Spectral decomposition with extracted seismic wavelet

Rongfeng Zhang*, Geomodeling Technology Corp., Calgary, Canada

Summary

In seismic interpretation, spectral decomposition has
rapidly evolved from a niche technique to a common tool at
work due to its distinguish advantages in channel
delineation, gas reservoir detection and thin-bed
interpretation. There are different approaches to do spectral
decomposition but they are basically using one or more
mathematical functions (sin/cosine, Morlet wavelet etc.) as
the basis to convolve with seismic sampling data. Very
often, the mathematical functions used deviate substantially
from the actual seismic response underground. This causes
the popular ringing effect in spectral decomposition results.
It could interfere with or, even worse, lead to false
interpretation. The author proposes a method that uses the
real seismic wavelet to replace the mathematic functions to
do spectral decomposition. The result from this method
shows much less ringing effect and therefore less ambiguity
for interpretation.

Introduction

Spectral decomposition has been extensively applied to
seismic interpretation and reservoir characterization. Since
it was formally introduced (Partyka 1999), there have
emerged several approaches, from the popular short time
Fourier transform (STFT) and continuous wavelet
transform (CWT) to less frequently used methods such as
matching pursuit, S-transform, chirprit transform and
wavelet packet transform. Each approach has its advantage
and disadvantage, but all approaches are common in that
they can be simplified as some kind of operation between
the seismic data and serial kernel functions with closed
form expressions. In STFT, sine/cosine and some window
functions are used; in CWT, a mathematic wavelet; in S-
Transform, the Gaussian function. The advantages of these
methods are their speed and capability to invert back to the
original time domain from the spectral decompositions. The
disadvantage of these approaches is that they may not
always suit the seismic data. A method is proposed that is
similar to CWT, however, instead of using a wavelet
derived from a mathematic expression, an actual wavelet is
extracted from the seismic data. Since the real wavelet does
not have a mathematic expression, we may not be able to
transform back from the spectral decomposition. However,
this is regarded as a limited deficiency, since inversion to
the original data may not be required in many cases. The
direct benefit, comparing to CWT, is that there are much
less ringing or wavelet side-lobe effects, while preserving
the high resolution characteristics of CWT. The proposed
method uses an algorithm similar to CWT, wherein the
seismic data is convolved with groups of dilated, squeezed,
and stretched seismic wavelets. If the closed form
expression of a wavelet is known, squeezing and stretching
can be easily done, but doing this with the discrete seismic
wavelet is challenging and special care is required. This
suggested approach is demonstrated in this study using a
3D seismic data set with promising results.

Theory and Method

In seismic, spectral decomposition is done by convolving a
seismic trace with a kernel function . It can be
expressed as
, (1)
where t represents time, represents frequency and
represents time shift. It gives localized spectra of the input
signal.
In STFT (Allen 1982),

, (2)
where W is the window function, can be boxcar, Hanning,
Gaussian etc. In S-transform (Stockwell 1996),

||

. (3)
In CWT (Mallat 2008),

, (4)
where is the mathematic wavelet that can have different
forms, among which Morlet wavelet (

is
the most popular, as is the frequently used Mexican hat
wavelet. Note that in equation (4) , which is called scale
or scale factor, is used instead of . They are linked by the
central frequency in practice. In CWT, is called the
mother wavelet. It can be squeezed and stretched simply by
changing the scale . This can be done in either the time
( or the frequency ( domains because of the
relationship
(

) ||. (5)
By the way, in equation (4) has to meet the so-called
admissibility condition (Mallat 2008),

||

||

, (6)
to guarantee the convergence and invertibility. Spectral
decomposition with CWT has some advantages comparing
to STFT, and is a preferred option during seismic
interpretation. However, the ringing or wavelet side-lobe
artifacts observed with this approach occur mainly because
the wavelets used in CWT do not suit seismic data in many
DOI http://dx.doi.org/10.1190/segam2013-0439.1 2013 SEG
SEG Houston 2013 Annual Meeting Page 2372
D
o
w
n
l
o
a
d
e
d

0
7
/
2
1
/
1
4

t
o

9
3
.
1
8
0
.
1
8
1
.
4
8
.

