Lin Thesis 2013

EXTRACTING POLAR ANISOTROPY PARAMETERS FROM
SEISMIC DATA AND WELL LOGS
A Thesis
Presented to
the Faculty of the Department of Earth and Atmospheric Sciences
University of Houston
In Partial Fulfillment
of the Requirements for the Degree
Master of Science
By
Rongrong Lin
May, 2013
Extracting polar anisotropy parameters from seismic data
and well logs
by Rongrong Lin _______________________
Approved by
_________________________
Dr. Robert Stewart, Committee chairman
_________________________
Dr. Leon Thomsen, advisor
_________________________
Dr. John Castagna, committee member
_________________________
Dr. Daniel Ebrom, Committee member
_________________________
Dean, College of Natural Sciences and Mathematics
ii
ACKNOWLEDGEMENTS
First of all, I would like to thank my advisor, Dr. Leon Thomsen, for his guidance and
encouragements. He inspired me with many good ideas, and also taught me how to be a
good computer programmer and a good writer. Although he is not a full-time professor,
he is very much responsible for my thesis research and I am always impressed by his
pursuit of excellence.
Second, I especially thank my committee members Dr. Robert Stewart, Dr. John
Castagna, and Dr. Daniel Ebrom for their useful comments and suggestions to my thesis
research. I would also like to specially thank Dr. Brian Russell, for his advice on using
Hampson-Russell software and providing me the data for my research. And I also want to
acknowledge Anoop William from Allied Geophysical Laboratories and Kevin Hall from
University of Calgary for their help in my first-stage data searching.
My thanks and appreciations also go to the graduate students, staff, and faculty of the
Dept. of Earth and Atmospheric Sciences for their help. The financial support from
University of Houston is also greatly appreciated.
Finally, I want to show my gratefulness to my family and my friends. Without their
support of my academic pursuit in the United States, the completion of my study would
not be possible.
iii
EXTRACTING POLAR ANISOTROPY PARAMETERS FROM
SEISMIC DATA AND WELL LOGS
An Abstract of a Thesis
Presented to
the Faculty of the Department of Earth and Atmospheric Sciences
University of Houston
In Partial Fulfillment
of the Requirements for the Degree
Master of Science
By
Rongrong Lin
May, 2013
iv
ABSTRACT
With Vp, Vs, and density well logs, we can do isotropic AVO forward modeling, using
isotropic Zoeppritz equations or its approximations, and a wavelet extracted from the
seismic data, to get synthetic seismic data. Compared with real seismic data, these
synthetic data are scaled differently, and ignore many propagation effects.
Conventionally, the seismic data are normalized to the isotropic synthetic data for
comparison; this procedure suppresses the anisotropy information in the real data.
In this study, I tested a new method, using log data of Vp, Vs, and density, together with
seismic pre-stack Common Depth Point (CDP) gathers near the well site, to deduce
anisotropy parameters from them. This dataset is from Colony Sand, Alberta, Canada. I
calculated the normalization function, as in the conventional procedure, but then filtered
it in frequency domain using a cut-off frequency to be determined empirically. The low-
frequency part of the normalization function was used to eliminate the propagation
factors in the real data, but not the reflection coefficients. After scaling, the difference
between the real data, thus normalized, and the isotropic synthetic data are attributed to
the anisotropic part of the reflection coefficients. The resulting distributions of  (z ) and
 (z ) are compared to the gamma-ray log. There is a positive correlation between the
anisotropy parameter and the gamma-ray log, which is an indication of the validity of the
method.
v
TABLE OF CONTENTS
ACKNOWLEDGEMENTS……………………………………………………………………….iii
ABSTRACT……………………………………………………………………………………….v
TABLE OF CONTENTS…………………………………………………………………….……vi
TABLE OF FIGURES……………………………………………………………………………vii
TABLE OF EQUATIONS…………………….……………………………….………………….ix
1. Background ........................................................................................................................... ix
1.1 Recent methods to get anisotropy parameters and their drawbacks ........................... 3
2. New method ........................................................................................................................... 5
2.1 Overview ............................................................................................................................. 5
2.2 Data preparation .................................................................................................................. 6
2.2.1 Well logs loading ......................................................................................................... 6
2.2.2 Seismic data loading ................................................................................................... 9
2.2.3 Synthetic generating .................................................................................................. 16
2.2.4 Synthetic and seismic correlation ............................................................................ 20
2.3 Methodology development .............................................................................................. 24
2.3.1 Anisotropy parameters in reflection coefficient equation .................................... 24
2.3.2 Connection between synthetic and real seismic amplitude .................................. 26
2.3.3 Using normalization factor in anisotropy parameter estimation ........................ 42
2.4 Test ..................................................................................................................................... 65
2.4.1 Anisotropy parameter with gamma ray log correlation ........................................ 65
2.4.2 Estimating the reliability of the inferred anisotropy ............................................. 67
2.5 Conclusions and discussion ............................................................................................. 70
REFERENCES.……………………………………………..……..…………………….73
APPENDIX A: Comparison of synthetic produced from Hampson-Russell software with
what is computed from logs…………..…………………………………………………………..75
APPENDIX B: Matlab codes for making synthetic using zero-phase wavelet and well logs
to extract synthetic A,B,C ………………………………………………………………………..83
APPENDIX C: Matlab codes for extracting anisotropy parameters delta and epsilon from
synthetic and seismic A,B,C ……..….…………………………………………………………..85
APPENDIX D: Correlation of the  parameter with gamma ray log from other literature…….87
vi
TABLE OF FIGURES
Figure 1. Vp,Vs, and density logs from Colony well……………………………...……………..8

Figure 2.Near well site P-wave reflections from anisotropic medium…………………………... 9
Figure 3.Geometry of Colony sand seismic data……………………………………………….. 10
Figure 4.CDP gathers for Colony sand seismic data…………………………………………… 11
Figure 5.CDP gathers for Colony sand seismic data with P-wave log posted………………… 12
Figure 6.Seismic wavelet, in time and frequency domains……………………………………..13
Figure 7.Seismic wavelet in time domain, converted to zero phase…………………………… 14
Figure 8.The frequency spectrum of the zero-phase wavelet………………………………….. 14
Figure 9.Supergather at CDP 330 deconvolved to zero-phase wavelet……………………….. 15
Figure 10.Synthetic created from colony well logs……………………………………………. 18
Figure 11.Synthetic converted to zero phase for colony well logs…………………………….. 19
Figure 12.Synthetic and seismic super-gather comparison at location CDP330………………. 21
Figure 13. Synthetic and seismic super-gather at location CDP330 in angle domain, with the
original seismic wavelet……………………………………………………………………. 22
Figure 14a.Synthetic and seismic super-gather at location CDP330 in angle gather display,
zero-phase…………………………………………………………………………………... 23
Figure 1514b.Horizons picked on both seismic and synthetic in angle domain, zero phase……. 31
Figure 1615a.Seismic horizon at 418ms and its Aki-Richards approximation………………….. 32
Figure 1715b.Seismic horizon at 436ms and its Aki-Richards approximation………………….. 33
Figure 1815c. Seismic horizon at 490ms and its Aki-Richards approximation…………………. 33
Figure 1915d. Seismic horizon at 582ms and its Aki-Richards approximation…………………. 34
Figure 2015e. Seismic horizon at 636ms and its Aki-Richards approximation…………………. 34
Figure 2115f. Seismic horizon at 688ms and its Aki-Richards approximation………………….. 35
Figure 2216. Amplitude intercept A ,gradient B and curvature C for the major seismic horizons,
zero phase…………………………………………………………………………………... 36
Figure 2317a. Synthetic horizon at 418ms and its Aki-Richards approximation………………... 37
Figure 2417b. Synthetic horizon at 436ms and its Aki-Richards approximation……………….. 38
Figure 2517c. Synthetic horizon at 490ms and its Aki-Richards approximation……………….. 38
Figure 2617d. Synthetic horizon at 582ms and its Aki-Richards approximation……………….. 39
Figure 2717e. Synthetic horizon at 636ms and its Aki-Richards approximation………………... 39
Figure 2817f. Synthetic horizon at 688ms and its Aki-Richards approximation………………... 40
Figure 2918. Amplitude intercept A, gradient B and curvature C for the major synthetic horizons,
zero phase…………………………………………………………………………………... 41
Figure 3019. Comparison of Asyn and Aseis……………………………………………………. 43
Figure 3120. Comparison of Bsyn and Bseis……………………………………………………. 44
Figure 3221. Comparison of Csyn and Cseis……………………………………………………. 44
Figure 3322. Normalization factor for intercept A: NA…………………………………………. 45
Figure 3423. Normalization factor for gradient B: NB………………………………………….. 46
vii
Figure 3524. Normalization factor for gradient C:
NC………………………………………………………………………………………….. 46
Figure 3625. High and low component of normalization factor N(,t)……………………………. 47
Figure 3726. NA and its linear least-square fit…………………………………………………... 48
Figure 3827. NB, the absolute value of NB and its linear least-square fit……………………… 49
Figure 3928. NC, the absolute value of NC and its linear least-square fit………………………. 49
Figure 4029. Bseislow and Bsyn comparison………………………………………………………. 51
Figure 4130. Cseislow and Csyn comparison……………………………………….………………. 51
Figure 4231. Plot of residual gradient ∆B(t)……………………………………………………………. 52
Figure 4332. Plot of residual curvature ∆C(t)…………………………………………………………...52
Figure 4433. Seismogram of the convolution of the wavelet and seismic pickings with seismic
pickings shown above………………………………………………………………………. 55
Figure 4534. Seismogram of the convolution of the wavelet and synthetic pickings with seismic
pickings shown above……………………………………………………………………… 56
Figure 4635. Comparison of one trace seismogram of the convolution of the wavelet
and B(t ) ………………………………………………………………………………….. 58
Figure 4736. Comparison of one trace seismogram of the convolution of the wavelet
and C (t ) ………………………………………………………………………………….. 59
Figure 4837. Gamma ray log for Colony well…………………………………………………… 61
Figure 4938. Time-Depth relationship…………………………………………………………… 64
Figure 5039. Anisotropy parameter  variation with depth for colony well…………………… 65
Figure 5140. Anisotropy parameter  variation with gamma ray log display…………………. 66
Figure 5241. Synthetic seismogram computed from anisotropic reflection coefficient…………. 68
Figure 5342. Comparison of the original seismic and synthetic seismogram computed from
anisotropic reflection coefficient…………………………………………………………… 69
Figure A-1. Amplitude intercept A, gradient B and curvature C for the major synthetic horizons,
zero phase……………………………………………………………………………………75
Figure A-2. Aref calculated from well logs ……………..………………………………………77
Figure A-3. Bref calculated from well logs ……………..………………………………………77
Figure A-4. Cref calculated from well logs ……………..………………………………………78
Figure A-5. Asyn calculated from correlation of seismic wavelet and Aref………………………79
Figure A-6. Bsyn calculated from correlation of seismic wavelet and Bref………………………79
Figure A-7. Csyn calculated from correlation of seismic wavelet and Cref………………………80
Figure A-8. Asyn comparison from log computation and HR software…………………………81
Figure A-9. Bsyn comparison from log computation and HR software…………………………81
Figure A-10. Csyn comparison from log computation and HR software……………………..…82
Figure D-1. Anisotropy parameter  variation with gamma ray log display………………...87
Figure D-2. Anisotropy parameter  variation with gamma ray log display ………...……...88
Figure D-3. Anisotropy parameter  (from 1250ft to 1750ft, equivalent to from 381m to
533.4m) ) correlation with well 10-14 gamma ray log (from 381m to 533.4m)…...…..…...90
viii
TABLE OF EQUATIONS
(1.1) V p ( )  V p 0 [1   sin 2 con 2   sin 4  ] ……………………………….…….….2

