See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/319167009

Seismic data gather conditioning for prestack seismic data inversion

Conference Paper · August 2017

DOI: 10.1190/segam2017-16928321.1

CITATIONS READS
2 619

7 authors, including:

Gang Yu Xing Liang

BGP Inc., China National Petroleum Corporation Shanghai University of Finance and Economics
91 PUBLICATIONS 390 CITATIONS 144 PUBLICATIONS 1,897 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Gang Yu on 05 May 2021.

The user has requested enhancement of the downloaded file.

 Seismic data gather conditioning for prestack seismic data inversion
Gang Yu*1, Yusheng Zhang1, Ximing Wang1, Xing Liang2, Benjing Liu1, Scott Singleton3 and Maggie Smith3,
Downloaded 04/17/20 to 210.12.194.136. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

1
BGP Inc., CNPC, P. R. China, 2Zhengjiang Oilfield, CNPC, P. R. China, 3RSI, Houston, USA.

Summary applies a spatial filter along local dip in either offset

direction or on offset cubes.
Most seismic data is processed to optimize image quality
for structural and stratigraphic interpretation, with little NMO Stretch Removal – the frequency characteristics of
regard to preserving characteristics essential for successful pilot trace formed from near offsets are analyzed via
seismic reservoir characterization. No matter how Gabor-Morley joint time frequency decomposition. The
sophisticated the inversion algorithm, use of inadequately spectral energy of each trace in the gather equalized to that
processed seismic data will severely impact the quality of of pilot trace, removing NMO stretch from far offsets
the final interpretation.
Gather Flattening – a differential timing correction is
The objective of this paper is to quantify the advantages of estimated in overlapping analysis windows. The absolute
performing pre-stack data conditioning prior to reservoir maximum value of cross correlation between adjacent pairs
characterization. Three specific seismic properties that will of traces is picked. The total correction for an offset is sum
be addressed are: (1) signal-to-noise ratio (SNR), (2) off of differential time shifts from first trace to that offset.
set-dependent frequency loss, and (3) gather alignment. Optional smoothing of time shifts may be applied before
application for greater stability of the process.
Seismic gather conditioning improved seismic data quality
prior to pre-stack inversion by improving signal/noise ratio, We believe that optimizing the quality of the data input to
removing NMO stretch and aligning reflection events. the interpretation process ensures a robust outcome. We use
Velocities from residual moveout (RMO) analysis on a comprehensive toolkit to provide a complete suite of
individual sectors were used as input to detection of seismic data conditioning steps to optimize the quality of
fracture orientation and anisotropy. In all of our integrated pre- and post-stack seismic data prior to use in impedance,
studies, our goal is to offer the greatest understanding of AVO and seismic facies analysis applications. We also use
the depositional environment, so that decisions can be our Geophysical Well Log Analysis to condition well logs
made with confidence. used for seismic modeling along with well tie analysis. Our
toolkit may be applied to pre- or post-stack seismic data
Introduction and involves several interlinked steps.

Reservoir characterization puts hefty demands on a seismic Gather-conditioning methods

data set. Asset teams are asking for more accurate seismic
estimates of hydrocarbon fluid fill and rock properties in We use the gather conditional workflow to condition
reservoirs. To be able to meet these increasingly stringent seismic data (specifically gathers) prior to seismic trace
demands, seismic data used for reservoir characterization inversion. Most seismic data are processed to optimize
need to be conditioned to remove as many undesirable image quality for structural interpretation, with little regard
effects as possible. Three wave-transmission effects that are to preserving characteristics essential for successful seismic
commonly removed or reduced in pre-inversion gather reservoir characterization. No matter how sophisticated the
conditioning are random noise, NMO wavelet stretch, and inversion algorithm, use of inadequately processed seismic
non-flat reflections. Most seismic datasets need additional data will adversely impact the quality of the final
processing. Decisions made on spec surveys are rarely ideal interpretation.
for a specific target interval, and amplitudes always need to
be calibrated to new wells. For quantitative interpretation, As a consequence, seismic data can rarely be used “as is”
gather conditioning is really a requirement (Singleton, for seismic reservoir characterization without substantial
2009). pre-conditioning. We use gather conditioning workflow
toolkit to provide a complete suite of seismic data
The gather conditioning procedures we used in our studies conditioning steps to optimize the quality of pre and post-
include proprietary techniques developed by RSI in the stack seismic data prior to use in impedance, AVO and
course of industry sponsored consortia and commercialized seismic facies inversion applications.
in Sharp Reflection’s PreStackPro software. Techniques
used in this project include the following: The data conditioning workflow is shown in Figure 1 and
includes signal to noise enhancement, passes of parabolic
Signal to Noise Enhancement – an edge and coherency and linear radon, removal of NMO stretch, residual
enhancing diffusion filter removes random noise and

