Chasing anisotropy in the Austin Chalk and Eagle Ford shale formations: azimuthal processing

challenges and considerations

Downloaded 08/22/17 to 80.82.77.83. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Peng Cheng, Mike Perz*, Femi Ogunsuyi, Frank Meng, and Xinxiang Li, Arcis Seismic Solutions, A TGS
Company; William Keller and Victor Kriechbaum, EnerVest, Ltd.

Summary Details of this comparison are presented in a companion

paper (Keller et al., 2017), but the main point in the context
A 3D land seismic survey from south central Texas was of the present work is that a strong correlation was observed
processed using an azimuth-friendly flow, and subsequent between production and the estimated azimuthal anisotropy
azimuthal velocity inversion (VVAZ) was performed in properties. In particular, regions exhibiting relatively low
order to characterize the subsurface anisotropy. The values of Vint_slow and/or values of strong anisotropy (i.e.,
resulting anisotropy maps show a strong correlation with as evidenced by the magnitude of Vint_fast minus
production data and therefore demonstrate successful Vint_slow) were observed to correspond to regions of high
application of the VVAZ technology. This paper examines production (mostly gas).
those key elements of the data processing and of the VVAZ
strategy which led to this successful result, namely multi- Not only does this happy observation of strong correlation
domain pre-migration noise attenuation, 5D interpolation to between VVAZ attribute maps and production data imply
an output grid which is regularly sampled across offset and that VVAZ inversion holds great promise as a tool for
azimuth coordinates, post-migration noise attenuation optimizing future production in the area, it also serves as
operating directly in the coordinate planes created by the implicit validation of the processing and inversion
above 5D interpolation, careful time-shift estimation and methodologies that were used to produce the maps. It is this
maximum incidence angle selection during the azimuthal latter consideration that constitutes the primary motivation
RMS parameter estimation process, and finally careful behind the writing of this paper, and in the following work
testing of the impact of the number of 5D-interpolated we discuss those elements of the processing flow and of the
azimuths on the quality of the azimuthal interval property VVAZ inversion which we believe contributed most
estimates. significantly to the successful outcome.

Introduction Description of key methodologies

The data set under study is a 50 sq. mile subset of a tightly- Pre-migration noise attenuation
sampled, 195 fold, 710 sq. mile 3D seismic survey which Pre-migration noise attenuation was performed in various
was acquired in 2015 over a portion of Giddings Field and domains using an AVO-compliant philosophy. Most
the eastern extension of the Eagle Ford shale play. The zone crticially, this noise attenuation avoided any multi-channel
of interest is between the top of the Austin Chalk and the top processes which risk smearing signal across the azimuth
of the Buda limestone, an approximately 1000 ft interval that domain. While it is relatively easy to naturally avoid such
spans both the Austin Chalk and Eagle Ford formations. At smearing at the linear noise and noise-burst suppression
the outset of the project, it was suspected that vertically stages, particular care is required at the random noise
pervasive fractures, local anomalies in the in-situ horizontal attenuation stage. To this end, our random noise attenuation
stress field, or both could place a significant control on approach entailed running fxy deconvolution in the cross-
hydrocarbon production, and accordingly an effort was spread domain, a domain for which the offset and azimuth
undertaken to characterize the interval velocity azimuthal coordinates vary slowly across neighboring traces within the
anisotropy using the surface seismic data. Processing was processing block. This fxy deconvolution was then
carried out according to an azimuthally-AVO-compliant combined with an adaptive signal addback scheme to ensure
framework which sought to preserve kinematic and preservation of subtle azimuthal signal signatures.
amplitude signal variations across both offset and azimuth
coordinates. After processing through anisotropic (VTI) pre- 5D interpolation and output geometry considerations
stack time migration (PSTM), the data were submitted to Implementation details of 5D interpolation can vary widely,
VVAZ inversion via the Generalized Dix Inversion but one common practice (and the one adopted here) is to
(Grechka et al., 1999), a methodology which estimates define the 5D internal computational grid in a mixed
azimuthal interval fast and slow velocities and fast-velocity Cartesian-polar coordinate system: (i.e., cmp-x, cmp-y,
orientation (Vint_fast, Vint_slow, _int, respectively) from offset, azimuth) according to the grid sampling
their corresponding RMS counterparts (Vrms_fast, recommended by Trad (2009). This choice of computational
Vrms_slow, _rms, respectively). The resulting interval grid naturally produces a high-quality interpolated set of
anisotropy maps were then compared to production maps. CMP gathers with regular sampling across both offset and