VP 0 2
(1.2) Vs  ( )  Vs 0 [1  ( ) (   ) sin 2 con 2 ] ……………………………….……..2
Vso
Vs|| ( )  Vs 0 [1   sin 2  ]
(1.3) ………………………………………………………….2
C11  C33
(1.4)  ………………………………………………………….…………..2
2C33
(C13  C44 )2  (C33  C44 )2
(1.5)  ..…………………………………………………..2
2C33 (C33  C44 )
C  C44
(1.6)   66 ………………………………………….………………………….2
2C44
x2 2 x 2
(1.7) t x 2  t0 2   …………………….…….4
V 2 NMO, SS V 2 NMO ,SS [t0 2V 2 NMO ,SS  (1  2 ) x 2 ]
 
(1.8)  …………………………………………………………………….…..4
1  2
(1.9) vNMO (0 )  VP (0 )(1  2 )1/2 …………………………………..…………………..4
(2.1) Vp1.16Vs1.36………………………………………………………………..6
(2.2) Vs  0.8621Vp  1.1724 …………………………………………………………..6
(2.3) s(t )  r (t ) * w(t ) …………………………………………………………………16
1  Z  1  Vp 2Vs 2  G   2 1  Vp  2
(2.4) R ( )       tan  sin  …….17
iso 2
   ( )    sin 
2  Z  2  Vp Vp  G   2  Vp 
       
(2.5) Raniso ( )  1  Z0   1  Vp0  ( 2Vs0 )2  G0   ( 2  1 )  sin 2   1  Vp0  ( 2  1 )  tan 2  sin 2  …..24
2  Z0  2  Vp0 Vp0  G0   2  Vp0 
1 1
(2.6) R aniso ( )  Riso ( )  R aniso ( )   2  1  sin 2    2  1  tan 2  sin 2  ……..26
2 2
(2.7a) s( , t )  I (t ) * P ( , t ) * r ( , t ) * P ( , t ) * w0 (t )S0 ( ) ………………………….…27
s( , t )  [ I (t ) * P ( , t ) * P ( , t ) * S0 ]w0 (t ) * r ( , t )
(2.7b) ………………..…….….…...27
 P( , t ) * w(t ) * r ( , t )
(2.8) S (t , )  RPP
iso
(t , )* w(t ) ……………………………………………...……..…….28
(2.9) N ( , t )  s( , t )  N ( , t )  P( , t ) * w(t ) * r ( , t )  w(t ) * RPP ( , t ) …………..…..28
iso
(2.10) N ( , t )  P( , t )  1 ………………………………………………………..……...28

(2.11) r ( , t)  R iso
pp ( , t ) ………………………………………………….……..……….28
ix
(2.12) Riso (t0 , )  A(t0 )  B(t0 )sin 2   C (t0 )sin 2  tan 2  ……………………………29
(2.13) S (t0 , )   A(t0 )  B(t0 )sin 2   C (t0 )sin 2  tan 2   * w(t0 ) …............................29
(2.14) s(t0 , )   Aseis (t0 )  Bseis (t0 )sin 2   Cseis (t0 )sin 2  tan 2   * w(t0 ) ………….….29
(2.15) S (t0 , )  Asyn (t0 )  Bsyn (t0 )sin 2   Csyn (t0 )sin 2  tan 2  ………………………41
Asyn  Asyn (t0 ) ;

(2.16) Aseis  Aseis (t0 ) ; ……………………………………………………………….42
Asyn
N0 
Aseis
A (t0 )  N 0 Aseis (t0 );
*
seis
*
(2.17) Bseis (t0 )  N 0 Bseis (t0 ); …………………………………………………………….43
*
Cseis (t0 )  N 0Cseis (t0 )
N A (t0 )  Asyn (t0 ) / Aseis
*
(t0 )
(2.18) N B (t0 )  Bsyn (t0 ) / Bseis
*
(t0 ) ………………………………………………………45
NC (t0 )  Csyn (t0 ) / Cseis
*
(t0 )
Aseislow (t0 )  N Alow (t0 )  Aseis
*
(t0 )
(2.19) Bseislow (t0 )  N Blow (t0 )  Bseis (t0 ) ……………………………………………...….50
*
Cseislow (t0 )  NClow (t0 )  Cseis

*
(t0 )
B(t0 )  Bseislow (t0 )  Bsyn (t0 )
(2.20) ……………………………………………….…...50
C (t0 )  Cseislow (t0 )  Csyn (t0 )
Aseislow  Asyn  ( Bseislow  Bsyn )sin 2   (Cseislow  Csyn )sin 2  tan 2 
(2.21)  0  B sin   C sin  tan 
2 2 2
………….……...…..53
1 1
 ( 2  1 )sin 2  * w(t )  ( 2  1 )sin 2  tan 2  * w(t )
2 2
 1 
 B(t )  2  (t )  w0(t) 
 
(2.22)   …………………………………………….………53
C (t )  1  (t )  w0(t) 

 2 

(2.23) e  B(t ) * w(t )  B(t )  B(t ) * w(t )  B(t ) * (t )  B(t ) * (w(t )  (t )) …..58
x
  2   1   1 
     
 3 2 2 
(2.24)   4   3   3  ……......................................................................................60
 ........ 
 
 n   n 1   n 1 
  2   1   1 
     
 3 2 2 
(2.25)  4   3   3  ……………………………………………………………..60
 ........ 
 
 n   n 1   n 1 
 Anew(t )  Asyn (t ) 
 
(2.26)  Bnew(t )  Bsyn (t )   (t )  …………………………………….…………….....67
Cnew(t )  C (t )   (t ) 
 syn 
1  Z 0  1  Vp0  2Vs0   G0   2

2
1  Vp0  2
R ( )  
iso
        sin     tan  sin  ….76
2
(A.1) 2  Z 0  2  Vp0  Vp0   G0   2  Vp0 
 
 A  B sin   C tan  sin 
2 2 2
1  Z 0  1  2Vp02  1Vp01 
A    
2  Z 0  2  2Vp02  1Vp01 
1  Vp0  2Vs0   G0  

2
(A.2) B       ……………………………………………76
2  Vp0  Vp0   G0  
 
1  Vp0 
C  
2  Vp0 
xi
1. Background
Anisotropy is the variation of a physical property depending on the direction in which it
is measured (Sheriff, 1972). This is a very common phenomenon in the science fields
such as chemistry, medical science, physics, and engineering. In the geophysics field, the
anisotropy mostly refers to seismic anisotropy. Seismic anisotropy is the dependence of
seismic velocity upon angle (Thomsen, 2002). It can come from the anisotropic rock
itself or the stress-induced anisotropy due to the difference of vertical stress and
horizontal stress caused by layers. The application of seismic anisotropy has improved
the exploration of hydrocarbons a lot, by modifying the velocity model from simple
isotropic to more realistically anisotropic. Seismic anisotropy has played roles in
applications such as the long offset seismic data with greater angles of incidence (the
angle-dependence of velocity is more evident), AVO (Amplitude versus offset)
quantitative analysis, and anisotropic migration. (Thomsen, 2002)
The study of seismic wave propagation leads to the study of rock elasticity tensors. For
the simplest realistic case, the elasticity is of type of hexagonal symmetry, which is also
called polar symmetry. In this case, the elasticity tensor has five independent elements
C11, C13, C33, C44, and C66. This polar anisotropy case applies to horizontal massive shales
and horizontal thin-bed sequences (Thomsen, 2002). Because its symmetry axis is
vertical, polar anisotropy is also referred to as vertical transverse isotropy (VTI).
1
To express the velocities of plane waves in polar anisotropy using elasticity tensor
generates complicated equations. Thomsen used the combinations of elastic modules to
make the expressions much simpler. If the material is of weak anisotropy, which is the
most common case, then:
V p ( )  V p 0 [1   sin 2 con 2   sin 4  ] (1.1)
VP 0 2
Vs  ( )  Vs 0 [1  ( ) (   ) sin 2 con 2 ] (1.2)
Vso
Vs|| ( )  Vs 0 [1   sin 2  ] (1.3)
There,
C11  C33
 (1.4)
2C33
(C13  C44 )2  (C33  C44 )2 (1.5)


2C33 (C33  C44 )
C66  C 44
  (1.6)
2C 44
The three anisotropic parameters δ,ε,γ can also be understood respectively as near-
vertical anisotropy, near-horizontal anisotropy, and SH anisotropy.
These three anisotropy parameters are useful because:
2
 They are dimensionless.
 The anisotropy parameters reduce to zero when degenerating to isotropic case.
 When the parameters are much less than 1, we can take such formations as
weakly anisotropic.
(Thomsen, 2002)
Nowadays the anisotropy has been developed to include azimuth anisotropy, which is
about HTI (horizontal transverse isotropy) media and TTI (tilted transverse isotropy)
media since it first came out during the 1970s. The theory development of anisotropy has
been making a lot of progress, while the real practice of getting anisotropy parameters is
relatively underdeveloped.
My thesis will try a new method to extract anisotropy parameters. Based on only the
simplest polar anisotropy case, I will extract the anisotropy parameters δ and ε from P-
wave propagation.
1.1 Recent methods to get anisotropy parameters and their drawbacks
In oil and gas reservoirs, shale is the lithology with the most significant anisotropy. The
most direct way to obtain anisotropy parameters of shale is to measure the shale sample
in a lab, such as using traveltime inversion and physical modeling. But in lab conditions,
the rocks are under different frequencies compared with seismic waves, and they are not
likely to exhibit its original anisotropy when buried underground.
3
Nowadays some other ways to measure anisotropy using P-wave well logs and seismic
data gathers have been developed as well. The seismic arrival-time methods (such as
those which measure hyperbolic and non-hyperbolic move-out (Tsvankin and Thomsen,
1994)) produce estimates of anisotropy which have low spatial resolution, and therefore
cannot be used in AVO studies, which have much higher spatial resolution (depending on
the seismic wavelet).
One method uses the hyperbolic move-out is based on the equation:
x2 2x 2
t x  t0  
2 2
(1.7)
V 2 NMO, SS V 2 NMO,SS [t 0 V 2 NMO, SS  (1  2 ) x 2 ]
2
(Alkhalifah and Tsvankin, 1995)
The v NMO,SS is the short spread normal move-out velocity using conventional velocity
analysis, by short-spread, we mean mute the CDP gather at offset=depth.  is dependent
on the anisotropy parameters δ and ε (Thomsen, 1986):
 
