace (Fig. 3) and prestack gather (Fig. 4) at the well location. seismic section based on the updated time–depth relation. The high
The seismic data reprocessing was done to optimize hydrate de- Vp anomaly ties to the seismic peak at w2020 ms, but Vp is prac-
tection and quantification as described in Part I (Dai et al., 2008). tically constant at the seismic peak at 2070 ms (Fig. 5, top panel). In
Two sources of time–depth relations were available to aid the well- comparison, the resistivity seems to tie to the seismic data at both
to-seismic tie. One is the time–depth curve from pre-drill PSWI, places (Fig. 5, bottom panel).
and the other is the checkshot data from the KC151-3 well. The
focus of the synthetic is at the interval where dipole sonic logs are 3. Gas hydrate concentration (Sgh) estimation from logs
available, between 1900 ms and above the level of the BSR (Figs. 3
and 4). For the synthetics, the impedance was calculated from the There is an extensive literature on the use of well logs for esti-
product of Vp and bulk density. At shallower and deeper zones mation of gas hydrate concentration (e.g., Collett et al., 1984, 1999b;

Fig. 3. Well-to-seismic tie for stack data. The seismic wiggle traces in between the synthetic and the seismic section are the extracted seismic response along the borehole
trajectory.
848 J. Dai et al. / Marine and Petroleum Geology 25 (2008) 845–859

Fig. 4. Well-to-seismic tie for prestack gather. The seismic wiggle traces in between the synthetic and the seismic gather are the stacked seismic trace of the gather.

Collett, 2001; Collett and Lee, 2004; Kleinberg et al., 2003, 2005; T ¼ Tr þ Tg *Depth (4)
Lee and Collett, 2008; Mathews, 1986; Guerin et al., 1999; Hyndman
et al., 1999). The most commonly used logs for hydrate concen- where Tg is the shallow geothermal gradient. Fig. 6 shows the for-
tration estimation include resistivity, sonic, and NMR logs. Below, mation water resistivity profiles with the thermal gradient chang-
we discuss the gas hydrate concentration estimates based on re- ing from 2.6 to 4.0  C/100 m. We chose a Tg of 3.8  C/100 m,
sistivity and sonic measurements at the KC151 wells. These esti- consistent with the near-seafloor shallow depth thermal gradient
mated results are then compared with the prediction from seismic (Hutchinson et al., 2008; Fig. 6).
measurements as presented in Part I.
3.1.2. Porosity estimation
3.1. Sgh estimation from resistivity logs Density-derived porosity and neutron porosity are available for
the studied well. Both porosities show coherent trends with depth,
Resistivity-based Sgh estimation includes two steps. It first in- but the neutron porosity is systematically higher than the density
volves solving for water saturation using Archie’s (1942) equation porosity (Fig. 1), possibly due to high shale content of the shallow
 1 unconsolidated rocks. In this work we calibrate the neutron po-
aRw n
rosity trend against the density porosity trend and use the porosity
Sw ¼ (1)
4m Rt as the average of the two porosities
The second step is to solve for Sgh on the assumption that the 1
porous medium contains only brine water and hydrates 4 ¼ ½4d þ 4n  (5)
2
Sgh ¼ 1  Sw (2) where 4d and 4n are the density-derived porosity and calibrated
neutron porosity, respectively. 4d is calculated as given below
In Eq. (1), 4 is the rock porosity, Rw is the formation water re-
rma  rb
sistivity, Rt is the formation resistivity, m is the cementation ex- 4d ¼ (6)
ponent, n is the saturation exponent, and a is a parameter related to rma  rw
rock tortuosity. where rma, rb, and rw are the densities for matrix (2650 kg m3),
bulk, and pore water (1036 kg m3), respectively. In the presence of
3.1.1. Rw estimation gas, the bulk density overestimates porosity, and the neutron data
Rw is estimated through Arps’ (1953) formula underestimate it. Thus, the averaging process helps reduce proba-
ble gas effects in porosity estimation.
Tr þ 21:5
Rw ¼ Rr (3) Review of literature indicates a rather large range of values
T þ 21:5 possible for various parameters needed in these calculations. The
where Rr (0.35 Um) and Tr (4  C) are the seafloor water resistivity most commonly used combination of parameter values are the
(in Um) and reference temperature (in  C), respectively. T is the Humble values, i.e., a ¼ 0.62; m ¼ 2.15. Collett et al. (1999b) used
formation temperature and it is calculated from the Humble values and n of 1.9386 (Pearson et al., 1983) for the
J. Dai et al. / Marine and Petroleum Geology 25 (2008) 845–859 849

Fig. 5. Left panel shows Vp overlay on seismic section. Right panel shows resistivity overlay on seismic section. The red arrow indicates the possible gas hydrate zone associated
with the strong seismic amplitude event on the right side of the well.

