Enhancing the resolution of seismic data using improved ...
Transcript of Enhancing the resolution of seismic data using improved ...
Enhancing the resolution of seismic data using improved time-frequency spectral modeling Huailai Zhou*, Changcheng Wang, State Key Laboratory of Oil and Gas Reservoir Geology and Exploration,
Chengdu University of Technology; Kurt J. Marfurt, University of Oklahoma; Yiwei Jiang, and Jianxia Bi,
Zhongyuan Oilfield of SINOPEC
Summary
Maximizing vertical resolution is a key objective in seismic
data processing. In this study, we propose an improved
time-frequency spectral modeling deconvolution method,
which firstly uses a generalized S-transform (GST) to
decompose each seismic trace, then models smoothly
varying spectra of the seismic wavelet using nonstationary
polynomial fitting (NPF) from the seismic trace, and finally
performs deconvolution and the inverse GST to obtain the
seismic data with enhanced resolution. Application to a
synthetic and a large 3D survey from onshore China shows
the proposed method is superior to conventional spectral
modeling method in improving vertical resolution, and
confirms that the proposed method is stable in practice and
robust in the presence of noise.
Introduction
The need to properly “land” horizontal wells in resource
plays has reinvigorated the need to improve the vertical
resolution of 3D surface seismic surveys. Conventional
enhancing resolution methods are based on a stationary
convolutional model (Robinson and Treitel, 1967) that
assumes the propagating wavelet does not evolve with
travel time. In reality, seismic energy spreads geometrically
through a heterogeneous, anelastic subsurface which gives
rise to both attenuation and dispersion effects, deforming
the seismic wavelet with travel time, and giving rise to a
nonstationary signal (Ziolkowski, 1991; Margrave and
Lamoureux, 2001). The problem of wavefront divergence
is commonly addressed by applying a simple geometric
spreading correction (Yilmaz, 1987, 2001). The more
challenging frequency-dependent attenuation is usually
referred to as the "Q-effect". While many authors,
including Futterman (1962), Kjartansson (1979), Hale
(1982), Hargreaves and Calvert (1991), Hargreaves (1992),
Zhang et al. (2007) and Wang (2010), have proposed the
design of inverse Q-filters to deal with the attenuation
component of nonstationarity, however, the estimation of
an accurate Q-model is difficult in practice. Margrave et al.
(2011) presented a time-varying deconvolutional method
that estimates the (attenuating) wavelet spectrum and
reflectivity spectrum directly from the Gabor transform
spectrum of the measured seismic data. Wang et al. (2013)
developed an adaptive spectral-broadening method to
improve the resolution of seismic data, based on the
assumption that the seismic data can be split into
approximately stationary segments.
Seismic interpreters routinely use the spectral response of
seismic reflections to detect subtle changes in stratigraphy.
Such nonstationary changes in the spectral response are
quantitatively estimated using a series of nonstationary
time-frequency analysis techniques including Gabor
transform (Gabor, 1946), the short-time Fourier transform
(STFT) (Partyka et al., 1999), wavelet transform (Rioul and
Vetterli, 1991), empirical mode decomposition (Huang et
al., 1998), and seislet transform (Fomel and Liu, 2010).
One of the more popular techniques is the S-transform
(Stockwell et al., 1996) which avoids many of the
shortcomings of the short-time Fourier transform. While
the original S-transform used a fixed mother wavelet,
McFadden et al. (1999), Pinnegar and Mansinha (2003) and
Gao et al. (2003) chose the decomposition parameters
based on the demands of the seismic data, giving rise to the
generalized S-transform (GST). Liu et al. (2011) proposed
a local attribute analysis method for nonstationary signals
using an iterative inversion framework, which calculates
time-varying Fourier coefficients by solving a least-squares
problem based on regularized nonstationary regression
(Fomel, 2007, 2009, 2013).
