GEOPHYSICS, VOL. 68, NO. 1 (JANUARY-FEBRUARY 2003); P. 286–296, 11 FIGS.

10.1190/1.1543214

Crooked-line 2D seismic reflection imaging

in crystalline terrains: Part 2, migration

Mladen R. Nedimović∗ and Gordon F. West‡

ABSTRACT flections; i.e., we attempt to construct a 3D image volume

of all reflectors viewed by the survey. Because reflectors
Seismic reflection surveys are frequently conducted that face away from the acquisition line cannot return
over very complicated geological structure, but survey- much wave energy from available sources to available re-
ing often must be confined to existing crooked roads or ceivers, they cannot be imaged even if they lie directly be-
tracks. Typically, data from such 2D crooked-line surveys neath the survey profile. Tests show that the cross-profile
are processed using standard common midpoint (CMP) spread of trace midpoints usually is sufficient to pro-
stacking techniques to obtain a 2D time section which is vide a useful degree of cross-line positioning of reflection
then 2D migrated. In Part I, we show that a reflector dip points. A very helpful image volume is thus obtained.
component across the processing line can cause serious Kirchhoff 3D prestack migration is computationally
problems for standard CMP stacking. We also propose laborious. A much quicker but less complete method is
a supplementary processing step in which cross-dip is to create the 3D migrated image volume from the 2D
determined locally and cross-dip moveout (CDMO) is optimum cross-dip stack and the associated set of cross-
removed from data to form an optimum cross-dip stack. dips. Robustness of migration methods to time errors in
However, a crooked-line survey is really a swath 3D sur- the prestack data traces such as poorly corrected statics
vey, and ideally we would like to obtain a 3D image of is also an issue. Tests show that in difficult cases, particu-
reflectivity surrounding the profile. larly where only 2D processing is warranted, migration
Here we investigate the potential of 3D prestack of trace absolute amplitude rather than standard phase
Kirchhoff migration to directly image all observed re- data may lead to a superior result.

INTRODUCTION subsurface reflectivity near the acquisition line. Optimistically,

because structures in crystalline terrain can be very complex,
In the companion paper (Nedimović and West, 2003; here- one would like to obtain some sort of 3D image of structure by
after, Part I), we discuss the problems of imaging reflectors in 3D migrating the seismic data. Of course, this could be imprac-
crystalline igneous metamorphic terrain from 2D seismic data tical because of the inherent limitations of the observational
acquired on crooked lines. We show that the cross-profile dip of data set. Swath 3D data have a very small cross-profile aper-
reflectors can severely damage CMP stacking of the data. We ture, an uneven distribution of source–receiver midpoints, and
also describe a process for determining reflector cross-dip by a a small range of source–receiver azimuths, so migration might
semblance-based method and then correcting for the crossdip be fruitless.
moveout (CDMO) in the stack. In this way we obtain what we Standard prestack 3D migration is a very cumbersome and
called a locally optimum cross-dip stacked section on which to expensive computational process. Therefore, in addition to
base interpretation. testing standard 3D prestack migration, we have also stud-
However, a high-fold 2D seismic reflection survey on a ied the feasibility of migrating the locally optimum cross-
crooked line is really a limited swath 3D investigation of the dip stack of Part I into a 3D volume. Furthermore, we have

Presented at the 70th Annual Meeting, Society of Exploration Geophysicists. Manuscript received by the Editor February 27, 2001; revised
manuscript received April 18, 2002.
∗
Formerly Geological Survey of Canada—Pacific, Natural Resources Canada; presently Columbia University, Lamont-Doherty Earth Observatory,
61 Route 9W, P.O. Box 1000, Palisades, New York 10964-8000.
‡University of Toronto, Geophysics Labs, Department of Physics, 60 St. George Street, Toronto, Ontario, M5S 1A7, Canada. E-mail: west@
physics.utoronto.ca.
°c 2003 Society of Exploration Geophysicists. All rights reserved.

