building tilted transversely isotropic depth models using...
TRANSCRIPT
Bl
A
sptmtpssHdbmw2
©
GEOPHYSICS, VOL. 75, NO. 4 �JULY-AUGUST 2010�; P. D27–D36, 16 FIGS.10.1190/1.3453416
uilding tilted transversely isotropic depth models usingocalized anisotropic tomography with well information
ndrey Bakulin2, Marta Woodward1, Dave Nichols1, Konstantin Osypov1, and Olga Zdraveva1
tltnpoiarmtsdcs
ABSTRACT
Tilted transverse isotropy �TTI� is increasingly recognized as amore geologically plausible description of anisotropy in sedi-mentary formations than vertical transverse isotropy �VTI�. Al-though model-building approaches for VTI media are well un-derstood, similar approaches for TTI media are in their infancy,even when the symmetry-axis direction is assumed known. Wedescribe a tomographic approach that builds localized anisotrop-ic models by jointly inverting surface-seismic and well data. Wepresent a synthetic data example of anisotropic tomography ap-plied to a layered TTI model with a symmetry-axis tilt of 45 de-grees. We demonstrate three scenarios for constraining the solu-tion. In the first scenario, velocity along the symmetry axis isknown and tomography inverts for Thomsen’s � and � parame-
ste
tpuiwructT2
d
[email protected], Saudi
D27
Downloaded 21 Aug 2010 to 198.36.39.3. Redistribution subject to S
ers. In the second scenario, tomography inverts for �, � , and ve-ocity, using surface-seismic data and vertical check-shot travel-imes. In contrast to the VTI case, both these inversions are nonu-ique. To combat nonuniqueness, in the third scenario, we sup-lement check-shot and seismic data with the � profile from anffset well. This allows recovery of the correct profiles for veloc-ty along the symmetry axis and �. We conclude that TTI is morembiguous than VTI for model building. Additional well data orock-physics assumptions may be required to constrain the to-ography and arrive at geologically plausible TTI models. Fur-
hermore, we demonstrate that VTI models with atypical Thom-en parameters can also fit the same joint seismic and check-shotata set. In this case, although imaging with VTI models can fo-us the TTI data and match vertical event depths, it leads to sub-tantial lateral mispositioning of the reflections.
INTRODUCTION
Vertical transverse isotropy �VTI� is a useful approximation to de-cribe anisotropic subsurface formations. However, widespread ap-lication of VTI depth imaging often suggests that the direction ofhe symmetry axis may not be vertical. Indeed, if sedimentary for-
ations were deposited in a layer-cake geometry and later folded byectonic forces, then tilted transverse isotropy �TTI� with the axiserpendicular to the bedding plane might be a more appropriate de-cription. Such media are sometimes referred to as demonstratingtructurally conformant transverse isotropy �Audebert et al., 2006�.owever, in an alternative geologic scenario, tectonic action andeposition may occur together, and the symmetry axis may no longere perpendicular to the bedding plane. In yet another scenario, sedi-ents may be subjected to anomalous stresses around salt bodies,hich can result in stress-induced anisotropy �Sengupta et al.,009�. In this case, because symmetry is controlled by principal
Presented at the 79thAnnual International Meeting, Society of ExplorationManuscript received by the Editor 31 December 2009; revised manuscript1WesternGeco/Schlumberger, Houston, Texas, U.S.A. E-mail: woodward22Formerly WesternGeco/Schlumberger; presently SaudiAramco, Dhahran2010 Society of Exploration Geophysicists.All rights reserved.
tress directions that are not vertical, horizontal, or perpendicular tohe bedding axis, general TTI or even an orthorhombic medium isxpected.
For any of these scenarios, we should be able to build an aniso-ropic model for depth imaging that is governed by a geologicallylausible anisotropic velocity field. In practical circumstances, thissually requires supplementing seismic data with some kind of wellnformation �Bear et al., 2005�. For a VTI case, constraining verticalell data in the form of a check-shot survey or depth markers usually
esolves the existing ambiguity �around the well� and delivers anique depth model that fits all of the data. Therefore, it has becomeommon industry practice to estimate anisotropic parameters withhe aid of borehole information �Sexton and Williamson, 1998;svankin, 2001; Morice et al., 2004; Bear et al., 2005; Huang et al.,007�.
The information content of the well and seismic data is well un-erstood for layered VTI models, and, from this case, most ap-
ysicists.d 24 February 2010; published online 2August 2010.om; [email protected]; [email protected]; [email protected]. E-mail: [email protected].
EG license or copyright; see Terms of Use at http://segdl.org/
psaastmtc
tia
1
2
TwmeiwawVlt
fihmP
uHafir
rpiudotatsa�
aato2aclmMuwfu
jrs
1
2
rTiclitjtnm
a
Ffrs
D28 Bakulin et al.
roaches and physical intuition originate. However, for TTI models,uch a practical foundation has not been established, even though annalytical description of various seismic signatures for TTI media isvailable �Tsvankin, 2001�. Figure 1 shows some typical practicalcenarios where dipping sediment layers may require a TTI descrip-ion. It is imperative that we be able to calibrate our models and esti-
ate TTI parameters in the presence of complex structures. In addi-ion, well data often come from deviated boreholes, thus requiring aalibration method to handle this additional complexity.
