You are on page 1of 7

technical article

first break volume 19.4 April 2001

Benefits of pre-stack time migration


in model building: a case history in
the South Caspian Sea
E. Robein1 & C. Hanitzsch2

Local context
The South Caspian Sea (Fig. 1) was only recently opened for
exploration to western companies. The number of offshore
wells drilled in the past has been limited. The seismic response
in the area is generally good, but the geology may be locally
complicated. In particular, the presence of undercompacted
overpressured shale layers has been reported. Reliable pieces
of information relevant to the geophysicist, such as sonic logs
or check shot surveys are sparse. Consequently, one often has
to rely on seismic data, and more precisely on stacking velocities to estimate the velocities needed for imaging (migration)
or for time-to-depth conversion later in the processing-interpretation process.
Unfortunately, stacking velocities often diverge dramatically from the needed geological propagation velocities. The
many reasons for this are well known and have been documented for a long time in the seismic literature (Al Chalabi
1994). The South Caspian Sea is no exception, to the rule and
strong stacking velocity anomalies were noticed over most of
the structures as soon as the first speculative surveys were
made available in the area.
In this context, improving the control over seismically derived velocities is an important issue in the preparation, acquisition and processing of newly acquired 3D seismic data
sets (Robein & Kerimov 1998).
This paper presents an approach to this problem applied
to a 3D survey shot in 1997 over the Lenkoran-Talysh structure in the Azerbaijan South Caspian Basin, offshore the city
of Lenkoran (Fig. 1).
To illustrate our point, Fig. 2 shows an in-line section extracted from the final 3D volume, which gives an idea of the
local geological complexity. Figure 3 presents the stacking velocity field measured after DMO in 2D along the correspond-

Figure 1 Location map.

Formerly Elf Petroleum Lenkoran-Talysh; Present address:


TotalFinaElf, CSTJF avenue Larribau 64018 Pau Cedex, France.
1

TotalFinaElf, CSTJF avenue Larribau 64018 Pau Cedex, France

Proofs to be sent to Etienne ROBEIN, TotalFinaElf, CSTJF avenue


Larribau 64018 Pau Cedex, France.
2000 EAGE

Figure 2 Line A. Typical seismic section shot in the dip direction


over the Lenkoran Talysh structure. One horizon is highlighted and
will be overlaid on other figures.

183

technical article

Figure 3 Stacking velocity field measured in 2D after DMO on line A


in Fig. 2. The time horizon picked on Fig. 2 and the iso-velocity
contour at 2600 m/s have been highlighted.

ing sail-line. Velocity picks are made every 500 m.


Strong lateral velocity changes show up below 1 s, especially over the top and on the left-hand flank of the structure.
These stacking velocity artefacts cannot be directly related to
the fairly smooth geological features.
These artefacts, although often related to surface anomalies, may in fact hide important velocity information, possibly
related to undercompaction. Any attempt to recover this information and to transform the seismic image into a depth
image would hardly be possible using the classical Dix formula. No classical editing or smoothing techniques can reliably transform such a velocity field into a geologically sound
one!

The strategy: from acquisition to depth


maps

184

The first issue in dealing with velocity estimation in the preparation of a 3D survey was the choice of the acquisition direction. Dip or strike acquisition (when the structure allows such
a clear categorization everywhere in the full 3D volume!) has
been a debate between geophysicists for many years.
In fact, acquisition direction and processing are closely
linked. Basically, one may claim that shooting strike helps to
improve the seismic image, because the hyperbolic assumption used in the NMO-DMO process is better honoured. Conversely, dip direction is acceptable when model based
techniques, such as pre-stack depth migration, are envisaged.
This oversimplified standpoint ignores any budget and acquisition constraints. Strike shooting implies a finer cross-line
interval, which may or may not be compensated for by fewer
boat turns and a better turnaround. Currents and streamer
feathering is another issue.
In the present case, where the northern tip of the survey
area was in shallow water, the coast geometry and the water

first break volume 19.4 April 2001

depth did not allow any discussions or testing: we had to


shoot dip anyway!
The second issue was processing. Cost and turnaround
considerations prevented us from considering pre-stack depth
migration, although this is probably an interesting option for
addressing the imaging and velocity issues in this part of the
Caspian Sea.
Testing pre-stack time migration (PreSTM) was an alternative which we planned to test right from the beginning of the
project. The dip shooting direction allowed us to test the
method in 2D. The expected benefits from PreSTM in the
present context were twofold: firstly, an improved image and,
secondly, a simpler stacking velocity field for time migration
and time to depth conversion. These two potential advantages
of PreSTM and the positive result of the actual tests are discussed in the next two paragraphs.
Finally, we will show that, although PreSTM was necessary, it is not sufficient by itself for a successful time to depth
conversion. This is why a model-based stacking velocity inversion was planned as a method for depth mapping the key
interpreted horizons.
In short, the chain of dip acquisition, PreSTMand modelbased stacking velocity inversion was forecast to be the best
way to address the velocity and imaging issues in this area.