We begin this study with a review of the expanded
generalized S-transform used to compute the time-
frequency components of the seismic trace. Next, we use
the improved spectral modeling method using NPF (Fomel,
2009) to estimate time-varying wavelet spectra from
seismic trace and perform the deconvolution for seismic
data. Finally, we apply this workflow to a synthetic and a
3D land survey acquired in China to illustrate the
adaptability and effectiveness in improving vertical
resolution.
Methodology
The generalized S-transform
Time-frequency analysis is powerful to characterize the
nonstationarity of seismic signal using the generalized S-
transform (GST), which is defined as 2
''2
( )
22
''ˆ( , ) ( ) e e
2
rr t f
i ft
GST
fx f x t dt
, (1)
Page 2656SEG Denver 2014 Annual MeetingDOI http://dx.doi.org/10.1190/segam2014-0353.1© 2014 SEG
Dow
nloa
ded
09/0
1/14
to 1
29.1
5.12
7.24
5. R
edis
trib
utio
n su
bjec
t to
SEG
lice
nse
or c
opyr
ight
; see
Ter
ms
of U
se a
t http
://lib
rary
.seg
.org
/
Enhancing the resolution of seismic data using improved time-frequency spectral modeling
where x(t) is the original seismic signal, is the time shift
parameter, ˆ( , )x f is the expression of a two-dimensional
time-frequency variable with regard to f and , '' and r are
adjustable parameters. Examining equation (1), we see that
the Gabor transform (Gabor, 1946) is a special case of the
GST when r=0. Similarly one obtains the S-transform
(Stockwell et al., 1996) by setting ''=1, and r=1.
Deconvolution using the improved spectral modeling
Based on the assumption of statistically white reflectivity,
Rosa and Ulrych (1991) represented to model the amplitude
spectra of the wavelet from seismic trace using a smooth
polynomial, in time-frequency domain (or generalize S-
domain), which can be defined as
0
( , ) exp ( ) , (2)N
k n
a m m n m
n
W f f a f
where Wa(,fm) denotes a smooth amplitude spectrum of a
time-varying wavelet, an() is the nth polynomial coefficient
independent on frequency at time , N is the polynomial
order, and fm is the mth frequency. The conventional method
is to use equation (2) to model the amplitude spectrum of a
wavelet. In our study, we think that the polynomial
coefficients varying with time and frequency represent the
nonstationarity of seismic data more precisely than
conventional fitting method. Hence, we can redefine the
equation (2) as
0
( , ) exp ( , ) , (3)N
k n
a m m n m m
n
W f f a f f
As illustrated in the context, we can estimate the
polynomial coefficients by solving the following least-
squares problem:
2
2
( , )1 1 0 2
min ln ( , ) ln ( , ) ( , ) , (4)n m
M M Nn
m m m n m m n ma f
m m n
A f k f a f f R a f
where 2
2denotes the squared L-2 norm of a function,
A(,fm) denotes the time-frequency amplitude spectra of
seismic trace, an(,fm) expresses the polynomial coefficients
varying with time and frequency, R denotes the
regularization operator, and εm is the misfit error to be
minimized. Here we use Fomel’s (2007) “shaping
regularization” which uses an adjustable radius Gaussian
smoothing operator to control the smoothness of the
coefficients an(, fm). We note that time-frequency
amplitude spectra Wa(,fm) of a wavelet, can be estimated
by iterating the procedure of equation (3) and (4) over all
the frequencies at every time-sample . As discussed above,
once getting the nonstationary polynomial coefficients an(, fm), we can model wavelet spectra between equation (3) and
the amplitude of seismic trace, and then design the
deconvolution operator to complete the processing of
resolution enhancement.
Application
In this section, we demonstrate the effectiveness and
advantage of the improved spectral modeling
deconvolution method proposed in this study on synthetic
data, a 2D seismic line, and a large 3D data volume
acquired from onshore China.
