t Special section: Seismic data conditioning

Target-oriented data conditioning for prestack inversion

in an unconventional reservoir: A Canadian case study
Jorge Estrada1, Peter Aaron1, and Richard Eden1

ABSTRACT
Target-oriented data conditioning is a key part of any reservoir characterization workflow. Data conditioning
is used to optimize the match between the synthetic data, used in the prestack inversion, and the real data. When
this is done correctly, the accuracy and confidence of inversion results may be greatly improved. This is proved
on prestack seismic inversion results from a resource play in Canada. The flow is broken down into prestack
gather conditioning, which improves the signal-to-noise and gather flatness, and poststack conditioning, which
further improves the coherency prior to applying spectral balancing. As a final key step, spatially variant am-
plitude balancing is used to calibrate the angle stacks to the expected background trend from the well syn-
thetics. The combination of all steps is demonstrated via V P ∕V S and mu-rho versus lambda-rho crossplots,
between the inverted results and the well measurements, to provide a significant improvement on the resolution
and accuracy of the final prestack inversion.

Introduction the following: flat gathers, optimal signal-to-noise ratio,

Since the classic article by Ostrander (1984), which spatially consistent amplitudes and bandwidth, as well
proposed a method to distinguish between gas-related as improved resolution. It is also recommended to ana-
and non-gas-related amplitude anomalies using Poisson’s lyze and compare the AVO background of the seismic
ratio, the analysis of prestack data has become increas- and synthetics at well locations and apply an appropri-
ingly common in reservoir characterization, first with ate angle scaling to compensate.
amplitude variation with offset (AVO) analysis (Chacko, A key part of any target-oriented data conditioning
1989; Rutherford and Williams, 1989; Castagna, 1993) workflow is the quality control (QC). It is important to
and more recently with prestack inversion (Hampson monitor and analyze the impact of each step on the AVO
et al., 2005). To understand the requirement to condition attributes and to check that a better correlation be-
data for prestack inversion, it is important to understand tween the seismic and well synthetics is being achieved.
the effects of wave propagation on seismic amplitudes In most cases, the conditioned seismic data will also
(O’Doherty and Anstey, 1971; Sheriff, 1975; Martinez, produce an improved stack for horizon interpretation,
1993) and the impact of algorithms and parameters ap- fault detection, and structural attribute computations
plied during seismic processing (Resnick, 1988). such as curvature and discontinuity.
Although the importance of applying AVO-compliant
algorithms during processing is well accepted, conven- Seismic data conditioning
tional prestack time migration (PSTM) and prestack A 3D wide-azimuth land seismic data set from
depth migration (PSDM) processing workflows must Canada is used to demonstrate the impact of carefully
try to honor a large area and a long time window usually executed target-oriented data conditioning on the pre-
encompassing several different target levels of varying stack inversion results and azimuthal anisotropy analy-
geophysical and geologic characteristics. As a result, sis. The gather conditioning workflow is described in
prestack inversion will generally benefit from a target- Figure 1. The input data are AVO-compliant, raw PSTM
oriented data conditioning workflow, which begins offset vector tile (OVT) gathers without any postmigra-
with AVO-compliant raw migrated gathers (Singleton, tion processing applied. Because the data are wide azi-
2009; Schmidt et al., 2013). muth and one of the requirements was to perform
Given the well-known assumptions of a prestack azimuthal anisotropy analysis, data conditioning had to
inversion, acceptable data conditioning will result in take into consideration the azimuthal amplitude and

1
Apache Corporation, Houston, Texas, USA. E-mail: jorge.estrada@apachecorp.com; peter.aaron@apachecorp.com; richard.eden@apachecorp.
com.
Manuscript received by the Editor 31 August 2015; published online 21 March 2016. This paper appears in Interpretation, Vol. 4, No. 2 (May
2016); p. SG11–SG18, 12 FIGS.
http://dx.doi.org/10.1190/INT-2015-0149.1. © 2016 Society of Exploration Geophysicists and American Association of Petroleum Geologists. All rights reserved.