© 2017 SEG Page 5074

SEG International Exposition and 87th Annual Meeting
 Seismic data gather conditioning

moveout analysis and a final pass of the gather alignment scaling) sum to the original energy envelope amplitude.
routine. This ensures that AVO character is not altered in this
Downloaded 04/17/20 to 210.12.194.136. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

process (Singleton, 2009).

INPUT

EPS3D – Edge Preserving Smoothing of Gathers

o
31 Angle Mute
NMO Stretch
Removal
The noise is part of the seismic signal not utilized by the
application being considered, which in this case is prestack

Updated Vrms
S/N Enhancement
Residual Moveout Elliptical impedance inversion. In order to reduce this undesired
Analysis Velocity
Inversion
component of the seismic signal, we assume that any
Parabolic Radon
“noise” remaining after conventional processing is random.
Gather Flattening
Filter If this were not the case (for instance, if dipping coherent
Linear Radon
energy were present), dip or spatial filters would need to be
Filter OUTPUT used prior to SNR enhancement.

S/N Enhancement To attack the random, uncorrelatable energy component,

we use a process developed by Tury Taner called “3D edge
Figure 1: Gather conditioning workflow. Gathers are output to preserving smoothing”, which is a methodology to enhance
angle stacks for simultaneous inversion and updated velocity fields S/N by means of noise reduction. This process works on
to elliptical velocity inversion. 3D gathers by first separating the data into common-off set
volumes so as not to interfere with the AVO signature of
Gather conditioning comprises the following programs: the gather. Then dip scanning is performed in a running
window on all traces surrounding the target trace to
SBALAN – Spectral Balancing of Gathers determine the dip with the greatest semblance. Once local
dip is determined, correlation coefficients of all
The object is to compensate for the loss of frequency with surrounding traces are calculated along this local dip. Edge
offset caused by NMO stretch. One method of removing detection is performed by eliminating correlations below a
stretch is by the use of a cos θ operator, since stretch user-specified amount, thus preserving discontinuities such
depends solely on the cosine of the reflection angle. Such as faults. The remaining trace samples are then summed in
methods seek to accurately calculate this angle, perhaps a Gaussian filter after they have been normalized and
with migration operators. weighted with their respective correlation coefficients.
Other researchers rely on a comparison between the near
and far-trace spectra to compensate for the loss of Before After
frequency. The effects of seismic data conditioning on
prestack simultaneous impedance inversion with offset
(Lazaratos and Finn, 2004; Xu and Chopra, 2007). Linear
spectral operators are designed based on the amount of
stretch at any time and off set as compared to the near off
set (Singleton, 2009).

This method uses the latter principle. Gabor-Morlet joint

time-frequency analysis (JTFA) is used to separate the
frequency spectra of each gather trace into a user-specified
number of sub-bands (Singleton et al., 2006). The sub-
bands are calculated using a running Gaussian-shaped
window, which gives a slowly varying amplitude profile of
each sub-band. Then, each sub-band spectrum is balanced Figure 2: Post-stack gather conditioning workflow results before
against the corresponding sub-band of a user-specified pilot (left) and after (right).
trace within the gather (Singleton, 2009).
This acts to preserve AVO effects because smoothing does
The primary advantage of this approach is two-fold: (1) the not take place between adjacent offsets within the gather. It
bandwidth of the gather at each time sample is determined takes into account spatial dip of coherent reflections so that
by the pilot trace, and (2) the total energy contained in each frequency and continuity are preserved. The “edge-
reflector is held constant by computing its energy envelope preserving” characteristics result from user-defined
and requiring that the energy of all sub-bands (after semblance cutoffs so that summation and filtering do not

© 2017 SEG Page 5075

SEG International Exposition and 87th Annual Meeting
 Seismic data gather conditioning

occur across traces with different reflector characteristics squares solutions are less stable but are required where
(i.e. faults, angular unconformities, channel boundaries, AVO class II anomalies are present or suspected
Downloaded 04/17/20 to 210.12.194.136. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

etc., Figure 2). (Singleton, 2009).

