Sie sind auf Seite 1von 9

ACQUISITION/PROCESSING

Downloaded 08/01/16 to 68.146.236.200. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Coordinated by William Dragoset

Processing/integration of simultaneously acquired 3D surface


seismic and 3D VSP data
SATINDER CHOPRA and VLADIMIR ALEXEEV, Core Laboratories Reservoir Technologies Division, Calgary, Alberta, Canada
ABHI MANERIKAR, Tiger Exploration Consulting Incorporated, Calgary, Alberta, Canada
ANDREW KRYZAN, ConocoPhillips, Calgary, Alberta, Canada

SP surveys use surface sources and borehole receivers and


record both downgoing and upgoing wavefields. Different
configurations of VSP surveys are possible and the various
types of VSP data sets are valuable to the interpreter in areas
where surface seismic data are of limited help.
Common applications of zero-offset VSPs are correlating
logs and seismic data, getting information about variations in
seismic velocity with depth, and detecting multiples. Offset
VSPs are usually recorded where subsurface structures are not
well known, or are complex. Their images, which are of higher
frequency than surface seismic, are used to obtain structural
detail around boreholes. Multioffset or walkaway VSPs are a
means of lateral reservoir definition away from the borehole.
They combine the advantages of VSP (including the possibility of anisotropy determination) with the continuous profiling aspects of surface methods.
However, these single or multioffset VSPs are usually
obtained as 2D sections and these sections, though useful, cannot be used to construct a 3D data volume around the borehole. Different acquisition geometries are needed to achieve
that objective.
The first 3D VSP acquisition was attempted in 1986 by
AGIP, and several have been acquired since then (Table 1).
Simultaneous acquisition of 3D VSP and 3D surface seismic
acquisition has been attempted on several occasions and the
results have been encouraging in terms of cost effectiveness
and the enhanced imaging of the subsurface. This innovative
acquisition and analysis offers exciting possibilities for determining data attributes such as anisotropy, Q, and measurement of reflection and transmission amplitudes at the wellbore
which can then be used to calibrate AVO attributes estimated
from surface seismic 3D data.
Although many 3D VSPs have been discussed in the literature, few articles go into detail about the processing (Gulati
et al., 2002). Therefore, a paper describing the complete processing sequence, the problems encountered, and how they
were overcome, is in order. This work describes the simultaneous acquisition of 3D surface seismic data and 3D VSP data,
and details the different steps used in processing the two data
volumes. Attempts were made to enhance the bandwidth of
the 3D seismic data using the VSP downgoing wavefield and
to integrate the two data volumes so that they could be interpreted together. Attempts at determination of data attributes,
however, have not been made yet, and will be investigated
separately.
Geologic background. Exploration in the area of Coyote in
the Western Canada Sedimentary Basin is principally for gas
in the Belly River, Viking, Upper Manville (Colony and
Glauconitic sandstone), Lower Mannville (Ostracod sandstone), and Banff formations. Oil is also found in the Lower
Mannville (Ellerslie sandstone) and Banff formations.
3D VSP and coincident 3D surface seismic data were
422

THE LEADING EDGE

MAY 2004

Table 1. Some highlights of 3D VSP history


Year
1986
1989
*
1993
1994
1995
1995
1995
1995
1996
1996
1997
1997
1998
1999
1999
1999
2000
2000
2000
2001
2001
2001
2002
2002

Company
Location
AGIP
Brenda
Phillips Pet.,Norway Ekofisk-K17

Receiver
8 levels
8 levels

Shell
Phillips Pet.,Norway
Phillips Pet.,Norway
Norsk Hydro
Phillips Pet.,Norway
PanCanadian
AGIP, Luisella
Petrozeit, Egypt
British Petroleum
Phillips Pet.,Norway
Chevron
Output Expl. Inc.
PanCanadian
Unocal
Crestar Energy
PanCanadian
-

5 levels
5 levels
5 levels
5 levels
5 levels
5 levels
8 levels
12 levels
40 levels
80/40 levels
80 levels
65 levels
80 levels
80 levels
23 levels, overall 86 levels
80 levels
80 levels
80 levels
80 levels
36 levels

