Вы находитесь на странице: 1из 10

Pitfalls in converting Time-to-Depth in complex Sub-Andean basins in Colombia

Antonio Velasquez (Colorado School of Mines), Hector Alfonso (Ecopetrol S.A.)


Copyright 2012, ACGGP.

incidence angle; and narrow effective angles of aperture


(Figure 1) caused by laterally-varying velocities and dipping
reflectors; just to mention some of them. As a result a poor
This paper was selected for presentation by an ACGGP Technical Committee following review of information contained in an abstract submitted by the author(s).
image or a good one but with high uncertainty is obtained.
Well data, however, can be used to apply necessary constraints
Copyright 2008, ACGGP.
Abstract
to control the non-uniqueness of seismic imaging. Since
explorationists need to get well prognosis and estimation of
This paper was
selected for presentation
Technical
Committee following
Velocity
variations
affect by
thean ACGGP
quality
of seismic
data review
and of information contained in an abstract submitted by the author(s).
reserves based on seismic data, we shall focuse on a better
hinder the interpreters from defining both structure and
understanting of the source of those uncertainties: The
lithology. Structural complexity, particularly in land seismic
velocity!
data, is responsible for both poor imaging and getting misRecent imaging techniques are able to handle very complex
positioned events in depth. In spite the advances in migration,
scenarios. For instance, the transition from poststack imaging
its ability to provide right depths is quite often limited because
to prestack depth migration (PSDM) allowed migration of
of there is a substantial difference between the migration
dipping conflicting events in a smoothly laterally-varying
velocity -the best for proper imaging-, and the vertical
velocity field (Bancroft, 2007). Even most recently Reverse
velocity, used to perform the time-to-depth conversion
Time Migration (RTM) and anisotropic depth imaging
(depthing). The main reason for this discrepancy is seismic
techniques, along with the availability of better computing
anisotropy, however sometimes it cannot be adecuately
devices, allow remarkable improvements in imaging of
included in data processing of complex areas; therefore
complex structures (Helbig and Thomsen, 2005; Sarkar and
depthing is still a crucial step in the interpretation process.
Tsvankin, 2006; Tsvankin et al, 2010; Alkhalifah, T., and P.
Several techniques for depthing are based on vertical velocity
Sava, 2010, among others). In particular, new PSDM
modeling from different approaches. Using study cases in
algorithms have appeared on the market that account for
areas of complex geology, we explain advantages and
complex wave propagation for imaging sub-salt plays. At any
limitations of the most common methods: 1) Time-Depth
case the quality of the image depends on the accuracy of the
Velocity Functions, 2) Layer Cake Methods, 3) Laterallyearth model and on the technique used for wavefield
Varying Layer Cake Modeling; and 4) Geostatistical Velocity
reconstruction (Gray et al., 2001; Sava and Biondi, 2004).
Modeling. There is no a unique recipe for chosing one of
This usually requires a very high quality of the data, i.e., high
those, but depending on the complexity and availability of
S/N ratio, enough migration apertures and even wide azimuth
high-quality data, different methods can be applied or cobined
geometries-. Unfortunately, those conditions are easier to get,
to obtain better results. Most of the pitfalls come from
in general, in marine than in land seismic data.
misunderstanding the type of velocity used for both imaging
In Colombia, throughout the time, the industry has done
and depthing. In general, a good velocity model is the one that
efforts in improving the quality of seismic data, through
better honor the data and reproduce the geologic trends at the
different seismic campaigns and applying different
same time. From the real case studies, we propose Laterallytechnologies. Alfonso (2011) shows some examples of
Varying Layer Cake and Geostatistical velocity models as
imaging in complex environments, and for specific structural
standar procedures for depthing in Andean complex structures.
features in Colombia he shows how we have used different
approaches to achieve the best result. Nowadays, the use of
Introduction
post stack migration procedures is just an intermediate step,
whereas Prestack Migration algorithms are the most common
The exploration in structurally complex areas has dramatically
procedures used regarding the complexity of data. New
increased over the past two decades in Colombia due to
methods like RTM, anisotropic depth migration and full-wave
important discoveries especially in the Llanos Foothills (e.g.
inversion are becoming the best option for imaging, but still
Martinez, 2005; Cooper et al., 1995). Several challenges have
they are not a common practice due to, among other factors,
to be solved by explorationits dealing with this sort of difficult
the computational cost and the low cost/benefit ratio.
enviroments. Particularly, geophysicists face different issues
from seismic acquisition to processing that must be taken into
Why depth conversion?
account: Topography and near surface velocity variations
create statics and shallow attenuation; low signal-to-noise ratio
While seismic imaging can image the earth, in theory, very
which makes the staking velocity estimation difficult and
well, its ability to provide precise depths at the same time is
produce a weak response of the migrated energy; seismic
not always achievable. In other words, the velocity models
anisotropy causes variation in stacking velocity depending on
Copyright
2008,
ACGGP.
This
paper was
selected
for presentation by an ACGGP Technical Committee following review
of information contained in an abstract submitted by the author(s).