Synthetic data example
In order to verify the effects of the improved spectra
modeling method based on the nonstationary polynomial
fitting (NPF) in resolution enhancement for seismic data,
we use the nonstationary convolution model (Margrave and
Lamoureux, 2001) to produce a synthetic for testing
purpose. Figure 1 shows the comparison results using
respectively stationary polynomial fitting (SPF) and
nonstationary polynomial fitting (NPF) deconvolution
(improved spectra modeling method) to process the
stationary (Figure 1b) and nonstationary synthetic (Figure
1b). Comparing Figure 1d, 1e, 1f and 1g in the red dashed-
line boxes, we find that the adjacent reflection events
centered about 200ms and 250ms cannot be distinguished
in Figure 1d and 1e processed by SPF deconvolution
method, but resolved in Figure 1f and 1g applied by NPF
deconvolution method. In comparison of these results with
the reflectivity (Figure 1a) we again confirm that the
improved spectra modeling deconvolution is superior to
conventional method based on the SPF.
Figure 1: A synthetic seismic trace example. (a) The
reflectivity spike series. (b) The stationary seismic trace
obtained by convolving (a) with a single wavelet. (c) The
nonstationary seismic trace obtained by convolving (a) with
a time-varying wavelet defined by a value of Q=50. Results
of SPF deconvolution method applied to (d) the stationary
and (e) nonstationary synthetic traces. Results of our
improved NPF deconvolution applied to (f) the stationary
and (g) nonstationary synthetic traces. Notice the
enhancement in resolution using our proposed method.
Page 2657SEG Denver 2014 Annual MeetingDOI http://dx.doi.org/10.1190/segam2014-0353.1© 2014 SEG
Dow
nloa
ded
09/0
1/14
to 1
29.1
5.12
7.24
5. R
edis
trib
utio
n su
bjec
t to
SEG
lice
nse
or c
opyr
ight
; see
Ter
ms
of U
se a
t http
://lib
rary
.seg
.org
/
Enhancing the resolution of seismic data using improved time-frequency spectral modeling
Field data examples
The previous synthetic example is noise free. In order to
compare the effectiveness and stability of the SPF (Figure
2b) vs the NPF (Figure 2c) deconvolution algorithm we
apply them to a vertical slice (Figure 2a) from a large 3D
poststack seismic volume acquired from an area in Western
China. Comparing the events within yellow ovals of
Figures 2a and 2b, we see an improvement in vertical
resolution that brings out more detailed reflections. The
thin beds indicated by the red arrow in Figure 2c are
unresolved in Figure 2b. While both SPF and NPF improve
the spectral content of the signal, NPF method which more
accurately accounts for the nonstationarity of the seismic
wavelet is significantly better. In addition, the NPF
deconvolution algorithm avoids increasing the spectral
content of the noise.
Figure 2: A vertical slice through a 3D seismic amplitude
volume (a) before and after deconvolution using the (b)
SPF and (c) NPF algorithms. Location of the vertical slice
is shown in Figure 4. Yellow ovals and red arrows indicate
areas of improved vertical resolution using the NPF
algorithm.
In Figure 3 we illustrate the comparison of the average
amplitude spectra for the three images shown in Figure 2.
As expected from our previous inspection of Figure 2, the
spectrum of the NPF deconvolution data (red dot-dashed
curve) is broader than that of the SPF deconvolution data
(green dot-dashed curve), and broader still than the original
data (black curve).
Seismic coherence is a measure of lateral changes in the
seismic response caused by variation in structure,
stratigraphy, lithology, porosity and the presence of
hydrocarbons (Marfurt et al., 1998). To see if we have
improved the lateral delineation of any thin stratigraphic
features, we calculate coherence for the three seismic
volumes shown in Figure 2. Figure 4 shows representative
time slices at t=2920 ms through the amplitude and
coherence volumes before and after SPF as well as NPF
deconvolution. Figures 4b and 4c show amplitude time
slices applying the conventional method and improved
method for input data (Figure 4a). We find that the
improved method (NPF deconvolution) provides higher
temporal and spatial resolution. Time slices from the
coherence attribute volumes shown respectively in Figures
4d, 4e and 4f corresponding to Figures 4a, 4b and 4c
manifest that the high resolution volume generated from the
improved NPF deconvolution serves to show the increased
details and more subtle structure characters.