286
 Migrating 2D Crooked-Line Seismic Data 287

tested migration of amplitude stacks, sections formed by av- practical. Certainly, there are fundamental limitations to what
eraging trace absolute amplitude in CMP gathers. A summary can be achieved from a 2D crooked-line data set, and there are
flowchart of the different approaches to processing crooked reasons to worry about the quality of any 3D image obtained
line data is shown in Figure 1. Our experiments indicate that from such data. We discuss these limitations and concerns be-
all three migration processes can be useful. low. Basically, they reduce to
Nevertheless, our tests on both synthetic and field data sug-
1) lack of observations from viewpoints that can generate
gest that the most effective approach to imaging 2D crooked-
Snell’s law reflections from whatever dipping interfaces
line survey data is first to produce an optimum cross-dip stack.
are in the ground;
Then, if cross-dip information can be extracted for many re-
2) lack of sufficient cross-line horizontal aperture in the data
flectors, and especially if overlapping events are abundant, in-
set to resolve the cross-line position of reflectors; and
vesting the CPU power to prestack-migrate the data into a 3D
3) irregular spatial distribution of the data.
volume will very likely be worthwhile.
The first and second points are fundamental to any swath
3D survey. The third condition results when shot and re-
The data ceiver positions are confined to a single (albeit crooked) survey
Our field survey example is the high-resolution Sturgeon line.
Lake line. It is a short but densely recorded profile acquired The Kirchhoff migration method seems, at least from an op-
for mineral exploration research. A fixed spread of 393 receiver erational view, to be usable with irregularly distributed data.
traces spaced at 20-m intervals recorded explosive shots spaced This summation process relies on interference to remove un-
every 40 m. Average CMP bin fold reached well over 100 traces wanted parts of the signal from the final image product. Ac-
per gather. The acquisition profile is locally very crooked, but cording to the traveltime of each datum value in a data trace,
it oscillates about a straight processing line. Figure 1a in Part I reflectivity amplitude is inserted into the image space at every
shows both the acquisition and the processing lines on a map space point that could possibly have generated that particu-
of the surface geology. lar datum. Then, when the putative reflectivity contributions
All synthetic survey data used in our tests were computed from all data have been included in the image, interference will
using a ray-Born method. The survey geometry of either the enhance the total reflectivity signal at locations where the re-
Sturgeon Lake line or Abitibi line 23 was assumed (see Part I). flections actually occurred and attenuate it elsewhere. For this
The reflectors modeled range from a few point diffractors to to happen, the survey data traces must be distributed evenly
groups of several rectangular planes. In all cases, the host enough throughout a large enough spatial range (aperture).
medium was assumed to have constant velocity. For more The required aperture is a function of wavelength and distance
details about the modeling and the field surveys, see the to the reflector. If the acquisition geometry does not provide
introduction and appendices A and B of Part I. the minimum required aperture, extended smears of reflectiv-
ity values will be left in the migrated image that indicate only
where the reflection points might be located, not where they
Migration of swath 3D data actually are.
If a straight-line 2D survey is conducted in geological terrain
The idea that a 3D structural image might be obtained from for which no assumptions can be made about cross-line struc-
a 2D crooked-line seismic data set is not novel. For instance, ture, the position of any observed reflectors will be completely
Geiger et al. (1995) and Bancroft et al. (1998) suggest the pos- ambiguous in a cylinder surrounding the acquisition/processing
sibility. However, we are unaware of published examples, and line. Also, no assurance can be given that all reflectors that cut
the concept appears to have remained more hypothetical than the vertical section containing the survey profile will be imaged
by the survey, because only those that directly face some part
of the survey line will generate appreciable signal.
In applying 3D migration to 2D crooked-line data, we are
hoping the survey geometry will provide enough cross-line
aperture so reflection points can be reasonably well positioned
in the cross-line direction. However, the degree of localization
achieved in the cross-line direction will likely be much less than
along the survey profile. Furthermore, it should be expected
to vary along the profile with the specific crookedness of the
acquisition line.
The 3D migration of the original prestack data volume with
a 3D laterally varying velocity model is a very computation-
ally expensive way to obtain a 3D image volume from the
data. The usual reasons for making such an effort are signif-
icant lateral velocity variations and very steep reflector dips.
However, velocity is not an important factor in our case. In
crystalline terrains, the average velocity over a path of several
FIG. 1. Processing flowchart summarizing both the standard hundred meters in the depth range from 0.5 to 10 km usu-
and the suggested alternative approaches to processing 2D ally varies less than a few percent, so constant velocity or rms
crooked-line data. velocity migration is often sufficient. With this simplification,
288 Nedimović and West