© 2017 SEG Page 458

SEG International Exposition and 87th Annual Meeting
 Azimuthal processing challenges and considerations

azimuth indices, and with fine sampling of the offset by both linear noise and multiple energy and Figure 2b
coordinate in particular. In the case of the present work, we shows the final denoised result. Clearly the denoise strategy
simply grouped these interpolated CMP gathers by common has worked very well: the underlying signal has been
Downloaded 08/22/17 to 80.82.77.83. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

offset-azimuth indices in order to produce single fold

common-offset-vector (COV) ensembles which were in turn
input to PSTM. Note that this process of direct grouping into
COV’s avoids a circuitous approach, adopted by many
practioners, comprising the following steps: (i) interpolation
using the above internal computational grid; (ii) casting the
interpolated data from (i) onto a “surface-referenced” grid
consisting of a set of finely sampled source and receiver
lines; and (iii) performing offset-vector-tiling (OVT) on
these surface-referenced data to produce single-fold,
azimuth-and-offset-localized data subsets which are in turn
input to PSTM. Although this “5D+OVT” approach is
suitable for creating the surface-consistent data
configurations required for certain imaging methods like
reverse-time-migration, it is unnecessary for the Kirchhoff
algorithm which is commonly used in azimuthal processing
workflows, including the present one (Perz and Cary, 2012).

An advantage of the 5D approach employed here (i.e., direct

grouping into COV’s) is that it produces a regular
distribution of polar offsets, which in turn permits the use of
some powerful post-migration noise suppression techniques
described in the next subsection. The differences in offset
and azimuth distributions between the present interpolation
scheme and the 5D+OVT approach are illustrated in Figure
1. The snail gather of the 5D+OVT approach shows an
irregular offset distribution (Figure 1a) and also a relative
dearth of near offsets (blue graph at top). By contrast, data
created by direct grouping of the 5D-interpolated data into
COV’s exhibit a perfectly regular, and finely sampled, offset
distribution (see Figure 1b for data display in common- Figure 1: Examples of CMP gathers after 5D interpolation + PSTM:
(a) snail gather after 5D+OVT+PSTM (west Texas, excerpted from
offset-common-azimuth (COCA) mode with primary sort
McCarthy et al., 2016); (b) COCA gather after direct migration of
key offset, and Figure 1c for same data displayed in CACO data from a 5D computational grid with 30° azimuth interval and 40
mode with primary sort key azimuth). The regular offset m offset interval (central Alberta, data courtesy of Arcis Seismic
distribution is particularly obvious in the CACO display, Solutions, A TGS Company); (c) same data as (b), except ordered
where it should be apparent that each of the four data subsets in CACO mode—note only 4 of the 6 azimuths are shown here .
correspond to a single constant azimuth “spoke”.
unveiled and exhibits very strong azimuthal variation as
Post-migration spoke-by-spoke noise attenuation evidenced by the pronounced “sinusoidal wobble” at far
Returning to Figure 1c, it is evident that the residual linear offsets; moreover, inspection of the underlying noise models
noise is particularly well-organized on each individual (i.e., which were subtracted from the input data) in Figures
azimuth spoke. We can exploit this fact by applying 2c and 2d show no hint of residual primary signal. We note
traditional 2D f-k filtering piecewise on each spoke (where, with interest, if not surprise, that these latter two noise
crucially, the fine offset sampling precludes spatial aliasing) models exhibit significant azimuthal variation themselves.
to attack the noise. In practice we combine the f-k filtering Finally, we note that a primitive 2D noise attenuation
with an adaptive signal addback scheme to ensure no operating across the offset coordinate in an “azimuth-blind”
distortion of primary reflections. Though not shown here, we fashion would surely produce very poor results as the
also observe strong organization of multiple energy on azimuthal variation in both signal and noise would be
azimuth spokes, paving the way for an analogous spoke-by- smeared in the process, resulting in poor noise removal and
spoke demultiple strategy. Figure 2 shows the application of signal distortion.
the spoke-by-spoke denoising to the test subvolume. Figure
2a shows the input COCA gather exhibiting contamination

© 2017 SEG Page 459

SEG International Exposition and 87th Annual Meeting
 Azimuthal processing challenges and considerations

45o, severe cycle-skipping is observed at the higher angles

(blue ellipse). Figure 3c shows the result of applying t’s
estimated via an alternative “sliding window” approach
Downloaded 08/22/17 to 80.82.77.83. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Figure 3: Comparisoin of t time shift estimation approaches .(a)

input data (migrated COCA); (b) after flattening with t’s computed
via AVO-projected pilot algorithm c) after flattening with t’s
computed via sliding window approach. Red dashed line shows 45°
angle. Seismic data is the proprietary property of Seitel, Inc.

which explicitly adapts to offset-dependent waveform

changes. Clearly it has done an excellent job of flattening,
even at the highest propagation angles (orange ellipse).