Interpretation / May 2016 SG11

Downloaded from http://pubs.geoscienceworld.org/interpretation/article-pdf/4/2/SG11/4841858/int-2015-0149.1.pdf

by CNOOC user
 traveltime variations present in the data. However, be- The first step used an automatic velocity picker to
cause prestack inversion does not recognize azimuthal derive high-resolution velocities at each common mid-
traveltime effects, the flow also needed to correct for point (CMP) location. To minimize the impact of multi-
these prior to generating the full-azimuth angle stacks. ples and coherent noise, it was critical to apply velocity
The workflow was designed so that the same offset gath- specific or “throw-away” conditioning that included a
ers were used to generate the azimuthal angle stacks for Radon, some band-pass filtering, and an automatic gain
analysis of amplitude variations with azimuth and the control to all azimuths. The smooth migration velocities
full-azimuth angle stacks for prestack inversion. Analysis were used as a guide function to further constrain the
of the velocity variations with azimuth was also carried update. The velocities were picked and applied at float-
out using the azimuthal traveltime correction information. ing datum, and the high-resolution velocities were
verified against the well data. Spatial smoothing was
applied to the interval velocities, parameterized to
statistically remove outliers while ensuring that the geo-
logic detail was preserved and preventing artifacts at
the far offsets. Figure 2 compares the original smooth
migration velocities to the high-resolution residual
velocities.
The updated velocities resulted in improved flattening
of the primaries in the CMP gathers, making it easier to
distinguish them from coherent noise and multiples. To
preserve azimuthal amplitude and timing effects, the
OVT CMP gathers were first binned to six 30° azimuth
sectors with 150 m offsets. Radon was applied to the azi-
muth sectored CMP gathers, making sure that the final
parameterization was carefully chosen based on study-
ing azimuthal angle stacks and gathers sorted to com-
mon offset, common azimuth (COCA) before and after.
Azimuthal traveltime variations were analyzed, and
the resulting time shifts were computed and applied
to correct for them. A mild 3D residual gather flattening
removed the remaining nonhyperbolic moveout effects
with offset and concluded the prestack portion of the
conditioning flow. Figure 3 shows a COCA gather be-
fore and after the full prestack conditioning.
For use in the prestack inversion, four full azimuth
angle stacks were generated from the conditioned CMP
gathers at angle centers 7.5°, 16.5°, 25.5°, and 34.5° by
recombining all offsets and azimuths. The angle stacks
were generated at floating datum using a smoothed
version of the high-resolution stacking velocities. Ac-
quisition footprint attenuation and structurally oriented
Figure 1. Data conditioning workflow. filtering were applied to attenuate coherent noise with-

Figure 2. Interval velocities for (a) original

migration velocities, (b) raw hi-res velocities,
and (c) hi-res velocities after conditioning for
gather flattening.

SG12 Interpretation / May 2016

Downloaded from http://pubs.geoscienceworld.org/interpretation/article-pdf/4/2/SG11/4841858/int-2015-0149.1.pdf

by CNOOC user
 in the individual angle stacks and resulted in a further balancing was used to normalize the wavelet over a
improvement in the S/N levels. Figure 4 shows a time background analysis window centered on target level.
slice through the 17.5° stack before and after footprint The same window was subsequently used to spatially
attenuation and structurally oriented filtering. calibrate the background amplitudes for each angle
QC plots of the rms and ”midfrequency” for a long stack to those expected from the well synthetics. The
window around the target level revealed spatial varia- window length was chosen to ensure that the ampli-
tions in background amplitudes and frequency content. tudes were representative of the background without
This is not uncommon for land data sets, and closer introducing false AVO effects. Figure 5 compares the
inspection, including maps taken over shallower win- average of three well synthetics versus the real data
dows, suggested they were caused by residual near- and lists the target rms values for each angle. Note that
surface and acquisition effects. Because prestack inver- care was taken to take the rms measurements over the
sion assumes a single wavelet for each angle, spectral same interval on the synthetic and real data. Figure 6

Figure 3. The COCA gather (a) raw data and

(b) after prestack conditioning.

Figure 5. Background analysis window compares the average

Figure 4. Time slice through the near stack (a) before and rms amplitudes of three well synthetics (green curve) versus
(b) after poststack footprint attenuation and structurally ori- the real seismic data before (red curve) and after (black curve)
ented filtering. angle scaling.

Interpretation / May 2016 SG13

Downloaded from http://pubs.geoscienceworld.org/interpretation/article-pdf/4/2/SG11/4841858/int-2015-0149.1.pdf

by CNOOC user
 Figure 6. Top: midfrequency maps taken on
long window over target interval for (a) raw
and (b) final conditioned 17.5° stack. Bottom:
rms maps over same interval for (c) raw and
(d) final conditioned 17.5° stack. Note that the
rms has now been scaled to a constant value
based on the calibration using the well syn-
thetics.