 (1.8)
1  2
The parameter  can be obtained from formula
v NMO (0  )  VP (0  )(1  2 )1 / 2 (1.9)
We can see that normally the traveltimes must be measured over thick intervals, so that
the small parameters  and δ may be estimated with confidence.
4
Another way to measure anisotropy is anisotropy logging, or cross-dipole logging. This
kind of log uses one monopole and two wideband, low-frequency, dipole transmitters
perpendicular to each other, it measures compressional (quasi-P), fast and slow shear-
slowness and fast-shear azimuth. The pitfall for this method is that it is very dependent on
the borehole environment, and it also has dispersion characteristics due to dipole flexural
waves (Patterson and Tang, 2005).
Here my study will focus on a method to extract anisotropy parameters using both well
logs and seismic data. I concentrate on amplitudes; hence the results will have the
resolution of the seismic wavelet.
2. New method
2.1 Overview
The goal of this study is to use log data of Vp, Vs, and density, together with seismic pre-
stack Common Depth Point (CDP) gathers near the well site, to deduce anisotropy
parameters from them, with the resolution of the seismic wavelet. This study will utilize
the Hampson-Russell AVO package to do the AVO modeling, and will calculate
normalization factors to compare log-derived synthetic data and real seismic data. Since
the synthetic is based on isotropic theory, the normalization factors are modified to
permit an estimate of anisotropy. The primary goal is to extract the and ε parameters, in
the logged section of the well.
5
The method involves the comparison of surface seismic data with synthetic data
calculated from logs. Before this comparison is feasible, certain conventional data
preparation steps are required. These are described next and the details of the new
method follow that.
2.2 Data preparation
The Colony Sand Dataset is suitable for this experiment because it has both seismic data
and most well logs needed in that seismic acquisition area. The Colony Sand Dataset is
acquired from Alberta, Canada and it is the default dataset for Hampson-Russell software
AVO package version 6.5 CE7/R4. In this experiment, I use this AVO package
extensively.
2.2.1 Well logs loading
The first step is to load the well logs from the Colony Dataset. In the dataset, we have one
well in LAS format, called Colony_well. In this well, there are logs for P-wave slowness
(1/velocity), density, gamma Ray, SP and resistivity. For this project, I still need an S-
wave log. Although the argument would be stronger with real data, since the real data is
not available, I estimate VS as follows. Assuming the log as a wet (brine-filled log), we
can use Castagna’s “mudrock” equation (Castagna, 1985) to create a VS log based on the
VP log. Castagna’s equation is:
Vp1.16Vs1.36(km/s) (2.1)
So
6
Vs  0.8621Vp  1.1724 (km/s) (2.2)
To this point, we have assumed that the logs are from a wet well, but according to the
information about the Colony data, there is gas sand present, so we still need to conduct a
fluid substitution to get the correct S-wave velocity. This procedure can be simply done
by performing Fluid Replacement Modeling in the Hampson-Russell AVO package. In
brief, the VP and density logs are converted to brine-filled equivalents, using separate
knowledge of which layers are gas-filled, and using standard isotropic Gassmann theory
(Gassmann, 1951). Then VS is estimated using (2.2), and all logs are converted back to
original fluids.
After all this process, the final logs for density, 1/VS, and 1/VP are shown in Figure.1:
7
Figure 1. Vp,Vs and density logs from Colony well
The stars indicate the logs used to compute the synthetic seismograms. The density log is
renamed as Density_FRM, which means density log after performing Fluid Replacement
8
Modeling, S-wave_cast_FRM means Vs log created by Castagna’s mudrock line equation
(2.1) and modified by Fluid Replacement Modeling. P-wave_corr_FRM means the P-
wave log has been stretched and squeezed to correlate to the seismic event time. This is
actually done after the synthetic is computed, we show the result here before the
discussion about the correlation later.
2.2.2 Seismic data loading
The required geometry of a well and a corresponding CDP gather is shown in Figure 2.
The synthetic seismic traces near the well site are generated from isotropic reflection
coefficients (using Vp , Vs , and density logs). However, the reflections recorded in the
real seismic data will contain anisotropy information, if this area has anisotropic media.
For example, in the CDP gather from Figure 2, if the medium is isotropic, the velocity of
the three reflection rays (S1-R1;S2-R2;S3-R3) will travel with same velocities, But if the
medium is anisotropic, even though they are reflected from the same interface, their
velocity is different. That is because their reflection angle is different ( 1   2   3 ).
Figure 2.Near well site P-wave reflections from anisotropic medium
9
The seismic data in the Colony dataset is called gathers.sgy. It is a pre-stack 2D line. For
the survey, there are 129 shots and the source/receiver offset interval is 40m. The CDP
gathers are numbered from 260 to 390; each gather has about 10 traces. For each trace,
the sample rate is 2ms and there are 250 samples in each trace. The seismic data begins
from 300ms.
The geometry display is simple:
Figure 3. Geometry of Colony sand seismic data
10
After loading in the seismic data, it displays as:
Figure 4.CDP gathers for Colony sand seismic data
As we will need to use the CDP gather near the well site, we should find the location of
the CDP gather which the well site is closest to. We can find this information easily in
Hampson-Russell when we load the well logs into the Geoview package. It is located at
the position of CDP 330.
We project the P-wave velocity log onto that CDP gather and the display is like this:
11
Figure 5.CDP gathers for Colony sand seismic data with P-wave log posted
In order to improve the signal/noise ratio, we use seismic CDP super-gathers in place of
the original seismic CDP gathers. Super-gathers are averages of the original gathers; in
this case we average over 5 CDP positions near the well.
The wavelet extracted from the seismic super-gather is shown here, along with its
spectrum (amplitude and phase).
12
Figure 6.Seismic wavelet, in time and frequency domains
Because seismic data are usually viewed with zero-phase wavelets, Figure 7 shows the
wavelet from Figure 6, shifted to zero phase. The zero-phase wavelet is extracted from
the original seismic wavelet (Figure 6) using wavelet dephase deconvolution, which can
shape the known input wavelet to a zero-phase wavelet with the same amplitude
spectrum as the input.
13
Figure 7.Seismic wavelet in time domain, converted to zero phase
Figure 8.The frequency spectrum of the zero-phase wavelet
14
Also, Figure 8 shows the amplitude spectrum of this wavelet.
As we can see, it differs from the spectrum in Figure 6. Evidently the HR module which
extracted this wavelet did some further wavelet-shaping; however, this is not a focus of
this thesis. We show below (Figure 9) how the seismic super-gather at CDP 330 looks,
deconvolved to this wavelet.
Figure 9.Super-gather at CDP 330 deconvolved to zero-phase wavelet
15
2.2.3 Synthetic generating
As the Vp, Vs, and density logs are ready, and the seismic wavelet is determined, we can
use them to generate an isotropic synthetic containing no propagation effects. In the
simplest version of the “convolutional model” of seismic wave propagation, the seismic
trace s(t ) is given by
s(t )  r (t ) * w(t ) (2.3)
It means one trace of the seismic reflection record is the convolution (*) of the reflection
coefficients r (t ) (of the layers the wave traverses) with the seismic wavelet w(t ) .The
reflectivity series r (t ) has amplitudes r, occurring at times t, both of which are affected by
anisotropy. For a synthetic trace created from well logs, this equation is true. But when it
comes to real seismic traces, it neglects propagation effects and noise. We will take these
factors into consideration in the anisotropy parameter calculation. Usually we can get the
wavelet by extracting it from seismic data. And for the best interpretation of the seismic
data, we usually convert wavelet into zero-phase.
The accurate isotropic algorithm for computing a synthetic would be the Zoeppritz
equation (Zoeppritz, 1919). But as we will use the Shuey’s approximation for Aki-
Richards equation (Aki and Richards, 1980, Shuey, 1985), which is an approximation for
Zoeppritz equation to extract anisotropy information, here we will use Shuey’s
16
approximation for Aki-Richards equation( referred to as Aki-Richards equation in the
following text) to create the isotropic synthetic also.
In the isotropic case, based on Aki-Richards’ approximation, the P-wave reflection
coefficient can be written as:
1  Z  1  Vp 2Vs 2  G   2 1  Vp  2

Riso ( )      ( )    sin     tan  sin 
2
(2.4)
2  Z  2  Vp Vp  G   2  Vp 
where Z  Vp is the P-wave impedance, G  Vs 2 is the shear modulus,  is density,
and the bar denotes the average of properties (above and below the interface). Vp is the
P-wave velocity and Vs is the S-wave velocity.  is the wavefront normal angle of the
incident P-wave. The synthetic gather generated (by HR) is a flattened gather, which uses
a model blocking of traveltime average to convert from depth to vertical traveltime t0
within the logged interval, and its output sample rate is 2ms. It selects a series of incident
angles , then constructs a spike series (t=2ms) using equation (2.4), and convolves
with a wavelet, as in equation (2.3), with r(t)  Riso(t0,), constructed as just described.
We found some problematic issues with the way this procedure is implemented by HR
(c.f. Appendix A), and so implemented a similar procedure in Matlab.
The synthetic gather is flat in time, for easy comparison with the NMO-corrected seismic
gather. The synthetic gather shown below is one with the wavelet extracted from seismic
data (Figure 6):
17
Figure 10.Synthetic created from colony well logs
The synthetic gather shown below is one with the wavelet (Figure 7) which was extracted
from seismic data, and converted to zero phase:
18
Figure 11.Synthetic converted to zero phase for colony well logs
19
2.2.4 Synthetic and seismic correlation
The synthetic and the real seismic differ in various ways. Some common reasons include
that the check shot correction is not applied and the time-depth curve for the synthetic is
not correct. And also if we use a default zero-phase wavelet to generate the synthetic, we
should expect the difference because the seismic wavelet is not supposed to be zero-
phase, and thus we need to extract it from the seismic.
After check shot analysis, log correlation, regenerating the synthetic with the seismic
wavelet extracted from the real seismic data, and converting to angle domain, the
synthetic and the seismic super-gather at location CDP330 display as in Figure 12.
20
Figure 12.Synthetic (right) and seismic super-gather(left) comparison at location CDP330
21
And after converting the seismic and synthetic data to angle gathers, the synthetic and the
seismic super-gather at location CDP330 display as in Figure 13.
Figure 13. Synthetic and seismic super-gather at location CDP330 in angle domain, with the
original seismic wavelet
Figure 14a shows the comparison between the zero-phase seismic data, and the zero-
phase synthetic in the angle domain:
22
Figure 14a.Synthetic and seismic super-gather at location CDP330 in angle gather display, zero-
phase
The synthetic and seismic traces thus constructed differ because the latter contain effects
of propagation and anisotropy, whereas the former do not. We address these issues next.
23
2.3 Methodology development
With well logs of Vp, Vs, and density, we can do isotropic AVO forward modeling, using
Zoeppritz equations or its approximations, to get synthetic seismic data. But we always
find that the synthetic data and real seismic data differ from each other. It is partly
because we do this modeling based on an isotropic assumption, whereas the seismic data
contain anisotropic effects. And, it is partly because the real data contain effects of
propagation, whereas the synthetic data do not. If we explore the differences between the
synthetic one and the real one, we are able to extract the anisotropic information from
these differences. We illustrate the theoretical development, using real examples from
the Colony dataset.
2.3.1 Anisotropy parameters in reflection coefficient equation
Here we discuss the anisotropic reflection coefficient for P-waves, since these are most
commonly used in exploration geophysics. Ruger (1997), based on Thomsen (1993),
gave the linearized equation for plane-wave P-wave reflection at a planar interface
separating two polar anisotropic media as:
1  Z  1  Vp0 2Vs0 2  G0   2 1  Vp0 