Mallik 2L-38 Sgh estimation, and the result matches well with the Sgh values as high as 30%. The result of upscaling to the seismic scale
core measurement. Hacikoylu et al. (2006) also published empirical shows more general hydrate anomalies, with the majority of Sgh
values for different lithologies, including shaly sand (a ¼ 1.65, lower than 20%. We will use the upscaled results for the compari-
m ¼ 1.33) and unconsolidated sand (a ¼ 0.62 and m ¼ 2.15). In this son with the pre-drill seismic gas hydrate concentration prediction
study, we adjust parameters a, m, and n to calculate Sw. We find that made in Dai et al. (2008, Part I).
a ¼ 0.90, m ¼ 1.90, and n ¼ 1.9386 produce Sw that follows the
background water-saturated intervals (in which Sw ¼ 1). We 3.2. Sgh estimation from P-wave velocity log
therefore use this set of values for the estimation of gas hydrate
saturation. To estimate gas hydrate saturation from Vp and Vs data, two
pieces of information are needed: knowledge of the elastic prop-
3.1.3. Sgh estimation erties of the background rocks and a rock physics model that ac-
The Sgh is calculated using Eq. (2). The Sgh curve (Fig. 7) reveals counts for the effect of gas hydrates on the host rocks. In this
two major Sgh zones in the interval from 2020 ms to 2120 ms, with section, we use P-wave velocity measurements from the KC151 well
850 J. Dai et al. / Marine and Petroleum Geology 25 (2008) 845–859

(Figs. 9 and 10). The sand in zone B may contain more shale lami-
nations. This shale lamination effect may have caused a lowering of
the background Vp that, in turn, lowered the estimated hydrate
saturation in this interval. In this case, the lower zone is still
probably hydrate bearing, but without much velocity increase due
to the laminated nature of the shale-prone host rocks. Another
possibility is that the rock properties at these twin holes may vary
due to heterogeneity, and the lower gas hydrate zone as diagnosed
from resistivity from KC151-2 may not exist at the KC151-3 location
may have been missed by sonic logging in this hole.

5. Comparison and calibration of seismic prediction with


log-derived data

5.1. Comparison with pre-drill seismic results

Fig. 11 shows the comparison of the log-based Sgh estimates with


Fig. 6. Formation water resistivity estimation using Arps’ relation. The water resistivity
the pre-drill seismic prediction (dashed lines). The log-based esti-
curves with depth shift to the left as geothermal gradient changes from 2.6 to 4.0  C/
100 m, with an increment of 0.4  C/100 m. mates are upscaled to the seismic scale as shown in Figs. 7 and 8.
Of the two major hydrate anomaly zones inferred to have been
penetrated by the KC wells at the interval between 2020 ms and
log and effective medium theory model to account for the gas hy- 2120 ms, the pre-drill prediction matches well with the lower zone
drate effect. Dai et al. (2004) and Xu et al. (2004) provide more B that is centered at about 2080 ms, especially through the re-
detailed discussions of this subject. sistivity method, but only shows very small hydrate saturation in
With the modeling, we estimated Sgh from the P-wave velocity the upper zone A that is centered at 2020 ms (Fig. 11). In addition,
of KC151-3 (Fig. 8). Note that the Sgh estimated from P-wave the pre-drill model also predicts high hydrate saturations at shal-
velocity ranges from 0 to 15% in seismic scale which is little lower lower intervals (1900–2000 ms), where the log-based estimation
but, in general, consistent with those estimated from resistivity. only shows small concentrations. The poor ties in the shallow in-
tervals (<2000 ms) may have been caused both by the errors in
4. Discussion on log-based hydrate concentration seismic prediction due to noise and unreliable log data in shallow
interval due to enlarged borehole conditions. The poor ties in the
The Sgh derived from resistivity data reveals two major zones of upper zone A centered at 2020 ms may be due to the absence of any
potential hydrate concentration in the interval between 2020 ms calibration and a good low-frequency background model in the
and 2120 ms, with Sgh ranging from 0 to 30% on the log scale and 0– pre-drill prediction. A re-examination of the pre-drill PSWI (Fig. 12)
15% in seismic scale (Fig. 7). The Sgh estimated from the P-wave does show a small peak in Vp at 2020 ms (zone A). However, the
velocity implies the presence of gas hydrates in both zones. How- absolute value of PSWI Vp at this event is not prominent, resulting
ever, the P-wave-based estimate in the lower zone is considerably in a possible under estimation of hydrate saturation in pre-drill
lower than that obtained by the resistivity method. The discrepancy predictions. We also note that the pre-drill model yields an ap-
between the two estimates may suggest that the upper hydrate preciable hydrate saturation event above the BSR at 2200 ms, but
zone can be more confidently assessed as hydrate bearing than the post-drill estimation from the resistivity log does not show any
lower zone. A close look at the RAB images of these two possible such indication (Fig. 11). The velocity logs were not available at this
hydrate zones may indicate a probable increase of shale content depth. It is possible that hydrate may be concentrated just above
from the upper hydrate zone (A) to the lower hydrate zone (B) the BSR, but it may be on the right side of the well, as indicated by
the strong reflection shown by the arrow in Fig. 5. The hydrate
anomaly above the BSR might not have been picked up by the
resistivity log due to its limited penetration of sediments away from
the borehole.
In summary, the comparison between pre-drill prediction and
log data based estimates suggests that the seismic prediction can
capture hydrate anomalies even with modest saturation, as low as
10–20% in this case. However, seismic prediction of low saturation
hydrates may be quite uncertain due to errors in seismic inversion,
in rock model of gas hydrates, in background rock properties, and
in well-positioning. The log-based calculation of gas hydrate
concentration also needs calibration of parameters, which is, in this
case, difficult to do due to low concentration of gas hydrates in
general and the lack of direct evidence of gas hydrates from coring
in particular.