British Petroleum
British Petroleum

Brent
Eldfisk
Ekofisk-K6
Oseberg
Ekofisk-K3
Blackfoot
Magnus Field
Ekofisk-C11a
Lost Hills Field, California
S. Louisiana, Salt Basin
Weyburn, Saskatchewan
Indonesia
Coyote, Alberta
Bakersfield, Calif.
Christina lake
West Texas
Wyoming
California
Alaska
Gulf of Mexico

Components
1
1
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3

* There is a reference by Shekhtman et al. from Neftegeofizika (EAGE meeting, 1993)


about a 3-level 3D, VSP shot in Kazakhastan, but not enough information is available.
Lost Hills is actually a 3D time-lapse VSP with a base survey in 1998 and repeat
surveys in 1998, 1999, 2000.

recorded around well 8-20-29-15W4 drilled by Crestar Energy


to target the Lower Mannville. This well is in the southwest
quadrant of a four-section 3D surface seismic survey. It encountered a gross Lower Mannville interval (Figure 1) that was 20.5
m thick (base Bantry shale to top Banff isopach), significantly
thicker than at adjacent wells 6-20 (11.5 m) and 8-21 (7.0 m).
This interval encountered a 5-m Ellerslie sand with log density porosity greater than 12%. Although the well logs indicated good permeability in this sand, on completion, which
included a sand frac, this interval swabbed dry with only
traces of oil produced. A thin (1 m) Ostracod sand interval,
which in five nearby wells has produced a total of 9.5 Bcf of
sweet dry gas, also swabbed dry on completion. The regional
Glauconitic sand in this well was completed successfully, with
flush production test rates in excess of 500 Mcf/d. This well
is expected to stabilize quickly at 200 Mcf/d and to produce
a total of about 0.3 Bcf, but this production is economic only
when sunk costs are ignored and because of the wells proximity to an existing gas gathering system.
The 3D seismic programs (surface and VSP) were recorded
MAY 2004

THE LEADING EDGE

0000

Downloaded 08/01/16 to 68.146.236.200. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Figure 3. Perspective view of elevation (in meters) in the 3D survey area.


Figure 1. Well logs showing different formation levels and the Ellerslie
sandstones which are the targets of interest.

Figure 4. Inline 85 passing through the well.

Figure 2. Geometry used for 3D surface seismic acquisition and the ring
geometry used for processing 3D VSP. W represents the position of the
well. The VSP data are sorted into rings or bands of radial offsets. The
wavefields within each ring then are similar to each other and so can be
separated with the same set of parameters.

to assist in defining sand presence and porosity development


in the Lower Mannville interval (Ellerslie sandstone) at 13001350 m. The objectives for the 3D VSP recording were to (a)
tie the seismic reflections to lithology and stratigraphic boundaries, (b) obtain a high-frequency image around the borehole
(the fixed receiver array and its proximity to the reservoir are
expected to improve image quality), and (c) obtain an
0000

THE LEADING EDGE

MAY 2004

Figure 5. Coherence time slices at different stages of processing.

improved subsurface velocity model.


Simultaneous data acquisition of surface 3D and threecomponent 3D VSP data was carried out by Crestar Energy
(now ConocoPhillips) in October 2000 in the Hanna area in
central Alberta. The geometry for the 3D survey was the Mega
Bin staggered brick pattern (Figure 2) with a source spacing
of 70 m and a receiver spacing of 140 m. Sample interval was
1 ms, record length was 3 s, and energy source was dynamite.
Using the same seismic source, Paulsson Geophysicals 80
level tool with 15-m interval, and a three-component geoMAY 2004