R aniso ( )   0    ( )    ( 2  1 )  sin     ( 2  1 )  tan 2  sin 2 
2  Z0  2  Vp0 Vp0  G0   2  Vp0 
(2.5)
This differs from Equation (2.4) in two ways:
24
 Z 0, G0, Vp0, Vs0, are respectively the vertical P-impedance, shear modulus, P-
velocity, and S-velocity; otherwise these terms are identical to the corresponding
terms in (2.4).
 are the anisotropy parameters (Thomsen, 1986) for the two
layers (layer 1 (above) and layer 2 (below the plane)). It is clear that the
anisotropic effect on amplitudes is a simple difference in the reflectivity. It is also
clear that, even though the anisotropic parameters are small compared to one, they
are not necessarily small compared to the other terms in the equation, and should
not be neglected.
The vertical P-wave velocity, S-wave velocity, and densities of the layers can be read
from the well logs; hence we may refer to the isotropic parts of Eqn. (2.5) as the “sonic-
band reflectivity”. As for , we use the incident phase angle (See Thomsen (1986) for
discussion of the difference between wavefront-angle and ray-angle in anisotropic media).
We can calculate by getting offset position from the SEG-Y file, and the depth of
reflectors from the logs or VSP data, or from the moveout velocity. Alternatively, we
could get velocity from Vp well logs, but they typically do not extend through the entire
overburden, and they don’t contain the anisotropy in the velocities as the moveout
velocities do, although the latter will be of poor vertical resolution. We can use ray
tracing to get the P-wave reflection angle; the ray tracing work can be done by Hampson-
Russell software, given a velocity function in the overburden. Hence, in the P-wave
reflection coefficient calculation, the only unknown quantities are the anisotropic
parameters.
25
Thus, the difference between anisotropic and isotropic reflection coefficients is, from (2.3)
and (2.5):
1 1
R aniso ( )  Riso ( )  R aniso ( )   2  1  sin 2    2  1  tan 2  sin 2  (2.6)
2 2
And we can see that the anisotropy information is all contained in the difference of the
synthetic and seismic, if the seismic data has been processed to only contain the primary
reflections, without propagation effect and other noise.
2.3.2 Connection between synthetic and real seismic amplitude
Besides the offset-dependence of reflection coefficient of the layers, there are many other
factors which affect the amplitude of the reflected wave recorded in the SEG-Y file, as
functions of offset. Some factors are from the subsurface, such as attenuation/dispersion,
geometrical spreading, random noise, transmission coefficients in the overburden,
focusing/defocussing, non-planar reflectors, surface ghost, receiver ghost, multiple
interference, etc.
Some other factors are from the recorded wavelet, such as the source radiation pattern,
the geophone response, and the array response. So the real seismic data used for
calculation must be processed to eliminate the factors mentioned above. But processing
itself can affect the amplitude, too. For example, the processing may include various
filtering and scaling operations that affect the amplitude as a function of offset.
26
All of these effects operate on the real wave, and they all affect the received amplitudes,
variably with offset. When we normalize the real data to the synthetic data, we are
implicitly correcting the real data for all of these, even without knowing any of the details,
leaving the normalized amplitudes representing only the reflectivity.
With this understanding, we can refine Eqn. (2.6). We use S 0 , P ( , t ) , P ( , t ) and I (t )
respectively to express, respectively the wavelet strength, downward wave propagation
operator, upward wave propagation operator, and instrumental effects, all of these
operating sequentially on the initial wavelet w0(t). Since many of these effects operate
differently on different frequency components, in the time domain it appears as a series
of convolutions:
s( , t )  I (t ) * P ( , t ) * r ( , t ) * P ( , t ) * w0 (t )S0 ( ) (2.7a)
Since convolution commutes, we can re-write this as
s( , t )  [ I (t ) * P ( , t ) * P ( , t ) * S0 ]w0 (t ) * r ( , t )
(2.7b)
 P( , t ) * w(t ) * r ( , t )
where all of these propagation and instrumental effects are included in the propagation
operator P( , t ) . Note that, as defined here, P operates on the wavelet w(t), which can be
determined from the seismic data using conventional techniques, rather than the initial
wavelet w0(t), which is not easily determined.
To find the propagation operator P( , t ) , the conventional procedure is to:
27
 pp ( z , ) , using equation (2.6).
Use logs to compute the isotropic reflectivity Riso
 Convert depth to time, for example using the functionality present in H-R
pp (t , ) .
software, producing Riso
 Convolve this with the seismic wavelet w(t ) (determined from seismic data) to
produce a synthetic seismic trace
S (t , )  RPP
iso
(t , ) * w(t ) (2.8)
 Define a normalization factor N(,t) (can be done with various methods),
converting seismic amplitudes (typically lying between +/- 1000) to isotropic
reflectivity (typically lying between +/- 0.1), band-limited by convolution with
the seismic wavelet,
 Multiply the seismic trace (2.7b) by N ( , t ) , forcing the seismic trace amplitudes
to match the isotropic reflectivity.
 Set the normalized seismic data (2.9) equal to the isotropic synthetic data (2.8):
N ( , t )  s( , t )  N ( , t )  P( , t ) * w(t ) * r ( , t )  w(t ) * RPP

iso
( , t ) (2.9)
This procedure clearly assumes that
N ( , t )  P( , t )  1 (2.10)
The procedure also assumes that the reflections are isotropic:
r ( , t)  R iso
pp ( , t ) (2.11)
The novelty introduced in the present proposal is two-fold:
28
a) We normalize the seismic to the synthetic in an angle-dependent way, as
described further below;
b) We separate the normalization into a low-frequency part and a high-frequency
part, and use only the former for normalization. As described further below, the
residuals, after this partial normalization, are attributed to anisotropy.
Taking these in order:
The isotropic reflectivity equation (2.3) can be written as Aki-Richards’

approximation, for an event at t0, as
Riso (t0 , )  A(t0 )  B(t0 )sin 2   C (t0 )sin 2  tan 2  (2.12)
where the coefficients A, B, C are given implicitly in equation (2.3). The Aki–Richards’
equation is valid for reflection angles up to about 40°, and for isotropic layers only. As
the angle range in the gathers is from 0° to 30°, we are safe to use it here.
As t0 varies through the logged interval, this makes a spike series, with the time-
resolution of the logs. After convolving with the wavelet (equation (2.8)), we have a
synthetic trace with the resolution of the wavelet:
S (t0 , )   A(t0 )  B(t0 )sin 2   C (t0 )sin 2  tan 2   * w(t0 ) (2.13)
In a similar way, we can parameterize the offset-dependence of the flattened seismic
gather, in the angle-domain as:
s(t0 , )   Aseis (t0 )  Bseis (t0 )sin 2   Cseis (t0 )sin 2  tan 2   * w(t0 ) (2.14)
29
If the seismic gather has zero phase, we can pick the major peaks (and troughs) of the
traces s(t0,θ), identify these with the major reflecting horizons, and find best-fit values of
Aseis, Bseis, and Cseis for each of these major events.
We are not able to detect every impedance layer from the synthetic and seismic, we can
only pick some obvious events from comparison of the seismic and synthetic. The picked
events may not be representative of all the reflection interfaces, so we only try to solve
the anisotropy for the picked horizons.
Here we pick the major events on synthetic and seismic zero phase. All the 6 horizons are
displayed in Figure.14b. The horizons are all picked using Hampson-Russel AVO
package ‘Horizon selection’ function.
30
Figure 1514b.Horizons picked on both seismic and synthetic in angle domain, zero phase
31
Figures 15a-f show the picked amplitudes, as functions of angle θ, for each of the six
major seismic events. They also show, as continuous lines, the fitted curves (Eqn.2.14)
for each event, constructed with the best-fit parameters using least-square method.
seismic horizon at 418ms and its Aki-Richards approximation

5000
original seismic horizon
4500 its Aki-Richards approximation
4000
3500
Amplitude
3000
2500
2000
1500
1000
500
0 5 10 15 20 25 30
angel(degree)
Figure 1615a.Seismic horizon at 418ms and its Aki-Richards approximation
32
6000
5000
4000
Amplitude
3000 original seismic horizon

its Aki-Richards approximation
2000
1000
0
0 5 10 15 20 25 30
angel(degree)
Figure 1715b.Seismic horizon at 436ms and its Aki-Richards approximation

5500
5000
4500
4000
3500
Amplitude
3000 original seismic horizon

2500
2000
1500
1000
500
0 5 10 15 20 25 30
angel(degree)
Figure 1815c. Seismic horizon at 490ms and its Aki-Richards approximation
33
9000
8000 its Aki-Richards approximation
7000
6000
Amplitude
5000
4000
3000
2000
1000
0
0 5 10 15 20 25 30
angel(degree)
Figure 1915d. Seismic horizon at 582ms and its Aki-Richards approximation
4
x 10 seismic horizon at 636ms and its Aki-Richards approximation
1.8
1.6
1.4
Amplitude
1.2
0.8
0.6
0 5 10 15 20 25
angel(degree)
Figure 2015e. Seismic horizon at 636ms and its Aki-Richards approximation
34
4
x 10 seismic horizon at 688ms and its Aki-Richards approximation
3
2.8
2.6
2.4
Amplitude
2.2
2
1.8
1.6
1.4
1.2
0 5 10 15 20 25
angel(degree)
Figure 2115f. Seismic horizon at 688ms and its Aki-Richards approximation
All of these figures (15a-f) show major deviations from the fitted curves, especially at
large angles. This confirms the common understanding that the curvature terms of the
Aki-Richards equation (2.12) are poorly determined, so that their interpretation is
problematic. The deviations may be due to interference from multiples, not accounted for
in prior processing. In any case, addressing this variation is outside the scope of the
present work. We carry forward the analysis of the curvature in any case, anticipating
that the derived values for  may be spurious.
Figure 16 shows the derived parameters Aseis, Bseis, Cseis.
35
5
x 10 InterceptA ,gradientB and curvatureC for major seismic horizons
2
-2
Aseis
Bseis
Amplitude
Cseis
-4
-6
-8
-10
400 450 500 550 600 650 700
time(ms)
Figure 2216. Amplitude intercept A ,gradient B and curvature C for the major seismic horizons,
zero phase
Of course these offset parameters are not directly comparable to reflectivity parameters
(2.12) because they contain the propagation effects as well as the reflection effects; we
address this issue with the normalization argument in section b).
Further, we cannot directly compare Aseis, Bseis, and Cseis, even after normalization,
with the parameters in equation (2.12) simply by reading off the values of A, B, and C
from the logs. The reason is that the convolution operation “folds together” the
36
contribution from nearby events; we need to estimate A, B, and C in a band-limited way,
on the synthetic, just as we did with the seismic. If the wavelet w(t0) has zero phase, we
pick the same major peaks (or troughs) of these functions A(t0)*w(t) etc. (as we did on the
seismic),and find best-fit parameters Asyn, Bsyn, Csyn for the major reflectors.
Figures 17a-f show the picked amplitudes, as functions of angle θ, for each of the six
major synthetic event. They also show, as continuous lines, the fitted curves (2.12) for
each event, constructed with the best-fit parameters.
synthetic horizon at 418ms and its Aki-Richards approximation