5.2. Post-drill calibration of PSWI results

Because of the band-limited nature of seismic data, the results


based on seismic inversion alone may be ambiguous and uncertain.
Fig. 7. Resistivity-derived Sgh (fraction of pore volume) versus two-way travel time
A major source of such uncertainty is the lack of a low-frequency
(ms). The solid curve is the original estimation and the dashed curve shows the results background trend that accounts for major lithologic variations. Due
upscaled to seismic scale. to lack of local well data during the pre-drill model construction
J. Dai et al. / Marine and Petroleum Geology 25 (2008) 845–859 851

Fig. 8. Sgh estimation from P-wave velocity log at KC151-3. The left panel shows the input Vp log (solid) and the background trend (dash). The nomogram in the middle panel shows
the modeled P-wave velocities with hydrate saturations ranging from 0 to 50% with an increment of 5% from left to right. The right panel shows the estimated hydrate saturation
(solid curve for log scale and dashed curve for seismic scale).

Fig. 9. KC151-2 logs with RAB images. Bright color in the upper RAB image for zone A reveals high resistivity, possibly due to both hydrate concentration and sandier sediments. The
sediments may become more shaly in the lower section as shown in the lower RAB image.
852 J. Dai et al. / Marine and Petroleum Geology 25 (2008) 845–859

Fig. 10. KC151-2 logs with RAB images. The color of the RAB images become more bright in the lower portion of zone B, possibly due to both gas hydrate concentration and more
sands in the clay-dominated environment.

phase, we used a generic lithology-independent trend based on The results from calibrated PSWI are shown in Fig. 13. We term
a rock physics model that is typical of deepwater sediments in the these the ‘‘post-drill inversion’’ results. The correlation coefficient
GoM (Dai et al., 2008, Part I). This may be a major source of error in between the actual seismic angle gather and the synthetic angle
our model prediction. To understand this further and to create gather is w0.91, suggesting a reliable match. We caution, however,
a reliable predictive model, we next calibrated the inversion results that the high correlation coefficient may not directly reflect the
using the well data. This entailed creating a new suite of low-fre- error/uncertainty of the solution due to the highly non-unique
quency Vp, Vs, and density models that more accurately describe nature of the methodology. The result should therefore always be
the earth model. This is shown in the well data in Fig. 12. We calibrated against log measurement if available. The post-drill PSWI
recomputed the PSWI with the new low-frequency model as the results now do show the presence of a high peak at event A, which
initial model. matches with the log measurement.
The highs and lows in the P-wave velocity data from inversion
reflect possible lithologic or fluid variations. At the shallow depth,
sandy sediments usually possess higher velocity than shale because
of lower porosity as well as higher grain velocity. Gas hydrate zones
have been known to possess high velocities above the BSR and low
velocity beneath the BSR. Based on these observations, we infer
that shallow sand and shale sequences, the BSR, and possible gas
hydrate anomalies can be recognized in the pseudo-well logs from
PSWI provided a reliable low-frequency trend is used to guide the
inversion process.
A comparison of the pre- and post-drill PSWI is given in Fig. 14.
Overall, both solutions are comparable, especially in P-wave
velocities, but there are also significant differences as discussed
above.
Similarly, the hydrate concentration estimate based on P-wave
velocity from post-drill PSWI is given in Fig. 15. It shows a high peak
of w20% of hydrate concentration at event A and a smaller peak
Fig. 11. Comparison of Sgh estimated from log data with pre-drill seismic prediction. The
blue and red curves are the Sgh estimates in seismic scale from resistivity and P-wave
of w10% concentration at event B. A comparison of log-based
velocity, respectively. The black dots are the result of pre-drill prediction as discussed Dai hydrate saturation with all versions of seismic estimates is given in
et al. (2008). Fig. 16, which shows coherency in zones A, B, and the interval
J. Dai et al. / Marine and Petroleum Geology 25 (2008) 845–859 853