Figure 2: Spoke-by-spoke noise attenuation. (a) input migrated (ii) Maximum angle in RMS parameter estimation
COCA; (b) final denoised COCA after linear noise and de-multiple Once computed, the t’s are input to an elliptical curve-
steps; (c) noise estimated from linear denoise; (d) multiples fitting process (Grechka and Tsvankin, 1998) to estimate
estimated from de-multiple process. Seismic data is the proprietary azimuthal RMS properties Vrms_fast, Vrms_slow, _rms.
property of Seitel, Inc.
One key input parameter in the curve-fitting process is the
maximum incidence angle for which the t’s are inverted
and its optimal selection proved surprisingly difficult in the
VVAZ inversion considerations
present work. Figure 4a shows a final migrated CIP gather
(i) Time shift estimation
in COCA mode, while Figure 4b shows its counterpart after
After spoke-by-spoke noise removal, the migrated common-
azimuthal NMO correction in the case that the associated
image-point (CIP) gathers were input to an algorithm which
azimuthal RMS inversion only considered t’s
estimates the t time shifts giving rise to the sinusoidal
corresponding to a maximum incidence angle of 40°. Clearly
wobble observed in Figure 2b. Estimation of these t’s
the wobble associated with mid-range offsets (blue ellipse)
required careful consideration on the test subvolume because is well-collapsed; however, the far offset wobble (red
of the large amount of azimuthal anisotropy together with
ellipse) has unfortunately been accentuated. Figure 4c shows
the uncommonly large incident angle range. While the large the azimuthal NMO correction based on a 65° maximum
range is ultimately a fortunate occurrence (because the inversion angle; in this case the far offset wobble is nicely
associated oblique ray angles are quite sensitive to the reduced while the mid-range wobble has been exacerbated.
effects of azimuthal anisotropy), it introduces a large amount
This tradeoff between optimal mid and near offset flattening
of systemic disparity between near and far offset waveforms
proved impossible to perfectly manage, and in the end it was
which can, in turn, pose challenges for t-estimation. Figure decided that a 60° angle be used in the production run.
3a shows a migrated CIP exhibiting extreme azimuthal Possible explanations for the existence of this tradeoff are:
anisotropy for which t time shifts are estimated via two (a) the fact that the elliptical moveout approximation loses
different approaches. The first approach is the “AVO- its validity at high propagation angles, (b) the existence of
projected pilot” technique of Zheng et al., 2008, an approach lateral velocity heterogeneity in the overburden or (c) a
which honors offset-dependent amplitude effects but not combination of both. Perhaps not surprisingly, the
waveform changes. Figure 3b shows the result after first downstream-generated azimuthal interval maps show
estimating then applying, t’s from the this approach (note significant variation depending on the choice of this
that a perfect result would imply perfect gather flattening). maximum angle parameter (Figure 5).
While the algorithm has done a good job of flattening up to

© 2017 SEG Page 460

SEG International Exposition and 87th Annual Meeting
 Azimuthal processing challenges and considerations

(iii) Impact of number of azimuths Next, for each post-5D output-azimuth configuration we
Lynn (2011) demonstrated via synthetic experiments that the estimated t shifts, performed the elliptical curve fitting to
quality of the azimuthal RMS parameter estimates can vary compute RMS azimuthal parameters, and finally executed
Downloaded 08/22/17 to 80.82.77.83. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

with the number of input azimuths (with reliability the Generalized Dix Inversion to estimate the interval
improving with increasing number), and, correspondingly azimuthal parameters. Results are shown in Table 1, where
that the number of azimuths may have a profound impact on it is clear that quality of the Generalized Dix Inversion result
the quality of the final azimuthal interval estimates. His after 5D interpolation does not improve with increasing
experimental approach, while scientifically sound, did not azimuth. In fact, quality actually degrades with increasing
consider the effects of 5D interpolation, a relatively new azimuth, an observation which is likely due to the fact that
inclusion in azimuthal processing flows which introduces a 5D interpolation struggles with extreme, and regular,
complex interplay between signal-to-noise enhancement, upsampling across any of its coordinates (in this case
interpolated image quality, and azimuth smearing. In order azimuth). Based on the results of this testing, 6 azimuths
to study the effect of the number of output azimuths from 5D were output from 5D interpolation in the production run.
interpolation on the quality of the estimated azimuthal

Table 1: Synthetic testing of azimuthal interval property estimation

for various trial post-5D output-azimuth configurations. Results are
compared against the “true” parameters defined in Lynn (2011).
Here “% aniso” is defined as (Vfast_int - Vslow_int)/Vfast_int.