Figure 3: Comparison of average amplitude spectra of three
images shown in Figure 2 for the original data (black dot-
dashed curve), data after SPF deconvolution (green dot-
dashed curve) and data after NPF deconvolution (red dot-
dashed curve). Note that the dominant frequency is higher
and bandwidth is broader for the seismic section processed
by NPF algorithm compared to the original and SPF
deconvolution data.
Page 2658SEG Denver 2014 Annual MeetingDOI http://dx.doi.org/10.1190/segam2014-0353.1© 2014 SEG
Dow
nloa
ded
09/0
1/14
to 1
29.1
5.12
7.24
5. R
edis
trib
utio
n su
bjec
t to
SEG
lice
nse
or c
opyr
ight
; see
Ter
ms
of U
se a
t http
://lib
rary
.seg
.org
/
Enhancing the resolution of seismic data using improved time-frequency spectral modeling
Figure 4: Time slices at t=2920ms through the (a) original
data volume and data volumes after (b) application of SPF
deconvolution, and (c) NPF deconvolution. (d)-(f) Time
slices through the corresponding coherence volumes. Black
line indicates location of the vertical slice shown in Figure
2. Notice the enhancement in reflection detail on the time
slice of amplitude and coherence attribute data after
applying the improved method.
Figure 5: Stratal slices through the (a) original data volume,
and after (b) application of SPF deconvolution and, (c) NPF
deconvolution. (d)-(e) Stratal slices through the
corresponding coherence volumes. Note that the stratal
slices from the amplitude and coherence volumes show
much more stratigraphic features (green arrows and red
ovals) and subtle faults (red arrows) with improved method
than the other two.
In Figure 5, we show stratal slices of amplitude volumes
from the original seismic data (Figure 5a) as well as
resolution-enhanced data generated using the SPF
algorithm (Figure 5b) and NPF deconvolution method
(Figure 5c). The stratal slices through the corresponding
coherence volumes are shown in Figures 5d-f. In
comparison areas (indicated by red ovals and green arrows),
we observe that the two methods can greatly improve the
seismic resolution and efficiently identify the faults. Note
that the result using the improved NPF deconvolution is
superior in delineating stratigraphic features and subtle
faults (show in red arrow) than SPF deconvolution method.
This example manifests that the enhanced resolution data
using the improved method is helpful to extract the better
seismic attribute and facilitate interpreters to accurately
depict the structural and stratigraphic features. Notice that
the amplitude and coherence slices show improved
resolution in the order stated above, with the highest lateral
resolution observed from slices computed on improved
spectral modeling algorithm.
Conclusions
Time-frequency analysis using the generalize S-transform
provides a basis for more precise and adaptive
deconvolution (or spectral modeling) algorithms. Previous
algorithms based on stationary polynomial fitting of the
spectrum assumed fitting coefficients to be invariable over
all the frequencies. By using nonstationary polynomial
fitting concepts described by Fomel (2009) we develop a
deconvolution algorithm that adjusts adaptively according
to nonstationary seismic wavelets. Unlike earlier
algorithms, our nonstationary polynomial fitting (NPF)
deconvolution algorithm circumvents the need to explicitly
choose spectral modeling parameters k and N and avoids an
explicit parameterization of Q. Application to synthetic and
field data show that the improved method not only
enhances the seismic resolution without boosting the noise
more effectively and also performs better compensation for
attenuation than conventional spectra modeling.
Furthermore, application to a large 3D survey from onshore
China shows an improvement in vertical resolution that
significantly facilitates stratigraphic interpretation, and also
confirms that the proposed method in practice is stable and
robust in the presence of noise.
Acknowledgement
This work is financially supported by the National Natural
Science Foundation of China (Grant No. 41204091), and
Science and Technology Support Program (Grant No.