Fig. 12. Pre-drill PSWI at KC151-3. The blocky curves in the left three panels are the inverted P-wave, Poisson’s ratio, and density profiles, respectively, resulting from PSWI. The
smooth curves are the corresponding input backgrounds. Red arrows show peaks in the seismic data, and the blue arrow indicates the BSR. The right two panels, respectively,
display the actual seismic angle gather and the synthetic angle gather after convergence during the iteration process. The correlation coefficient of the two gathers is about 0.90.

immediately above the BSR at about 2200 ms. The Sgh from log data model parameters, they differ substantially in forward modeling.
does not show a significant hydrate response near 2200 ms; how- While the PSWI forward modeling relies on the full elastic wave
ever, the seismic estimates show the anomaly consistently. In- propagation method using the Kennett propagator matrix (Mallick,
spection of the seismic section does reveal a significant event at the 1995; Kennett, 1983), the forward modeling in simultaneous
2200 ms level just at the north side of the well, and we postulate inversion is based on the Linearized Zoeppritz equation. The
that the well may have missed this hydrate-bearing zone. Zoeppritz equation-based forward modeling renders the simulta-
neous inversion method very fast. There are also other additional
5.3. ISIS simultaneous inversion for Sgh estimation differences, including differences in the optimization procedures
and wavelet extraction. The optimization in simultaneous inversion
ISISÔ-based simultaneous inversion (Rasmussen et al., 2004) is done over a 3D volume while the optimization in PSWI is for
(hereafter referred to as simultaneous inversion) is a seismic a single CMP. Mallick (2007) compared PSWI results with those of
method for inverting prestack seismic data for elastic parameters. Zoeppritz based 2-term and 3-term synthetic seismograms using
The method uses multiple traces and a global error minimization well data containing gas hydrates.
algorithm to robustly invert for elastic parameters. Preconditioned Because simultaneous inversion is fast and robust, it is very
seismic data are input as multiple angle stacks. Prior models for Vp, suitable for efficiently scanning the elastic parameters in a 3D
Vs (or Poisson’s ratio), and density are the initial low-frequency seismic volume and for subsequent quantification of hydrates. In
background models for elastic parameters and form a basis for the the present study, we used simultaneous inversion to generate
objective and cost functions for inversion. The prior models are acoustic and shear impedance volumes at and around well KC151-3
derived from seismic velocity, interpreted seismic horizons, and with both raw and preconditioned seismic data. We then compared
available well information. A simulated annealing method is used the results at the well with those of the well data and PSWI results
to generate and update model parameters. The forward modeling is and transformed the inverted acoustic impedance profile at the
done using the linearized Zoeppritz equation-based reflection well location to gas hydrate concentration. The results are shown in
coefficient series and convolution of wavelets. The wavelets may Figs. 17 and 18.
vary spatially and temporally for each angle stack. The post-drill PSWI and simultaneous inversion-based Vp and
Although both simultaneous inversion and PSWI assume a local Sgh are generally consistent with each other and with the sonic Vp.
1D horizontally layered earth model for inversion of earth’s elastic Thus simultaneous inversion results can be used for rapid scans of
854 J. Dai et al. / Marine and Petroleum Geology 25 (2008) 845–859

Fig. 13. Post-drill PSWI at KC151-3. The blocky curves in the left three panels are the inverted P-wave, Poisson’s ratio, and density profiles, respectively, resulting from PSWI. The
smooth curves are the corresponding input backgrounds, and the pink curves are the actual logs from the wells. Red arrows show peaks in the seismic data, and the blue arrow
indicates the BSR. The right two panels, respectively, display the actual seismic angle gather and the synthetic angle gather after convergence during the iteration process. The
correlation coefficient of the two gathers is about 0.91.

