You are on page 1of 4

Downloaded 05/18/16 to 128.210.126.199. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.

org/

Macro velocity model estimation through model-based globally-optimized S13.4


residual-curvature analysis
Bin Wang*, Keh Pann, and Robert A. Meek, Mobil Exploration & Producing Technical Center
SUMMARY each reflector image becomes single valued. Migration also
In the past few years, significant progress has been made on collapses the diffraction energy. Also migration enhances the
new velocity analysis algorithms. In the first part of this paper, we coherent reflection events and suppresses random noise, thereby
will briefly summarize recent advances on velocity analysis. Then improving the signal-to-noise ratio.
we describe a new model-based globally-optimized residual Migration velocity analysis
curvature analysis algorithm we have just developed. Like There are two commonly used migration velocity analysis
conventional residual curvature analysis, the algorithm is based on methods: Depth Focusing Analysis (DFA) and Residual Curvature
the principle that after prestack migration with a correct velocity Analysis (RCA). DFA (Yilmaz and Chambers, 1984; Faye and
model, an image in the common image point (CIP) gather is Jeannot 1986; McKay and Abma, 1992) is based on the principle
aligned horizontally regardless of structure. Unlike conventional that if the true velocity model is used for migration then multi-
residual curvature analysis, this algorithm uses not only the offset reflection energy will be well focused at the migration
interpreted CIP gathers, but also the interpreted migrated depth depth. If a wrong velocity is used for migration, then reflection
section as input. The algorithm is model-based, and uses model- energy will be focused at a depth different from migration depth.
based CIP ray tracing to relate residual moveouts in CIP gathers to Migration error, which is the difference between focusing depth
errors in the velocity model. Residual moveouts measured in CIP and migration depth, is measured by picking the location of a
gathers are globally used in the optimization process for updating focusing point in a depth focusing panel. RCA (Al-Yahya, 1989;
the whole velocity model. Also model-based normal incident ray Willis, 1990; Lafond and Levander, 1993) is based on the
tracing is used for updating the reflector boundaries. principle that after prestack migration with the correct velocity, a
REVIEW OF RECENT ADVANCES OF VELOCITY reflection event in the CIP gather should be aligned horizontally
ANALYSIS regardless of structure. The failure of an event to line up
Recent advances of velocity analysis can be summarized into horizontally is an indication of error in the velocity model on
two categories: 1) traveltime inversion; 2) migration velocity which the imaging is based.
analysis. As pointed out by Liu and Bleistein (1994), conventional
migration velocity analysis methods have the following
Traveltime inversion assumptions: (1) lateral velocity homogeneity; (2) small offset; (3)
Traveltime Inversion (TI) (Bishop et al., 1985; Stork and horizontal reflector. In addition, they are all local methods. Each
Clayton, 1991) estimates a depth velocity model from traveltimes time only a single CIP gather (for RCA) or a single depth
picked from prestack data. The main advantage of TI is that it is focusing panel (for DFA) is used for updating the local velocity.
formulated as an optimization problem and therefore model Figures 1A and 1B schematically show that they cannot handle the