VELASQUEZ & ALFONSO

that are good for stacking and afterwards for imaging are not
necessarily constrained by geology and enough well data
control. The anisotropy is usually invoked as the main reason
for these discrepancies. Even looking at the simplest earth
model of flat layers (Figure 1) we realize that the horizontal
velocity should be greater than the vertical velocity VH VV ,
because of loses of seismic energy across the interfaces with
different acoustic impedance. Indeed, Grechka, V. (2009)
states that the fundamental implication of anisotropy arises
from the fact that seismic waveforms are actually nonspherical which has the following consequences for seismic
imaging: 1) the migration velocity, which is derived from
stacking-velocity analysis, is different from the actual vertical
velocity, derived from, for instance, check-shot data. 2) For
this reason there is a trade-off between depthing and imaging,
since by using the vertical velocities for imaging we will get a
poorly focused image, and conversely, by using migration or
imaging velocities for depthing will lead to depth misties. 3)
Wavefronts of P-waves usually deviates also from an
ellipsoid, which causes that isotropic migration and imaging is
no longer capable of imaging different dips because focusing
of those dips would require different isotropic velocities.
Therefore, usually the processing velocities are grater that the
vertical velocities, i.e. VSTK V0 . Nonetheless, even after
anisotropic imaging, underlying problems with the quality of
the interpretation, which is used also for vertical velocity
modeling; and the lack of well control, can also cause
significant mis-tie problems (MacKay et al, 2006).
Furthermore, the fundamental difficulty of estimating
velocities in the subsurface from surface data, lithological
heterogeneities and poorly selected processing parameters are
another reason for this trade-off.
S

VH
VV

vertical wave propagation velocities. In regions characterized


by complex subsurface structure, wave-equation depth
migration is a powerful tool for accurately imaging. This is
based on the reconstruction of the wavefields along
propagation paths (Gray et al., 2001, Sava and Biondi, 2004,
Sava and Vlad, 2011). Thus, the velocities that do the best job
focusing the migrated energy are not vertical.
Therefore, since structural complexity and poor well control in
do not allow accurate results even in terms of pure imaging,
we conclude that time-to-depth conversion still remains as a
key stage in the interpretation process, not only for performing
depthing but also as measurement of the velocity uncertainty,
which is very important in risk analysis.
Types of velocity
Sometimes using velocities from different sources may look
confusing since those have to be combined in such a way that
any mathematical operation among them must have
geophysical sense. We established above that generally
seismic processing and imaging provide velocities different
from true vertical velocities, reason why depth migration
techniques do not achieve the correct depth, even though such
imaging procedures can come up with very accurate lateral
positioning of seismic events. Now, the next question would
be: What kind of information we can use to get vertical
velocities? Since extracting those velocities directly from
reflection seismic seems unreliable, we have to use the only
additional source of data available: The well logs. In fact the
zero-offset VSP/Check-Shot experiment apparently provides a
direct measurement of the vertical velocity for a vertical well
(e.g. Balch et al, 1982). A group geophones located inside the
borehole is set up at different depths. Each one measures first
arrivals from downgoing waves traveling closely at vertical
paths from a source located near or at the surface. Velocity
data from VSP and/or Check-Shots are ideal for seismic
applications since the frequency is completely equivalent.

: Angle of aperture

Earth: Actual depth

Depth

Image TWT or z

Figure 1. Simplest earths model of flatten layers of different


velocities (left). Image obtained after wavefield reconstruction
of the ray paths (right).
Likewise, the basic question: Why depth imaging is not depth
conversion? Can be answered keeping in mind the same
problems listed above. Etris, N., (2001) clearly establishes
conceptually the difference between processing and vertical
velocities. Imaging velocities come from seismic processing
and their objective is a proper focusing and lateral positioning
of the seismic energy, whereas depthing is a pure vertical
transformation from time domain to actual depth, using true

Depth

Depth
Calibrated DT

Twt

Twt

Figure 2. Example of sonic log calibration

Actual DT

SPE

Pitfalls in converting Time-to-Depth

On the other hand, sonic logs are often used as a source of


vertical velocity. However, the sonic-derived velocity needs
an additional calibration (Figure 2) in order to be compared to
actual seismic values, because unlike those velocities from
VSP and/or Check-Shots sonic data come from a source with a
higher frequency content than any actual seismic source, and
also because the wave propagation characteristics are different
in sonic logs. Typically the maximum frequencies in seismic
are < 100 Hz, while the sonic slowness (DT) is averaged over
the interval covered by sonic-log receivers, which is of the
order of 1m, and it is measured typically at frequencies from 2
to 20 kHz. This implies that the sonic log calibration (e.g.,
Figure 2) is a necessary step even before performing synthetic
seismograms to get the well-seismic tie.