Conclusions

Careful azimuth-friendly processing and VVAZ inversion

were performed on a data set exhibiting a wide range of
incident angles, resulting in successful characterization of
Figure 4: COCA’s before and after azimuthal NMO correction with
anisotropy. Key steps in processing were: azimuthally-
45° angle shown in dashed green. (a) data before correction; (b) after
correction via azimuthal RMS property inversion with maximum AVO-compliant pre-migration noise attenuation, 5D
angle of 40°; (c) same as (b), except a maximum angle of 65° was interpolation to a set of regularly sampled azimuth spokes,
used. and spoke-by-spoke post-migration noise attenuation. Key
considerations in VVAZ inversion were: careful attention in
interval properties, we devised a synthetic experiment. time shift estimation, use of a relatively large incident angle
in azimuthal RMS property estimation, and finally choosing
Specifically we considered the same azimuthally anisotropic to output 6 azimuths after 5D interpolation. Future work is
earth model as Lynn (2011) and we performed forward aimed at comparing the present PSTM-based anisotropy
modeling to create synthetic traces (i.e., containing characterization to one obtained through orthorhombic
PSDM. The possible benefits of using PSDM instead of
PSTM in azimuthal velocity characterization have been
discussed by numerous authors, including Goodway (2016),
Belguermi et al. (2016) and Brown et al. (2017).

Acknowledgements

Figure 5: Azimuthal interval maps of Vint_fast computed using two We thank Walt and Heloise Lynn for their help and guidance
different choices for maximum incident angle in the upstream throughout the project. We thank Enervest, Ltd. and Seitel,
azimuthal RMS parameter estimation.. (left) 40° maximum angle;
Inc. for granting permission to publish this work.
(right) 60° maximum angle. Images courtesy of Lynn Inc.

anisotropic traveltime effects) based on the real survey

geometry of the test subvolume. We added random noise to
this synthetic data set and submitted the noisy data to 5D
interpolation for several trial output-azimuth configurations.

© 2017 SEG Page 461

SEG International Exposition and 87th Annual Meeting
 EDITED REFERENCES
Note: This reference list is a copyedited version of the reference list submitted by the author. Reference lists for the 2017
SEG Technical Program Expanded Abstracts have been copyedited so that references provided with the online
metadata for each paper will achieve a high degree of linking to cited sources that appear on the Web.
Downloaded 08/22/17 to 80.82.77.83. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

REFERENCES
Belguermi C., A. Zarkhidze, N. Meddour, H.F. Zeidan, C. Chironi, H. Harkas, and A. Widjiastono, 2016,
Apparent azimuthal anisotropy resolved by depth imaging: 86th Annual International Meeting,
SEG Expanded Abstracts, 328–331, http://dx.doi.org/10.1190/segam2016-13948733.1.
Brown, M., D. Waibel, S. Schapper, and J. Starr, 2017, Marcellus and Utica Shale PSDM Case Study –
Improved Structural Imaging and Rock Properties: 23rd Annual RMAG and DGS 3D Seismic
Symposium.
Goodway, W., G. Purdue, and M. Perez, 2016, Pre-stack depth imaging to remove the effects of lateral
velocity heterogeneity on time imaging that masquerade as azimuthal anisotropy: Midale EOR
Case Study: 5th Annual CSEG Symposium.
Grechka, V., and I. Tsvankin, 1998, 3-D description of normal moveout in anisotropic inhomogeneous
media: Geophysics, 63, 1079–1092, http://dx.doi.org/10.1190/1.1444386.
Grechka, V., I. Tsvankin, and J. Cohen, 1999, Generalized Dix equation and analytic treatment of normal
moveout velocity for anisotropic media: Geophysical Prospecting, 47, 117–148,
http://dx.doi.org/10.1046/j.1365-2478.1999.00120.x.
Keller, W., R. Mott, A. Jumper, H. Lynn, W. Lynn, and M. Perz, 2017, Correlation of azimuthal velocity
anisotropy and seismic inversion attributes to Austin Chalk production, a south central Texas case
study: 87th Annual International Meeting, SEG, Expanded Abstracts, submitted.
Lynn, W., 2011, Azimuthal interval velocity uncertainty: 81st Annual International Meeting, SEG,
Expanded Abstracts, 279–283, http://dx.doi.org/10.1190/1.3627776.
McCarthy, A., A.B. Ahmadi, and J. Schneider, 2016, Advanced seismic processing: a case study on
imaging in the Delaware Basin: CSEG Geoconvention.
Perz, M., and P. Cary, 2012, 5D interpolation and COV migration: A perfect marriage: CSEG
Geoconvention.
Trad, D., 2009. Five-dimensional interpolation: Recovering from acquisition constraints: Geophysics, 74,
no. 6, V123–V132, http://dx.doi.org/10.1190/1.3245216.
Zheng Y., J. Wang, and M. Perz, 2008, Pitfalls and tips for seismic fracture analysis: 78th Annual
International Meeting, SEG, Expanded Abstracts, 1531–1535,
http://dx.doi.org/10.1190/1.3059205.

© 2017 SEG Page 462

SEG International Exposition and 87th Annual Meeting