3D migration becomes feasible using only workstation-level evenly distributed grid of image data traces aligned with the
computers. 2D processing (slalom) line.
As a more economical alternative to 3D prestack migra- To minimize the computational load of the prestack migra-
tion, we have explored poststack migration of the 2D optimum tion, and also to smooth out some of the great bin density
cross-dip stack into a 3D volume. This is accomplished by us- variations of trace midpoints inherent to crooked line geom-
ing the cross-slowness information provided in the composite etry, we first perform an NMO-corrected partial stack on the
cross-dip image together with the actual cross-line spread ex- cleaned-up prestack observational data. In any CMP bin gather
tent. In effect, a 3D presentation of the 2D data section is pro- along the processing line, there is a strong correlation of mid-
duced together with some indication of the inherent positional point cross-line offset with source–receiver offset; so partial
ambiguity. stacking into narrow ranges of source–receiver offset only av-
Although useful results can be obtained with the post- erages traces with quite similar cross-line offsets. Thus, all re-
stack procedure, it has fundamental limitations. For instance, flection signals should remain well preserved for migration.
where the reflection responses of the imaged events over- When designing the output migration image grid, it is very
lap in the section, only the stronger will remain in the helpful to have already obtained the optimum cross-dip stack
optimum cross-dip stack and the 3D image; also, to inde- for the survey. It helps decide how extensive the output grid
pendently determine NMO, DMO, and CDMO, an azimuthal needs to be so run time can be minimized.
approximation must be made. Signal loss from these limita- Unlike the schematic shown in Figure 2 and our Sturgeon
tions/approximations cannot be recovered by any poststack Lake field data, most crooked-line surveys are sufficiently long
3D migrator.
In Part I, we showed that converting partially stacked CMP
trace data into absolute amplitudes and then stacking to ob-
tain an amplitude section is effective in some cases where time
errors in the data are too abundant or too unpredictable to be
corrected before stack. In this paper, we show that poststack
migration can be successfully applied to amplitude stacks. Sim-
ilarly, absolute amplitude trace data of partially stacked CMP
gathers can be migrated by a prestack process. As with ampli-
tude stacking, amplitude migration can sometimes be useful as
a process of last resort.

METHODOLOGY
FIG. 3. Geometry for 3D poststack migration. The origin of
3D prestack migration.—We have carried out our 3D space coordinates (0, 0) is the location on the processing line
prestack migrations by applying a standard 3D rms veloc- of the selected trace in the 2D stacked section. The image point
(xi , yi , z i ) is at distance R from it. Points Y1 (0, y1 ) and Y2 (0, y2 )
ity Kirchhoff migration method to the observational data. mark the cross-line range of individual trace midpoints in the
As shown in Figure 2, the input 2D crooked-line survey has CMP bin gather of the stacked trace.
data traces distributed unevenly in a swath around the ac-
quisition line; the output 3D prestack migrated volume is an

FIG. 4. Images obtained by migrating the signal from a point

FIG. 2. Midpoint geometry of a crooked-line survey with the diffractor. (a) Migrated standard stack. (b) Migrated amplitude
output data array for 3D migration (schematic). stack. The results are similar.
 Migrating 2D Crooked-Line Seismic Data 289

to require a multisegment processing line. This can complicate 3D poststack migration.—A more economical but poten-
the 3D prestack migration process since separate 3D output tially less effective method of obtaining a 3D reflectivity
volumes and separate migration processes for each grid may volume from 2D crooked-line data is to apply a 3D post-
be needed. stack migration to the traces of the locally optimum cross-dip