ALIGN – Pre-stack Data Alignment XTEND – Spectral Extension (Post-Stack)

A fundamental assumption made in AVO inversion is that Data bandwidth is a well-known constraint on the resolving
primary reflection events are horizontally aligned (flat) power of seismic data, with implications for both structural
across each CDP gather. Unfortunately, this usually only and stratigraphic interpretation. The objective of this
happens with synthetic gathers. The presence of residual process is to enhance seismic resolution by extending data
move-out, non-hyperbolic move-out, random noise, bandwidth beyond the limits of current seismic processing
multiples, tuning effects etc., will violate this assumption techniques such as spectral whitening or deconvolution.
and introduce noise (uncertainty) into the AVO results. The
two basic approaches to ensure recorded seismic data The process extends seismic bandwidth by first performing
conform to these assumptions might be termed “velocity- a form of sparse inversion, and then creating a bandwidth
based” and “statics-based”. Velocity-based methods extended version of the input trace from the sparse
assume that nonflat reflectors are caused by residual NMO inversion result. Sparse inversion is implemented by means
(RMO), and thus can be corrected by high-resolution of an orthogonal matching pursuits algorithm, which is
estimation of the second and fourth-order RMS velocity comparable to constrained least-squares modeling of the
field. Statics-based methods assume local velocity input data, with constraints limiting the number of non-zero
perturbations in the seismic raypath cause random values of the derived reflectivity (Figure 3).
undulations in gather reflectors. These cannot be removed
using an overall velocity field, so they are treated as static
errors (Hinkley, 2004; Gulunay et al., 2007). The ALIGN
module within gather conditioning workflow corrects for
various residual move-out effects, resulting in a much-
improved AVO signature.

ALIGN flattens gathers based on a conditional

minimization of a reflector’s least-squares fit error by
determining a local statics shift on each gather trace. The
“condition” that is minimized can be either semblance or
AVO fit. Semblance is the most robust but it is only
applicable for AVO Class I and III anomalies. AVO Class
II phase rotations (especially those that are not full phase Figure 3: An example of seismic data conditioning results before
(left) and after (right) using the gather conditioning technique.
reversals) present a special problem that require great care
in addressing. For this case, we minimize the fit error to Our Processing Toolkit includes Radon demultiple, gather
either a 2-term Shuey or 3-term Aki & Richards equation. flattening, spectral balancing, and dip-steered random noise
These least-squares solutions are less stable but are removal (Figure 4).
required where AVO Class II anomalies are present or
suspected. The gather conditioning workflow was used to condition
the 3D seismic data from the Zhao-104 shale gas
Our algorithm takes the “statics-based” approach. It exploration and development field in Sichuan Basin of
minimizes a least-squares (L2 norm) error in a reflector by China. Following application of the three algorithms
determining a time-variant statics shift on each gather trace. described previously, reflectors appear continuous and flat
Shifts are calculated in a running window, typically 50–200 from near to far offsets and stretch has been removed.
ms long, and are vertically smoothed to prevent jumps. The Overall, it is readily apparent that the conditioned gather is
“condition” that is minimized can be one of two choices— in much better shape to perform an AVO or impedance
semblance or AVO fi t. Semblance is the most robust and is inversion.
the one described in the articles referenced above.
However, it is only applicable for AVO class I and class III Gather-conditioning results
anomalies. AVO class II phase rotations (especially those
that are not full phase reversals) present a special problem Before and after the gather conditioning results along with
that requires great care in addressing. For this case, we their difference are shown in Figure 5. The conditioned
minimize the L2 norm fit error in either a two-term (Shuey) gathers are significantly cleaner than the original and the
or three-term (Aki and Richards) equation. These least-

© 2017 SEG Page 5076

SEG International Exposition and 87th Annual Meeting
 Seismic data gather conditioning

difference plot shows that most of what has been removed non-flat reflections. The gather conditioning workflow and
is linear and random noise. processing results demonstrated that data that might be
Downloaded 04/17/20 to 210.12.194.136. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

considered acceptable for normal structural interpretation

could still be prone to large errors when subjected to
prestack impedance inversion without first being
conditioned (Singleton, 2009).