Figure 7. Arbitrary line through the 26° stack for (a) raw data
and (b) after the data conditioning workflow. The orange and
green horizons bound the target interval.

Figure 8. Original logs and the synthetic

created using the original logs in black. The
blocked logs and the synthetic created using
the blocked logs in red.

SG14 Interpretation / May 2016

Downloaded from http://pubs.geoscienceworld.org/interpretation/article-pdf/4/2/SG11/4841858/int-2015-0149.1.pdf

by CNOOC user
 Figure 9. Comparison of the inversion results before seis-
mic conditioning in blue and after seismic conditioning in
red. The black curves are the blocked well logs.

Figure 10. Inverted V P ∕V S ratio versus V P ∕V S ratio

from the blocked logs for (a) without and (b) with
target-oriented conditioning.

Interpretation / May 2016 SG15

Downloaded from http://pubs.geoscienceworld.org/interpretation/article-pdf/4/2/SG11/4841858/int-2015-0149.1.pdf

by CNOOC user
 shows the rms and midfrequency plots, taken over Prestack inversion results
the target interval on the 17.5° stack before and after To demonstrate the benefit of this workflow, pre-
the full data conditioning flow, and Figure 7 shows an stack inversions were carried out on data sets with
arbitrary line through the corresponding data. and without the target-oriented data conditioning. The
unconditioned data set was generated directly from the
raw OVT PSTM gathers, outputting the same full-azi-
muth angle stacks. For consistency, the same velocities
were used for the angle decomposition. To make the
comparison fair, identical background models and lith-
ology constraints were used. The only difference be-
tween the two inversions are the input seismic, the
statistically derived angle depended wavelets extracted
from each of the data sets, and the global scalars com-
puted during the inversion process.
Comparison of the inversion result versus the well
data was done using a blocked version of the well logs.
Figure 8 shows the original and the blocked well logs
from well B. The numbers of blocks were selected
based on a synthetic comparison using the original logs
and the minimum amount of blocks require to match
the synthetic created with the original logs. The wavelet
used to create the synthetics was a statistical wavelet
extracted from the near stack over the zone of interest.
The blocked logs are representative of the expected
seismic resolution from the inversion results. In this
case, blocking enabled us to mimic the seismic band-
width, while preserving the rapid changes in rock prop-
erties seen in the well logs compared with a filtering
method.
Figure 9 shows the comparison of the inversion re-
sults before and after the target-oriented conditioning at
well B. Note the inversion results using the conditioned
data at the zone of interest (between the tops H1 grid
and base pay) show better matching with the blocked
well logs for all the inversions attributes (P-impedance,
S-impedance, V P ∕V S ratio, lambda-rho [LMR], and
mu-rho).
Using the data from wells A, B, and C, Figure 10
shows the crossplot comparison, taken over the zone
of interest between the tops H1 grid and base pay, of
the inverted V P ∕V S ratio with and without the target-ori-
ented seismic conditioning versus the V P ∕V S ratio from
the blocked logs. Note the improvement in correlation
from 80% to 93%, showing that the target-oriented data
conditioning has had a significant impact on the AVO
gradient-related attributes.
Figure 11 shows a comparison using mu-rho versus
LMR crossplots. The LMR crossplot in Figure 11a uses
data from the blocked well logs and identifies the zones
corresponding to above the reservoir and the pay zone.
The same zones are then shown overlain on the corre-
sponding crossplots from the respective inversion re-
sults with and without conditioning.
Figure 12 shows the projection of the crossplot zones
on an arbitrary 2D section going through wells A, B, and
Figure 11. Interpreted zone of the LMR crossplot (a) using
C using the inverted attributes before and after seismic
the blocked well data, (b) using the inversion results before conditioning. The data conditioning has improved the
seismic conditioning, and (c) using the inversion results after continuity of the pay zone and given a better correlation
seismic conditioning. with the thickness of the pay zone at the well locations.