FIG. 5. Imaging of four point scatterers by 3D prestack migration. (a) Plan and perspective view of model A and the image slices.
(b, c, d) In-line slices (1), (2), and (3) in variable density format. The color palette is given in Figure 9. (e, f, g, h) Close-ups of
the upper event under the processing line. Parts (e) and (f) are the result of standard and amplitude migration, respectively; parts
(g) and (h) are the same except that time shifts simulating static errors have been included.
290 Nedimović and West

stack. Contributions to the migrated image from each of the series:

2D stacked traces are computed as if a stacked trace were the ¡ ¢1
average of a group of individual traces uniformly spaced over R + δ R = xi2 + (yi − yc )2 + z i2 2
the cross-line width of the CMP bin and as if the events on µ ¶1
those traces shift in time with cross-line offset according to 2yi yc yc2 2
= R 1− + (1)
the cross-slowness found for each event by the optimum cross- R2 R2
dip analysis. Modifying the standard Kirchhoff poststack mi- µ ¶
2yi yc yc2
gration to do this is straightforward and is described below. ' R 1− + .
The output 3D grid for poststack migration is chosen in the R2 2R 2
same way as for 3D prestack migration. It may be necessary Thus, the total time shift equals
to apply a bandpass and coherency filter to the input opti-
2yi yc y2
mum cross-dip stack to prevent spatial aliasing on the output δt = 2 pt yc − + c . (2)
grid. vrms R vrms R
Figure 3 shows the geometry for one image point in the By substituting y1 and y2 instead of yc , the extreme time shifts
3D image space and one trace from the optimum cross-dip are found. Data in the time window (t + δt)max to (t + δt)min
stacked data. The datum value to be migrated to the image are averaged and added to the image point (xi , yi , z i ). The av-
point at (xi , yi , z i ) is found on the data trace at (0, 0) at time eraging acts as a focusing filter for migration. The process is
t = 2R/vrms , where vrms is the root mean square velocity and R repeated for each output image point and for all data samples
is the distance from the stacked trace to the image point. The of every trace of the optimum cross-dip stack.
event on the (imaginary) traces distributed over the cross-line
Y1 (0, y1 )–Y2 (0, y2 ) and contributing to the stacked trace would Amplitude migration.—Migration of trace absolute ampli-
have been found at time t shifted by δt, the product of the cross- tude has rarely been contemplated, perhaps because absolute
line offset yc and the optimum cross-slowness pt estimated for trace amplitudes do not behave linearly in sums of signals.
each t in the trace at (0, 0) by the optimum cross-dip anal- Thus, combining absolute amplitudes does not have a basis
ysis. The contributions to the image point by all (imaginary) in wave theory. However, geometrical migration of the density
traces in the gather are summed. The change of R to R + δ R or opaqueness of event lines or curves has a perfectly good
for each trace can be approximated by using the binomial foundation in Radon transform theory.

FIG. 6. Imaging of dipping reflectors by 3D prestack migration. (a) Six reflecting plates dipping along the processing line. (b, c) Slice
of the processing line from the 3D prestack migrated image volume at medium and high plot sensitivity. (d) Six cross-line dipping
reflectors that face toward the acquisition/processing line and lie approximately northwest of it. (e, f) Some in-line and cross-line
slices of the image volume.
 Migrating 2D Crooked-Line Seismic Data 291