Bakulin et al. �2009a� propose using conventional gridded reflec-ion tomography to invert seismic and well data jointly for a local an-sotropic VTI depth model. We extend this technique to TTI mediand attempt to answer two questions:
� Is it feasible to invert for a local TTI model using certain combi-nations of seismic and well data?
� Can we distinguish TTI and VTI models using a joint seismicand well data set alone?
he answers obviously depend on the model geometry. Therefore,e focus on the simplest case of a layered model where we can deter-ine analytically if the inverse problem is unique and which param-
ters can be estimated. We start by describing the approach of local-zed gridded anisotropic tomography that inverts a joint seismic andell data set. Then we apply it to a synthetic example where we buildlocal TTI model. We observe some ambiguities and explain themith an analytical description of seismic signatures. Next, we build aTI model that fits the same data set and explain why this does not
ead to obvious contradictions. At the end, we discuss the implica-ions of imaging a TTI subsurface with a VTI velocity model.
LOCALIZED REFLECTION TOMOGRAPHY
Postmigration anisotropic gridded reflection tomography of sur-ace seismic data is a well-established tool to build velocity modelsn depth �Woodward et al., 1998, 2008; Zhou et al., 2004�. For bore-ole data, anisotropic traveltime tomography in the premigrated do-ain is a more typical model-building approach �Chapman andratt, 1992; Pratt and Chapman, 1992�. Common industry practice
Well VSP
Acousticlog
TTI
Well markers
Salt
TTI
Seismic horizons
TTIWell
) b)
igure 1. Sketch of various scenarios that involve a complex subsur-ace and deviated wells, which would require estimation of TTI pa-ameters aided by well information: �a� sediment anticline; �b� sub-alt targets.
Downloaded 21 Aug 2010 to 198.36.39.3. Redistribution subject to S
ses reflection tomography to build anisotropic velocity models.owever, typically we only invert for velocity along the symmetry
xis; the remaining anisotropic velocity field parameters are keptxed. They are derived beforehand using manual well calibrations,egional trends, or other nontomographic methods.
In the case of VTI media, Thomsen parameters � and � can be de-ived near a well location using a manual layer-stripping approach,rovided the seismic data are supplemented by check-shot or markernformation. Then, such parameters are propagated into a volumesing geologic horizons as a guide. In the case of TTI media, two ad-itional parameters such as tilt and azimuth of the symmetry axis areften derived from the seismic dip field under the assumption thathe symmetry axis is perpendicular to the geologic beds �Audebert etl., 2006� or some other simple geometric relationships. Such prac-ices stem from an understanding that inverting simultaneously foreveral parameters of the anisotropic velocity field is almost alwaysn underconstrained problem when seismic data alone are usedTsvankin, 2001�.
Quite simply, we must acknowledge that it is impossible to derivenisotropic parameters away from well control using seismic datalone.As a result, practical focus shifts toward designing techniqueshat estimate anisotropic parameters with the help of well data. Vari-us methods have been proposed to handle this problem �Tsvankin,001; Morice et al., 2004; Bear et al., 2005; Huang et al., 2007; Wangnd Tsvankin, 2009�; however, most of them are designed for verti-al wells and VTI media. In addition, most of them do not easily re-ate to the common postmigration approaches used for conventional
odel building. Although manual 1D layer-stripping inversion byorice et al. �2004� and 1D joint inversion by Huang et al. �2007�
se prestack depth-migrated data, they still apply only to VTI mediaith flat layers. Therefore, most existing approaches cannot be used
or anisotropic calibration in the typical 3D scenarios shown in Fig-re 1.
Bakulin et al. �2009a� propose using localized multiparameteroint inversion of seismic and well data using conventional griddedeflection tomography. Multiparameter inversion becomes con-trained and feasible under three conditions:
� The seismic data set is inverted jointly with one or more typesof well data �velocity measured in wells, traveltimes measuredfrom check shot or vertical seismic profiling �VSP�, markers, orother�.
� The inversion is performed in a local volume around the wellwhere well data are assumed to remain valid.
We must emphasize that independent inversion for multiple pa-ameters at each grid cell away from the well is highly nonunique.herefore, we add a shape constraint to the parameter update by us-
ng preconditioning to smooth and steer the solution along geologi-al layers away from the well. This strategy prevents uncontrolledateral variation of the anisotropic parameters that would make thenverse problem highly unstable. There are multiple practical advan-ages to using conventional reflection tomography for simultaneousoint inversion. The most important is that the constraints can be in-roduced in a flexible way using the same gridded model without theeed to rebuild or regrid the model every iteration, as is the require-ent in many other sequential techniques.
EG license or copyright; see Terms of Use at http://segdl.org/
sTTTa1tt
afwocsao
tVsrdl
tUef
1
23
Tc
T
dVpoAstcusi
Tc
w1tleltb
tphorot
TcT
kpak
W
liTs�
Htnmqa
zaenual
Localized TTI tomography D29
SYNTHETIC EXAMPLE
Let us apply anisotropic tomography with well constraints to aimple deepwater model. The subsurface is represented by layeredTI sediment with a uniform symmetry-axis tilt of 45° �Figure 2�.he model has smooth vertical variation of velocity and anisotropy.wo pronounced velocity inversions are present in the model. Theyre also accompanied by anisotropy reductions. A cable length of2 km is assumed. A prestack gather computed with anisotropic rayracing is shown in Figure 2c. Reflected events from 49 density-con-rast interfaces are located every 200 m.