seismic data for hydrate detection and quantification. Comparison existing data was to provide resolutions higher than that available
of the hydrate concentrations predicted from surface seismic mea- through the standard processing schemes and to make sure that all
surements with the results obtained during drilling indicates that significant processing steps were amplitude preserving. When well
pre-drill seismic prediction can capture hydrate anomalies even at data are available, we usually validate amplitude characteristics
modest saturation of w10% (Fig. 18). Although possible ambiguities and scales through seismic-to-well synthetic ties. This is an im-
and uncertainties involved in each estimation step increase at low portant step to limit processing errors and pitfalls, but seldom
saturation levels, meaningful relative changes may be captured. available in an exploration setting and within the zone of gas
hydrate stability. For the analysis of possible errors we confine
6. Seismic detectability of hydrates in the GoM ourselves to those arising from the elastic inversion, rock models
for shallow sand and shale, and the corresponding gas hydrate
We have showed that seismic methods that rely on a properly models for shallow sand and shale.
calibrated low-frequency background model can be used for gas As discussed in Part I (Dai et al., 2008) and Sections 4 and 5 of
hydrate detection and quantification. However, predictions based this paper, the most important elastic parameter in seismically
on seismic data do have a degree of uncertainty, particularly in the estimating Sgh is the P-wave velocity, Vp (or impedance Ip). The S-
absence of local well log data. This is particularly true if the inferred wave velocity, Vs (or impedance, Is) may be also used either
gas hydrate saturation is small, as is the case for this present study. independently or jointly with Vp (or Is). Usually the error in Vs
In this section we discuss possible errors in the estimation of Sgh inversion is more (wroughly 1.5–2 times) than that in Vp or (Ip). In
and try to find a minimum value of Sgh for which seismic-based our experience the error in Vp or Ip in deepwater sediments near
estimation might be reliable for the GoM. the mudline is about 2–5%. By the same token, the error in porosity
The error in the estimation of gas hydrate saturation may arise modeling, when well calibration is available, is about 3–4%. In the
from many factors, including data acquisition geometry, seismic absence of well calibration, the porosity error may be higher. The
data processing, elastic inversion, estimation of porosity profiles, density and porosity functions that are needed are usually modeled
and gas hydrates modeling. The estimation of error due to in- to be consistent with the P-wave velocity. From our experience
adequate data acquisition geometry is beyond the scope of the with seismic data and from rock physics principles, we know that
present study. We started out with data that were already acquired the low-frequency trends for shale and sand differ substantially at
and partially processed for exploration and production of hydro- shallow depths. For example, in the GoM, the velocity of sands at
carbon reservoirs in the area. The objective of reprocessing of a given depth in the shallow section is typically higher than those
J. Dai et al. / Marine and Petroleum Geology 25 (2008) 845–859 855

Fig. 14. Comparison between logs and PSWI results. The blue and red curves are the pre-drill and post-drill PSWI-derived Vp and Vs, respectively. The black and green curves in the
three panels are the initial input models and corresponding log measurements, respectively. Both solutions are comparable especially in P-wave velocities. The high Vp below
2000 ms from post-drill PSWI (red curve in the left panel) corresponds to possible hydrate in zone A.

for shales. Since the estimation of Sgh is based on the deviation of estimation of Sgh is expected to be reliable when the inverted P-
the P-wave velocity from the low-frequency trend, the main error wave velocity exceeds that of the sand trend. Any reversal in the
may come from the modeling of low-frequency trends for shales trend may be due to changes in the lithology at that depth. On this
and sands. Thus in the sand-shale environment in the GoM the basis we calculate a threshold Sgh value above which the estimation

Fig. 15. Hydrate concentration estimation based on P-wave velocity from post-drill PSWI. The left panel shows the input Vp from PSWI (blocky) and the smooth background trend.
The middle panel plots Vp and its background trend against the nomogram of modeled P-wave velocities with hydrate saturations ranging from 0 to 50% with an increment of 5%
from left to right. The right panel shows the estimated hydrate saturation (Sgh).
856 J. Dai et al. / Marine and Petroleum Geology 25 (2008) 845–859