This suggests that the 2D amplitude stacks discussed in part I, or receiver skips. The data were stacked in the normal way.
as well as the absolute amplitudes of partially stacked CMP Absolute amplitudes were raised to a power of 1.5 before am-
trace data, might be successfully migrated in the same fash- plitude stacking. After stacking, both the standard and the am-
ion as standard data. However, amplitude envelope traces will plitude section were 2D migrated using a standard phase shift
need to be frequency filtered before migration. The filtering algorithm. To achieve the good focusing shown in the figure,
must remove the zero frequency component or dc shift of the bandpass filtering was applied twice to the amplitude data: first
amplitude trace while retaining low frequencies that corre- to the input amplitude stack (0-20-120-150 Hz) and then to
spond to time variations of the trace envelope. Traces may the migrated section (5-15-40-50 cycle/km). The same postmi-
also be smoothed to eliminate rectification noise. This can be gration bandpass filter was applied to the migrated standard
achieved with a trapezoid bandpass filter in which the low cut stack.
is set at 0 Hz and the low-frequency corner is set lower than for
standard phase data. High cut and high pass frequencies can MIGRATION OF SYNTHETIC DATA
be determined experimentally for each data set.
Before migration, the trace amplitudes should be normalized Point diffractors.—In our first test of 3D prestack migration,
by automatic gain control (AGC). If spuriously large amplitude the modeled data consist of the seismic response of four point
events in the stacked data remain after AGC, raising the val- diffractors. The crooked-line geometry of the Sturgeon Lake
ues to some arbitrary positive power less than unity tends to line was assumed. Two of the diffracting points are approxi-
improve the migrated amplitude stack. mately under the processing line, and two are offset ∼1500 m
Figure 4 provides a simple demonstration of migration ap- to one side. The shallower diffractors are buried 1000 m; the
plied to amplitude traces. It shows 2D poststack migrated stan- deeper ones, 3000 m. The model and results are shown in
dard and amplitude stacks of a single point diffractor. Both Figure 5. The main goal was to learn if standard and/or am-
results exhibit approximately the same focusing. plitude 3D Kirchhoff migration could clearly differentiate be-
The diffractor was positioned directly below the middle of tween a diffractor that lies below the acquisition/processing
the line at a depth of 2400 m and a 10- to 150-Hz Klauder line and one that is laterally offset from it. Only three planes
wavelet was applied. Unlike all the other synthetic data shown of the migrated data volume were computed.
in this paper, this set was produced by using a straight-line Figures 5b–d show three vertical slices of the migrated im-
survey geometry and a regular surface sampling with no shot age parallel to the processing line and 750 m apart. The first

FIG. 7. Imaging reflectors with both in-line and cross-line dip. (a) Four plate-like reflectors that dip south at 15◦ , 30◦ , 45◦ , and 60◦
across the southwest directed processing line. (b, c, d) Three perspective views of slices of the 3D prestack migrated image volume.
The vertical slice has been rotated to show the different parts of the reflector images. All reflectors are easily distinguished and well
delineated.
292 Nedimović and West

and third image slices contain the actual diffraction points. appreciable artifacts and smearing. Although better imaging is
Slice 2 is positioned midway between to check on the expected usually to be expected for the reflectors with mild or no cross-
cross-line smearing of the images. To make artifacts visible, line dip, all events are imaged comparably. This is because of
plot amplitude was significantly increased, so dark parts are the positioning of the reflectors in the model relative to the
oversaturated. Slices 1 (Figure 5b) and 2 (Figure 5c) show im- acquisition line. The steeply dipping reflectors face the middle
ages in accord with what 2D in-line migration theory predicts. of the survey line, and the horizontal and mildly dipping ones
The events in slice 1 are well focused at their true subsurface are found at the edges of the survey.
positions. As expected, there are artifacts in slice 2 between
the diffractors. However, they are much weaker than the main In-line and cross-line dip.—The next model (Figure 7a) was
events. The focusing in slice 3 (Figure 5d) at the true position of designed to make the imaging process challenging. Four reflec-
the diffractors is surprisingly good, considering the irregularity tors with similar components of in-line and cross-line dip were
of the survey geometry and the small average swath width of
the survey (∼350 m).
To see if the result is robust to time errors and to com-
pare standard and amplitude migration, a second test was con-
ducted. In it, the static corrections estimated for the Sturgeon
Lake line field data were added to the synthetic data before
the partial stack, as if they were uncorrectable. Figures 5e–h,
show magnified images of the shallow diffraction point closest
to the processing line after standard and amplitude migration
on data with and without time errors. With no time error, the
image of the scatterer focuses well in both standard (Figure 5e)
and amplitude data (Figure 5f). With the simulated static er-
ror, focusing is seriously damaged regardless of the type of
migrated data [standard (Figure 5g) or amplitude (Figure 5h)].
Nevertheless, the amplitude migration focuses better, which
again suggests that a migration of amplitude data may be more
robust to error.