This result provides in advance two very important results.


First of all, we can notice from the definition of average
velocity that this is the right velocity to compute the mapping
from time-to-depth; in other words, the average velocity is the
one to be used for depthing. Besides, as long as we have a
model of interval velocities we can compute the average
velocity in a straightforward way. Notice that the form of (2)
is on purpose written in such a way that the weights are given
by interval travel times t instead of actual thicknesses z ,
which has immediately an application in depthing: The
velocity model is built in time domain and afterwards depth
conversion is performed by a simple multiplication.
Datum

Interval Velocity
z1

This velocity can be defined as the representative value of the


speed of propagation of seismic waves in the vertical direction
within a given rock interval. For a particular layer with depths
zi and zi 1 at the bottom and the top, respectively, we measure

Topo

V1 Vr

t1

P1
t2

z2

V2

z j

Vj

Pj

t j

zn

Vn

Pn

tn

P2

the interval thickness ttzi - zi 1 , e.g from VSPs/check-shots or


calibrated sonic logs. The interval velocity is given by

VINT

z z
V i i 1 .(1)
ttzi ttzi 1

The notation in equation (1) means that unless we specify


otherwise V is vertical interval velocity.

Average Velocity
The average seismic velocity at certain point P, is the depth
divided by the vertical travel time at such point, assuming
depths and times measured from the same reference datum(Figure 3). In the case of a VSP/Check-Shot we have a direct
measure of this quantity; however, for modeling velocities
away from well control, which is the main challenge in
depthing, we have to compute it generally from a model of
interval velocities. For instance, in the simple earth model
composed by flat layers with different thickness and velocities
V (Figure 3), at P1 the average velocity is equal to the interval
velocity, but below this point we have to consider the
influence additional layers with different interval velocities.
One way to accomplish this is by doing layer stripping which
leads to take a simple weighting average of the interval
velocities above the target point, for instance at Pn we get the
generalized expression
n

V t
j

Vavg

j 1
n

t
j 1

.....(2)
j

Figure 3. Average velocity approximated by layer stripping


leads to weighted average of interval velocities.
Instantaneous Velocity
This is, in general, the actual velocity at any point. Physically,
it can be thought as the interval velocity of an infinitesimal
thick layer, and mathematically it is the derivative of a depthtime function evaluated at certain depth. Usually, analytical
expressions describing the instantaneous velocity with depth
are developed. The simplest of those approximations is
assuming that the velocity varies linearly, from a reference
value V0 , with depth at certain rate k. That is,

Vinst

dz
V0 kz (3)
dt

Marsden (1992) re-defines this simplest, but popular and


perhaps the oldest method used in industry, as Vo-k method,
and provides guidelines on its applications and limitations in
practice. Naturally, any velocity profile that violates this
assumption of linearity cannot be modeled using (3). We can
actually obtain easily V0 and k for each formation (or rock
interval of interest) from calibrated sonic logs (Figure 4) as

VELASQUEZ & ALFONSO

long as the basic assumption holds. Selecting linear segments


and computing the slope we get k whereas V0 is just the value
of the sonic velocity at the top of each interval.

V
V0
k1
Z1

V1
k2

Z2

V2

Z3

k4

k3
V3

V4

Figure 4. Vo-k method of instantaneous velocity. Different


layers having different values of k (left), and an example
(modified from Marsden, 1992) of a sonic velocity showing
some linear segments (right).
There is common source of error especially in commercial
applications for structural modeling, where values of V0 and k
for each stratigraphic unit have to be provided as an input to
perform a time-to-depth conversion of interpreted horizons
and faults. In complex compressive areas the fact of assuming
a linear behavior for the same geologic unit leads to important
mistakes because the same unit can be presenting in different
thrust sheets. Since the velocity is strongly controlled by
burial and finally by tectonic history, there is no reason for
using the same velocity function everywhere. Likewise, the
faults and stratigraphic contacts may create velocity
distortions in short lateral distances which are no necessarily
true for seismic waves propagation. To enhance how sensitive
this approximation might be, lets integrate the equation (3) to
obtain depth as a function of travel time