We assume a certain orientation of the symmetry axis and applynisotropic reflection tomography �Woodward et al., 2008� to solveor a local anisotropic model using joint inversion of seismic andell data. Well data may come in various forms, but in this study wenly examine velocity measured along the well �acoustic log� orheck-shot survey. We apply a mute of 50° to the data before inver-ion. This limits the useable offsets to less than 8 km for eventsbove 6 km; however, even with a maximum offset of 12 km, half-pening angles remain less than 40° for events deeper than 8 km.
First, we attempt to build a TTI velocity model assuming that therue orientation of the symmetry axis is known. Second, we attempt aTI inversion of the seismic and check-shot data, pretending that the
ymmetry axis is vertical. In each case, we benchmark tomographicesults against an analytic description of seismic signatures and vali-ate the results. Finally, we contrast TTI and VTI results and high-ight the consequences of replacing TTI media with VTI media.
BUILDING A LOCAL TTI MODEL
It is a common practice not to invert for orientation of the symme-ry axis. Let us assume that the true orientation is known and fixed.nder this assumption, we proceed with building the local TTI mod-
l using seismic and well data. We consider three different scenariosor the well data:
� accurate knowledge of symmetry-axis P-wave velocity VP0 de-rived from an acoustic log in a deviated well, a virtual checkshot, or an offset well
� check-shot survey in a vertical well� check-shot survey in a vertical well and correct profile of
Thomsen’s � from an offset well.
First, we present the tomographic results for all three scenarios.hen we attempt to explain them using theoretical analysis and dis-uss the differences between these scenarios.
wo-parameter inversion after fixing VP0
Let us assume that well data are available from a deviated wellrilled along the TTI symmetry axis �45° to the vertical in our case�.elocity along the well can be estimated from acoustic logs or byerforming virtual check shots �Mateeva et al., 2006�. Alternatively,ne may utilize a VP0 profile derived with the help of an offset well.fter fixing VP0 to its correct values, we attempt tomographic recon-
truction of � and � from long-offset reflection seismic data. Al-hough such an inversion would be unique and stable for the VTIase, for this TTI case only an approximate model is recovered �Fig-re 3�. Individual values of the Thomsen parameters �� in particular�how errors of 0.05 and more. Nevertheless, the final model providesmage gathers that appear as flat as in the true model �Figure 10b�.
Downloaded 21 Aug 2010 to 198.36.39.3. Redistribution subject to S
hree-parameter inversion of seismic and verticalheck-shot data
In the second scenario, we assume the availability of a verticalell with a check-shot survey acquired every 50 m from.5 to 11 km. We jointly invert seismic and check-shot data forhree parameters �VP0, �, and � � around the well. Because we haveong-spread data, such an inversion would result in a unique recov-ry of the true model in a VTI case. To our surprise, TTI inversioneads to a different model �Figure 4� that provides a reasonable fit tohe check-shot survey �Figure 5� and that flattens the image gathersut that has geologically implausible � and � profiles.It may seem that the unfortunate model we arrived at is caused by
he poor initial guess of anisotropy. To investigate it further, let us re-eat the same inversion but start with a much better initial model thatas ��0.08 and � �0.03. In this case, we again can easily match allf the data and this time arrive at a model with positive Thomsen pa-ameters; however, this model is still quite far removed from the truene �Figure 6�. These results suggest deeper underlying reasons forhe observed ambiguity.
wo-parameter inversion of seismic and verticalheck-shot data after fixing the correct profile ofhomsen’s � from an offset well
In the third scenario, we supplement the vertical check shot withnowledge of the correct profile of � from an offset well. Tomogra-hy performs a two-parameter inversion �VP0 and �� of the seismicnd check-shot data and recovers excellent estimates of the un-nown parameters at all depths �Figure 7�.
eak-anisotropy analysis of the results
Why do such different models provide similar fits to this seeming-y complete data set? To obtain analytical insight into the problem, its instructive to obtain weak-anisotropy expressions for all P-waveTI signatures at hand. For a single horizontal TTI layer with a 45°ymmetry-axis tilt, these signatures are expressed as followsTsvankin, 2001; Pech et al., 2003�:
VNMOTTI �VP0
TTI�1�1.25�TTI�0.75� TTI�, �1�
VvTTI�VP0
TTI�1�0.25��TTI�� TTI��, �2�
A4�2�TTI
tP02 �VP0
TTI�4 . �3�
ere, VP0TTI, � TTI, and �TTI are the three independent Thomsen parame-
ers that describe the TTI velocity field; �TTI��TTI�� TTI; VvTTI de-
otes velocity in the true vertical direction; VNMOTTI describes the nor-
al moveout �NMO� velocity from a horizontal reflector; and A4 is auartic moveout coefficient describing P-wave traveltime behaviort long offsets �Tsvankin, 2001�.
Various numerical coefficients arise after substituting values ofero reflector dip and tilt of the symmetry axis �45°�. In the first ex-mple �Figure 3�, the information constrained by the seismic data isquivalent to equations 1 and 3. Indeed, we observe that the combi-ation �1.25��0.75� � that controls VNMO
TTI is best determined in Fig-re 3d. Parameter �TTI is less well determined. This observation isnalogous to a VTI case where the trade-off between VNMO and A4
eads to substantial uncertainty in the quartic coefficient and thus in
EG license or copyright; see Terms of Use at http://segdl.org/
Fmsd�t
F�temt�idr
Fsseebtpweores�s�
D30 Bakulin et al.