R
e
d
i
s
t
r
i
b
u
t
i
o
n

s
u
b
j
e
c
t

t
o

S
E
G

l
i
c
e
n
s
e

o
r

c
o
p
y
r
i
g
h
t
;

s
e
e

T
e
r
m
s

o
f

U
s
e

a
t

h
t
t
p
:
/
/
l
i
b
r
a
r
y
.
s
e
g
.
o
r
g
/
Spectral decomposition with seismic wavelet

cases. This study uses the real seismic wavelet instead of
the closed-form mathematic wavelets. The real seismic
wavelet is actually extracted from the seismic data and it is
expected to vary with different seismic data. The extracted
seismic wavelet is squeezed, stretched and then convolved
with the seismic trace. The amount of squeeze and stretch
is determined by equation (5). For example, a seismic
wavelet with the dominant frequency of 32Hz, needs to
be equal to 2 to compute spectral decomposition at 64Hz. If
is not an integer number, the squeeze and stretch cannot
be easily done. As an alternative approach, I choose to use
an interpolate-resampling technique. The wavelet is first
densely interpolated and then resampled according to the
amount of squeeze and stretch desired.

(a) The extracted wavelet


(b) Morlet wavelet
Figure 1: The extracted wavelet from Blackfoot P-wave
seismic data (a) and Morlet wavelet (b).

Figure 1 (a) shows an extracted wavelet of Blackfoot
(Blackfoot field near Strathmore, Alberta, Canada) P-wave
seismic data in both the time (red) and frequency domains
(blue). Figure 1 (b) shows a Morlet wavelet in both the time
and frequency domains. Morlet wavelet is a complex
wavelet, so both the real parts (red) and the imaginary parts
(green) are shown. Figure 2 shows the extracted wavelet
(blue) and its many squeezed and stretched versions in the
time domain. Excessive squeezing will undoubtedly cause
distortion, compared to stretching which is relatively much
safer. Once the extracted wavelet has been squeezed and
stretched, based on the desired frequencies used in spectral
decomposition, they are subsequently convolved with the
seismic trace to produce the spectra. It can be done either in
the time or frequency domains.


Figure 2: The extracted wavelet (blue) and its squeezed and
stretched versions.

In practice, the illustrated method seems to produce fewer
problems. However, in theory the mathematic functions
used in CWT have to meet admissibility condition given by
equation (6). Otherwise, the method will still be possible to
run into problems with divergence. Though difficult to
prove that all the extracted seismic wavelet will meet this
condition, it is worth trying numerically. Figure 3 shows
||

|| for the extracted wavelet depicted in Figure

1(a). Here, the area between the displayed curve (blue) and
the bottom horizontal axis is the left or integral term in
equation (6). Since the area or the integral is finite,
conservatively this extracted wavelet does meet
admissibility condition.


DOI http://dx.doi.org/10.1190/segam2013-0439.1 2013 SEG
SEG Houston 2013 Annual Meeting Page 2373
D
o
w
n
l
o
a
d
e
d

0
7
/
2
1
/
1
4

t
o

9
3
.
1
8
0
.
1
8
1
.
4
8
.

R
e
d
i
s
t
r
i
b
u
t
i
o
n

s
u
b
j
e
c
t

t
o

S
E
G

l
i
c
e
n
s
e

o
r

c
o
p
y
r
i
g
h
t
;

s
e
e

T
e
r
m
s

o
f

U
s
e

a
t

h
t
t
p
:
/
/
l
i
b
r
a
r
y
.
s
e
g
.
o
r
g
/
Spectral decomposition with seismic wavelet


Figure 3: ||

|| of the extracted wavelet

Examples

To illustrate the new approach, the spectral decomposition
at 72Hz using three methods on a cross line in Blackfoot P-
wave data are displayed in Figure 4. Figure (a), (b), and (c)
display results using CWT, the new approach, and STFT
respectively. Compared to STFT, both the new approach
and CWT clearly exhibit higher time resolution at high
frequency and the new approach displays a clear advantage
over the CWT method in that less ringing effect is
observed, especially in the three high energy event groups
(corresponding to warm colours). Similar improvements
can be observed at other frequencies.