0.023
original synthetic horizon
0.022
0.021
Amplitude
0.02
0.019
0.018
0.017
0 5 10 15 20 25 30 35
angel(degree)
Figure 2317a. Synthetic horizon at 418ms and its Aki-Richards approximation
37
0.028
0.026
0.024
Amplitude
0.022
0.02
0.018
0.016
0 5 10 15 20 25 30 35
angel(degree)
Figure 2417b. Synthetic horizon at 436ms and its Aki-Richards approximation

0.04
0.035
Amplitude
0.03
0.025
0 5 10 15 20 25 30 35
angel(degree)
Figure 2517c. Synthetic horizon at 490ms and its Aki-Richards approximation
38
0.054
0.052
0.05
0.048
Amplitude
0.046
0.044
0.042
0.04
0 5 10 15 20 25 30 35
angel(degree)
Figure 2617d. Synthetic horizon at 582ms and its Aki-Richards approximation

0.25
0.24
0.23
0.22
Amplitude
0.21
0.2
0.19
0.18
0 5 10 15 20 25 30 35
angel(degree)
Figure 2717e. Synthetic horizon at 636ms and its Aki-Richards approximation
39
0.345
0.34
0.335
Amplitude
0.33
0.325
0.32
0.315
0 5 10 15 20 25 30 35
angel(degree)
Figure 2817f. Synthetic horizon at 688ms and its Aki-Richards approximation
As anticipated, the fits here are excellent (by construction), in strong contrast with figures
(15a-f). Figure 18 shows the derived parameters Asyn, Bsyn, Csyn. Note that Asyn and Csyn
are positively correlated, (because of the correlation between impedance and velocity),
whereas in Figure 16, Aseis and Cseis are anticorrelated.
40
intercept A ,gradient B and curvature C for the major synthetic horizons
0.6
Asyn
0.5 Bsyn
Csyn
0.4
0.3
Amplitude
0.2
0.1
-0.1
-0.2
400 450 500 550 600 650 700
time(ms)
Figure 2918. Amplitude intercept A, gradient B and curvature C for the major synthetic horizons,
zero phase
Then convolution of these values with the wavelet produces an isotropic synthetic gather:
S (t0 , )  Asyn (t0 )  Bsyn (t0 )sin 2   Csyn (t0 )sin 2  tan 2  (2.15)
41
2.3.3 Using normalization factor in anisotropy parameter estimation
2.3.3.1 Getting normalization factor for intercept A, gradient B and curvature C
Now that we have characterized, the AVO behavior, separately for synthetic and seismic,
we have to normalize them, in order to compare them. We can see that the seismic
amplitude is of much bigger magnitude than that of synthetic. The amplitude for seismic
is around +/-10000 while the synthetic amplitude is around +/-0.1.
We do this in several stages. First, we convert the seismic amplitudes in (2.14) to
approximate reflectivity amplitudes by normalizing (2.14) with a constant normalization
factor N0, based upon the average of the A(t0), function, so that they may be displayed
together with the synthetic values. Specifically, we calculate
Asyn  Asyn (t0 ) ;

Aseis  Aseis (t0 ) ;
Asyn
N0 
Aseis
(2.16)
where the angle brackets indicate the arithmetric average, over the selected major events.
We calculate with the absolute values, as shown, since we want for N0 to be a positive
number, adjusting all events (both + and -) only for this scale factor. For the Colony
dataset, as prepared above, we obtain N0= 1.1007e-005.
42
Then we adjust each of the seismic AVO parameters by this scale factor:
*
Aseis (t0 )  N 0 Aseis (t0 );
*
Bseis (t0 )  N 0 Bseis (t0 );
*
Cseis (t0 )  N 0 Cseis (t0 )
(2.17)
Now the seismic and synthetic AVO parameters have comparable scales, so that we can
plot them together, preparing for further adjustment. Figure 19 shows both Asyn (t0 ) and
*
Aseis (t0 ) on the same plot for the major events of the Colony dataset; you can see that
they still have significant differences. Figures 20 and 21 show the same information for
the gradient parameters B and the curvature parameters C.
comparison of Asyn and seismic normalized Aseis

0.35
Asyn
Aseis
0.3
0.25
0.2
amplitude
0.15
0.1
0.05
0
400 450 500 550 600 650 700
time(ms)
Figure 3019. Comparison of Asyn and Aseis
43
comparison of Bsyn and seismic normalized Bseis
2
Bsyn
Bseis
1.5
1
amplitude
0.5
-0.5
-1
400 450 500 550 600 650 700
time(ms)
Figure 3120. Comparison of Bsyn and Bseis
comparison of Csyn and seismic normalized Cseis

2
-2
Csyn
-4 Cseis
amplitude
-6
-8
-10
-12
400 450 500 550 600 650 700
time(ms)
Figure 3221. Comparison of Csyn and Cseis
44
The differences evident between seismic and synthetic parameters include both
propagation effects, and anisotropy effects. If there were no anisotropy effects, we
could completely account for the propagation effects by constructing normalization
factors
N A (t0 )  Asyn (t0 ) / Aseis

*
(t0 )
N B (t0 )  Bsyn (t0 ) / Bseis
*
(t0 ) (2.18)
NC (t0 )  Csyn (t0 ) / Cseis
*
(t0 )
Figures 22, 23, and 24 show these normalization functions, for the Colony dataset.
normalization factor for intercept A: NA

2
1.8
1.6
1.4
1.2
0.8
0.6
0.4
0.2
400 450 500 550 600 650 700
time(ms)
Figure 3322. Normalization factor for intercept A: NA
45
Normalization factor for gradient B: NB
0.7
0.6
0.5
0.4
0.3
0.2
0.1
-0.1
-0.2
400 450 500 550 600 650 700
time(ms)
Figure 3423. Normalization factor for gradient B: NB
Normalization factor for curvature C: NC

0.05
0.04
0.03
0.02
0.01
-0.01
-0.02
-0.03
400 450 500 550 600 650 700
time(ms)
Figure 3524. Normalization factor for gradient C: NC
46
It is clear that applying these angle-specific normalization factors to the seismic gathers
(2.12) would convert them into the isotropic synthetic gather (2.13), and that this
constitutes a refinement of the “conventional” procedure described above. However, we
want to avoid this assumption of isotropy.
2.3.3.2 Low-cut the normalization factor NB and NC
However, we recognize that these normalization functions NA(t0), NB(t0), and NC(t0), are
measured over the entire logged interval, and can be analyzed as Fourier series, with
spectra NA(f), NB(f), and NC(f). Because reflection amplitude (calculated from either
anisotropic or isotropic expressions) varies rapidly as a function of depth or time, this
contributes to the high-frequency portion of these spectra. However, the propagation
effects mentioned above (except for the reflection coefficient) accumulate progressively
as the wave propagates; hence they are represented by the low-frequency portion of these
spectra.
Figure 3625. High and low component of normalization factor N(,t)
47
Therefore we low-cut filter the normalization functions, and call these filtered functions
N Alow ( , t0 ), N Blow ( , t0 ), NClow ( , t0 ) .
In practice, for the Colony dataset, the logged interval is too short, and the major
reflectors too sparse to make a well-behaved filter operation. So, we have implemented
the following procedure, equivalent for this dataset: we find the least-squares best-fit
linear function, which accounts for the low-order time-variation in each of the
normalization functions above. Specifically, we compute NAlow(t0) as the straight line
which best approximates (in the least-squares sense) the function N A (t0 ) as defined
above. This straight line NAlow(t0) is also shown in Figure 26. Similarly, we compute
NBlow(t0) and NClow(t0); these straight lines are shown in figures 27 and 28.
NA and its least-square linear approximation

2
1.8 NA
linear fit of NA
1.6
1.4
1.2
0.8
0.6
0.4
0.2
400 450 500 550 600 650 700
time(ms)
Figure 3726. NA and its linear least-square fit
48
NB, absolute value of NB and its least-square linear approximation
0.7
NB
0.6
linear fit of abs(NB)
abs(NB)
0.5
0.4
0.3
0.2
0.1
-0.1
-0.2
400 450 500 550 600 650 700
time(ms)
Figure 3827. NB, the absolute value of NB and its linear least-square fit
NC,absolute value of NC and its least-square linear approximation

0.05
NC
linear fit of abs(NC)
0.04
abs(NC)
0.03
0.02
0.01
-0.01
-0.02
-0.03
400 450 500 550 600 650 700
time(ms)
Figure 3928. NC, the absolute value of NC and its linear least-square fit
49
This procedure approximately removes the (low-frequency) propagation effects, without
damaging the (high-frequency) reflectivity effects, which contain the anisotropy.
Then we construct low-cut seismic parameters
Aseislow (t0 )  N Alow (t0 )  Aseis

*
(t0 )
Bseislow (t0 )  N Blow (t0 )  Bseis
*
(t0 ) (2.19)
Cseislow (t0 )  NClow (t0 )  Cseis
*
(t0 )
These are observable quantities, calculable from the data, as described above.
As is evident from Figure 26, the normalization factor NA does contain some (minor)
high-frequency variation. Since this is a normal-incidence parameter, this high-frequency
variation cannot be explained by neglected anisotropy. In fact, within the present
assumptions, there is no satisfactory explanation of this high-frequency variation in
NA. A revision of these standard assumptions is beyond the scope of this M.S. thesis, as it
would constitute a deep revision in conventional thinking about the reflection process. So,
for now, we ignore this variation, keeping in mind that it might affect subsequent
conclusions.
1) Get residual gradient ∆B and residual curvature ∆C
Now we already have NBlow and NClow. We can get the ∆B and ∆C for the final calculation
about anisotropy parameters  and  .
B(t0 )  Bseislow (t0 )  Bsyn (t0 )

(2.20)
C (t0 )  Cseislow (t0 )  Csyn (t0 )
50
The plot of the Bseislow and Bsyn is:
Comparison of the Bsyn and low-cut Bseis

0.25
Bsyn
0.2 Bseislow
0.15
0.1
amplitude
0.05
-0.05
-0.1
-0.15
-0.2
400 450 500 550 600 650 700
time(ms)
Figure 4029. Bseislow and Bsyn comparison
The plot of the Cseislow and Csyn is:
Comparison of the Csyn and low-cut Cseis

0.3
Csyn
0.2 Cseislow
0.1
0
amplitude
-0.1
-0.2
-0.3
-0.4
400 450 500 550 600 650 700
time(ms)
Figure 4130. Cseislow and Csyn comparison
51
The difference of Bseislow and Bsyn is ∆B, its plot is:
residual Bseis-difference of low-cut Bseis and Bsyn

0.15
0.1
0.05
0
amplitude
-0.05
-0.1
-0.15
-0.2
400 450 500 550 600 650 700
time(ms)
Figure 4231. Plot of residual gradient ∆B(t)
The difference of Cseislow and Csyn is ∆C, its plot is:
residual Cseis-difference of low-cut Cseis and Csyn