24
68
10
2500 50007500 10000
Tim
e(s
)
Offset (m)0
2000
4000
6000
8000
10,000
12,0002000 3000 4000
Dep
th(m
)
V (m/s)P0
0
2000
4000
6000
8000
10,000
12,000–0.1 0 0.1 0.2Thomsen parameters
Dep
th(m
)
a) b) c)
η
�ε
igure 2. Deepwater horizontally layered TTIodel used for tomography: �a� velocity VP0 �along
ymmetry axis, which is at 45° to the right of theownward vertical axis�; �b� Thomsen parameters;c� prestack gather. Reflections arise from 49 densi-y interfaces. Water depth is at 1500 m.
0
2000
4000
6000
8000
10,000
12,000–0.1 0 0.1
Dep
th(m
)
0 0.05 0.1 0.15
0
2000
4000
6000
8000
10,000
12,0000 0.05 0.1
η
0
2000
4000
6000
8000
10,000
12,000
0
2000
4000
6000
8000
10,000
12,0000 0.05 0.1 0.15
a) b) c) d)
InitialTrue1st2nd3rd
InitialTrue1st2nd3rd
InitialTrue1st2nd3rd
InitialTrue1st2nd3rd
igure 3. Results of a two-parameter TTI inversion� and � � of seismic data after fixing velocity alonghe symmetry axis �VP0�. Anisotropy profiles afterach iteration are shown with initial �zero� and trueodels: �a� � ; �b� �; �c� � ; �d� parameter combina-
ion �1.25��0.75� � that controls NMO velocityequation 1�. Although this parameter combinations tightly constrained as seen in �d�, � and � areetermined with errors. In particular, � is not wellesolved.
δ ε 1.25 ε –0.75 δ
0
2000
4000
6000
8000
10,000
12,000–400 –200 0
Dep
th(m
)
V Vinitial- (m/s)–0.2 –0.1 0 0.1
δ ε η
0
2000
4000
6000
8000
10,000
12,000
0
2000
4000
6000
8000
10,000
12,000–0.2 0 0.2
0
2000
4000
6000
8000
10,000
12,0000 0.05 0.1
a) b) c) d)
InitialTrue1st2nd3rd
InitialTrue1st2nd3rd
InitialTrue1st2nd3rd
InitialTrue1st2nd3rd
igure 4. Results of a three-parameter TTI inver-ion �VP0, �, and � � of seismic and vertical check-hot data. Velocity and anisotropy profiles afterach iteration are shown with initial and true mod-ls: �a� update in velocity shown as the differenceetween current velocity at each iteration and ini-ial velocity profile; �b� � ; �c� �; �d� � . Whereas thearameter combination � ���� is relativelyell constrained, tomography recovers one of the
quivalent models with incorrect VP0, � , and �. Inur case, � plays the same role as � in the VTI case,elating vertical velocity and VNMO according toquation 4. Even though vertical velocity Vv is con-trained by the check shot, we are unable to resolve
and � individually because, in our case, bothhort- and long-spread moveouts are controlled by�see equations 3 and 4�.
Downloaded 21 Aug 2010 to 198.36.39.3. Redistribution subject to SEG license or copyright; see Terms of Use at http://segdl.org/
�w�r
iea
Intctca
pe
natlst�
W
fi
a
a
Localized TTI tomography D31
�Tsvankin, 2001�. We expect the individual parameters to be lessell determined because combinations �1.25��0.75� � and ��� � from equations 1 and 3 are too similar to constrain them sepa-
ately, thus creating additional ambiguity.For the second case when check-shot data are available, we add
nformation equivalent to equation 2. It is instructive to combinequations 1 and 2 and thus rewrite equation 1 in the following weak-nisotropy form:
VNMOTTI �Vv
TTI�1��TTI� . �4�
t becomes obvious that three measurements �VNMOTTI , Vv
TTI, and A4� doot constrain all three parameters �VP0
TTI, �TTI, and � TTI� because equa-ions 4 and 3 constrain only �TTI, whereas equation 2 constrains aombination of all three desired quantities. In the absence of addi-ional information, tomography retrieves an equivalent model withorrect �TTI and Vv
TTI but incorrect individual parameters VP0TTI, �TTI,
nd � TTI. When � TTI is additionally constrained as in the third exam-
0
2000
4000
6000
8000
10,000
12,000−10 0 10
Check-shot misfit (ms)
1st2nd3rd
0
2000
4000
6000
8000
10,000
12,0000 50 100 150Check-shot misfit (ms)
Initial
Dep
th(m
)
) b)
0
2000
4000
6000
8000
10,000
12,000–100 0 100 200 300
Dep
th(m
)
V V– (m/s)P0 P0,initial
0
2000
4000
6000
8000
10,000
12,0000 0.1 0.2 –0.05
ε δ
0
2000
4000
6000
8000
10,000
12,0000 0.05 0.1
) b) c)
InitialTrue1st2nd3rd
InitialTrue1st2nd3rd
InitialTrue1st2nd3rd
Downloaded 21 Aug 2010 to 198.36.39.3. Redistribution subject to S
le �Figure 5�, then tomography correctly recovers VP0TTI and �TTI, as
xpected from the weak-anisotropy equations above.