updating is very effective and efficient. However, in areas of complex overburden problem. Figure 1A shows that migration
complex geological structure, picking prestack traveltimes in error at a dipping reflector is not caused by the velocity model
surface seismic data is almost unfeasible. Picking prestack component which is directly above it, but by other parts of
traveltimes may have the following problems: 1) in the case of velocity model. This indicates that a local updating scheme may
complex reflector geometry, seismic energies reflected from cause velocity corrections to a wrong spatial location. Figure 1B
different parts of a reflector may arrive at the same receiver shows that if there is complex overburden, such as a syncline, then
location; 2) reflection arrivals contaminated by diffraction energy; conventional migration velocity analysis will not work well,
3) low signal-to-noise ratio often associated with complex because velocity updating is based on the raypath length
structure. difference of different offsets assuming a flat layered model.
In the past few years, a few researchers tried to solve the Obviously, if there is complex overburden, raypath lengths in each
traveltime picking problem. IFP (Institut Francais du Petrole) layer can be significantly different from those for a flat layered
developed a method called SMART (Sequential Migration Aided model.
Reflection Tomography) (Delprat-Jannaud and Lailly, 1993) to TO solve the above stated problems of migration velocity
solve the traveltime picking problem. The main idea of SMART is analysis, attempts (Stork, 1992, 1994, Lafond and Levander, 1993,
to use an approximate velocity model to migrate seismic data, then Liu and Bleistein, 1994, 1995, Wang et al., 1995) have been made
pick the imaged reflectors in the cube of migrated shot gathers, to formulate velocity updating as an optimization process. To
finally trace rays that propagate in the same velocity model as the improve RCA, Liu and Bleistein (1994, 1995) proposed a method
one used for the migration and that are reflected on the picked called migration velocity analysis by perturbation. They
imaged reflectors. They claimed that because the ray tracing parameterized a velocity model as a blocky model, which consists
undoes what the migration has done, even with an approximate of different layer boundaries and each layer has different velocity
velocity model they can recover traveltimes. More recently function. Liu and Bleistein’s method relaxes all three assumptions
DATAID (1994) used a similar approach as IFP, but performed of conventional RCA. Stork (1992, 1994) developed a method
migration and raytracing in the common-offset gather instead of called tomographic migration velocity analysis. Like reflection
common-shot gather. The main point of these approaches is that tomography, he parameterized the velocity model as a grid model,
instead of directly picking events in the time domain, picking is used ray tracing to relate residual moveouts in CIP gathers to
done after depth migration, and ray tracing is used to recover the velocity model, and used inversion techniques to update the whole
“observed traveltimes” for TI. After migration, different branches grid model. Both Liu and Bleistein’s approach and Stock’s
of reflection events are migrated to the appropriate reflector. As approach can handle complex overburden problems. Also both of
long as a velocity model is not too far away from the true model, their approaches used migration errors measured for all CIP
1084
Model-based globally-optimized residual-curvature 2