Expected benefits and limits of PreSTM


What we mean by PreSTM here is described by the sequence
schematized in Fig. 4.
The method is very close to the MOVES technique first
introduced by Marcoux et al. 1987). A simplified description
of the method splits the process into the following steps:
Apply NMO-DMO corrections (with an initial velocity
field) to data sorted in constant offset blocks;
Apply zero-offset time migration to each individual block,
generally using a single velocity function;
Apply inverse NMO corrections to the whole data
volume;
Re-sort data as common mid-point gathers and carry out
stacking velocity analyses (where needed) in order to
establish the final stacking velocity field.
Apply NMO corrections to the whole data set using this
final stacking velocity field, and stack;
Time demigrate this stack and apply the final migration,
or, optionally, apply a cascaded residual time-migration.
This procedure allowed us to carry out both the stacking
velocity analysis and the stack itself in the time-migrated domain. It was expected that this would improve the seismic image and, at the same time, facilitate considerably the velocity
picking. The reason for this is illustrated in Fig. 5.
This cartoon shows schematically what may happen in the
classical syncline case. Various reflection points show up in
the same common mid-point gather after NMO-DMO (the
so-called triplication or multiple arrival phenomenon). Re 2000 EAGE

first break volume 19.4 April 2001

technical article

Figure 5 Schematics of the PreSTM benefits in the standard syncline


case. In this typical situation, reflections corresponding to various
points in depth may appear on the same gather after DMO, making
stacking velocity picking ambiguous. After PreSTM stacking velocity
analysis is carried out in a migrated position. It should be noted
however, that the same area in the subsurface (highlighted in blue
for point A) influences the kinematics of reflection times.

Figure 4 Schematic flow chart of Pre-stack time migration as


applied in this case.

flected arrivals are mixed up, making the velocity picking ambiguous. Generally in real life the situation is even worse than
depicted here, due to the velocity heterogeneity and the limitation of constant velocity DMO. The situation is clarified
markedly by PreSTM: now geographically distinct reflection
points appear in different gathers. The point we want to make
in this figure is twofold.
Firstly, in conventional post-DMO velocity analysis the reflection point is close to the end of the normal ray emerging at
this location (Fig. 5) which is generally not vertically below
the surface location of the gather under investigation. When
stacking velocity analysis is, instead, carried out after
PreSTM, the data analysed has migrated towards this expected vertical location. This goal may not be reached perfectly because of the basic assumptions of time migration, but
the closer the migration velocity is to the true velocity, and
the less lateral velocity variation there is in the subsurface, the
better the approximation. In any case, the problem of multiple arrivals is at least partially solved, yielding one expected
benefit of PreSTM: velocity picking is less ambiguous in complex geology and each single event may be better stacked
without a compromise in the choice of its own optimal stacking velocity.
2000 EAGE

Secondly, Fig. 5 clarifies the statement often made that


PreSTM migrates stacking velocities. In effect, the stacking
velocity value of, say, reflection point A in Fig. 5, will appear
in the stacking velocity field at its migrated position. In this
respect, the stacking velocity value is migrated through
PreSTM and consequently the stacking velocity field is simplified. It should be realized, however, that the actual value of
this stacking velocity, besides being modified by the kinematic
effects of NMO, DMO and time migration processes, is bearing information on propagation velocities mainly in the region delineated by the fan of offset rays around the normal
ray ending in A (which is not vertical in a geologically complex environment).
This simple reasoning has been quantified by Robein et al.
1997) who showed that stacking velocities after PreSTM may
be significantly different from vertical Vrms, even in a very simple geological context. In other words, the stacking velocity
field after PreSTM is simplified but is generally not the desired
vertical rms average propagation velocity field. A straight application of the Dix formula to stacking velocities after
PreSTM, although probably more stable because of the simplification of the velocity field, is still not theoretically justified.
There are several variations in the implementations of
PreSTM. Bloor & Deregowski (1995) introduced a special
one under the name of demigration to zero offset or DZO. It
has the particularity of using the same constant velocity as
Initial stacking velocity, intermediate migration velocity and