BUILDING LOCAL VTI MODEL
Now let us assume that the true orientation of the symmetry axis isot known. We further assume that the axis of symmetry is verticalnd invert vertical symmetry-axis velocity directly from check-shotraveltimes recorded in a vertical well. After fixing the vertical ve-ocity, we perform a two-parameter inversion �� VTI and �VTI� of theeismic data and recover the resultant profiles �Figure 8�. This solu-ion has positive � VTI but almost zero �VTI and therefore negative
VTI.
eak-anisotropy analysis of VTI inversion
To understand this result, let us predict what VTI model can bestt our TTI data �Figure 9�. To satisfy the same check-shot survey, we
igure 5. Misfit in check-shot traveltimes for �a� the initial modelnd �b� all subsequent tomography iterations for the second scenariof joint three-parameter inversion of seismic and check-shot data.isfit is computed as the difference between experimental and pre-
icted traveltimes.
η
0
0
0
0
0
0
00 0.05 0.1
InitialTrue1st2nd3rd
Figure 6. Same as Figure 4 but for a different initialmodel with � �0.03 and ��0.08. Results of athree-parameter TTI inversion �VP0, �, and � � ofseismic and vertical check-shot data. Velocity andanisotropy profiles after each iteration are shownwith initial and true models: �a� update in velocityshown as a difference between current velocity ateach iteration and initial velocity profile; �b� �; �c�� ; �d� � .
FaoMd
200
400
600
800
10,00
12,00
d)
EG license or copyright; see Terms of Use at http://segdl.org/
mae
watsc�m
zvcir
L
dt
F�aawatv
Faaoe�t
D32 Bakulin et al.
ust require that VP0VTI�Vv
TTI. To match the same seismic data, welso require that VNMO
VTI �VNMOTTI . Comparing normal moveout �NMO�
quation 4 with the corresponding VTI equation
VNMOVTI �VP0
VTI�1�� VTI�, �5�
e derive that � VTI��TTI. Finally, we deduce that the horizontalnd vertical VTI velocities must be equal to each other �VP90
VTI�VP0VTI�
o match the TTI velocity surface that is symmetric around the 45°ymmetry axis �Figure 9�. This implies that �VTI�0. Therefore, wean conclude that the VTI medium with parameters VP0
VTI�VvTTI,
VTI��TTI, and �VTI�0 has these three signatures identical to a TTIodel with a 45° tilt of the symmetry axis: velocity along vertical
0
2000
4000
6000
8000
10,000
12,000−400
V V−P0
Dep
th(m
)
a)igure 7. Results of a two-parameter TTI inversionVP0 and �� of seismic and vertical check-shot datafter fixing the � profile to true values. Velocity andnisotropy profiles after each iteration are shownith initial and true models: �a� update in velocity
long symmetry axis shown as a difference be-ween current velocity at each iteration and initialelocity profile; �b� �; �c� � .
−0.1
0
2000
4000
6000
8000
10,000
12,000
Dep
th(m
)
a)igure 8. Results of a two-parameter inversion ��nd � � of seismic data after assuming a VTI modelnd fixing vertical velocity to correct values basedn the check shot. Anisotropy profiles after each it-ration are shown with initial and true models: �a�; �b� �; �c� � . Note that recovered � is quite close
o the true � , whereas � is close to zero.
Downloaded 21 Aug 2010 to 198.36.39.3. Redistribution subject to S
-axis, moveout velocity, and horizontal velocity. Profiles of the in-erted parameters in Figure 8 are quite consistent with these analyti-al predictions, thus validating the numerical solution. These find-ngs explain the excellent fit of the same data by TTI and VTI tomog-aphic solutions.
DISCUSSION
evel of ambiguity in model building
As expected, all of the TTI and VTI models we derived fit the jointata set consisting of seismic and well data. Figure 10 verifies thatheir migrated common-image-point gathers are not only flat but
0(m/s)
lε η
0
2000
4000
6000
8000
10,000
12,0000 0.05 0.1 0.15
InitialTrue1st2nd3rd
0
2000
4000
6000
8000
10,000
12,000−0.05 0 0.05 0.1
b) c)
0.1 −0.2 0 0.2ε
0
2000
4000
6000
8000
10,000
12,000−0.1 0 0.1
η
InitialTrue1st2nd3rd
0
2000
4000
6000
8000
10,000
12,000
b) c)
−200
P0, initia
δ0
EG license or copyright; see Terms of Use at http://segdl.org/
aoiwl
vmtlrfsvnacsms
C
utasat
4os“tVtpV
fldrw
uegvaa7f
Vtz
C
nm
misnlthm1eT
sA
Fvsflt
Fmmn�m
Localized TTI tomography D33
lso have reflections positioned at the same depth with an accuracyf about 10 m. Therefore, they are indistinguishable from the focus-ng and vertical positioning points of view. In mathematical terms,e can say that they all are in the null space of the tomography prob-
em.It is further obvious from the theoretical analysis and from the
ariability of the TTI solutions that these are probably not the onlyodels that fit the data. In other words, we did not fully characterize
he TTI null space — we just sampled several members of a mucharger set. If we were to use more sophisticated tools such as tomog-aphy with uncertainty analysis, we would be able to obtain an entireamily of kinematically equivalent TTI models that make up the nullpace �Osypov et al., 2008; Bakulin et al., 2009b�. An obvious ad-antage of such an approach is the ability to reveal a high level ofonuniqueness using only a single inversion. For instance, in the ex-mple of a TTI inversion of seismic and check-shot data, we easilyould have detected that Thomsen parameters are not well con-trained �Bakulin et al., 2009b� and then have determined additionaleasurements or information required to obtain a more constrained
olution.
an we distinguish TTI or VTI?