Due to the lack of good-quality logs for sand events, we obtain


porosity and density trends for sand based on the regional GoM
data from Gregory (1977). We then calculate sand velocity based on
a mixed shale (50%) and sand (50%) properties to account for the
shaliness.
We then model the P-wave velocity responses of hydrate-
bearing shales. The colored curves in the left panel of Fig. 20 are
a nomogram with hydrate concentration ranging from 0 to 50%
with an increment of 5%. The threshold hydrate concentration is the
calculated hydrate concentration at which the hydrate-bearing
shale will possess a P-wave velocity that is equal to that of sand.
Hydrate saturations larger than this threshold saturation could not
be caused merely by lithologic change, providing that the models
used are relevant. Our modeling shows that the threshold satura-
tion varies from about 30% near the seafloor to about 20% at a depth
of 500 mbsf.

7. Summary
Fig. 16. Comparison of hydrate concentration estimates from both pre- and post-drill
PSWI (Sgh–Seis-Pre-Drill, Sgh–Seis-Post-Drill) with those estimated from log mea-
surements (Sgh–R from resistivity, and Sgh-Vp from P-wave velocity). The comparison We presented a detailed analysis of estimation of gas hydrate
shows coherency in zones A, B, and the interval immediately above the BSR. concentration based on log data in KC151 wells and compared these
estimates with pre-drill and post-drill seismic predictions. Our log-
based estimation implies that low hydrate saturation (<20%) zones
may be considered reliable within the context of possible errors in were penetrated by drilling in KC151 at the interval between 2.0
the estimation of velocity or impedance. and 2.1 s. The log estimates are based on resistivity and sonic log
We first obtain the porosity, density, and P-wave velocity trends (Vp) information. The resistivity-based estimates yield a larger Sgh
of shale based on the measurements in the KC151 wells (Fig. 19). than those based on sonic logs alone. Also, sonic-based results are

Fig. 17. Acoustic impedance inversion results from simultaneous inversion. The blue curves are the inverted acoustic impedance at the well location, and the red curves are the
corresponding result from post-drill PSWI. The green curves are the initial prior models. The panels show the consistent inversion results from both raw and conditioned seismic
data.
J. Dai et al. / Marine and Petroleum Geology 25 (2008) 845–859 857

Fig. 18. Hydrate concentration (Sgh) estimation based on P-wave velocity from simultaneous inversion. The left panel shows the input Vp curve from simultaneous inversion and the
smooth background trend. The middle panel plots Vp and its background trend against the nomogram of modeled P-wave velocities with hydrate saturations ranging from 0 to 50%
with an increment of 5% from left to right. The right panel shows the estimated hydrate saturation based on the input Vp curve.

not consistent with the possible presence of hydrates in zone B, one Comparison of log-based estimates with the pre-drill seismic
of the two zones inferred to contain hydrate based on an analysis of prediction and subsequent post-drill model calibration is encour-
the resistivity logs. aging. It indicates that the current seismic method can capture gas
Two pressure core samples were recovered within the interval hydrate concentration anomalies, even at low to moderate satu-
of interest. Degassing experiments released methane consistent rations. However, quantitative seismic assessment of gas hydrates
with 6% and 1.5% hydrate by volume at depths of 236 mbsf and in low hydrate-concentration zones may be quite error-prone.
383 mbsf, respectively (JIP coring update in 2006). These results Gas hydrate saturations from seismic estimation are systemat-
are, in general, consistent with seismically based post-drill esti- ically lower than those estimated from downhole logs. We believe
mates discussed in this paper. that this is mainly due to thin layer effects. When the log is upscaled

Fig. 19. Background models for shale and sand constrained by the KC151 drilling result. The blue curves are the porosity, density, and P-wave velocity models for shale, and the red
curves are the corresponding models for sand. The back dashed lines are from the well logs.
858 J. Dai et al. / Marine and Petroleum Geology 25 (2008) 845–859

Fig. 20. Seismic detectability of natural hydrates. In the left panel, the dashed line with lower values shows the P-wave velocity for the shale trend, and the dashed line with larger
values displays the sand trend. The group of solid curves represents hydrate concentrations from 0 to 50% with an increment of 5% from left to right, starting with the shale
background. The solid curve in the right panel shows hydrate concentration interpolated from the sand P-wave velocity trend.