THE LEADING EDGE

423

Downloaded 08/01/16 to 68.146.236.200. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

phone cable in well 8-20, data were simultaneously recorded


in the borehole. The acquisition required approximately 3000
shots.
Processing of 3D surface data. The quality of the seismic data
was reasonably good. Figure 3 is a perspective view of the
elevations in the area which vary from 937 to 993 m. All values are measured with respect to mean sea level. The standard processing sequence included:
1) First break picking, shot and receiver statics applied to
replace the weathering layer with bedrock, and placing the
shot and receivers on a floating datum corresponding to a
smoothed version of the topographic surface.
2) Spherical divergence correction.
3) Four-component surface consistent designature deconvolution.
4) Velocity analysis and NMO.
5) Datum correction to move shots and receivers to flat datum.
6) Surface-consistent shot and residual statics.
7) Residual statics correction to step 4.
8) Inverse of step 4 to move shots and receivers back to floating datum.
9) Application of inverse NMO using velocities of step 4.
10) Velocity analysis and NMO.
11) CDP residual statics
12) Datum correction to move the shots and receivers to a flat
datum.
13) Rebinning from 35 70 m to 35 35 m.
14) Mutes and stack (referenced to flat datum 1000 m above
mean sea level).
15) Kirchhoff time migration.
Figure 4 shows an inline from the final processed
(migrated) volume. To provide QC, coherence cubes (Chopra,
2002) were generated at different stages in the processing flow.
Each processing step is expected to add value to the previous
step by better focusing events. Should this improvement not
be evident on the coherence slices, the processor would go
back and fine tune some parameters. Figure 5 compares coherence slices of a channel running N-S. Notice how, as the processing advances from brute stack through residual statics and
migration, the edges of the channel are better defined.
Processing of 3D VSP data. Since, as stated earlier, this is an
area that is not routinely discussed in the literature, the processing flow is presented in detail.
1) Extraction of the z-component (vertical) from the three-component data. This was the input to the subsequent processing.
2) A very significant aspect in VSP data processing is that the
VSP wavefield structure is strongly dependent on the shot-receiver
distance and dip/azimuth. Consequently, for 3D VSP it becomes
difficult to use the same set of parameters for wavefield separation. For flat geology, in order to keep the parameter description simple, data are sorted into rings or bands of radial
offsets The wavefields within each ring are similar and can
be separated conveniently with the same set of parameters.
So, next, the data were sorted into different rings (marked
in Figure 2) according to source distance from the well (not
necessarily using a cylindrical coordinate system). Distance
between rings was fixed at 50 m for 32 rings or nearly 1600
m from the well. However, the amplitudes and signal-to-noise
ratio were too low on the farther rings to allow any meaningful conclusions. Figure 6 shows some representative records
from different rings (offsets). Wavetrains having relatively
high amplitudes, and low frequency are seen crossing the
wavefields, especially the nearer rings. These are the tube
424