Methods of depthing: Case studies in Colombia


According to Etris (2001) we can divide the depthing methods
is two main groups: 1) Direct time-depth conversions and 2)
Velocity-modeling-based time-depth conversions. Direct timedepth conversions are crude computations of depths from
travel times, taking into account that we know the actual
depths in the wells. So, we force the result to match the actual
well markers without regard to the structure of the velocity,
but bringing some implicit velocities in the process. On the
other hand, Velocity-modeling-based time-depth conversions
do care about the structure of the velocity field, therefore we
can obtain the vertical velocity that reproduces the best not
only the expected depths at the well location but also the
controls that geology exerts on velocity field, allowing more
reliable models and more accurate predictions far away from
well control points.
It seems logic that we focus mostly on the second group,
however it is important to analyze the direct methods first
since they are still a common practice, and we can learn some
conceptual pitfalls from it. First of all, direct methods can still
be valid tools in some circumstances. For example, if the
lateral variation of velocity is negligible, the interpretation is
feasible and we are drilling a well very close to a previous
one, perhaps performing a velocity modeling is not worth it
for depthing purposes since we only care about the prognosis
at depth of the target horizon and a better velocity model does
not guarantee a better forecast. Indeed, Etris (2001) describes
a paradox between a good velocity model and the accuracy of
the prognosis (Figure 5). The poor velocity model does not
reproduce the actual data but still it can come up with a good
result in the prognosis at the target depth (direct conversion).
Conversely, increasing the accuracy of the velocity model
makes the error at the tie point rises as well.

Vo(ekt 1)
..(4)
Z
k
Here, it is clear that in spite the linear behavior of the velocity
with depth; the actual depth is predicted using a non-trivial
exponential function of the travel time, which makes the k
factor a potential source of error. Thereby, this approach must
work fine if this parameter is properly constrained.

Time

Time

Model T-D

Model T-D

Actual T-D

Non-Zero error
Depth

Acceptable velocity model

So far, we have provided the basic understanding about why


we need to perform depthing of migrated seismic data. We
interpret on a migrated image, in time or depth, but we have to
resize the image to actual depths using vertical average
velocities and ultimate well control for both time-to-depth and
depth-to-depth conversions. We also discussed the types of
vertical velocities currently used and how to obtain them. Now
we shall focus on the methods to perform depthing.

Actual T-D

Zero depth error


Depth

Poor velocity model

Figure 5. Paradox between velocity modeling and depth


estimation (modified from Etris, 1992).
Direct T-D conversion: Time-Depth velocity function
The Figure 6 shows an interpreted 2D migrated seismic
section. The exploratory well is highlighted in red, and there is
another well pretty close. Sometimes is hard to define whether
or not the other well is close. At the end, it depends upon the

SPE

Pitfalls in converting Time-to-Depth

stratigraphic column in both well trajectories rather than a


pure distance criterion. In this example we take advantage of
the relatively simple geology above the unconformity to
assume the lateral variation in velocity is negligible. Then,
using the Check-Shot data from the correlation well, we
compute an analytical function, a fourth order polynomial in
this case, via regression in the sense of least squares (Figure
7). This regression has a correlation coefficient of 0.99; which
in principle adjusts the actual data very well. Thus, the depth
conversion of the horizon of interest at the location of such
well must be almost perfect. Another important thing to be
noticed is that the geometry of the structure does not change
after depth conversion because functions with this form
Z f (Twt ) , only consider the vertical variation of the
velocity, so the lateral changes are not being taking into
account at all. This T-D function is continuous, therefore it has
implicit an instantaneous velocity function according to the
definition given in equation (3).

Figure 6. Seismic line interpreted showing the exploratory


well and its target, the green horizon in a faulted anticline
under an angular unconformity.

measure of the error of the depth conversion. At target depths


(blue rectangle) the difference was about 30 ft., which is
equivalent to less than 0.4%. However, this outstanding result
contradicts what we observe above and below the target,
where the error increases rapidly. This is a good example of
the paradox discussed above; since we have a poor velocity
model that does not consider any lateral variation, although
the accuracy in the prognosis was almost perfect.
Velocity modeling: Layer Cake method
Layer cake is an adopted term that actually refers to the fact
that we can compute depths by layer stripping as long as we
know the interval velocities. Thus, the approximation of the
average velocity given by the equation (2) is all the foundation
behind this method. The problem here is how to get a feasible
model of interval velocity, and which velocities we should
use. Here we establish the following two steps: 1) Compute
representative values of interval velocity, from VSP/CheckShots/Sonic, for each geologic unit and/or structural sheet.
The interval velocity no necessarily has to be a constant value.
For instance, remember that linear trends in the calibrated
sonic logs can be used to define instantaneous velocity
functions (equation 3) instead. 2) Since usually the well
control available is poor, for the units and/or structural blocks
without control we need to use seismic processing velocities to
find those missing values. Seismic data does not provide
enough resolution to identify any instantaneous velocity
function, but the interval velocities can be approximated from
the processing velocities by using the well-known Dix
equation (see: Dix, 1955; Yilmaz, 1987)

Funcin Tiempo-Profundidad. Pozo Llanito - 1.