We also examined a key decision in model building: whether tose VTI or TTI parameterization. We have demonstrated on a syn-hetic example that, for a general TTI medium when the symmetry-xis tilt is not orthogonal to bedding, such a decision may be impos-ible to make based on data alone, even when well information isvailable. Indeed, locally, we were able to focus and correctly posi-ion seismic data in depth with either a VTI or a TTI model.
Although we have analyzed this ambiguity for a special case of5° tilt, these conclusions should be revisited for a more general casef the symmetry-axis orientation. Does it mean we should opt for aimpler model, i.e., VTI that fits the data? The answer is a resoundingno.” Good kinematic fit �data focusing� and accurate vertical posi-ioning do not guarantee correct lateral positioning; wrongly using aTI model that fits seismic moveout and vertical check-shot travel-
imes will result in mispositioning and apparent azimuthal anisotro-y. For example, if we were to image even our flat TTI data with theTI instead of the TTI model, any amplitude anomalies along the
°45
VTI
TTI
VTI TTIV V=P0 v
igure 9. How to find a best-fit VTI model for a given TTI medium.Aertical cross section of the TTI phase-velocity surface �black� isymmetric around a 45° axis of symmetry. The phase velocity sur-ace for the best-fit VTI medium �gray� matches the vertical TTI ve-ocity in order to match the check shot. It has the same curvature nearhe vertical direction to match the observed NMO velocity.
Downloaded 21 Aug 2010 to 198.36.39.3. Redistribution subject to S
at reflectors would be laterally mispositioned. VTI migration of ourata will always propagate energy symmetrically around a verticaleflection axis; for any offset, an image point will always fall mid-ay between the source and receiver.In TTI media, angles of incidence and reflection are different �Fig-
re 11a�. Therefore, TTI migration of the same data will propagatenergy asymmetrically. Figure 11b quantifies the amount of this an-ular asymmetry as the deviation of the opening angle bisector fromertical. It is clearly controlled by local anisotropy values and can bes high as 8° �Figure 11b�. Because of this asymmetry, image pointsre displaced to the left of their common-midpoint locations by up to00 m �Figure 11c�, which represents large lateral mispositioningrom a practical standpoint.
If more complete borehole data are available, such as walkawaySP, then VTI and TTI cases may be easily distinguished: the travel-
ime curve minimum in the TTI case will be shifted away from theero-offset location.
omplexities of wave propagation in TTI models
In general, we expect more complex wave-propagation phenome-a in TTI media, but in particular we want to discuss two effects thatay be important for model building and imaging.First, we want to emphasize the asymmetric propagation that weentioned earlier. Figure 12 shows actual ray trajectories computed
n our TTI model at hand from a deep reflector. One can clearly ob-erve the effect of ray-trajectory bias to the left; in particular, one canotice that zero-offset rays are not vertical but rather are tilted to theeft. This is easily explained by analyzing Snell’s law in an aniso-ropic medium �Figure 13�. By definition, zero-offset rays have zeroorizontal slowness. To derive the corresponding ray direction, oneust find a normal to the slowness surface �Tsvankin, 2001�. Figure
3 shows that a tilted slowness surface leads to nonvertical rays, thusxplaining the nonvertical zero-offset ray propagation observed inTI media.Second, even in a 1D layered medium, ray trajectories for a check-
hot survey are not vertical, even if the well is vertical �Figure 14�.s above, this deviation can be easily explained by anisotropic
2
4
6
8
10
5000 10,000 5000 10,000 5000 10,000 5000 10,000 5000 10,000
Offset (m)
Dep
th(k
m)
a) b) c) d) e)
igure 10. Common-image-point gathers for various anisotropicodels derived by localized tomography: �a� true model; �b� TTIodel obtained after fixing VP0 �see Figure 3�; �c� TTI model with
egative � and � obtained from seismic and check-shot inversionFigure 4�; �d� TTI model with positive � and � obtained from seis-ic and check-shot inversion �Figure 6�; �e� VTI model.
EG license or copyright; see Terms of Use at http://segdl.org/
Fmtssm
Fert
D34 Bakulin et al.
Opening anglebisector
Normalto
reflector
Horizontalreflector
ϕ
0
–2000
–4000
–6000
–8000
–10,000
–12,000–5 0 5 10
15,000
10,000
5000
0
Dep
th(m
)
Angle from vertical (°)
0
–2000
–4000
–6000
–8000
–10,000
15,000
10,000
5000
0
Offs
et(f
t)
a)
b) c)
–12,000–1000
Reflection point deviation (m)–500 0
igure 11. Location of subsurface reflection points and angles in the true TTI model as predicted by ray modeling. �a� Sketch explaining asym-etric ray propagation caused by tilt of the symmetry axis. �b� Deviation of opening angle bisector from vertical direction; angle bisector divides
he opening angle between incident and reflected rays in half. For VTI media, the angle bisector is always zero, whereas for TTI media we ob-erve a consistent shift of the angle bisector for all offsets including the zero offset. Both attributes are color coded by source-receiver offset. Ob-erve the strong correlation between �b� and the anisotropy profiles �Figure 2�. �c� Shift of actual reflection points with respect to source-receiveridpoint locations, as varying with depth and offset.
a
Fph�g
Zero-offsetray
Reflection pointMidpoint
igure 12. Ray trajectories from a deep reflector computed in the lay-red TTI model at hand. The zero-offset ray is not vertical, and allays have a general bias to the left. Also, the reflection points all fallo the left of their source-receiver midpoints.