2011GZ0244) from Science and Technology Department of
Sichuan Province of China. The first author appreciates the
Chinese Scholarship Council for financial support. We also
thank Puguang Branch of Zhongyuan Oilfield of SINOPEC
for permission to present these results and want to honestly
thank Ph.D. students Shiguang Guo, Fangyu Li and Tao
Zhao in the University of Oklahoma for their inspiring
discussions.
Page 2659SEG Denver 2014 Annual MeetingDOI http://dx.doi.org/10.1190/segam2014-0353.1© 2014 SEG
Dow
nloa
ded
09/0
1/14
to 1
29.1
5.12
7.24
5. R
edis
trib
utio
n su
bjec
t to
SEG
lice
nse
or c
opyr
ight
; see
Ter
ms
of U
se a
t http
://lib
rary
.seg
.org
/
http://dx.doi.org/10.1190/segam2014-0353.1 EDITED REFERENCES Note: This reference list is a copy-edited version of the reference list submitted by the author. Reference lists for the 2014 SEG Technical Program Expanded Abstracts have been copy edited 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. REFERENCES
Fomel, S., 2007, Shaping regularization in geophysical-estimation problems : Geophysics, 72, no. 2, R29–R36, http://dx.doi.org/10.1190/1.2433716.
Fomel, S., 2009, Adaptive multiple subtraction using regularized nonstationary regression: Geophysics, 74, no. 1, V25–V33, http://dx.doi.org/10.1190/1.3043447.
Fomel, S., 2013, Seismic data decomposition into spectral components using regularized nonstationary autoregression: Geophysics, 78, no. 6, O69–O76, http://dx.doi.org/10.1190/geo2013-0221.1.
Fomel, S., and Y. Liu, 2010, Seislet transform and seislet frame: Geophysics, 75, no. 3, V25–V38, http://dx.doi.org/10.1190/1.3380591.
Futterman, W. I., 1962, Dispersive body waves: Journal of Geophysical Research, 67, no. 13, 5279–5291, http://dx.doi.org/10.1029/JZ067i013p05279.
Gabor, D., 1946, Theory of communication: Journal of the IEEE, 93, no. 26, 429–441, http://dx.doi.org/10.1049/ji-3-2.1946.0074.
Gao, J. H., W. C. Chen, Y. M. Li, and F. Tian, 2003, Generalized S transform and seismic response analysis of thin interbeds: Chinese Journal of Geophysics, 46, no. 3, 526–532 (in Chinese).
Hale, D., 1982, Q-adaptive deconvolution: Stanford Exploration Project, 30, 133–158.
Hargreaves, N. D., 1992, Similarity and the inverse Q filter: Some simple algorithms for inverse Q filtering: Geophysics, 57, 944–947, http://dx.doi.org/10.1190/1.1443307.
Huang, N. E., Z. Shen, S. R. Long, M. C. Wu, H. H. Shih, Q. Zheng, N. C. Yen, C. C. Tung, and H. H. Liu, 1998, The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis: Proceedings of the Royal Society of London, Series A: Mathematical, Physical and Engineering Sciences, 454, no. 1971, 903–995, doi: 10.1098/rspa.1998.0193.
Kjartansson, E., 1979, Constant Q-wave propagation and attenuation: Journal of Geophysical Research, 84, no. B9, 4737–4748, http://dx.doi.org/10.1029/JB084iB09p04737.
Liu, G. C., S. Fomel, and X. H. Chen, 2011, Time-frequency analysis of seismic data using local attributes: Geophysics, 76, no. 6, P23–P34, http://dx.doi.org/10.1190/geo2010-0185.1.
Marfurt, K. J., R. L. Kirlin, S. L. Farmer, and M. S. Bahorich, 1998, 3-D seismic attributes using a semblance-based coherency algorithm: Geophysics, 63, 1150–1165, http://dx.doi.org/10.1190/1.1444415.
Margrave, G. F., and M. P. Lamoureux, 2001, Gabor deconvolution: Consortium for Research in Elastic Wave Exploration Seismology (CREWES) Research Report 13, 241–276.