TT(ms) @ 50m
-100

100

300

500

700

900

1100

1300

1/ 2

T jVRMS 2j T j 1VRMS 2j 1
Vj

T j T j 1

1500

-500
0

500

TWT

5K

1000
1500

.(5)

2000
2500

3000
3500

Where the travel times are given by: T j

4000
4500

5500
6000
6500
7000

TVD

TVD (Ft) @ 50msnm

t
k 1

5000

Time structure

Structure map

7500

Thickness

8000
8500
9000
9500

Target

10000
10500
11000
11500
12000
12500
13000

y = -1E-09x4 + 2E-06x3 + 0,0024x2 + 6,9891x - 157,5


R2 = 0,9995

13500
14000

Figure 7. T-D velocity model using a correlation well (left),


structure maps (upper-right corner), and the results after
drilling (right-down). The blue spot in the maps is the location
of the correlation well.
Check shot

Polinmica (Check shot)

Finally, lets observe the results after drilling in the table


shown in Figure 7. The last column is the difference between
the prognosis and the actual drilled depths; hence it is a direct

The Dix equation relates the interval velocity with the rootmean-square velocity VRMS . Traditionally, this velocity is used
erroneously for stacking velocity which is true only if the
offset approaches to zero, velocity layering are parallel and the
layers are isotropic. Using the Dix equation to get vertical
interval velocities from RMS velocities is a very common
pitfall in depth conversion since anisotropy is generally
present, and parallel layering is just a very simplistic
representation of the earth. Nonetheless, some assumptions
can be done in order to use the processing velocities in some
way. In terms of depthing we take care of anisotropy during
the time-to-depth conversion. Besides, we interpret migrated
data so we have a condition of zero-offset data somehow.
Thus, we can make this rough approximation VRMS VSTK as
long as we are always aware about the nature of the Dix
interval velocities (5), which are inaccurate and do not

VELASQUEZ & ALFONSO

represent actual vertical velocities either. An adequate way to


use processing (migration) velocities is to compute the Dix
interval velocities, and then to find tendencies and relative
differences with actual vertical velocities, e.g. comparing with
check-shots, to estimate the amount of change in zones
without well control. We shall see below that the best way to
do this properly is by applying Geostatistical methods in the
average velocity domain, but for now lets see how this
approach works in a real example.

well. So, in this case definitely this would be the worst


velocity model to perform a time-to-depth conversion.

Figure 8. Seismic interpretation of a complex transpressive


basement-involved structure. Notice a potential pull-up
effect in the foot-wall of the faulted basement.
Figure 8 shows an example in a structurally-complex setting
dominated by high-angle basement-involved reverse faults.
The block in purple is basement so we expect a high velocity
within it. Notice that the surface anticline and the anticline in
the foot-wall of the main fault have almost the same
wavelength and their crests - the highest points- are laterally
positioned in the same spot (black arrow), what makes this
structure a classical candidate for being a pull-up. Well, layer
cake modeling is a very useful way to test the impact of the
velocity on the integrity of a potential structure (velocity
sensitivity analysis). Thus, if we have the interval velocity of
the basement we can change this value within a feasible range
to see how such a change affect the geometry of the depthconverted structure. The figure 9 shows both the layer cake
model and the average velocity model obtained from it using
equation (2). Although this velocity model looks fancy and
geologically correct, its ability to predict depths accurately
might be very limited because the layer cake model comes
from the interpretation and therefore the average velocity is
biased by a subjective input. In complex areas like this one we
do not want to have velocity models biased by interpretation
since that is one of the strongest sources of uncertainty.
It is important to enhance the fact that we do not have
velocities from well logs everywhere; therefore we obtained
interval velocities propagating the differences between actual
interval velocities, from wells, and Dix interval velocities.
Unlike this model, in the figure 10 the average velocity is
obtained using only the processing velocities. As we discussed
above the velocity field is inaccurate and usefulness because
these are not vertical velocities, but in addition we see that the
average velocity field does not resemble the geology quite