0.1
-0.1
-0.2
amplitude
-0.3
-0.4
-0.5
-0.6
400 450 500 550 600 650 700
time(ms)
Figure 4332. Plot of residual curvature ∆C(t)
52
2.3.3.3 Solving residual anisotropy parameters from residual gradient and curvature
According to Eqn. (2.6,2.14,2.15), we get
Aseislow  Asyn  ( Bseislow  Bsyn )sin 2   (Cseislow  Csyn )sin 2  tan 2 

 0  B sin 2   C sin 2  tan 2  (2.21)
1 1
 ( 2  1 )sin 2  * w(t )  ( 2  1 )sin 2  tan 2  * w(t )
2 2
That is, the gradient residual ∆B is half the convolution of the residual delta anisotropy
 and wavelet w(t) while the curvature residual ∆C is half the convolution of the
residual epsilon anisotropy  and wavelet w(t). However, by picking only the peaks and
troughs on the seismic data, we lose the shape of the wavelet, thus the convolution is not
maintained any more.
For the optimal situation, the reflection interfaces are so sparsely distributed in depth and
the wavelet is so close to a spike that the peaks and troughs of a seismic trace can be
expressed as the multiplication of the reflection coefficient and the wavelet peak. Then,
Eqn. (2.21) can be modified as:
 1 
 B(t )   (t )  w0(t) 
 2 
  (2.22)
C (t )  1  (t )  w0(t) 

 2 

where w0 is the peak value of the wavelet, for zero-phase wavelet.
53
Here we conduct a posteriori test on whether the data we have meets the optimal situation.
Suppose it is the optimal situation, then a seismic trace can be expressed as the
multiplication of the reflection coefficient and the wavelet peak. As the peak value of the
zero-phase wavelet is 1, then the reflection coefficient at that depth is equal to the seismic
picking value there. Thus the wavelet convolved with the seismic picking value should
have the same value with the seismic picking at its depth. The result is shown in Figure
33, the seismogram colored black is the convolution result of the zero-phase wavelet and
* * *
the seismic picking value(reconstructed by normalized Aseis Bseis Cseis using Aki-Richards’
equation), the red curve displays the seismic picking value itself. The equation for the
seismogram is Eqn.(2.14).
And a similar figure for the synthetic is also shown here (Figure 34), the seismogram
colored black is the convolution result of the zero-phase wavelet and the synthetic
picking value (reconstructed by normalized Asyn Bsyn Csyn using Aki-Richards’ equation),
the red curve displays the synthetic picking value itself.
54
55
Figure 4433. Seismogram of the convolution of the wavelet and seismic pickings with seismic pickings shown above
56
Figure 4534. Seismogram of the convolution of the wavelet and synthetic pickings with seismic pickings shown above
We can find by the comparisons of the convolution and the picking itself that they are not
identical, thus it is not an optimal situation.
We can try some method to make the optimal assumption more valid such as using a kind
of wavelet deconvolution that can attempt to shape the known wavelet to a spike. But the
biggest drawback of this method is that it will introduce more high frequencies in the data,
which will influence the spectrum of the NB and NC, and further more the anisotropy
parameters  and  .
If we accept it as the optimal situation and the relationship between the residual gradient
∆B and residual delta anisotropy  meets Eqn. (2.22). As the peak value of the zero-
phase wavelet is 1, then the residual gradient ∆B is half the value of residual delta
anisotropy  . Similarly, the residual curvature ∆C is half the value of residual epsilon
anisotropy  .
To explore how much error will be in the  and  parameters using this optimal
assumption, we plot the B(t ) with the convolution of B(t ) and the wavelet in Figure
35, and also the C (t ) with the convolution of C (t ) and the wavelet in Figure 36.
57
delta B and deltaB convolved with wavelet
300
350
400
450
500
time(ms)
550
600
650
700
750 deltaB convolved with wavelet

deltaB
800
-0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15
Amplitude
Figure 4635. Comparison of one trace seismogram of the convolution of the wavelet and B(t )
The difference of these two curves is:
e  B(t ) * w(t )  B(t )  B(t ) * w(t )  B(t ) * (t )  B(t ) * (w(t )  (t )) (2.23)
here (t ) is a spike in the time domain with the same amplitude of the wavelet.
B(t ) * ( w(t )  (t ))

The relative error of  (t ) is .
B(t ) * (t )
From the data in Figure 35, the average value of error in  (t ) is -0.408%.
58
For the curvature residual C (t ) , the comparison of C (t ) and the convolution of C (t )
and the wavelet is shown in Figure 36.
deltaC and deltaC convolved with wavelet

300
350 deltaC convolved with wavelet

deltaC
400
450
500
time(ms)
550
600
650
700
750
800
-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3
Amplitude
Figure 4736. Comparison of one trace seismogram of the convolution of the wavelet and C (t )
C (t ) * ( w(t )  (t ))
Similarly, the error of  (t ) is . From the data in Figure 36, the
C (t ) * (t )
average error in  (t ) is 0.246%.
59
2.3.3.4 Getting anisotropy parameters for each layer
The anisotropy jump  and  can be used to calculate the anisotropy parameters  and
 for each layer, as long as we know the  and  for one arbitrary layer.
Suppose the  parameter for layers from layer1 to layer n is {  1 ,  2 ,  3 …  n }, then
  2   1   1 
     
 3 2 2 
 4   3   3  (2.24)
 ........ 
 
 n   n 1   n 1 
Similarly, assume the  parameter for layers from layer1 to layer n is {  1 ,  2 ,  3 …  n },
then
  2   1   1 
     
 3 2 2 

 4   3   3  (2.25)
 ........ 
 
 n   n 1   n 1 
We now have  (t ) and  (t ) , if we find a layer of low-shale component sand, we can
assume that sand layer is isotropic and its  =0 and its  =0.
60
We can locate such a sand layer based on the gamma ray curve.
The gamma ray curve for colony well is:
Figure 4837. Gamma ray log for Colony well
61
From this log plot, we see that the lowest gamma ray reading is at 650ms, indicating sand
layer with lowest shale component. Thus we assign  and  parameter at this time to be
zero. As  (t ) and  (t ) is sampled at 2ms interval from 418ms to 688ms, we assign  6
=0 and  6 =0. Then we substitute  6 =0 and  6 =0 into Eqn. (2.37) and Eqn. (2.38), we can
know all the values of parameter  and  in the sequence. This is an intuitive description
of the process known as seismic inversion, converting interface parameter-jumps to layer
parameters.
Solving equations (2.24) and (2.25) we get the anisotropy parameters  and  in chart 1
and chart 2
Time(ms) Parameter 
413 0.2345
427 -0.0286
463 -0.0043
536 0.2220
609 0.1800
662 0
693 0.2258
Chart 1. Anisotropy parameters  for layers from time 391ms to 739ms
62
Time (ms) Parameter 
413 0.7682
427 0.7782
463 0.7117
536 0.5806
609 0.5392
662 0
693 -1.0535
Chart 2. Anisotropy parameters  for layers from time 391ms to 739ms
We can see that the  parameter is in a reasonable range; its variation is discussed further
below.
By contrast, the  parameter doesn’t make a lot of sense. A realistic  should be > 0
(Thomsen, 1986) and within the range of 0~1. These  parameters we have here are not
consistent with these rules and they don’t satisfy the weak anisotropy assumption either.
We can explain it as because the C parameters we got from seismic are not reliable (c.f.
Figures 15a-f). Hence, we do not discuss  further, in this report.
The time-depth relationship is shown on Figure 38:
63
Figure 4938. Time-Depth relationship
Transforming from the time domain to the depth domain, we get plots of anisotropy
parameter vs. depth( Figure 39):
64
anisotropy parameter delta variation with depth
1000
1200
1400
1600
depth(ft)
1800
2000
2200
2400
-0.2 0 0.2 0.4 0.6 0.8
Figure 5039. Anisotropy parameter  variation with depth for colony well
2.4 Test
2.4.1 Anisotropy parameter with gamma ray log correlation
The anisotropy parameter is a combination of intrinsic anisotropy of rocks and the layer-
induced anisotropy. The result may be either bigger or smaller than the intrinsic
anisotropy. But we can still use the intrinsic anisotropy value estimated from rock physics
information from logs to see whether the calculated anisotropy parameter makes sense.
65
As the gamma ray log is from 550ms to 690ms, we can compare the anisotropy parameter
 from 550ms to 690ms with the gamma ray log. Converted to depth, 550ms to 690ms is
corresponding to 1780ft to 2340ft (equivalent to 542.5m to 713m).

1000
1200
1400
1600
depth(ft)
1800
2000
2200
2400
-0.2 0 0.2 0.4 0.6 0.8
Figure5140. Anisotropy parameter  variation with gamma ray log display
We can see that the anisotropy parameter  is coherent with the gamma ray readings.
When the gamma ray value is high, which indicates high shale component percentage is
above zero, with relatively high absolute values. This further adds to the reliability of
these estimated anisotropy parameters.
66
2.4.2 Estimating the reliability of the inferred anisotropy
Furthermore, to test the reliability of the calculated anisotropy parameter, we can use the
calculated anisotropy parameters to generate an anisotropic synthetic data and compare it
with the processed real seismic data, if the anisotropic synthetic ties with the real seismic
data better than the isotropic synthetic, it is reasonable to believe the inferred anisotropy
is reliable.
To compute the anisotropic seismic data using the anisotropy parameters  and 
(although the  is not reliable) we get,
the new reflection coefficient is:
 Anew(t )  Asyn (t ) 
 
 Bnew(t )  Bsyn (t )   (t )  (2.26)
Cnew(t )  C (t )   (t ) 
 syn 
where the A(t), B(t), and C(t) is from the log computed isotropic reflection coefficients.
Using the zero-phase wavelet shown in Figure 12, the new synthetic seismogram
computed for the anisotropic reflection coefficient is:
67
68
Figure 5241. Synthetic seismogram computed from anisotropic reflection coefficient
The co-display of the new synthetic and the original seismic and is shown in Figure 42:
Figure 5342. Comparison of the original seismic and synthetic seismogram computed from
anisotropic reflection coefficient
69
In comparison with Figure 13, which is the co-display of the original synthetic and
seismic, we can see that the new anisotropic seismogram is closer to the seismic data. Of
course, this new anisotropic synthetic does not have any propagation effects in it, so we
do not expect an exact match.
2.5 Conclusions and discussion
The anisotropy parameter  we get is within the range of -0.0286~0.2345, and the
anisotropy parameter  is within the range of -1.0535~0.7782. The parameter  is in the
reasonable range 
of anisotropy parameters, but the value of  parameter doesn’t make
 
sense, even though, this new method is still reliable because the problematic  value is

due to the fact that we can’t get the true curvature from the Aki-Richards’ approximation

of the amplitude versus angle. By comparison with the gamma ray log, which is an
indication of the shale component in the layers, the  anisotropy parameters is well
correlated to the gamma ray log: when the gamma ray reading is smaller, the  has
smaller absolute value, that is, the anisotropy isweaker.