Conclusions and discussions

A new spectral decomposition method using the extracted
seismic wavelet, rather than the mathematic wavelets, is
proposed. The extracted wavelet is numerically squeezed,
stretched, and then convolved with seismic data samples.
The result of this study display less ringing or side-lobe
effects compared to results from CWT; but with the same
high resolution characteristics as CWT.

Since the squeeze and stretch of the wavelet in the new
approach has to be done numerically, the computation cost
would increase. However, the additional calculations need
only be done once for the entire data set, as the extracted
wavelet will be used throughout the entire data set.
Therefore the increased computation cost can be
minimized. Where multiple seismic wavelets are used, the
increased computation cost would still be very small.


(a) CWT result at 72Hz


(b) Result via the new approach at 72Hz


(c) STFT result at 72Hz
Figure 4: Comparison of the three spectral decomposition
methods on a cross line in Blackfoot data set.




Acknowledgments

The author thanks Geomodeling Technology Corp. for
support and permission to present this work. The author
also thanks the sponsors of the Blackfoot data set. The
author also thanks Les Dabek for reviewing the abstract.
DOI http://dx.doi.org/10.1190/segam2013-0439.1 2013 SEG
SEG Houston 2013 Annual Meeting Page 2374
D
o
w
n
l
o
a
d
e
d

0
7
/
2
1
/
1
4

t
o

9
3
.
1
8
0
.
1
8
1
.
4
8
.

R
e
d
i
s
t
r
i
b
u
t
i
o
n

s
u
b
j
e
c
t

t
o

S
E
G

l
i
c
e
n
s
e

o
r

c
o
p
y
r
i
g
h
t
;

s
e
e

T
e
r
m
s

o
f

U
s
e

a
t

h
t
t
p
:
/
/
l
i
b
r
a
r
y
.
s
e
g
.
o
r
g
/
http://dx.doi.org/10.1190/segam2013-0439.1

EDITED REFERENCES
Note: This reference list is a copy-edited version of the reference list submitted by the author. Reference lists for the 2013
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
Allen, J. B., 1982, Application of the short-time Fourier transform to speech processing and spectral
analysis: IEEE Transactions on Acoustics, Speech, and Signal Processing, 82, 10121015.
Mallat, S., 2008, A wavelet tour of signal processing: Academic Press.
Partyka, G. A., J. Gridley, and J. Lopez, 1999, Interpretational applications of spectral decomposition in
reservoir characterization: The Leading Edge, 18, 353360, http://dx.doi.org/10.1190/1.1438295.
Stockwell, R. G., L. Mansinha, and R. P. Lowe, 1996, Localization of the complex spectrum: The S
transform: IEEE Transactions on Signal Processing, 44, 9981001,
http://dx.doi.org/10.1109/78.492555.
DOI http://dx.doi.org/10.1190/segam2013-0439.1 2013 SEG
SEG Houston 2013 Annual Meeting Page 2375
D
o
w
n
l
o
a
d
e
d

0
7
/
2
1
/
1
4

t
o

9
3
.
1
8
0
.
1
8
1
.
4
8
.

R
e
d
i
s
t
r
i
b
u
t
i
o
n

s
u
b
j
e
c
t

t
o

S
E
G

l
i
c
e
n
s
e

o
r

c
o
p
y
r
i
g
h
t
;

s
e
e

T
e
r
m
s

o
f

U
s
e

a
t

h
t
t
p
:
/
/
l
i
b
r
a
r
y
.
s
e
g
.
o
r
g
/