Downloaded 21 Aug 2010 to 198.36.39.3. Redistribution subject to S
Reflectedray
Interface
Incidentray
Zero-offsetray
pz
px
p = 0x
Interface
VfastVslow
(Vp0)
) b) c)
igure 13. Schematics of Snell’s law applied in a TTI medium, ex-laining the nonvertical nature of zero-offset rays reflecting from aorizontal reflector: �a� cross section of the phase-velocity surface;b� cross section of the corresponding slowness curve; �c� ray dia-ram. For the zero-offset ray horizontal slowness �p � should vanish.
xEG license or copyright; see Terms of Use at http://segdl.org/
Sptcsipdb
wsclweittmwt1odvco
Fcvs
FvStcnr
Frmbt
Localized TTI tomography D35
nell’s law. If media above and below an interface are the same, thenreserving horizontal slowness leads to vertical rays orthogonal tohe slowness surface �Figure 15a�. However, if the bottom mediumhanges slightly, then the tilted slowness surface deforms; for theame horizontal slowness, the ray direction becomes different, bend-ng the initially vertical ray �Figure 15b�. This is exactly what hap-ens in our model with a tilted symmetry axis: smooth variation withepth effectively creates a set of many tiny interfaces where smallending occurs and accumulates along the ray.
Can we disregard this small effect?Although deviations out of theell may seem insignificant at first, their cumulative effect may be
ubstantial enough to affect check-shot traveltimes computed withommonly used approximations. One of the approximations typical-y used is related to the perturbation technique �Červený, 2001�,hereby a ray trajectory is first traced in the inhomogeneous refer-
nce isotropic medium and then traveltimes are computed using re-ntegration in the proper anisotropic model along this ray. If we wereo apply such an approximation for computing check-shot travel-imes in our example, then the difference between true and approxi-
ate traveltimes could be as large as 12 ms at depth �Figure 16�,hich is enough to bias the model-building process and interpreta-
ion. From Snell’s law �Figure 15� and observed trajectories �Figure4�, we can see that rays deviating from the vertical direction onlyccur when the symmetry axis is not orthogonal to the interface. In-eed, for horizontal reflectors in isotropic or VTI media, rays remainertical. The amount of ray bending is likely to be proportional to theontrast in velocity and anisotropic parameters as well as the changef symmetry-axis direction across the interface �Figure 15�.
igure 14. �a� The tilted symmetry axis in a TTI model leads tourved rays in a 1D medium with source and receiver located along aertical borehole. �b� Deviation of the rays is controlled by thetrength of the anisotropy in the model.
igure 15. Snell’s law diagram explains refraction of the initiallyertical incident ray across the interface between two TTI media. �a�nell’s law requires preservation of horizontal slowness px. Thus, if
wo TTI media above and below are identical, then rays remain verti-al. �b� A slight change in the slowness surface leads to a differentormal direction for the same value of the horizontal slowness andesults in a nonvertical ray after transmission.
0
2000
4000
6000
8000
10,000
12,000–0.1 0 0.1 0.2
Dep
th(m
)
Thomsen parameters
~100 m
10km
~60 m
a) b)δεη
Symmetry axisat 45°
Slownesssurfacechanged Transmittedslightly ray deviates
a) b)Incident Incidentray ray
Slowness surface Slowness surface
++p pxx Interface Interface––p p
xx
Ray Ray
Transmittedray
Downloaded 21 Aug 2010 to 198.36.39.3. Redistribution subject to S
0 5 10 15
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000
11000
Traveltime difference (ms)
Dep
th(m
)
igure 16. Traveltime difference between approximate and exactay-traced traveltimes for a vertical check-shot survey in the TTIodel of interest. Approximate times are computed using a pertur-
ation technique of reintegrating along vertical isotropic ray trajec-ories.
EG license or copyright; see Terms of Use at http://segdl.org/
bmoackmctptpcVo
dVTTlm
uwactmAupi
mmaciw
S
A
B
—
B
ČC
H
M
M
O
P
P
S
S
T
W
W
W
Z
D36 Bakulin et al.
CONCLUSIONS
We applied localized anisotropic tomography with well data touild a locally layered TTI model with a constant 45° tilt of the sym-etry axis around a single borehole. In contrast to the VTI case, we
bserved that supplementing seismic data with velocity measuredlong the vertical or deviated well may still leave the problem under-onstrained, even when the true orientation of the symmetry axis isnown. As a result, we were able to construct several different TTIodels that fit the seismic and well data, yet they were far from the
orrect model. In a VTI case, all such inversions would result in therue model. We were able to explain these results from first princi-les using weak-anisotropy approximations for reflection signa-ures. Because such an increased ambiguity is observed for the sim-lest case of layered models, we expect even more ambiguity in a 3Dase. We conclude that TTI model building is more challenging thanTI model building and would likely require more well informationr other constraints to arrive at a geologically plausible solution.