185

technical article

inverse NMO velocity. This constant velocity approach has


the theoretical advantage of applying a fully reversible operator before stacking velocity analysis. It has the disadvantage
of only partially migrating the data.
PreSTM tests on the actual data
Both time-variant and constant velocity Pre-stack Time
Migration options were tested in 2D on the real data.
In this test, the initial NMO correction of the V(T)
PreSTM option was done using stacking velocities picked
onboard the acquisition vessel. A constant velocity DMO operator was applied with 25 constant offset sections as an output. Each section underwent a Stolt time migration using a
time variant, but laterally invariant velocity function Vmig(T).
This function was defined at each time sample as about 90%
of the minimum value of the initial stacking velocities over the
whole line. This intermediate velocity function is chosen to
avoid any over migration effects and to allow a future residual migration option.
Figure 6 presents the stacking velocity field of line A of
Fig. 1 after 2D V(T) PreSTM.
For the DZO option, the unique velocity was chosen as
1800 m/s.
The resulting stacking velocity field for this constant velocity PreSTM test is shown in Fig. 7.
The results of both tests met the theoretical expectations of
the technique. The improvement in terms of simplifying the
velocity field is clear when comparing Fig. 3 (post-DMO
stacking velocities) with Figs 6 and 7.
We also noted improvements in the images produced from
these sequences, especially for steeply dipping events.
Based on the results of these tests and on turnaround and
budget considerations, it was decided to apply DZO in 3D as
part of the full processing sequence. Figure 8 presents the final
stacking velocity field along in-line A after 3D production
DZO.

first break volume 19.4 April 2001

Figure 7 Stacking velocity field of line A, computed in 2D, after


demigration to zero offset (DZO).

Figure 8 Stacking velocity field of line A, computed after production


3D demigration to zero offset (DZO). Note that the colour code is
different from Figs 3, 6 and 7, but the iso-velocity 2600 m/s has
been highlighted.

Once again, note the added value of 3D imaging: the


stacking velocity field is further simplified compared with
Fig. 7, thanks to 3D migration effects.
Incidentally, we may claim that surface static anomalies
are not strongly affecting stacking velocities since most of the
high frequency artefacts observed in the velocity field of
Fig. 3 are removed by the PreSTM operators.

Building the time migration velocity field

186

Figure 6 Stacking velocity field of line A, computed in 2D, after V(T)


pre-stack time migration.

One spin-off of simplifying the stacking velocity model using


PreSTM, is that it makes it more consistent with time migration, since it is well-known that a fairly smooth migration velocity field is required for time migration. Smoothing the
stacking velocity field in a reasonable way is often difficult.
PreSTM did it quite nicely and intelligently for us in the
present example, as may be seen from Fig. 8. In this case the
migration of the stacking velocity field (in the sense explained above), was effective. A very gentle lateral smoothing
using a large running window-averaging operator was ap 2000 EAGE

first break volume 19.4 April 2001

plied to the stacking velocity field. Next, fine-tuning of migration velocities was carried out as usual by extensive testing of
the sensitivity of the 3D migrated image to migration velocities. Only a few local adjustments not exceeding a few per
cent, were actually necessary. This confirmed another potential benefit of PreSTM.
The final migrated seismic data set was of very good quality. It was interpreted and a set of time-migrated maps was
picked at the horizons of interest. The next step was then to
convert these time maps into depth maps. As stated in the introduction, this step was carried out using model-based stacking velocity inversion.

From time maps to depth maps: modelbased stacking velocity inversion


There are basically two main issues in time-to-depth mapping.
The first is the estimation of the propagation velocities. Here
we have to use stacking velocities since there were no well
data. The simplest inversion approach would be to make use
of the Dix formula, but there are more accurate alternatives.
The second issue was choice of appropriate time-to-depth migration technique. The two most popular ones are straightforward vertical stretching and image-ray migration.
We discussed earlier why the use of the Dix formula
should be discarded in the first place even if PreSTM had been
applied previous to stacking velocity picking. It is clear from
Fig. 8 that applying the Dix formula vertically to the stacking
velocity field shown cannot yield an interval velocity model
consistent with geological horizons.
Considering the second issue, vertical stretching is known
to be a rough approximation in most cases. Image ray migration instead, is often considered to be more accurate. It does
not, however, take into account and compensate for the dip
and migration-velocity dependence of time migration artefacts.
Stacking velocity inversion, as implemented in
TotalFinaElfs Superdix proprietary software, actually addresses both issues simultaneously with no theoretical
shortcuts: it estimates propagation velocities and correctly
depth migrates time maps in a single step. The technique
makes use of ray tracing and inversion theory (Sexton 1998).
The goal is to build a layer cake depth-velocity model in 3D
which is in agreement with all available measurements: time
maps resulting from the interpretation of the time migrated
seismic data set, the corresponding actual time-migration velocity field; stacking velocities; and possibly well ties when
these are available, which was not the case here.
Interfaces of the layer cake model are depth-migrated versions of the interpreted time maps. Propagation velocities between these interfaces are parameterized in various ways,
which are specified below.
Because of the point discussed above, time to depth con 2000 EAGE