gathers globally. the following equation:


Downloaded 05/18/16 to 128.210.126.199. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Upon improving DFA, recently Wang et al (1995) developed


a new migration velocity analysis method called Model-based (2)
Globally-optimized Depth Focusing Analysis (MGDFA) method, From reflection tomography, we have the following traveltime
MGDFA uses the focusing panel to pick the migration errors, but derivative formula:
disregards the existing 1-D velocity updating formulas and uses
ray tracing and an inversion technique to update the velocity = +
model globally. Comparisons among these new TI, RCA, and DFA
methods are summarized in Table 1. where v is the interval velocity right above the reflector being
imaged.
MODEL-BASED GLOBALLY-OPTIMIZED RESIDUAL From equations (2) and we have the following equation:
CURVATURE ANALYSIS at
Algorithm of the method (4)
i
(1) Model Parameterization This equation is similar to equation 6 of Liu and Bleistein
We use a similar model parameterization as Liu and Bleistein (1994). Equation (4) shows that the derivative of image-depth to
(1994, 1995). We use a blocky velocity model instead of a grid velocity is related to the derivative of traveltime to velocity by a
velocity model (Stock, 1992, 1994), even though the grid model factor which can be easily calculated
representation has more flexibility for representing complex through ray tracing. Therefore, any existing traveltime inversion
velocity models. The main reason for choosing the blocky model software can be modified for the calculation of the derivative of
representation is that it results in fewer unknowns, and inversion is image-depth to velocity.
stable. If the grid model is used, the velocity value at every grid
point becomes an unknown, and the inversion problem becomes The objective of our global inversion is to make the reflection
either singular or ill-conditioned, To solve these under- image flat for all CIP gathers. In addition, damping and
determined inversion problems, constraints are needed, and the smoothing constraints are incorporated into the inversion process.
inversion results are very dependent upon the kind and the (3) Updating Layer Boundary
amount of constraints added. If constraints are not used Liu and Bleistein (1994, 1995) used a layer-striping approach
appropriately, the inversion may not be stable or may result in for their velocity analysis. For each velocity model update layer
inversion artifacts. In our experience, we need a more highly boundary locations are not updated. The new layer boundary
trained user for grid model inversion compared with blocky corresponding to the new velocity is obtained by another
model inversion. Also, by using the blocky velocity model migration. To improve the convergence speed, we develop a way
representation, depths to certain reflectors are more easily tied to to update reflector positions for each iteration, also layer stripping
well information. is not always necessary. In our approach, for each iteration two
(2) Derivative Calculation sets of information are needed. One is the interpreted residual
moveout picked from each CIP gather. The other is the
In Liu and Bleistein’s approach, for each iteration, two separate interpreted migrated depth section, which is used for updating
Kirchhoff migrations are needed, and amplitude information of layer boundaries.
migration results are used for their derivative calculation. We use
the kinematic condition of zero-time imaging to derive the The layer boundaries of the starting velocity model for
derivative calculation formula which turns out to be similar to Liu inversion are built based on a depth-migrated image. According
and Bleistein’s formula. By taking a different perspective, we to the SMART approach (Delprat-Jannaud and Lailly, 1993),
develop a new interpretation of the formula. The new which states that raytracing undoes migration, if normal incident
interpretation is that the derivative of image-depth to velocity is rays are traced for each reflector, then the calculated traveltimes
related to the derivative of traveltime to velocity by a factor which along these normal incident rays will approximate the true
can be easily calculated through ray tracing. “observed zero-offset traveltimes”. Therefore we can have another
set of equations associated with these normal incident rays.
If we start prestack migration with an initial velocity model Ideally, this set of equations and those equations describing
which typically is not correct, we expect an image in a CIP gather flatness criterion can be solved simultaneously for velocity and
not to be aligned horizontally. After we perturb the initial velocity reflector unknowns. In our current implementation, only those
model, image-depths associated with different offsets will change parameters representing velocity functions are set to be unknown
differently. The inversion process tries to find the correct velocity and inverted, and the reflector positions are updated after the
perturbation, with which the migration depths are same for all velocity has been updated. Reflectors are updated using the
offsets. Therefore, we need to know the derivatives of migration- following equation:
depth to velocity for each offset.
In Figure 2, SPR is the ray path through an initial velocity +( (5)
model, and SP’R is the ray path through the perturbed velocity i
model. According to zero-time image condition, the traveltimes where is the updated reflector position at the point where
associated with these two ray paths should be the same. In other normal incident ray is fired, is the corresponding reflector
words, the traveltime difference between these two ray paths position of the old model, is calculated using equation (3) for
should be zero.
at normal-incident rays.
(1)
i Testing with synthetic data
where pi’s are model parameters representing the velocity model, A synthetic seismic data set was generated for testing the new
m is the number of model parameters, AZ is the depth velocity analysis method. The velocity model used for modeling
perturbation of current reflector as shown in Figure 2. seismic propagation is shown in Figure 3A. This is a three layer
By definition of partial derivative, from equation (l), we have model, with the second layer reflector geometry highly irregular.

1085
3 Model-based globally-optimized residual-curvature

We used finite difference acoustic modeling to generate synthetic domian: Geophysics, 57 (5), p. 680-692.
Downloaded 05/18/16 to 128.210.126.199. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