In the procedure of getting the anisotropy jumps  and  between layers, the validity
of the assumption that the convolution of anisotropy jump and wavelet can be simplified
 
as the multiplication of the anisotropy jump and the peak of the wavelet depends on two
factors: the sparseness of the certain anisotropic layers and the similarity of the seismic
wavelet to a spike. If the seismic resembles a spike and the certain anisotropic layers are
spaced sparsely enough, then the assumption is valid and the method we use to deduce
the anisotropy jumps  and  is reliable. Furthermore, wavelet deconvolution
70
 
attempting to suppress the wavelet to a spike should be applied with concerns, because
this kind of wavelet shaping can introduce more high frequencies in the data to interfere
with the signal spectrum.
Besides the two major influence mentioned above, which are simplification of the
convolution to a multiplication and the unreliable curvature in Aki-Richards’ equation,
other factors can contribute to the errors in the anisotropy parameters we get as well.
They are:
1) The shear wave velocity is made using mudrock line equation, not a real log
measurement.
2) The wavelet we extracted from the seismic data in comparison of the synthetic
may not be the true wavelet in the real seismic data.
3) We converted the offset domain to angle gather using an isotropic ray theory,
which is not consistent with the anisotropy assumption.
4) The noise in seismic data, either low frequency or high frequency, will affect the
normalization factor low-cut.
5) We assume the minimum value of the gamma ray we have is corresponding to an
isotropic layer location with  =0 and  =0, this may not be the real situation.
In a brief summary, the procedures to get the anisotropy parameters  and  using this
 
method are listed as below:
 
1) Get the intercept A, gradient B and curvature C of each major horizon using Aki-
Richards’ equation for both the synthetic and seismic data
71
2) Divide the average of synthetic A by the average of seismic A to get the quotient
as seismic normalization factor NA.
3) Apply the NA to make the seismic B and C the scale of synthetic B and C.
4) Divide the synthetic B and C by the scaled seismic B and C to get the quotient as
normalization factor NB and NC.
5) Low cut the NB and NC to eliminate the propagation influence in the seismic B
and C
6) After scaling and low-cut, the difference of the seismic B and C and synthetic B
and C will be the anisotropy parameters  and  convoluted with the seismic
wavelet

7) Solve for the parameters  and  .
 
72
REFERENCES
1. Aki, K. and Richards, P. G., 1980, Quantitative Seismology-Theory and Methods,
vol.I: W.H. Freeman and Co., San Francisco.
2. Alkhalifah, T. and Tsvankin, I., 1995. Velocity analysis for transversely isotropic
media: Geophysics, vol.60, p.1550–1566.
3. Castagna, J. P., Batzle, M. L., and Eastwood, R. L., 1985, Relationships between
compressional-wave and shear-wave velocities in clastic silicate rocks:
Geophysics, vol.50, p.571–581.
4. Gassmann, F., 1951, Elasticity of porous media: Uber die elastizitat poroser
medien, Vierteljahrsschrift der Naturforschenden Gesselschaft, vol.96, p.1-23.
5. Patterson, D.J. and Tang, X. 2005, Pit Falls In Dipole Logging - Anisotropy:
Cause Of Discrepancy In Borehole Acoustic Measurements: Offshore Technology
Conference, 2 May-5 May 2005, Houston, Texas, Paper 17644.
6. Ruger, A.,1997. P-wave reﬂection coefﬁcients for transversely isotropic models
with vertical and horizontal axis of symmetry: Geophysics, vol. 62, No.3, p.713-
722.
7. Sheriff, R. E., 2002, Encyclopedic Dictionary of Applied Geophysics, 4th ed.,
Tulsa, OK: Society of Exploration Geophysicists, p.13.
8. Shuey, R.T., 1985. A simplification of the Zoeppritz equations: Geophysics,
vol.50, p.609-614.
73
9. Thomsen L., 1993, Weak anisotropic reflections, in Castagna J.P., Backus M.,
eds., Offset-dependent reflectivity: Theory and practice of AVO analysis:
Investigations in Geophysics 8: Tulsa, OK, Society of Exploration Geophysicists,
p. 103–111.
10. Thomsen, L., 1986, Weak elastic anisotropy: Geophysics, vol. 51, No.10, p.1954-
1966.
11. Thomsen, L., 2002. Understanding Seismic Anisotropy in Exploration and
Exploitation: SEG/EAGE Dist. Instrctr. Series No.5, Tulsa, OK : Society of
Exploration Geophysicists, p.1-1~1-34
12. Tsvankin, I. and Thomsen, L., 1994. Nonhyperbolic reflection moveout in
anisotropic media: Geophysics, vol.59, p.1290–1304.
13. Zoeppritz, K., 1919. Erdbebenwellen VIII B, On the reflection and penetration of
seismic waves through unstable layers: Goettinger Nachrichten, vol.I, p.66-84.
74
APPENDIX A
Comparison of synthetic produced from Hampson-Russell software with what is
computed from logs
This Appendix concerns only the synthetic traces, as constructed using the HR software.
We will see that their calculation raises some interesting issues.
The co-display of the Asyn, Bsyn, and Csyn picked from HR generated synthetic is (the
same as in the main text) :
intercept A ,gradient B and curvature C for the major synthetic horizons

0.6
Asyn
0.5 Bsyn
Csyn
0.4
0.3
Amplitude
0.2
0.1
-0.1
-0.2
400 450 500 550 600 650 700
time(ms)
Figure A-1. Amplitude intercept A, gradient B and curvature C for the major synthetic
horizons, zero phase
75
We may notice that the range of A parameter is from about 0.02 to 0.35, the range of B
parameter is from about -0.02 to -0.2, and the range of C parameter is from about 0 to 0.6.
The range of the C parameter doesn’t seem to be right, because based on equation
1  Z 0  1  Vp0  2Vs0   G0   2

2
1  Vp0  2
R ( )        sin     tan  sin 
iso 2
 
2  Z 0  2  Vp0  Vp0   G0   2  Vp0  (A.1)
 
 A  B sin   C tan  sin 
2 2 2
We get:
1  Z 0  1  2Vp02  1Vp01 
A    
2  Z 0  2  2Vp02  1Vp01 
1  Vp0  2Vs0   G0  

2
B      (A.2)
2  Vp0  Vp0   G0  
 
1  Vp0 
C  
2  Vp0 
In which Z0  Vp0 , G0  Vs02 , where the notation is the same as in the main text.
Compare the formula for the parameter A and C, we find that they should be around the
same value if the densities of two adjacent layers are not drastically different. However,
in Figure A-1, clearly the difference between A and C is serious for deeper horizon.
Evidently, their calculation is not as we have interpreted it to be.
Thus we choose another method to get the synthetic A, B, C parameters---we calculate
them directly from logs. Based on Eqn. (A.1), using well logs Vp, Vs and density. These
well logs begins from 135 ms to 690 ms. We get parameter Aref, Bref, Cref as :
We get parameter Aref, Bref, Cref as :
76
0.25
0.2
0.15
0.1
0.05
Aref
-0.05
-0.1
-0.15
-0.2
100 200 300 400 500 600 700
time (ms)
Figure A-2. Aref calculated from well logs
0.5
0.4
0.3
0.2
Bref
0.1
-0.1
-0.2
-0.3
100 200 300 400 500 600 700
time (ms)
Figure A-3. Bref calculated from well logs
77
0.2
0.15
0.1
0.05
Cref
-0.05
-0.1
-0.15
100 200 300 400 500 600 700
time (ms)
Figure A-4. Cref calculated from well logs
After exporting the zero phase wavelet converted from the seismic extracted wavelet,
which has been shown in Figure 7, to a readable file, we can convolve the wavelet with
parameter Aref, Bref, Cref. Then we get Asyn, Bsyn, Csyn as :
78
Synthetic A paramters computed from logs
0.5
0.4
0.3
0.2
0.1
-0.1
-0.2
0 100 200 300 400 500 600 700 800
time(ms)
Figure A-5. Asyn calculated from correlation of seismic wavelet and Aref
Synthetic B parameters computed from logs

0.3
0.2
0.1
-0.1
-0.2
-0.3
-0.4
0 100 200 300 400 500 600 700 800
time(ms)
Figure A-6. Bsyn calculated from correlation of seismic wavelet and Bref
79
Synthetic C paramters computed from logs
0.3
0.25
0.2
0.15
0.1
0.05
-0.05
-0.1
-0.15
-0.2
0 100 200 300 400 500 600 700 800
time(ms)
Figure A-7. Csyn calculated from correlation of seismic wavelet and Cref
By comparison, the Asyn and Csyn are almost the same, which makes them more reliable.
And if we plot the Synthetic ABC from the synthetic picks together with the ABC
calculated from correlation of seismic wavelet and logs-derived reflection coefficients for
comparisons, the results are shown in figures A-8~A-10:
80
Asyn comparison from log computed and HR picked
0.5
0.4
0.3
0.2
amplitude
0.1
-0.1
-0.2
0 100 200 300 400 500 600 700 800
time(ms)
Figure A-8. Asyn comparison from log computation and HR software
Bsyn comparison from log computed and HR picked

0.3
0.2
0.1
0
amplitude
-0.1
-0.2
-0.3
-0.4
0 100 200 300 400 500 600 700 800
time(ms)
Figure A-9. Bsyn comparison from log computation and HR software
81
Csyn comparison from log computed and HR picked
0.6
0.5
0.4
0.3
amplitude
0.2
0.1
-0.1
-0.2
0 100 200 300 400 500 600 700 800
time(ms)
Figure A-10. Csyn comparison from log computation and HR software
We find that the Asyn parameter from both the HR software generated and log
computation is almost identical, but the Bsyn and Csyn parameter is different for these two
methods. As we are more confident with the ones computed from the logs, in this thesis,
we adopt the log computed ones instead of the HR software-generated ones.
82
APPENDIX B
Matlab codes for making synthetic using zero-phase wavelet and well logs to extract
synthetic A,B,C
function [A,B,C,synlogABC]=syn0wav
% Well Name : Colony_WELL
% Kelly Bushing Elevation : 0 m
% Surface Elevation :0m
%X:0m Y:0m
% First Sample at : 135.000 ms
% Last Sample at : 690.000 ms
% Sample Rate : 1.00000 ms
% Number of Samples : 556
%
% Column 1 : Domain ms
% Column 2 : Density_FRM (Density) - Units : g/cc
% Column 3 : P-wave_corr_FRM (P-wave) - Units : us/m
% Column 4 : S-wave_cast_FRM (S-wave) - Units : us/m
logs=load('logs');% read the logs