In-line dipping reflectors.—The next model contains six in-

line dipping platelike reflectors in the geometry of the Sturgeon
Lake line (Figure 6a). All reflectors are squares with 1500-m
sides and a top edge at 1000 m. They dip in the in-line direction
at 0◦ , 15◦ , 30◦ , 45◦ , 60◦ , and 75◦ .
Figures 6b and 6c show slices of the migrated data along
the processing line at medium and high sensitivity. In spite
of the shortness of the acquisition line (∼8 km), standard 3D
prestack migration accurately captures most of the in-line dip-
ping events below the slalom line. A somewhat longer sur-
vey would be necessary to completely image the two steepest
events. Even at high plot sensitivity, artifacts are not strongly
visible. However, we have not shown image slices on either
side of the reflectors. The 3D image of the reflectors may be
laterally smeared from insufficient cross-line aperture or be
attenuated before reaching the true edges of the reflectors be-
cause of problems in facing direction.

Cross-line dipping reflectors.—The model consists of six

cross-dipping reflectors and the geometry of the Sturgeon Lake
line (Figure 6d). The reflectors are the same size and depth as in
the in-line case and were placed to northwest of the survey line
so they would face the survey profile. The data were prestack
migrated to a 3D data volume of 8.8 × 8.0 × 3.0 km. Due to
CPU power limitations, the output volume was coarsely sam-
pled (100 m), so some aliasing/undersampling is evident in the FIG. 8. Comparison of prestack and poststack 3D migration.
result. Vertical and horizontal slices through the migrated 3D (a) The model consists of two pairs of reflectors in the Litho-
volume presented in Figures 6e and 6f are positioned to show probe line 23 geometry. (b, c) Cross-line slices through the
first pair of reflectors from the 3D prestack migrated and 3D
images of the 15◦ and 60◦ dipping reflectors. poststack migrated data, respectively. A mild Gaussian blur
Remarkably, migration was able to position the cross-line and toning were applied to these images to make them visible
dipping reflectors quite accurately in space, but not without at reduced scale.
 Migrating 2D Crooked-Line Seismic Data 293

placed near the survey profile so at least large parts of them face the image is incomplete wherever there is event interference.
some part of the survey line. The reflectors are square plates Moreover, the approximations made to compute and remove
with sides 2000 m long. Their apparent in-line and cross-line NMO, DMO, and CDMO are not sufficiently accurate when
dips are ∼10.5◦ , ∼20.7◦ , ∼30.0◦ , and ∼37.8◦ . Because the plates all of these moveouts are very large, so this is a test case where
are large and planar, only parts of their surfaces are viewable. 3D migration would seem to be needed.
The variety of in-line and cross-line dips in the model ensures The data were prestack depth migrated to a 3D (6.8 × 3.0 ×
that large-magnitude NMO, DMO, and CDMO are present 2.0 km) data volume. The results are shown Figures 7b–d. All
in 2D processing of the data. Because the top parts of the of the reflectors have been well indicated, despite the fact that
reflectors were placed very close to each other, some of the CPU limitations forced migration of the data to a coarse 3D
reflection events overlap in the top of the 2D stacked time grid (50-m sample rate) in which aliasing can occur. However,
section. The optimum cross-dip stack provides a much better smearing and attenuation because of facing problems certainly
image of reflectors than does the standard stack. However, are evident.