THE LEADING EDGE

MAY 2004

waves or the high-amplitude Stoneley waves that travel along


the cylindrical fluid-rock boundary for the length of the borehole.
3) An obvious processing objective at this point is to extract the
upgoing and downgoing wavefields and prevent the contamination
of the data by the tube wave. Separation of these waves from useful data was quite a challenge. In addition to the tube waves,
some VSP gathers in each ring (about 15% in all) were very
noisy and had to be rejected. Separate noisy traces within each
gather were edited using surgical muting and bandpass filtering. Figure 7 shows a representative total VSP wavefield
and its separated components.
4) Source statics corrections (derived from surface seismic data)
were now applied to the VSP gathers to bring them to a common
datum.
5) First breaks were picked automatically, but then corrected manually on each VSP gather to harmonize any stray picks.
6) Maps for the first arrival times were prepared and analyzed/edited to estimate lateral velocity changes and/or determine
an anisotropy model. As part of the QC, these arrival times for
different rings and channels were plotted. Figure 8 shows the
arrival times for channel (depth level) 21 (550 m). It was found
that the arrival times formed concentric rings as expected and
did not exhibit any significant deviation that could have been
due to azimuthal anisotropy. Minor changes (traveltime deviations up to 4 ms) have been ignored.
7) Another interesting observation was that, when the arrival
times for increasing distance from the well were plotted, they did
not create a smoothly varying curve but rather a jittery curve about
a mean trend (red curve in Figure 9). The variation goes from 25 ms to +45 ms. After refraction statics derived from seismic
were applied and the arrival times again plotted with increasing distance from the well, the curve was much smoother. The
time variation was 2-4 ms. This could be expressed as the application of refraction statics harmonized the jitter.
8) Zero-offset VSP processing. The sources near the recording well were used to generate a zero-offset VSP that was then
used for initial calibration of well logs and the surface data.
The raw zero-offset VSP data traces were edited and sorted
by receiver depth. This was followed by wavefield separation,
waveshaping deconvolution, and corridor stack. The separation of the total wavefield into component wavefields (downgoing and upgoing P waves) was done by an optimization
inversion method (see Appendix) which contains several steps.
In the first step, the different wavetrains are identified and
marked. After subtracting these waves, the residual wavefield
may contain irregular noise and there may also be some regular wavetrains, which could be extracted from the total wavefield. The residual wavefield structure shows the quality of
separation and adequacy of the models relationship to the
real data. Figure 7 shows the different component wavefields.
It is comforting to see such noise being removed as the data
are being processed. The corridor stack trace is a zero-offset
trace; that is, it represents the wavefield (primary reflections)
recorded along the well bore (corridor chosen shown in Figure
10). The corridor stack trace can therefore be compared directly
to drift-corrected logs (in terms of time).
Figure 10 shows correlations between the sonic and gamma
ray logs, the upgoing wavefield, and the VSP corridor stack.
The display polarity convention is such that an increase in
impedance is represented as a peak.
A synthetic VSP was then generated to identify the multiple activity in the broad zone of interest and also understand
the paths followed by the source pulse to generate the multiples. The left side of the sonic log was used to derive the
acoustic boundaries and then, simulating zero-offset VSP
geometry, a synthetic VSP was generated using ray tracing.
MAY 2004

THE LEADING EDGE

0000

Downloaded 08/01/16 to 68.146.236.200. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Figure 6. Tube waves on records from different rings.

As seen in Figure 11, the events corresponding to the main


formation tops on the sonic log tie the real VSP corridor stack
very well. This suggests that the corridor stack picked for stacking was a good choice and that no multiple activity is included.
This corridor stack, as a result, can now be confidently correlated with the 3D surface seismic.
Figure 12 shows the correlation between logs, the upgoing VSP wavefield, and the surface data. Key formation tops
are tracked across logs, the real VSP data (with the same depth
index), the corridor stack (time), and the surface seismic.
Clearly, the frequency content of the VSP is much higher than
that of the surface seismic section. A band-pass filtered corridor stack was also generated to compare frequencies with the
surface data. However, even for a 5-45 Hz band pass, the surface data lack the reflection detail on the corridor stack around
the zone of interest (marked in Figure 12). Due to the known
phase and polarity of the VSP trace, the correlation is still sufficient to lend confidence to the surface seismic interpretation
but, at the same time, also suggests that the frequency content of the seismic data is insufficient for optimal interpretation of the zone of interest.
9) Offset VSP processing. A procedure called the VSP-CDP
transform was used to map the traces of the upgoing wavefield from the domain of receiver-depth versus VSP time (i.e.,
original traveltime versus depth) to the domain of lateral-offset (with respect to the well head) versus two-way vertical
reflection point traveltime. This mapping of reflected events
from the VSP domain into the CDP domain requires a velocity model for the subsurface.
Traveltime inversion was used to create a layered velocity model for each ring. First, an isotropic layered velocity
model was created from zero-offset VSP first arrivals. This
model consisted of 18 layers with thicknesses ranging from
50 to 150 m. Figure 13 shows the actual wavefield and model
time arrivals (for zero offset) which fit the data well. For different offsets, the first arrivals times were calculated and overlain on the real data. As seen in Figure 14, the isotropic velocity
model derived from zero-offset VSP first arrivals does not fit
offset data. This can be explained in the following way.
In a far-offset VSP wavefield, the velocity in the shallow
layers is usually relatively slow. This implies that the traveltime in these layers is longer. As seen in Figure 15, a ray path
from source S to a receiver R1 (in the shallow level) takes
longer (receivers depths 207 to 1285.5 m corresponding to
receivers 64 to 70 in Figure 14). When the receiver is below
the shallow layer (e.g., R2 in Figure 15), the ray in these layers becomes close to vertical and has a bigger angle of incidence in the higher velocity layers underneath. This implies
that, although the receiver is deeper, the time from the source
to the receiver (t2 + t3) is shorter. In an anisotropic medium,
0000