to the seismic scale, the agreement with the estimated hydrate the Mallik 2002 Gas Hydrate Production Research Well, Mackenzie Delta,
Northwest Territories, Canada. Geological Survey of Canada Bulletin 585.
saturations is found to be much better. Post-drill analysis indicates
Collett, T.S., Godbole, S.P., Ecomomides, C.E., 1984. Quantification of in-situ gas
that PSWI results agree with those based on analysis of the well log hydrates with well logs. In: Proceedings of the 35th Annual Technical
Vp measurements after upscaling the sonic data to seismic Meeting of the Petroleum Society of CIM, June 10–13, Calgary, Canada, pp.
frequencies. In addition, adjusting the hydrate nomogram to the 571–582.
Collett, T.S., Lee, M.W., Dallimore, S.R., Agena, W.F., 1999a. Seismic- and well-log-
background trend of the input velocity data enhances the reliability inferred gas hydrate accumulations on Richards Island. Geological Survey of
of hydrate saturation estimation from seismic measurements. Canada Bulletin 544, 357–376.
We also tested simultaneous inversion technology and pro- Collett, T.S., Lewis, R.E., Dallimore, S.R., Lee, M.W., Mroz, T.H., Uchida, T., 1999b.
Detailed evaluation of gas hydrate reservoir properties using JAPEX/JNOC/GSC
duced P- and S-wave impedance volumes. The results from this Mallik 2L-38 gas hydrate research well downhole well-log displays. Geological
process matched those from the PSWI approach and the sonic log at Survey of Canada Bulletin 544, 295–311.
the well. This implies that the inversion technology used here is Dai, J., Xu, H., Snyder, F., Dutta, N., 2004. Detection and estimation of gas hydrates
using rock physics and seismic inversion: examples from the northern deep-
adequate for seismic characterization of gas hydrates. Because the water Gulf of Mexico. Leading Edge 23, 60–66.
simultaneous inversion technology does not use full waveform Dai, J., Snyder, F., Gillespie, D., Koesoemadinata, A., Dutta, N., 2008. Exploration for
matching in the inversion algorithm, it is very fast and well-suited gas hydrates in the deepwater northern Gulf of Mexico: part I. A seismic
approach based on geologic model, inversion and rock physics principles. Mar.
for inversion of large volume of 3D seismic data. Petr. Geol. 25, 830–844.
Finally, inference of significant (w10–20% Sgh) hydrates in two Diaconescu, C.C., Kieckhefer, R.M., Knapp, J.H., 2006. Geophysical evidence for gas
zones (A and B) in the well coincided with sand-rich intervals of hydrates in the deep water of the south Caspian Basin, Azerbaijan. Mar. Petr.
Geol. 18, 209–221.
a predominantly shale section. Out of these two intervals, zone A
Dutta, N.C., 2002. Deepwater geohazard prediction using prestack inversion of large
probably has a higher percentage of sand and, based on our results, offset P-wave data and rock model. Leading Edge, 193–198.
is more likely to contain hydrate than zone B. Ecker, C., Dvorkin, J., Nur, A., 1998. Sediments with gas hydrates: internal structure
from seismic AVO. Geophysics 63, 1659–1669.
Ellis, M., Evans, R.L., Hutchinson, D.R., Hart, P.E., Gardner, J., Hagen, R., 2008.
Acknowledgements Electromagnetic surveying of seafloor mounds in the Gulf of Mexico. Mar. Petr.
Geol. 25, 960–968.
We thank A. Dev and M. Eissa for their help with the study. This Gardner, G.H.F., Gardner, L.W., Gregory, A.R., 1974. Formation velocity and density-
the diagnostic basics for stratigraphic traps. Geophysics 39, 770–780.
paper was prepared with the support of the U.S. Department of Guerin, G., Goldberg, D., Meltser, A., 1999. Characterization of in situ elastic prop-
Energy. However, any opinions, findings conclusions, or recom- erties of gas hydrate-bearing sediments on the Blake Ridge. J. Geophys. Res. 104,
mendations expressed herein are those of the authors and do not 17,781–17,795.
Gregory, A.R., 1977. Aspects of rock physics from laboratory and log data that are
necessarily reflect the views of the DOE. important to seismic interpretationSeismic Stratigraphy – Applications to
Hydrocarbon Exploration. AAPG Memoir 26, 15–45.
References Hacikoylu, P., Dvorkin, J., Mavko, G., 2006. Resistivity–velocity transforms revisited.
Leading Edge, 1006–1009.
Archie, G.E., 1942. The electrical resistivity log as an aid in determining some Hutchinson, D.R., Hart, P.E., Ruppel, C.D., Snyder, F., Dugan, B., 2008. Seismic and
reservoir characteristics. Petr. Trans. AIME 146, 54–62. thermal characterization of a bottom simulating reflection in the northern Gulf
Arps, J.J., 1953. The effect of temperature on the density and electrical resistivity of of Mexico. In: Collett, T.S., Johnson, A., Knapp, C., Boswell, R. (Eds.), Natural Gas
sodium chloride solutions. Petr. Trans. AIME 198, 327–330. Hydrates: Energy Resources, Potential and Associated Geologic Hazards. AAPG
Collett, T.S., 2001. A review of well-log analysis techniques used to assess gas- Special Publication.
hydrate-bearing reservoirs. In: Paull, C.K., Dillon, W. (Eds.), Natural Gas Hyndman, R.D., Moran, K., Yuan, T., 1999. The concentration of deep sea gas
Hydrates; Occurrence, Distribution and Detection. American Geophysical Union hydrates from downhole resistivity logs and laboratory data. Earth Planet. Sci.
Monograph, vol. 124. AGA, Washington, D.C., pp. 189–210. Lett. 172, 167–177.
Collett, T., Lee, M., 2004. Archie well log analysis of gas hydrate saturations in the Kennett, B.L., 1983. Seismic Wave Propagation in Stratified Media. Cambridge
Mallik 5L-38 well. In: Dallimore, S.R., Collett, T.S. (Eds.), Scientific Results from University Press, Cambridge, UK, 126 pp.
J. Dai et al. / Marine and Petroleum Geology 25 (2008) 845–859 859