Figure 9. Interpretation-based layer cake interval velocity


model (above), and computed average velocity (below).
Velocity modeling: Laterally-Varying Layer Cake
This is a variation of the previous method including lateral
variation within the layers. The most common used approach
is the horizon-guided interpolation where the interpolation of
velocities is controlled by interpreted horizons and faults, so
the interpolation between two points is constrained to follow
some structural trend. The advantage of this approach is that
we can include lateral variations but the problem is still the
same, as long as we have uncertainties in the seismic image
and in the interpretation any horizon-guided interpolation will
result in a biased version of the data. The figure 11 shows a
typical example of laterally-varying layer cake velocity model
in a relative quite zone. We can observe a time structure map

SPE

Pitfalls in converting Time-to-Depth

at the top of some particular unit and some dispersed wells


with velocities interpolated following the interpretation. This
king of model is, however, much better that the models in
figures 9 and 10 since it looks geological, respect the data and
includes lateral variations.

Figure 10. Average velocity model obtained from processing


velocities.

Figure 11. Horizon-guided interpolation. Notice that the


velocity comes from a poor well control.
There is another way to combine these two layer cake
methodologies and it consists in creating vertical pseudo wells
within the seismic survey. At some selected CDP locations we
measure the travel times for each interpreted horizon and/or
fault which have assigned certain interval velocity value from
the actual VSP/Check-Shots surveys. Measuring travel times
is often time-consuming and laborintensive depending on the
density of sampling. Finally, we get a collection of pseudo
wells with interpreted points and computed average velocities
covering the entire area of interest. The difference from the
layer cake is that here we do not built a 3D structural model
for modeling velocities, like the one showed in figure 9,
instead, we interpolate directly the velocities to populate the
entire 3D model. Likewise, the difference from the
conventional laterally varying layer cake velocity model is

that the input from the interpretation is given previously while


reading the travel times, but the interpolation is not horizonguided anymore. This variation is useful when we have, for
instance, a poor quality of seismic data and therefore high
uncertainty in the interpretation, but at the same time when we
need to capture some dominant structural features. As usual, to
see how it works lets analyze the next example.

Figure 12. Seismic line interpreted showing a complex folded


and thrusted belt in the Colombian Andes. The white arrow is
a scketched projection of an exploratory well.
We have a complex set of stacked thrust sheets developed
inside a triangle zone in a typical deformation front in the
Colombian Andes (Figure 12). The seismic data is a poor 2D
PSTM, but there are structural reasons to believe that this
interpretation is actually feasible. In fact, even the seismic data
helps to define major structural features. The main target is a
reservoir located in the shallower thrust sheet, between 1200
and 1700 ms. We performed the average velocity modeling in
pseudo-wells instead of using layer cake and the interpolate
those values using the common inverse of distance
interpolation (Figure 13). On the other hand, we do not need to
create any pseudo wells at the actual well locations.
Furthermore, the velocity model is not biased by the complex
interpretation but it does follow the major structural feature:
The triangle zone with high velocities in the middle. The final
velocity model was refined by using geostatistics instead of
distance-based interpolation (Figure 14). If we compare the
velocity model with the actual average velocities in one
particular well we conclude that the accuracy reached is
outstanding. Below we cover some general aspects about
velocity modeling using geostatistics in more detail.
Most importantly, we can include uncertainties to each value
according to the degree of confidence of the source (figure
13), thus we can associate distributions of probability to each
velocity, and then come up with prognosis given in ranges
rather than fixed values. The structure of interest was depth
converted and the prospect was drilled afterwards. Thereby,
we can see the feedback from nature in the figure 15. The
vertical axis is the uncertainty of the prognosis value in feet
and the horizontal axis is the actual depth. The bars represent
each geologic marker and the most likely range of variation of

VELASQUEZ & ALFONSO

the predicted depth in feet, according to the maximum and


minimum variation of the modeled velocities.

ranges the depth conversion can be consider successful


because it was able to predict both depths and uncertainties.
For this particular model (Figure 14) we have very good
results above 8200 ft., but below that point the difference
between expected and drilled depths increases rapidly,
although it remains inside the predicted error.
2000

Uncertainty (ft)

1000

-1000

-2000
0

4000

8000

12000

16000

20000

Depth (ft)

Figure 15. Results after drilling of the structure in figure 12.


Red bars represent geologic markers and their expected
variations due to velocity uncertainties. The dashed blue line
is the difference between prognosis and well markers.
Velocity modeling: Geostatistical methods

Figure 13. Interpretation-based pseudo-wells with computed


average velocities (above). Each value has an uncertainty
associated modeled with a distribution of probability. And
interpolated average velocity (below).