THE LEADING EDGE

MAY 2004

Figure 7. Total and component wavefields in Ring 4.

velocity increases with the angle so, for the shallow layers,
the time would be less than for the isotropic case.
To get a better match, an anisotropic layered velocity model
was constructed. For each ring, a vertical axis of anisotropy
is created with elliptical anisotropy associated with each layer.
The elliptic anisotropy coefficients VH/VV (VH = horizontal
velocity, and VV = vertical velocity) are determined by minimizing the difference between real first-break times and those
of the model. The minimum of a misfit function was found
by using a Monte Carlo approach (global optimization) followed by local optimization. After this exercise, the average
difference between the real and model arrivals was less than
1.5 ms. Figure 16 shows the actual wavefield and first breaks
for the anisotropic model. The anisotropic coefficients were
typically between 0.85 and 1.15 but, in the shallow part, they
increased to 1.25. Figure 17 shows the variation of VH/VV
with depth.
10) The velocity model so generated was used to calculate the
upgoing and multiple downgoing traveltimes. These were input
into each VSP gather to help wavefield separation for each
ring. Model traveltime paths are shown in Figure 18.
11) The total wavefield was separated into component wavefieldsdowngoing, upgoing, downgoing tubewaves, and upgoing
tubewaves. Since the records from different rings can be treated
as offset VSPs, the wavefields of interest have oblique incidence angles at the receivers. Since we have a three-component recording, it is possible to separate the mode-converted
shear waves. The axes chosen are HS, a horizontal axis from
MAY 2004

THE LEADING EDGE

425

Downloaded 08/01/16 to 68.146.236.200. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Figure 9. Blue curve = arrival times at a single channel from all


sources. Green curve = the same arrival times after application of
refraction statics. Notice how the variation about the mean trend (red
curves) is harmonized with the application of refraction statics.

Figure 8. Planar view of first arrival times (in ms) for channel 21.

Figure 10. Correlation of stratigraphy, well logs (in depth), VSP upgoing wavefield, VSP corridor stack, and surface seismic data.

426

THE LEADING EDGE

MAY 2004

MAY 2004

THE LEADING EDGE

0000

Downloaded 08/01/16 to 68.146.236.200. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Figure 11. Correlation of stratigraphy, well logs (in depth), synthetic VSP, real VSP upgoing wavefield, VSP corridor stacks, and surface seismic data.

well to source (radial component); HT, a horizontal axis at right


angles to HS (transverse component); and Z, vertically downward (vertical component). The effect of the tool rotation as
it moves up the well on the horizontal geophones is removed
by data rotation. Figure 18 shows the combined VSP wavefield. To make full use of the recorded information, this z-component VSP wavefield on the raw data channels was separated
into P and S upgoing and downgoing wavefields using the
optimization approach (Goltzman and Troyan, 1967; Blias and
Katkov, 1990) described in Appendix.
12) VSP deconvolution was run for both downgoing and upgoing wavefields. This step restored frequencies of the upgoing
wavefield using the downgoing wavefield.
13) A close look at the downgoing wavefields in Figure 6 reveals
the interesting observation that the wavelets embedded in these data
do not appear uniform. This set us thinking about the causes of
the nonuniformity in the wavelets. The shot point depths
were plotted (Figure 19) and found to be nonuniform; there
is a significant oblique N-S trend in the depth pattern. To try
to equalize the embedded source wavelets, matched filtering
was done between a model wavelet and the source wavelets
embedded in the data. The model wavelet was the average
source wavelet in the consistent source-depth zones in Figure
19. This enabled us to have a globally consistent wavelet in
the data.
14) VSP-CDP time-to-depth transformation for every ring was