FIG. 9. The 3D prestack migration of Sturgeon Lake line field data. (a, b) Views at the migrated image volume from the
north–northeast and southwest–south, respectively. (a) A perspective of variable density slices showing the location of the strongest
events. (b) An isosurface presentation of the image data in 3D space with the isosurface threshold set in the midnegative signal
range. Opaque variable density slices are inserted to help the reader visualize the position of the main clusters of events. The
processing line is superimposed on both images. The appropriate color palette is given for each part of the figure.
294 Nedimović and West

Comparison of prestack and poststack 3D migration.—Our this reason, it may be used when the resources for a prestack
final test is a comparison of 3D prestack and 3D poststack migration are unavailable.
migration. The model for it contained two pairs of reflecting
plates and used the survey geometry of the northern third of THE FIELD EXAMPLE
the Lithoprobe Abitibi–Grenville transect, line 23 (Figure 8a).
Reflectors 1 and 2 of pair I are relatively far away from the ac- The field data of the Sturgeon Lake line, after being sorted
quisition line. They dip at 0◦ and 45◦ , respectively. Reflectors 3 into 200-m offset windows and partial stacking, consisted of
and 4 of pair II are closer to the acquisition line and dip at 45◦ . 342 CMP bin gathers with a maximum fold of 53. The original
For practical purposes, reflector 2 exhibits a pure cross-line dip, data were sampled at 1 ms and had useful frequency content of
while reflectors 3 and 4 dip in both directions equally. All reflec- up to ∼300 Hz and traces of 3 s length. Our fastest computer
tors are square planes with sides 3000 m long. Their reflective at the time we planned to 3D prestack migrate the data was
responses overlap in the 2D stacked section on the chosen pro- a 330-MHz Sun UltraSparc station 10. To hold the migration
cessing line. A 10–56-Hz Klauder wavelet was applied in the run time to something realizable, the size of the output data
modeling. set was set at 9.7 × 6.0 × 6.0 km, a volume somewhat smaller
We have applied both 3D prestack and 3D poststack am- than originally desired. The spatial sample rate in the image
plitude migration to the synthetic data to obtain two coarsely space was set at 18 m in all three directions. Migration rms
sampled (100 m) image volumes (15.0×20.0×16.0 km). Vertical velocity ranged from 6000 to 6500 m/s. To prevent aliasing in
slices of the image volume through reflector pair I are shown in the image volume, a high-cut frequency filter (10-30-80-100 Hz)
Figures 8b and 8c. As expected, a much better image was pro- was applied to the input data.
vided by the 3D prestack migration. The poststack migration Two isosurface/slice images of the main events in the 3D
provides a useful image at substantially lower CPU effort, but prestack migrated data are displayed in Figure 9. Most of the
there is a break in event 1, and event 2 is imaged only partially. reflection events migrate south from the northeast–southwest
The information loss about the overlapping events that occurs processing/acquisition line, and there they display an east–west
in producing the optimum cross-dip stack is permanent and strike and a northerly dip. The strongest groups of events
necessarily affects the poststack migrated image. The advan- are labeled A, B, C, and D. All exhibit both in-line and
tage of the poststack migration is its computational speed. For cross-line dip (Figures 9 and 10). The same groups of events

FIG. 10. Close-ups of the main groups of events in the 3D prestack migrated volume of Sturgeon Lake line data. (a, b, c, d)
Perspective views from the west of vertical variable density sections through event groups A, B, and C, parallel and perpendicular
to the processing line. The labeled reflectors clearly exhibit both in-line and cross-line dip. Views (a) and (b) are magnified more
than views (c) and (d).
 Migrating 2D Crooked-Line Seismic Data 295