time=logs(:,1);
density=logs(:,2);
Vp=1000000./logs(:,3);% The unit for Vp and Vs is converted to m/s
Vs=1000000./logs(:,4);
A=zeros(556);% sampled 1ms from 135ms t0 690ms, there are 556 points in total
for i=2:555
A(i)=(density(i+1).*Vp(i+1)-density(i-1).*Vp(i-1))./(density(i+1).*Vp(i+1)+density(i-
1).*Vp(i-1));% reflection coefficient intercept A
end
B=zeros(556);
for i=2:555
shearmodulus2=density(i+1).*Vs(i+1).^2;
shearmodulus1=density(i-1).*Vs(i-1).^2;
B(i)=(Vp(i+1)-Vp(i-1))./(Vp(i+1)+Vp(i-1))-((shearmodulus2-
shearmodulus1)/(shearmodulus2+shearmodulus1)).*(2.*(Vs(i+1)+Vs(i-
1))./(Vp(i+1)+Vp(i-1))).^2;% reflection coefficient gradient B
end
C=zeros(556);
for i=2:555
C(i)=(Vp(i+1)-Vp(i-1))./(Vp(i+1)+Vp(i-1));% reflection coefficient curvature C
end
83
wav=load('wave_zero_seis');% load in the zero-phase seismic extracted wavelet
[i,j]=size(wav);% i records the length of the wavelet when sampled 2ms
Maxwav=max(wav);% normalize the wavelet to make its peak is 1
wav=wav./Maxwav;
synalog=conv(A(1:2:555),wav(1:i));% because the wavelet is sampled 2ms, so we need

resample the A,B,C by 2ms.the conv is the convolution function in matlab
synblog=conv(B(1:2:555),wav(1:i));
synclog=conv(C(1:2:555),wav(1:i));
[T,ABCsyn]=loadsyn;% load in the ABC calculate from Synthetic pickings and their
time locations in T
for i=1:6 % we already know there are 6 horizons in the synthetic picks
synlogABC(i,:)=[synalog((T(i)-36)./2+1),synblog((T(i)-36)./2+1),synclog((T(i)-
36)./2+1)]% 36ms is the beginning time location of the convolved synthetic, from the
header of the 'wave_zero_seis', the wavelet begins at -99ms,135-99=36
end
Restrictions
Copyright, Allied Geophysical Laboratories University of Houston, Houston TX,
USA Royalty free use for AGL sponsors and co-investigators for use in research,
exploration, with partners, host governments, and for provision of
processing/interpretation service to sponsor clients Redistribution, sale, or
inclusion of this software in software products outside the sponsor worksite
requires a separate commercialization agreement with the University of Houston.
Author
Rongrong Lin, University of Houston. Version April, 2013.
84
APPENCIX C
Matlab codes for extracting anisotropy parameters delta and epsilon from synthetic
and seismic A,B,C
function [Bseis,Cseis]=anisoparameter
[A,B,C,ABCsyn]=syn0wav;% load in synthetic A,B,C from logs computed synthetic in

function syn0wav, as well as the reflection coefficient intercept, gradient and curvature
A,B,C
[T,ABCsei0]=loadseis;% load in seismic A,B,C and their time locations in T
NA=mean(ABCsyn(:,1))/mean(ABCsei0(:,1));% NA is the ratio of the average synthetic
amplitude and the average seismic amplitude for each horizon
for i=1:6 % we chose 6 horizons in total

ABCseis(i,:)=ABCsei0(i,:).*NA;% the new seismic is normalized by the scale factor
end
NB=ABCsyn(:,2)./ABCseis(:,2);% the normalization factor NB

NC=ABCsyn(:,3)./ABCseis(:,3);% the normalization factor NC
NBabs=abs(ABCsyn(:,2)./ABCseis(:,2)); % the absolute value of normalization factor

NB
NCabs=abs(ABCsyn(:,3)./ABCseis(:,3)); % the absolute value of normalization factor
NC
%Get the least-square linear solution for aT+b=NBabs, convert the equation to
Gb*mb=NBabs,solve for mb, then Gb*mb will be the least square linear approximation
of NBabs, which is the low cut of NB-NBlow
Gb=[T,ones(6,1)];
mb=Gb\NBabs;
NBlow=Gb*mb;
% Solve for NClow using the same method applied to NBlow

Gc=[T,ones(6,1)];
mc=Gc\NCabs;
NClow=Gc*mc;
Ti=(T-418)./2+1;%Ti is an index of time location,418 is the beginning time.

Bseis=ABCseis(:,2).*NBlow(Ti)';% Bseis is the low-cut seismic B
dB=real(Bseis-ABCsyn(:,2));% dB is the difference between low-cut seismic B and
synthetic B
85
Cseis=ABCseis(:,3).*NClow(Ti)';% Cseis is the low-cut seismic C
dC=real(Cseis-ABCsyn(:,3));% dC is the difference between low-cut seismic C and
synthetic C
ddelta=2*dB;% ddelta is the delta delta parameter
depsilon=2*dC;% depsilon is the delta epsilon parameter
D=zeros(7,1);%D stores the value of Delta parameter

D(6)=0;% gamma ray displays smallest value at 650 ms, as the time range for the
horizons is 413mms to 693ms, it is at the location of the 6th horizon
for i=1:5
D(6-i)=D(7-i)-ddelta(6-i);
end
for i=6
D(i+1)=D(i)+ddelta(i);
end
E=zeros(7,1);%E stores the value of Epsilon
E(6)=0;
for i=1:5
E(6-i)=E(7-i)-depsilon(6-i);
end
for i=6
E(i+1)=E(i)+depsilon(i);
end
T2=zeros(7,1); % T2 records the time location for anisotropy parameters D and E

T2(1)=T(1)-5;
T2(7)=T(6)+5;
for i=2:6
T2(i)=0.5*(T(i-1)+T(i));% use the average of time locations of two adjacent anisotropy
jumps as the time for the anisotropy itself
end
Restrictions
Copyright, Allied Geophysical Laboratories University of Houston, Houston TX,
USA Royalty free use for AGL sponsors and co-investigators for use in research,
exploration, with partners, host governments, and for provision of
processing/interpretation service to sponsor clients Redistribution, sale, or
inclusion of this software in software products outside the sponsor worksite
requires a separate commercialization agreement with the University of Houston.
Author
Rongrong Lin, University of Houston. Version April, 2013.
86
APPENDIX D
Correlation of the parameter with gamma ray log from other literature
From the discussion in the main text, we see that the  parameters we get correlate well
with the gamma ray log we have in Colony well. (See Figure D-1, which is the same as
Figure 40 in the main text.)

1000
1200
1400
1600
depth(ft)
1800
2000
2200
2400
-0.2 0 0.2 0.4 0.6 0.8
Figure D-1. Anisotropy parameter  variation with gamma ray log display
87
However, the gamma ray log is only from 1750ft to 2250ft, we don’t have any gamma
ray log information for above 1750ft, so we look for the gamma ray log for Colony sand
in other literature.
In the CREWES annual report ’Exploitation of an oil field using AVO and post-stack
rock property analysis methods’ by Andrew J. Royle (2001), we find a well 10-14 (See
Figure D-2) drilled in the Colony sand area with a similar Colony sand layer as in our
well log in the Colony sand dataset provided in the Hampson-Russell software as a
default dataset.
Figure D-2. Anisotropy parameter  variation with gamma ray log display
88
Although this well is at a different location in the Colony field, we may use the data to
extend our analysis to shallower depth, by analogy, since the stratigraphy in the area is
quite flat. Figure D-3 shows a highlighted box (381m to 533m) of this gamma ray log,
compared with our calculated  function in the corresponding depth interval (1250ft-
1750ft, equivalent to 381m to 533.4m). The minimum in the gamma ray reading near
440m correlates with the minimum in  function at 1400ft(equivalent to 426.72m),
giving more confidence to our method.
However, we note that the correlation is flawed, since we have not considered the
possibility of lateral variation. Other logs from the area show different trends and the
present analysis should be repeated in full, for each of the logs, for a satisfactory
conclusion.
89
1000
1200
1400
1600
depth(ft)
1800
2000
2200
m
2400
-0.2 0 0.2 0.4 0.6 0.8
Figure D-3. Anisotropy parameter  (from 1250ft to 1750ft, equivalent to from 381m to
533.4m ) correlation with well 10-14 gamma ray log (from 381m to 533.4m)
90

Lin Thesis 2013

Uploaded by

Copyright:

Available Formats

Lin Thesis 2013

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Lin Thesis 2013

Uploaded by

Copyright:

Available Formats

EXTRACTING POLAR ANISOTROPY PARAMETERS FROM

SEISMIC DATA AND WELL LOGS

by Rongrong Lin _______________________

University of Calgary for their help in my first-stage data searching.

University of Houston is also greatly appreciated.

Finally, I want to show my gratefulness to my family and my friends. Without their

it in frequency domain using a cut-off frequency to be determined empirically. The low-

Figure 1. Vp,Vs, and density logs from Colony well……………………………...……………..8

(1.1) V p ( )  V p 0 [1   sin 2 con 2   sin 4  ] ……………………………….…….….2

(2.10) N ( , t )  P( , t )  1 ………………………………………………………..……...28

Asyn  Asyn (t0 ) ;

Cseislow (t0 )  NClow (t0 )  Cseis

1  Z 0  1  Vp0  2Vs0   G0   2

1  Vp0  2Vs0   G0  

Anisotropy is the variation of a physical property depending on the direction in which it

anisotropy mostly refers to seismic anisotropy. Seismic anisotropy is the dependence of

isotropic to more realistically anisotropic. Seismic anisotropy has played roles in

angle-dependence of velocity is more evident), AVO (Amplitude versus offset)

quantitative analysis, and anisotropic migration. (Thomsen, 2002)

vertical, polar anisotropy is also referred to as vertical transverse isotropy (VTI).

generates complicated equations. Thomsen used the combinations of elastic modules to

most common case, then:

V p ( )  V p 0 [1   sin 2 con 2   sin 4  ] (1.1)

Vs|| ( )  Vs 0 [1   sin 2  ] (1.3)

(C13  C44 )2  (C33  C44 )2 (1.5)

vertical anisotropy, near-horizontal anisotropy, and SH anisotropy.

These three anisotropy parameters are useful because:

 The anisotropy parameters reduce to zero when degenerating to isotropic case.

1.1 Recent methods to get anisotropy parameters and their drawbacks

likely to exhibit its original anisotropy when buried underground.

the seismic wavelet).

One method uses the hyperbolic move-out is based on the equation:

(Alkhalifah and Tsvankin, 1995)

analysis, by short-spread, we mean mute the CDP gather at offset=depth.  is dependent

on the anisotropy parameters δ and ε (Thomsen, 1986):

The parameter  can be obtained from formula

v NMO (0  )  VP (0  )(1  2 )1 / 2 (1.9)

the small parameters  and δ may be estimated with confidence.

waves (Patterson and Tang, 2005).

resolution of the seismic wavelet.

the logged section of the well.

method follow that.

2.2 Data preparation

2.2.1 Well logs loading

VP log. Castagna’s equation is:

by performing Fluid Replacement Modeling in the Hampson-Russell AVO package. In

(2.1) and modified by Fluid Replacement Modeling. P-wave_corr_FRM means the P-

discussion about the correlation later.

2.2.2 Seismic data loading

velocity is different. That is because their reflection angle is different ( 1   2   3 ).

Figure 2.Near well site P-wave reflections from anisotropic medium

The geometry display is simple:

Figure 3. Geometry of Colony sand seismic data

Figure 4.CDP gathers for Colony sand seismic data

the position of CDP 330.

this case we average over 5 CDP positions near the well.

spectrum (amplitude and phase).

spectrum as the input.

Figure 8.The frequency spectrum of the zero-phase wavelet

deconvolved to this wavelet.