technical article

version of interfaces is not carried out by image-ray techniques. It uses instead kinematic time de-migration followed
by normal ray depth migration through the current model
(Whitcombe 1994). This method allows us to compensate accurately for any mispositioning of a picked horizon due either
to time migration effects or to inaccuracies or
oversimplifications of the time migration velocity field. The
correction makes the depth conversion almost insensitive to
the choice of the time migration velocity field, as long as the
latter is sufficiently correct to avoid any erroneous picking by
interpreters.
The propagation velocity in each layer between two interfaces may be described in various ways. For example in the
present case, we chose to describe the instantaneous velocity
function in each layer as:
Vlayer(x, y, z) = V0_layer(x, y) + klayerZ
V0_layer(x, y) is a reference velocity at the surface, described
by a polynomial function in x and y. Thus the velocity was
parameterized by the coefficients of the polynomial and klayer,
which is a linear vertical gradient, unique for a single layer
but different in each layer.
For each interface, these velocity parameters were inverted
in order to minimize the difference between the measured
stacking velocities and the synthetic ones computed through
the current model. V0_layer coefficients and klayer were inverted
simultaneously, layer per layer.
In the inversion process, the computation of synthetic
stacking velocities (forward modelling) used 3D ray tracing
through the current model. It was imperative to take into account the kinematics of all processing steps which the data
had undergone before stacking velocity analysis: in this case
PreSTM. The corresponding technique was first published for
DMO and extended to the case of PreSTM (Robein et al.
1995, 1997). The idea is to search for the reflection point in
depth such that the reflected signal recorded with a given offset will migrate through DMO or PreSTM to the location of
the gather under investigation.
The technique works best when stacking velocities have
been picked horizon consistently along the interface of interest. This shows again how important it is to know, as early as
the processing stage, what technique of depth mapping will be
used later on. If this is not the case, velocity interpolation may
be used for reconstructing horizon consistent stacking velocities.
To recapitulate, here are two advantages of model-based
stacking velocity inversion, compared to other techniques:
The impact on stacking velocities of the geometry of
interfaces and velocity variations in the layers are correctly
taken into account during the ray tracing step. No
simplifying assumption is made on the model geometry.
The influence of data processing on stacking velocities is
taken care of.
Both points were particularly important in this case study.

187

technical article

A spin-off of the method is that we could keep track of the


distance between locations of any time-migrated and depthmigrated points of the gridded maps. This made possible a
posterior assessment of the seismic image distortions induced
by time migration. It also gave us a method for locating well
trajectories in the time migrated data far superior to the conventional vertical stretch method.

Result of the time to depth transform


The choice of the degree of smoothness of the velocity model
is an important step in the inversion process. Both testing and
some degree of interpretation are required to find the best
compromise between geological soundness and simplicity of
the model and inversion efficiency and stability. Introducing
too many details in the velocity field may well explain most of
the high frequency stacking velocity features, but these may
in fact be only noise! On the other hand, to have too simple a
model may lead to overlooking some important geological
features.
In the present case, two things helped us significantly in
making this choice. Firstly, the fact that PreSTM had simplified the stacking velocity field facilitated the decision on how
many details of the stacking velocity field needed to be explained by inversion. Secondly, inversion of the klayer parameter was particularly stable thanks to the wide range of depths
in each layer. A first or second order polynomial was found to
be a good compromise for all layers.
Figure 9 gives a perspective view of the depth map resulting from stacking velocity inversion for an horizon close to
the one shown on Fig. 1.
Two cross sections through the instantaneous velocity
model are also shown in this picture. The smooth lateral velocity changes and the vertical gradient show up in the colour

188

Figure 9 Result of stacking velocity inversion: depth maps and


velocity model. Note the low velocity layer (undercompaction) in the
cross-section through the velocity model in the background.