Figure 4: Data conditioning processing Toolkit and results.

In a final step 4 angle stacks were formed in ranges 3-12

degrees, 9-18 degrees, 15-24 degrees, 21-31 degrees.
Overlapping angle ranges improved the signal to noise ratio Figure 6: Angle ranges for angle stacks. From left – full 0-32o, 3-
of the angle stacks which have approximately equal fold at 12o, 9-18o, 15-24o, 21-31o. ZOI ~1.54s has approximately equal
interval of interest (Figure 6). fold in each.

c
Input After AVATAR

d Figure 7: Angle Gather Sector 3 (90-143 degrees) stack before,

left, after gather conditioning process, right. Signal to noise ratio
Figure 5: Gathers before, left, after conditioning, center and and event continuity have been considerably improved.
difference. Signal to noise ratio and event continuity have been
considerably improved. Panels a-d are azimuthal sectors 0-37o, 37- The conditioned gathers are significantly cleaner than the
90o, 90-143o and 143-180o respectively. Horizon is TS. original and the difference plot shows that most of what has
been removed is linear and random noise. Signal to noise
Figure 7 is the Angle Gather Sector 3 3 (90-143 degrees) ratio and event continuity have been considerably
stack before and after gather conditioning process. Signal improved. Overlapping angle ranges improved the signal to
to noise ratio and event continuity have been improved noise ratio of the angle stacks which have approximately
significantly. equal fold at interval of interest. This processing step
significantly improved the seismic data gather quality and
Conclusions
provided much more reliable seismic data for pre-stack as
Reservoir characterization puts hefty demands on a seismic well as post-stack inversion process.
data set. Asset teams are asking for more accurate seismic
Acknowledgements
estimates of hydrocarbon fluid fill and rock properties in
reservoirs. To be able to meet these increasingly stringent We would like to thank Zhejiang Oilfiled and CNPC for
demands, seismic data used for reservoir characterization allowing the publication of this article. We would also like
need to be conditioned to remove as many undesirable to thank BGP management for their support and guidance
effects as possible. Three wave-transmission effects that are during the data acquisition, processing and interpretation
commonly removed or reduced in pre-inversion gather phases of this project.
conditioning are random noise, NMO wavelet stretch, and

© 2017 SEG Page 5077

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 04/17/20 to 210.12.194.136. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

REFERENCES
Gulunay, N., M. Magesan, and H. Roende, 2007, Gather flattening: The Leading Edge, 26, 1538–1543,
https://doi.org/10.1190/1.2821939.
Hinkley, D., G. Bear, and C. Dawson, 2004, Prestack gather flattening for AVO: 74th Annual
International Meeting, SEG, Expanded Abstracts, 271–273, https://doi.org/10.1190/1.1839719.
Lazaratos, S. and C. Finn, 2004, Deterministic spectral balancing for high-fidelity AVO: 74th Annual
International Meeting, SEG, Expanded Abstracts, 219–223, https://doi.org/10.1190/1.1851212.
Singleton, S., M. T. Taner, and S. Treitel, 2006, Q estimation using Gabor-Morlet joint time-frequency
analysis techniques: 76th Annual International Meeting, SEG, Expanded Abstracts, 1610–1614,
https://doi.org/10.1190/1.2369829.
Singleton, S., 2009, The effects of seismic data conditioning on prestack simultaneous impedance
inversion: The Leading Edge, 28, 772–781, https://doi.org/10.1190/1.3167776.
Xu, Y., and S. Chopra, 2007, Improving AVO fidelity by NMO Stretching and offset-dependent tuning
corrections: The Leading Edge, 26, 1548–1551, https://doi.org/10.1190/1.2821941.

© 2017 SEG Page 5078

SEG International Exposition and 87th Annual Meeting
View publication stats