SG16 Interpretation / May 2016

Downloaded from http://pubs.geoscienceworld.org/interpretation/article-pdf/4/2/SG11/4841858/int-2015-0149.1.pdf

by CNOOC user
 Martinez, R., 1993, Wave propagation ef-
fects on amplitude variation with off-
set measurements: A modeling study:
Geophysics, 58, 534–543, doi: 10.1190/
1.1443436.
O’Doherty, R. F., and N. A. Anstey, 1971,
Reflections on amplitudes: Geophysical
Prospecting, 19, 430–458, doi: 10.1111/j
.1365-2478.1971.tb00610.x.
Ostrander, W. J., 1984, Plane-wave reflec-
tion coefficients for gas sands at nonnor-
mal angles of incidence: Geophysics, 49,
1637–1648, doi: 10.1190/1.1441571.
Resnick, J. R., 1988, Seismic data process-
ing for amplitude versus offset analysis:
58th Annual International Meeting, SEG,
Expanded Abstracts, 1354.
Rutherford, S., and R. H. Williams, 1989,
Amplitude-versus-offset variations in gas
sands: Geophysics, 54, 680–688, doi: 10
.1190/1.1442696.
Schmidt, D., A. Veronesi, F. Delbecq, and
J. Durand, 2013, Seismic data precondi-
Figure 12. Crossplot zones projection on an arbitrary 2D section: (a) before tioning for improved reservoir characteri-
seismic conditioning and (b) after seismic conditioning. zation (Inversion and fracture analysis):
Presented at the GeoConvention.
Conclusion Sheriff, R. E., 1975, Factors affecting seismic amplitudes:
Target-oriented data conditioning provided a signifi- Geophysical Prospecting, 23, 125–138, doi: 10.1111/j
cant impact on the inversion results: It improved the .1365-2478.1975.tb00685.x.
accuracy and the geologic meaning of the inversion Singleton, S., 2009, The effects of seismic data conditioning
attributes at the wells by better delineating the bounda- on prestack simultaneous impedance inversion: The
ries of the potential reservoir zone. Leading Edge, 28, 772–781, doi: 10.1190/1.3167776.
Gather conditioning workflows are data dependent
and reservoir oriented; therefore, the algorithms and
parameters are different for each study and must be
carefully tested to make sure that the seismic signal is Jorge Estrada received a B.S. (2003)
preserved. in geophysical engineering from
UNAM in Mexico and an M.E. (2005)
in reservoir geoscience and engineer-
ing from the IPF-School in Paris,
Acknowledgments
France. He is a senior geophysicist in
We would like to thank J. Lee and T. Maksymchuk
the Technology Group with Apache
from Apache Canada for their valuable contributions Corporation in Houston. His research
plus the EPT geophysics group in Houston for their sug- interests include the application of
gestions and feedback on the paper. In addition, we seismic inversion, AVO, and other seismic attribute technol-
would like to thank Seitel for allowing us to show their ogies for exploration risk reduction and assessment.
data.

References Peter Aaron received a B.Sc. in phys-

Castagna, J. P., 1993, Petrophysical imaging using AVO: ics from the University of Birmingham
The Leading Edge, 12, 172–178, doi: 10.1190/1.1436939. (2000) and an M.Sc. in exploration
geophysics from the University of
Chacko, S., 1989, Porosity identification using amplitude
Leeds (2001). He joined Petroleum
variation with offset: Examples from South Sumatra: Geo-Services in 2001 and worked on
Geophysics, 54, 942–951, doi: 10.1190/1.1442737. the development and commercializa-
Hampson, D., B. Russell, and B. Bankhead, 2005, Simulta- tion of seismic processing software.
neous inversion of pre-stack seismic data: 75th Annual Since 2013, he has been a senior geo-
International Meeting, SEG, Expanded Abstracts, 1633– physicist within the Technology Group with Apache Cor-
1637. poration in Houston. His research interests include in land

Interpretation / May 2016 SG17

Downloaded from http://pubs.geoscienceworld.org/interpretation/article-pdf/4/2/SG11/4841858/int-2015-0149.1.pdf

by CNOOC user
 and marine seismic acquisition, processing and imaging,
AVO analysis, inversion and seismic attributes.

Richard Eden received a B.Sc. in geo-

physics from the University of East
Anglia (1999) and an M.Sc. in geophys-
ics from the University of Durham
(2000). In 2000, he joined Petroleum
Geo-Services as a Depth Imaging Geo-
physicist and worked in the locations
of UK and USA, then he moved to Aus-
tralia in 2009, to manage Petroleum
GeoServices’ Data Processing Center. In 2012, he returned
to the U.S. and joined Apache as a senior staff geophysicist
where he is responsible for a small team focusing on spe-
cialized seismic data processing and inversion projects in
support of their domestic and international regions.

SG18 Interpretation / May 2016

Downloaded from http://pubs.geoscienceworld.org/interpretation/article-pdf/4/2/SG11/4841858/int-2015-0149.1.pdf

by CNOOC user