first break volume 19.4 April 2001

coded instantaneous velocities. Note that the fourth layer


presents a distinct velocity decrease compared to the layer just
above it: stacking velocity inversion detected undercompaction, which to a great extent explains the velocity
anomaly of Figs 3, 6, 7 and 8. It is clear from this figure that
a simple vertical application of the Dix formula could not
have explained such a model dependent effect, even after
PreSTM.
It should be realized, however, that the technique does not
provide the interval velocity model with sufficient resolution
for a reliable pore-pressure prediction exercise.
Stacking velocity inversion is based on minimizing the difference between actual stacking velocities and the modelled
ones computed through the depth-velocity model, taking all
processing steps into account. For the layer shown in Fig. 9,
residuals do not exceed 100 m/s with an overall average rms
value of 60 m/s which was around 3% or less of the actual
stacking velocities.
It is always possible to attempt further improvement in the
residuals by pushing the inversion process, for instance by introducing higher degrees of lateral velocity variations. We
considered, on the contrary, that the present result was the
best achievable compromise because it was consistent with the
intrinsic accuracy of stacking velocity picks.
Again, it should be borne in mind that this depth-velocity
model was derived from seismic data only. These seismic
depths are probably different from the absolute depths which
will be measured in the first exploration well to be drilled in
this area. Then we will be in a position to estimate the regional anisotropy and calibrate depth maps (Williamson
et al. 1996).

Conclusions
We feel that the technology described above is well suited to
exploration in the South Caspian Basin.
The sequence of PreSTM more specifically here
demigration to zero offset followed by model-based stacking velocity inversion was very beneficial in the case presented.
The use of PreSTM improved the stacked image and made
the crucial choice of time-migration velocity very easy. This
resulted in a high quality final seismic image.
Of equal importance was that method we chose enabled us
to interpret and make use of a stacking velocity field which
initially appeared rather complex. This was done in two steps.
Firstly, 3D DZO simplified drastically the stacking velocity
field and made the velocity picking easy and accurate. Secondly, the stacking velocity inversion that takes into account
the kinematic effects of PreSTM processing, allowed the
building of a reliable macro depth-velocity model. Depth
maps of the key interpreted horizons were part of this model.
Assessment of the residuals in the inversion allowed us to
2000 EAGE

first break volume 19.4 April 2001

make sure that the accuracy of the model was consistent with
the accuracy of the input stacking velocity picks.
However, we note that the depths in the model are seismic
depths. They need to be calibrated with regional knowledge
of anisotropy. Unfortunately, a reliable calibration will only
be possible when the first exploration well is drilled. The good
news is that we are ready to carry out the stacking velocity
inversion calibrated at wells by introducing anisotropy into
the model as soon as the well data are available.

Acknowledgements
The authors wish to thank ELF Petroleum Lenkoran-Talysh
and its Partners, Wintershall, Fina Neftegaz SA, OIEC, SOA
and Total E&P Azerbaijan for authorization to publish this
paper.
Caspian Geophysical in Baku carried out the tests and production data processing.

References
Al Chalabi M. [1994] Seismic velocities: a critique. First Break 12,
589595.

technical article

Bloor, R. and Deregowski, S. [1995] Demigration to zero offset.


Abstracts of the Annual SEG Conference.
Marcoux, Godfrey and Notfors [1987] Migration for Optimum
Velocity Evaluation and Stacking (MOVES). Abstracts of the
Annual EAGE Conference.
Robein, E. and Kerimov, T. [1998] Pre-stack time migration and
vemlocity estimate in the Lenkoran Talysh area. Presented at the
Annual International Geophysics Conference in Baku.
Robein, E., Reynaud, B. and Sexton, P. [1995] Accurate simulation
of the kinematics of DMO in travel time inversion. Abstracts of the
Annual EAGE International Conference.
Robein, E., Sexton, P. and Philandrianos, F. [1997] The significance
of stacking velocities after 3D prestack time migration. Abstracts of
the Annual EAGE International Conference.
Sexton, P. [1998] 3D velocity-depth model building using surface
seismic and well data. PhD Dissertation, University of Durham, UK.
Whitcombe, D.N. [1994] Fast model building using demigration
and single-step ray migration. Geophysics 59, 439449.
Williamson, P., Sexton, P., Robein, E. and Berthet, P. [1996]
Stacking velocity inversion in anisotropic media. Abstracts X010 of
the Annual International EAGE Conference.

189
2000 EAGE

You might also like