Vavg

Figure 14. Geostatistical model of average velocities from


pseudo-wells and data from actual VSP/Check-Shot surveys.
Further, the dashed blue is the result after drilling. We can say
now that as long as this line does not go outside the predicted

According to what was mentioned above, the best velocity


model for depthing is the one that respects the hard data from
wells, but also can use seismic processing velocities, and is
geologically constrained at the same time. After all those
examples it seems like we would need a lot of high-quality
information to be able to build one robust velocity model for
depth conversion, but actually we can make it by being aware
of the value and limitations of both the velocities and, of
course, our geological model. The previous model is the proof
of that: we have just a couple of wells, 2D seismic data, a very
complex structure, and by combining the inputs in adequate
proportions we were able to take advantage of some
Geostatistical method to get a fair enough result.
Geostatistics gathers a group of optimization techniques to
process spatial data. Invented for solving mining problems,
geostatistics uses powerful mathematical tools that attempt to
streamline and quantify the model building process. The main
problem with distance-based interpolations is that all data is
considered statistically independent and the weights only
depend upon the distances. That is why, for instance in the
model of the figure 13, we lose some geological
characteristics. Unlike distance-based methods, geostatistics
uses variograms to determine the interpolation weights.
Hence, if the velocity is correlated better in some direction the
variograms will show such preference therefore the weights

SPE

Pitfalls in converting Time-to-Depth

are greater in that particular direction. The simple kriging


estimator to find the value of the variable Z at some point X0 ,
using n measurements is given by (Davis, 2002)

Z ( X0 ) m

Z (X ) m .(6)
i

i 1

( Xij ) i ( X0 ) .

Here m is the mean and the weights i are given by the matrix
system above, which depends on the variograms of the data
( Xij ) and of the estimation vector ( X0 ) . Inverting this
matrix ( Xij ) is computationally expensive. Nonetheless, if
we approximate the variograms by simple analytical functions,
then this big matrix is guaranteed to be positive definite,
which can be inverted faster. That is part of the routine, we
have to compute variograms from our velocity data, either
from wells or seismic, and then approximate them by simple
functions. The figure 16 shows the 2D variograms used to
interpolate the velocity model in the figure 14. Maximum
levels of correlation in this particular case are oriented 150
degrees of azimuth, and minimum correlation distances were
found at 90 degrees.

hard data comes from the well velocity data, while the drift
data come from all the pseudo-wells.
Conclusions: Summary of Pitfalls
The most common pitfalls in converting time-to-depth in
complex areas, like Sub-Andean basins in Colombia, generally
come from misunderstanding of imaging and depthing, and
therefore from our lack of knowledge about the velocities
involved in those processes. Here we summarize some
erroneous but common depth-conversion practices: 1)
Interpreting depth migrated data without any additional
correction or sensitivity analysis. 2) Applying direct depth
conversion techniques without regard to the lateral structure of
the velocity. 3) Modeling velocities from seismic processing
to perform depth conversions directly as if they were actual
vertical velocities. Likewise, encourage processing guys to use
vertical velocities for imaging. 4) To believe that the velocity
model that resembles the most the geological features is the
best for depthing, this is not always true. 5) Using linear
functions of instantaneous velocity, Vo-k method, in complex
structures, assuming constant values of Vo and k in different
thrust sheets. For instance, commercial applications for
structural modeling have this tool available but their values are
usually fixed in a stratigraphic framework.
Conclusions: Summary of methods

Figure 16. Variograms computed in different azimuths using


the pseudo wells of the figure 13.
Geostatistic procedures can benefit from a wealth of data.
Particularly, the so-called kriging with external drift can
handle two types of data sets. One is the hard data, generally
limited and punctual, and the other one is the drift data, which
is less accurate but much more available and widely
distributed in space. In velocity modeling we have naturally
hard data of vertical velocities from well logs (VSP, CheckShots and calibrated sonic logs), and drift data from seismic
processing velocities. The hard data have more weight since
they are the values to be interpolated, while the drift data will
guide the interpolation but unlike horizon-guided
interpolation, the weights from the system (6) play a
fundamental role as well. This is exactly the procedure
followed to get the final velocity model in the figure 14. The

There is no a recipe or an ultimate method for depth


conversion. First of all, it is necessary to spend more time and
make the biggest effort in getting the best imaging result. In
the best scenario depthing may be avoided if we have a good
image obtained from anisotropic depth imaging (e.g. PSDM)
and enough well control that allows us to resize the image to
actual depths being consistent with well tops.
Nonetheless, generally depth-conversion is still necessary.
After a cost/benefit analysis some depthing method must be
chosen. Each method applies better in certain circumstances,
although depth-conversions based on velocity modeling are
strongly recommended, even poor velocity models can
eventually come up with good results. In this regard, a
velocity model-driven depth-conversion procedure should be
controled by a modeling process that properly honors
structural detail. Thereby, depthing must be performed in
order to get not only accuracy in the prognosis, especially far
away from well control, but also geological sense and the
possibility to quantify the uncertainties.
The best velocity model is the one that honors the hard data,
uses other sources of information, and is geologically
constrained at the same time. Although Geostatistical methods
seem to be more robust than any other technique, sometimes
characteristics of a good model can also be found in laterallyvarying velocity models, being the last one, indeed,
computationally less expensive.