done to obtain vertical VSP profiles. The velocity model determined in step 9 was used. All depth images were stacked after
making automatic shift determinations. This prevented any
loss of higher frequencies and a crisp image was obtained for
different offsets.
15) Ring images were stacked to obtain a 3D VSP volume in
depth.
16) The 3D VSP depth volume was transformed to a time volume. Figure 20 shows a cutaway section from the 3D VSP volume. Figure 21 shows some vertical planes cutting through
the 3D VSP time volume. Finally, this volume is merged with
the seismic volume.
Reconciling VSP to surface data. Generally, the seismic reflection interpretation forms the basis for integrating the two different data sets, which becomes especially valuable when
subtle variations are sought. The coordinate system origin
and orientation of both the VSP and surface data were made
identical. The bin size for the surface data was 35 35 m, and
the bin size for the VSP data was 10 10 m. This inconsistency is apparent in Figure 22 where a profile from 3D VSP is
inserted into surface inline 85.
The seismic volume was interpolated to have the same bin
size as the VSP and the two volumes were then integrated.
Figure 23 shows a profile through the integrated volume superior to that generated from either data set. The latter has betMAY 2004

THE LEADING EDGE

427

Downloaded 08/01/16 to 68.146.236.200. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Figure 12. Correlation of stratigraphy, VSP upgoing wavefield, well logs (in depth and time), VSP corridor stack, and surface seismic data. The
magnified view (compared with Figure 10) indicates the seismic signature in the zone of interest.

final results lead us to recommend that such simultaneous


surveys, when they can be conducted at an incremental cost
(and thus remain commercially feasible), could provide more
detailed images around the borehole.

Figure 13. Real VSP data with the modeled time arrivals for zero offset
overlain.

ter reflection continuity and focusing and enhanced resolution at the level of interest.
Conclusions. The processing of 3D VSP was an enriching
experience, though quite labor intensive and time consuming. The vertical profiles through the 3D VSP volume exhibited higher resolution with detailed and focused reflections
especially at the target level. The deployment of a large number of borehole seismic receivers was expected to yield larger
reflection coverage around the borehole and seems to have
paid off in this respect. The results of this project have demonstrated the benefit of the integrated approach pursued which
seeks to optimize the image of exploration objectives.
The experience gained through processing, integrating
the two data sets recorded simultaneously, and evaluating the
428

THE LEADING EDGE

MAY 2004

Suggested reading. Optimization algorithms for selecting three


components VSP wavefield by Blias and Katkov (Automated
System of Geological-Geophysical Data Acquisition and Processing,
Riga, 1990, in Russian). Difficulties and clues in 3D VSP processing by Boelle et al. (SEG 1998 Expanded Abstracts). An integrated 3D tomographic inversion-application to multisurvey
VSP data by Cao et al. (SEG 2002 Expanded Abstracts). 2D versus 3D VSP imagingA North Sea example by Clochard et al.
(EAGE 1999 Extended Abstracts). Coherence Cube and beyond
by Chopra (First Break, 2002). Simultaneous acquisition of 3D
surface seismic data and 3C, 3D VSP data by Constance et al.
(SEG 1999 Expanded Abstracts). Processing of 3D VSP data from
a permanent borehole array by Cornish et al. (SEG 2000 Expanded
Abstracts). Shallow 3D seismic and a 3D borehole profile at
Ekofisk Field by Dangerfield (AAPG Memoir 42, 1991). First constant angle 3D circular profile on Ekofisk by Dangerfield et al.
(SEG 1999 Expanded Abstracts). Optimum algorithms of the seismic interference wavefield separation by Goltsman and Troyan
(Acad. News of USSR, Physics of Solid Earth, N8, 1967). 3C-3D VSP
analysis by Gulati et al. (GEOPHYSICS, 2002). Coherency calculations in the presence of structural dip by Marfurt et al.
(GEOPHYSICS, 1999). Analyzing 3C-3D VSP data: The Blackfoot,
Alberta survey by Stewart et al. (SEG 1998 Expanded Abstracts).
3D walkaway VSP, enhancing seismic resolution for development optimization of the Brent Field by VanDerPal et al. (First
Break, 1996).