Margrave, G. F., M. P. Lamoureux, and D. C. Henley, 2011, Gabor deconvolution: Estimating reflectivity by nonstationary deconvolution of seismic data: Geophysics, 76, no. 3, W15–W30, http://dx.doi.org/10.1190/1.3560167.
Page 2660SEG Denver 2014 Annual MeetingDOI http://dx.doi.org/10.1190/segam2014-0353.1© 2014 SEG
Dow
nloa
ded
09/0
1/14
to 1
29.1
5.12
7.24
5. R
edis
trib
utio
n su
bjec
t to
SEG
lice
nse
or c
opyr
ight
; see
Ter
ms
of U
se a
t http
://lib
rary
.seg
.org
/
McFadden, P. D., J. G. Cook, and L. M. Forster, 1999, Decomposition of gear vibration signals by generalized S transform: Mechanical Systems and Signal Processing, 13, no. 5, 691–707, http://dx.doi.org/10.1006/mssp.1999.1233.
Partyka, G., J. Gridley, and J. Lopez, 1999, Interpretational applications of spectral decomposition in reservoir characterization: The Leading Edge, 18, 353–360, http://dx.doi.org/10.1190/1.1438295.
Pinnegar, C. R., and L. Mansinha, 2003a, Time local spectral analysis for non-stationary time series: The S-transform for noisy signals : Fluctuation and Noise Letters, 3, no. 3, L357–L364, http://dx.doi.org/10.1142/S0219477503001439.
Pinnegar, C. R., and L. Mansinha, 2003b, The S-transform with windows of arbitrary and varying shape : Geophysics, 68, 381–385, http://dx.doi.org/10.1190/1.1543223.
Rioul, O., and M. Vetterli, 1991, Wavelets and signal processing: IEEE Signal Processing, 8, no. 4, 14–38, http://dx.doi.org/10.1109/79.91217.
Robinson, E. A., and S. Treitel, 1967, Principles of digital Wiener filtering: Geophysical Prospecting, 15, no. 3, 311–332, doi: 10.1111/j.1365-2478.1967.tb01793.x.
Rosa, A. L. R., and T. J. Ulrych, 1991, Processing via spectral modeling: Geophysics, 56, 1244–1251, http://dx.doi.org/10.1190/1.1443144.
Stockwell, R. G., L. Mansinha , and R. P. Lowe, 1996, Localization of the complex spectrum: The S transform: IEEE Transactions on Signal Processing, 44, no. 4, 998–1001, http://dx.doi.org/10.1109/78.492555.
Wang, L. L., J. H. Gao, W. Zhao, and X. D. Jiang, 2013, Enhancing resolution of nonstationary seismic data by molecular-Gabor transform: Geophysics, 78, no. 1, V31–V41.
Wang, Y. H., 2010, Multichannel matching pursuit for seismic trace decomposition: Geophysics, 75, no. 4, V61–V66, http://dx.doi.org/10.1190/1.3462015.
Yilmaz, Ö., 1987, Seismic data processing: SEG.
Yilmaz, Ö., 2001, Seismic data processing, 2nd ed.: SEG.
Zhang, X. W., L. G. Han, F. J. Zhang, and G. Y. Shan, 2007, An inverse Q-filter algorithm based on stable wavefield continuation: Applied Geophysics, 4, no. 4, 263–270, http://dx.doi.org/10.1007/s11770-007-0040-9.
Ziolkowski, A., 1991, Why don't we measure seismic signatures?: Geophysics, 56, 190–201, http://dx.doi.org/10.1190/1.1443031.
Page 2661SEG Denver 2014 Annual MeetingDOI http://dx.doi.org/10.1190/segam2014-0353.1© 2014 SEG
Dow
nloa
ded
09/0
1/14
to 1
29.1
5.12
7.24
5. R
edis
trib
utio
n su
bjec
t to
SEG
lice
nse
or c
opyr
ight
; see
Ter
ms
of U
se a
t http
://lib
rary
.seg
.org
/