10

VELASQUEZ & ALFONSO

Acknowledgments
We want to thank Ecopetrol S.A. for providing, during the last
years, all this the data set, to Colorado School of Mines for its
support, Tom Davis and Ilya Tsvankin for advising and
technical discussions, and to all our colleagues, geophysicists
and interpreters in Ecopetrol S.A., who contributed
enormously to this work.

References
Alfonso, H., 2010. Evolution of Seismic Imaging in Colombia.
72nd EAGE Conference & Exhibition incorporating SPE
EUROPEC.

Marsden D., 1992. Vo K method of depth conversion. TLE.


Martinez, J., 2005. Structural evolution of the Llanos foothills,
Eastern Cordillera, Colombia.. Journal of South American
Earth Sciences 21 (2006) 510520.
Sarkar, D., and I. Tsvankin, 2006. Anisotropic migration velocity
analysis: Application to a data set fromWest Africa:
Geophysical Prospecting, 54, 575587.
Sava, P., and B. Biondi, 2004. Wave-equation migration velocity
analysisI: Theory: Geophysical Prospecting, 52, 593
606, doi: 10.1111/j.1365-2478.2004.00447.x.
Sava, P., and I. Vlad, 2011. Wide-azimuth angle gathers for waveequation migration: Geophysics, 76, S131S141,
doi:10.1190/1.3560519.

Alkhalifah, T., and P. Sava, 2010, A transversely isotropic medium


with a tilted symmetry axis normal to the reflector:
Geophysics, 75, A19-A24.

Tsvankin, I., 2005. Seismic signatures and analysis of reflection data


in anisotropic media: Elsevier Science Ltd.

Balch, A. H., Lee, M. W., Miller, J. J., and Rider, R. T., 1982. The
use of Vertical Seismic Profiles in investigation of the
earth: Geophysics, 47, P. 906-918.

Tsvankin, I., J. Gaiser, V. Grechka, M. van der Baan, and L.


Thomsen, 2010 Seismic anisotropy in exploration and
reservoir characterization: An overview: Geophysics, 75,
no. 5, A15A29, doi: 10.1190/1.3481775.

Bancroft, John C., 2007. A Practical Understanding of Pre- and


Poststack Migrations: SEG Course Notes Series No. 14.
Volume 2: Prestack.

Yilmaz, ., 1987. Seismic data processing: Society of Exploration


Geophysics.

Cooper, M.A., Addison, F.T., Alvarez, R., Coral, M., Graham, R.H.,
Hayward, A.B., Howe, S., Martinez, J., Naar, J., Pen as,
R., Pulham, A.J., Taborda, A., 1995. Basin development
and tectonic history of the Llanos Basin, Eastern Cordillera,
and Middle Magdalena Valley, Colombia. A.A.P.G. Bull.
79 (10), 14211443.
Davis John C., 2002. Statistics and Data Analysis in Geology. John
Wiley & Sons.
Dix, C. H., 1955. Seismic velocities from surface measurements:
Geophysics, 20. p. 68-86.
Etris, E. L., Crabtree, N., J., Dewar, J. 2001. True Depth
Conversion more than a pretty picture. CSEG recorder.
Gray, S. H., J. Etgen, J. Dellinger, and D. Whitmore, 2001. Seismic
migration problems and solutions: Geophysics, 66, 1622
1640, doi:10.1190/1.1487107.
Grechka, V., 2009. Applications of seismic anisotropy in the oil and
gas industry: European Association of Geoscientists and
Engineers.
Helbig, K., and L. Thomsen, 2005. 75plus years of anisotropy in
explorationand reservoir seismics: A historical review of
concepts and methods: Geophysics, 70, no. 6, 9ND
23ND.
MacKay, S., H. R. Jimnez, J. San Martin Romero, and M. Morford,
2006. Calibrating prestack depth migration volumes with
well control: SEG, Expanded Abstracts, 25, 530534,
doi:10.1190/1.2370315.

Zhou, H.,D. Pham, S.Gray, and B.Wang, 2004. Tomographic


velocity analysis in strongly anisotropic TTI media: 73rd
Annual International Meeting, SEG, ExpandedAbstracts,
23472350.

Вам также может понравиться