Kleinberg, R.L., Flaum, C., Straley, C., Brewer, P.G., Malby, G.E., Peltzer, E.T., Mallick, S., 2007. Amplitude-variation-with-offset, elastic impedance, and wave
Friederich, G., Yesinowski, J.P., 2003. Seafloor nuclear magnetic resonance assay equation synthetics, a modeling study. Geophysics 72, C1–C7.
of methane hydrate in sediment and rock. J. Geophys. Res. 108, 2137, doi:10. Mathews, M., 1986. Logging characteristics of methane hydrate. Log Analyst 27,
1029/2001JB000919. 26–63.
Kleinberg, R.L., Flaum, C., Collett, T.S., 2005. Magnetic resonance log of Mallik 5L 38: Pearson, C.F., Halleck, P.M., McGuire, P.L., Hermes, R., Mathews, M., 1983. Natural gas
hydrate saturation, growth habit, and relative permeability. In: Dallimore, S.R., hydrate deposit: a review of in situ properties. J. Phys. Chem. 87, 4180–4185.
Collett, T.S. (Eds.), Scientific Results from the Mallik 2002 Gas Hydrate Rasmussen, K.B., Bruun, A., Pedersen, J.M., 2004. Simultaneous Seismic Inversion.
Production Research Well, Mackenzie Delta, Northwest Territories, Canada. In: 66th Meeting of European Association of Geoscientists and Engineers. EAGE
Geological Survey of Canada Bulletin 585. P165.
Kvenvolden, K., Barnard, L., 1983. Hydrates of natural gas in continental margins. In: Shipley, T.H., Houston, M.H., Buffler, R.T., Shaub, F.J., McMillen, K.J., Ladd, J.W.,
Watkins, J., Drake, C. (Eds.), AAPG Memoir, 34, pp. 631–640. Worzel, J.L., 1979. Seismic evidence for widespread possible gas hydrate
Lee, M., Collett, T., 2008. Integrated analysis of well logs and seismic data to esti- horizons on continental slopes and rises. AAPG Bulletin 12, 2204–2213.
mate gas hydrate concentrations at Keathley Canyon, Gulf of Mexico. Mar. Petr. Wood, W.T., Hart, P.E., Hutchinson, D.R., Dutta, N., Snyder, F., Coffin, R.B., Gettrust, J.F.,
Geol. 25, 924–931. 2008. Gas and gas hydrate distribution around seafloor seeps in Mississippi
Mallick, S., 1995. Model-based inversion of amplitude-variations-with-offset data Canyon, Northern Gulf of Mexico, using multi-resolution seismic imagery. Mar.
using a genetic algorithm. Geophysics 60, 939–954. Petr. Geol. 25, 952–959.
Mallick, S., 1999. Some practical aspects of pre-stack waveform inversion using Xu, H., Dai, J., Snyder, F., Dutta, N., 2004. Seismic detection and quantification of gas
a genetic algorithm: an example from East Texas Woodbine gas sand. hydrates using rock physics and inversion. In: Taylor, C.E., Kwan, J.T. (Eds.),
Geophysics 64, 326–336. Advances in the Study of Gas Hydrates. Kluwer, New York, pp. 117–139.

You might also like