Downloaded 08/01/16 to 68.146.236.200. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Figure 14. First arrival times for Ring 21 (1050 m offset) computed
using an isotropic velocity model and overlayed (blue curve) on real
data. The misfit indicates that the isotropic velocity model derived from
zero-offset VSP is not satisfactory.

Figure 16. First arrival times for Ring 21 computed using an anisotropic
velocity model and overlayed (blue curve) on real data. The good fit
between the two indicates the adequacy of the derived velocity model.

Figure 15. Travel paths and their traveltimes through shallow and
deeper layers.

Figure 17. Variation of anisotropic coefficient with depth.

Figure 19. Planar view of shot point depths (in meters).


Figure 18. Typical z-component of wavefield on VSP records.
MAY 2004

THE LEADING EDGE

429

Downloaded 08/01/16 to 68.146.236.200. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Figure 22. Overlay of a 3D VSP vertical plane and seismic inline 85.

Figure 20. 3D VSP seismic volume.

Figure 23. Seismic inline 85 through the integrated volume (seismic


volume integrated with the 3D VSP volume). Light blue represents the
area over which the VSP data were available for integration.

Figure 21. Vertical planes in the 3D VSP volume.

Appendix. This approach is based on the nonlinear minimization of the residual wavefield. Consider the mathematical model that describes the actual VSP three-component data:
Ri(tj) = adi fd(tj - di) + aui fu(tj + ui) + bdi gd(tj - di) + bui gu(tj
+ ui) + ij, i=1,2,...,n.
Here Ri(t) = (Rxi(t), Ryi(t), Rzi(t)), is a three-component trace
in receiver number i, t = tj, j=1,2,...,M is discrete time; adi, aui
are vector amplitudes of downgoing and upgoing P-waves
respectively, bdi, bui are vector amplitudes of downgoing and
upgoing S-waves; di, ui are time arrivals of downgoing and
upgoing P-waves; di, ui are time arrivals of downgoing and
upgoing P-waves; fdi(t), fui(t), gdi(t). gui(t) are wave shapes; and
ij is additive noise. A least-squares procedure is used to solve
for wave parameters. The minimum value of the following
function has to be found:
M

F=
j=1 i=1

430

||Ri(tj) - adi fd(tj -

THE LEADING EDGE

di

) - aui fu(tj +

MAY 2004

ui

) - bdi gd(tj -

di)

- bui gu(tj +

We are trying to make the residual wavefield (after subtracting all regular determined waves) as small as possible.
Here we have to find unknown amplitude vectors adi, aui, bdi,
bui, times di, ui, di, ui, and wave shapes functions fd(t), fu(t),
gd(t), gu(t). An iterative method has been developed to find
the solution of this problem. It is based on improving different wave parameters and performing wavefield separation in
the time domain. The structure of function F makes it necessary to perform nonlinear minimization only for time arrivals
di, ui, di, ui. It can be shown that to find amplitudes adi, aui,
bdi, bui we have to find the eigenvalues of the covariance
matrix. Waveshapes fd(t), fu(t), gd(t), gu(t) are improved one
after another. TLE
Acknowledgments: Mega Bin is a trademark of EnCana Corporation and
Coherence Cube is a trademark of Core Laboratories. We are indebted to Bob
Hardage and Rob Stewart for editing the manuscript thoroughly, which
improved the quality of the content. We thank ConocoPhillips for release of
data and thank both ConocoPhillips and Core Laboratories, Canada for permission to publish this paper. Authors from Core Laboratories thank
Vasudhaven Sudhakar for constant encouragement and helpful discussions.
Corresponding author: schopra@corelab.ca

ui)||

Das könnte Ihnen auch gefallen