seismic data. We purposely designed the second reflector to be Stork, C., 1994, Demonstration of MVA tomography with controls
highly irregular. With reflector geometry like this, conventional and constraints for determining an accurate velocity model for
stacking velocity analysis and traveltime inversion methods will prestack depth migration: 64 th Ann. Intemat. Mtg., Soc. Expl.
Geophys., Expanded Abstracts, 1338-1342.
not work well because of complex traveltime pattern (multiple Wang, B., Pann, K., and Malloy, J.E., 1995, Macro velocity model
arrivals). Figure 3B shows an example of the synthetic data in a estimation through model-based globally-optimized depth
common offset gather. focusing analysis: 57 th Annual EAEG meeting. Expanded
We start the velocity analysis by using a constant velocity Abstracts, in press.
model with the velocity equal to the first layer velocity of 2000 Willis, M.E., 1990, Private communications.
m/s to perform the prestack depth migration. Notice that Yilmaz, O., and Chambers, R., 1984, Migration velocity analysis
compared with the true velocity model, this initial velocity model by wavefield extrapolation: Geophysics, 49, 1664-1674.
has about 30% lower velocity for the second layer, and 50% lower
velocity for the third layer. Figure 4 shows one example of a CIP
gather. Figure 5A shows an example of CIP ray tracing, which is
used to relate the residual moveouts measured from CIP gathers to
the velocity model. Figure 5B shows normal incident ray tracing
which is used for updating reflector positions. Figure 6A is the
output of the second iteration. Compared with the true velocity
model (Figure 3A), the velocity model obtained after two
iterations works reasonably well for migration. Figure 6B is the
seismic depth migrated section using the velocity model obtained
by two iterations of this velocity analysis.
CONCLUSIONS
A new model-based globally-optimized residual curvature
analysis method has been developed. This method needs an
interpreted migrated depth section in addition to the traditional
interpreted CIP gathers as input. CIP ray tracing is used to relate
the residual moveouts measured from CIP gathers to the whole
velocity model, and normal incident rays are traced for updating
reflector positions. Testing on prestack synthetic seismic data
indicates that this method provides a practical means of solving
velocity problems with complex structures, and importantly the
convergence speed is fast.
ACKNOWLEDGMENTS
We would like to thank Rick Cook, Dave Hinkley, and Mark
Willis for many helpful discussions. Thanks also go to John
Anderson and Dana Jurick for reviewing this paper and for their
constructive comments. We would also like to thank Terry Young
for his support and encouragement of this work, and Mobil
management for permission to publish this work.
REFERENCES
Al-Yahya, K., 1989, Velocity analysis by iterative profile
migration. Geophysics, 54, 718-729. Figure 1. Ray path diagram showing the limitations of
Bishop, T.N., Bube, K.P., Cutler, R.T., Langan, R.T., Love, P.L., conventional RCA and DFA. (A) migration errors for a dipping
Resnick, J.R., Shrey, R.T., Spindler, D.A., and Wyld, H.W., 1985, reflector are not caused by part of the velocity model directly
Tomographic determination of velocity and depth in laterally above it; (B) problems caused by complex overburden.
varying media: Geophysics, 50, 903-923.
DATAID, 1994, private communications.
Delprat-Jannaud, F., and Lailly, P., 1993, Tomography with
multiple arrivals: how to handle noise corrupted data? 63 rd Ann.
Internat. Mtg., Soc. Expl. Geophys., Expanded Abstracts,
11911194. 587-590.
Faye, J.P., and Jeannot, J.P., 1986, Prestack migration velocities
from focusing depth analysis: 56th Ann. Internat. Mtg., Soc.
Expl. Geophys., Expanded Abstracts, 438-440.
Lafond, C.F., and Levander, A.R., 1993, Migration moveout
analysis and depth focusing: Geophysics, 58, 9 l-100.
Liu, Z., and Bleistein, N., 1994, Velocity analysis by perturbation:
64 th Ann. Internat. Mtg., Soc. Expl. Geophys., Expanded
Abstracts, 11911194.
Liu, Z., and Bleistein, N., 1995, Migration velocity analysis: theory
and an iterative algorithm: Geophysics, 60, 142-l53,
Stork, C., and Clayton, R.W., 199 1, Linear aspects of tomographic
velocity analysis: Geophysics, 56, 483-495. Figure 2. Schematic diagram showing ray paths of initial and
Stork, C., 1992, Reflection tomography in the postmigrated perturbed velocity models.

1086
Model-based globally-optimized residual-curvature 4
Downloaded 05/18/16 to 128.210.126.199. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Figure 5. (A) model-based CIP ray tracing is used to relate


residual moveouts of CIP gathers to velocity model; (B) model-
based normal-incident ray tracing is used for updating reflector
positions.

Figure 3. (A) velocity model used for generating synthetic seismic


data; (B) an example of synthetic seismic data in a common-offset
gather.

Figure 6. (A) velocity model obtained after two iterations of this


Figure 4. An example of an interpreted CIP gather, which is sorted velocity analysis; (B) migrated depth image using the velocity
from partial images of common offst migration. model shown by Figure 6A.

1087

You might also like