were shown in the optimum cross-dip stack example of Part I standard stack are saturated with reflection events, both sec-
(Figure 8b). The results obtained by the two approaches agree tions are misleading because most of the imaged events do
with each other. Thus, the basic geologic interpretation given not lie in the plane of the profile. This is confirmed by the 3D
in Part I for the Sturgeon Lake line data and based on the prestack migration result, which shows that most of the reflec-
optimum cross-dip stack is largely unchanged by the 3D re- tion events do not cross the plane of the processing line (see
sult. Many details were better imaged by the 3D prestack mi- Figures 9 and 10) and much of what is imaged in this section
gration, but their interpretation is beyond the scope of this is migration noise (Figure 11c). Similar results are obtained
paper. by 3D poststack migration of the optimum cross-dip stack. In
To further emphasize the importance of extracting 3D struc- view of the above, the main goal when processing swath 3D
tural information from 2D crooked-line survey data by using data is to extract the true reflector positions via the optimum
the methods presented here and in Part I, we compare various cross-dip stack and/or 3D prestack migration. The conven-
results obtained by migrating the Sturgeon Lake line data. Part tional and amplitude stacks and their migrated counterparts
of the migrated amplitude stack is shown in Figure 11b. The cor- should be used only when extracting 3D structural information
responding migrated standard stack is shown in Figure 11a. The fails.
migrated amplitude stack gives a better rendition of the relative
strength of the reflectors and has better event continuity. The
nominal resolution in the standard data is far superior, but most SUMMARY AND CONCLUSIONS
events have a multicyclic nature that makes any real structural
In Part I, we explored methods for dealing with cross-dip
resolution comparable to or worse than it is on the amplitude
problems in 2D processing of crooked-line data. Here we have
section. Although the migrated amplitude stack and migrated
explored the ability of 3D Kirchhoff prestack migration to solve
the problem of correctly positioning all of the observed reflec-
tors in 3D space. Our experiments on synthetic and field data
indicate that, despite the limited cross-profile aperture of a
typical crooked-line survey and the irregular distribution of
midpoints in space, the method works quite well. Its principal
disadvantage is the very large computational cost relative to
any kind of 2D processing.
We also compared 3D migration of the full prestack data set
with a simpler 3D poststack migration of the 2D locally opti-
mum cross-dip stack of the data. Tests on synthetic data showed
a high correlation between results from the two methods, both
delineating the 3D geometry of the test reflectors quite well.
But dropouts arise when the reflection responses from var-
iously oriented reflectors overlap on the optimum cross-dip
stack. Thus, one can expect to obtain better images by using
the full prestack migration, but at much higher CPU cost.
Furthermore, we have shown that 2D amplitude sections are
imaged by using standard 2D migration techniques. Amplitude
stacking and migration seem to be more tolerant of time errors
in the data than the usual methods. However, we only suggest
amplitude processing of 2D crooked-line data as a last resort
when extraction of 3D information by other techniques has
failed.
Finally, there is no doubt that additional source points lo-
cated off the acquisition line on any available side roads could
improve the 3D imaging. Since adding a few source points
is fairly inexpensive, such a strategy should be considered by
future surveyors.

ACKNOWLEDGMENTS

We thank Larry Petrie and Larry Matthews of Noranda Inc.

for providing the Sturgeon Lake line data for our experiments.
Costs of the research were covered by a grant to G.F.W. from
the Natural Sciences and Engineering Research Council of
FIG. 11. Comparison of various migration results obtained Canada. Dr. Roy Hyndman generously permitted M.R.N. to
from Sturgeon Lake line data. (a) Standard stacking and mi-
gration. (b) Amplitude stacking and migration. (c) The corre- prepare the manuscript during a research fellowship at Geo-
sponding part of the slice across the processing line extracted logical Survey of Canada–Pacific. Reviews by Carlos Calderón
from the 3D prestack migrated data volume. and Reinaldo Michelena helped improve the manuscript.
296 Nedimović and West

REFERENCES tion of crustal data using common scatterpoint (CSP) migration tech-
nique: Nat. Conv., Can. Soc. Expl. Geophys., Expanded Abstracts,
Bancroft, J. C., Geiger, H. D., and Margrave, G. F., 1998, The equivalent 49–50.
offset method of prestack time migration: Geophysics, 63, 2042– Nedimovic, M. R., and West, G. F., 2003, Crooked-line 2D seismic
2053. reflection imaging in crystalline terrains: Part 1, data processing:
Geiger, H. D., Bancroft, J. C., and Foltinek, D., 1995, Prestack migra- Geophysics, 68, 274–285, this issue.