On the other hand, if we assume VTI symmetry, then the sameata set of seismic and check-shot data can be accurately fit with aTI model. Therefore, we demonstrated that the choice betweenTI and VTI models can also be difficult. Even though both VTI andTI models can image the data and correctly position them vertical-
y, there are negative implications when we image TTI data with VTIodels that may lead to significant lateral mispositioning.We conclude that localized tomography which simultaneously
pdates velocity and Thomsen parameters should always be usedith great care. Despite supplementing seismic data with well data
nd despite restricting the shape of the updates to geologic layers, lo-al multiparameter inversion may still remain nonunique, as seen inhe examples presented here. To arrive at a plausible TTI model, one
ust incorporate additional data or make additional assumptions.nother way to proceed is to analyze the null space of the solutionsing tomography with uncertainty analysis and to select anisotropicarameters based on a priori information from the area or rock-phys-cs measurements.
Our study highlights challenges associated with TTI velocityodel building from narrow-azimuth surveys. We expect wide-azi-uth data to provide additional constraints and to reduce observed
mbiguity for TTI models. We anticipate that this approach of well-onstrained tomography can be applied to inversion for anisotropyn 2D and 3D TTI models and would allow anisotropic calibrationith deviated wells.
Downloaded 21 Aug 2010 to 198.36.39.3. Redistribution subject to S
ACKNOWLEDGMENTS
We thank DmitryAlexandrov �summer intern from St. Petersburgtate University� for help with generating the synthetic data set.
REFERENCES
udebert, F. S., A. Pettenati, and V. Dirks, 2006, TTI anisotropic depth mi-gration — Which tilt estimate should we use?: 68th Conference & Techni-cal Exhibition, EAGE, ExtendedAbstracts, P185.
akulin, A., M. Woodward, D. Nichols, K. Osypov, and O. Zdraveva, 2009a,Localized anisotropic tomography with well information in VTI media:79thAnnual International Meeting, SEG, ExpandedAbstracts, 221–225.—–, 2009b, Anisotropic model building with uncertainty analysis: 79thAnnual International Meeting, SEG, ExpandedAbstracts, 3720–3724.
ear, L. K., T. A. Dickens, J. R. Krebs, J. Liu, and P. Traynin, 2005, Integrat-ed velocity model estimation for improved positioning with anisotropicPSDM: The Leading Edge, 24, 622–634.
ervený, V., 2001, Seismic ray theory: Cambridge University Press.hapman, C. H., and R. G. Pratt, 1992, Traveltime tomography in anisotropicmedia — I. Theory: Geophysical Journal International, 109, 1–19.
uang, T., S. Xu, and Y. Zhang, 2007, Anisotropy estimation for prestackdepth imaging — A tomographic approach: 77th Annual InternationalMeeting, SEG, ExpandedAbstracts, 124–127.ateeva, A., A. Bakulin, P. Jorgensen, and J. Lopez, 2006, Accurate estima-tion of subsalt velocities using virtual checkshots: Offshore TechnologyConference, Proceedings, OTC 17869.orice, S., J.-C. Puech, and S. Leaney, 2004, Well-driven seismic: 3D dataprocessing solutions from wireline logs and borehole seismic data: FirstBreak, 22, 61–66.
sypov, K., D. Nichols, M. Woodward, O. Zdraveva, and C. E. Yarman,2008, Uncertainty and resolution analysis for anisotropic tomography us-ing iterative eigendecomposition: 78th Annual International Meeting,SEG, ExpandedAbstracts, 3244–3249.
ech, A., I. Tsvankin, and V. Grechka, 2003, Quartic moveout coefficient: 3Ddescription and application to tilted TI media: Geophysics, 68,1600–1610.
ratt, R. G., and C. H. Chapman, 1992, Traveltime tomography in anisotropicmedia — II.Applications: Geophysical Journal International, 109, 20–37.
engupta, M., R. Bachrach, and A. Bakulin, 2009, Relationship between ve-locity and anisotropy perturbations and anomalous stress field around saltbodies: The Leading Edge, 28, 598–605.
exton, P., and P. Williamson, 1998, 3D anisotropic velocity estimation bymodel-based inversion of pre-stack traveltimes: 68thAnnual InternationalMeeting, SEG, ExpandedAbstracts, 1855–1858.
svankin, I., 2001, Seismic signatures and analysis of reflection data in an-isotropic media: Elsevier Science Publ. Co., Inc.ang, X., and I. Tsvankin, 2009, Stacking-velocity tomography with bore-hole constraints for tilted TI media: 79th Annual International Meeting,SEG, ExpandedAbstracts, 2352–2356.oodward, M., P. Farmer, D. Nichols, and S. Charles, 1998, Automated 3Dtomographic velocity analysis of residual moveout in prestack depth mi-grated common image point gathers: 68th Annual International Meeting,SEG, ExpandedAbstracts, 1218–1221.oodward, M., D. Nichols, O. Zdraveva, P. Whitfield, and T. Johns, 2008, Adecade of tomography: Geophysics, 73, no. 5, VE5–VE11.
hou, H., D. Pham, S. Gray, and B. Wang, 2004, Tomographic velocity anal-ysis in strongly anisotropic TTI media: 74th Annual International Meet-ing, SEG, ExpandedAbstracts, 2347–2350.
EG license or copyright; see Terms of Use at http://segdl.org/