Advanced processing of long-offset seismic data for sub-basalt ...

Advanced processing of long-offset seismic data for sub-basalt imaging in the Faeroe-Shetland
Basin
Hassan Masoomzadeh* and Penny J. Barton, University of Cambridge, Satish C. Singh, IPG Paris
Summary
A specialized strategy for long-offset data processing is
applied to a dataset from the Faeroe-Shetland Basin, North
Atlantic, in order to image the sub-basalt structure. A nonstretch
third term moveout correction is implemented both
in the time-offset and in the tau-p domain to exploit wideangle
information. Non-stretch processing in the tau-p
domain provides remarkable advantages over the
conventional processing sequence in the time-offset
domain.
Introduction
In the Faeroe-Shetland Basin of the northeast Atlantic
Margin, there was significant igneous activity at about 55
Ma, when the proto-Icelandic plume initiated. The typical
setting in the area includes the water column, near surface
sediments, basalt extrusions, Tertiary sediments and the
basement. The geological structure beneath the basalt flows
in the area of the Faeroe-Shetland basin has been a serious
concern for the oil companies active in the region. The
hydrocarbon exploration industry is seeking for the traps in
Mesozoic and lowermost Cenozoic sediments covered by
massive basalt extrusions (Spitzer and White, 2003).
Recently gathered long-offset seismic data sets have shown
great potential for providing information from the deeper
parts of the sub-basalt zones. There are important questions
still remaining unanswered such as the thickness of the
basaltic layers, and the structure and fluid content of the
sub-basalt sediments. Conventional seismic exploration
methods fail to provide reliable answers to these questions.
Seismic imaging in the Faeroe-Shetland basin faces the
common sub-basalt imaging difficulties. Strong reflectivity
caused by the significant velocity contrast between the top
basalt and the young sediments above reflects a major part
of the seismic energy upward. The weak transmitted part of
energy travels through thick, highly absorbing and
scattering basaltic layers, and then gets reflected from
sediments and basement below the basalt, which makes
imaging of the sediments below the basalt layer difficult.
Furthermore, the reflected part of energy from the seafloor
and the sediments above the basalt generates strong short
and long period multiples, which arrive at about the same
time as primary reflection from the sediments below the
basalt, and mask overwhelmingly these weak signals.
Refraction from the top of the basalt is another source of
difficulty in the data processing. At the far offset zone, the
strong refraction from the top of the basalt, accompanied by
the linear noise of ringing and multiplication, arrive
tangentially at about the time of the reflection signal from
the sub-basalt targets. Such a contamination causes severe
problems during the velocity analysis and stacking process.
There are certain requirements for acquisition, processing
and interpretation of long-offset data in order to image
structures below the basalt. In this case study we implement
a specialized processing strategy utilizing our new method
of non-stretch stacking (Masoomzadeh et al., 2004).
Processing strategy
The processing strategy implemented in this case study is
summarized in Figure 1. After data preparation there are
two alternative routes. Route A follows the processing
sequence in the time-offset domain and can produce either
conventional stack or non-stretch stack. Route B on the
other hand, is carried out in the tau-p domain and produces
two more alternative stack sections, one after the
downward continuation, and the other one after non-stretch
shifted-elliptical moveout correction. The resulting stacked
sections are then subjected to post-stack processing
including time variant band pass filter, f-x deconvolution,
Figure 1: Flowchart of processing strategy applied to the longoffset
data set from the Faeroe-Shetland Basin. Processing is
performed in the tau-p domain as well as the t-x domain, and four
alternative stack sections are produced (showing in white boxes).

trace mixing, automatic gain control and migration.
Data preparation
The data was recorded on a 12 km streamer with 12.5 m
group interval and 37.5 m shot interval. After trace
decimation the offset interval increased to 25 m and each
CMP gather contained 160 traces with 75 m offset
increment. Since this value of offset increment was
producing spatial aliasing distortion during the Radon and
tau-p transforms, three adjacent CMP gathers were
combined in order to include 480 traces with 25 m offset
spacing in each super gather. CMP spacing then increased
to 37.5 m. In this way, signal-to-noise at each CMP
increases for the price of lateral resolution, which is
acceptable when targeting the deep interfaces.
High-resolution Radon filter
Conventionally, it is assumed that after moveout correction,
primary hyperbolas are corrected to flat lines, and multiples
are converted into parabolas. Hence a parabolic Radon
filter is normally used after an initial velocity analysis and
NMO correction. In long-offset data processing, however,
this procedure is impractical because the whole process of
NMO + filter + de-NMO is potentially destructive to the
wide-angle data. An exact hyperbolic Radon filter without
any NMO and de-NMO corrections is performed as a
preferred alternative. The method we adopted was to
convert the data from the time domain to the time-squared
domain, where hyperbolas convert to parabolas, so that the
parabolic Radon filter can be applied to remove multiples
without any NMO correction. Possible remaining multiples
in near offset zone can be suppressed using inner trace
mute.
Offset optimization
Since the recorded reflection events from the deep targets
usually have low signal-to-noise ratios, it is important to
concentrate the stacking process on the offsets that contain
higher s/n ratios. Top and bottom mutes with appropriate
tapering are used to exclude the parts of the gathers
containing lower s/n ratios. Muting and weighting are used
to reduce multiple strength in the stacked section. A trace
weighting can strengthen those parts of the gathers that
contain higher s/n ratios and can therefore result in an
enhanced image. The square root of offset is multiplied by
the trace amplitudes as a complimentary for the amplitude
recovery process.
Figure 2 shows the top and bottom mutes applied to a super
CMP gather. The dashed arrow highlights the wide-angle
zone containing valuable information from the deep targets,
normally excluded from the stack by a conventional 30%
Sub-basalt imaging in the Faeroe-Shetland Basin
stretch mute. Non-stretch stacking on the other hand,
exploits this information for which the long-offset
acquisition is carried out.
Figure 2: Offset optimization on a sample super CMP gather after
standard NMO. Conventional mute (30% stretch) eliminates the
valuable wide-angle information. The aim of non-stretch stacking
is to exploit wide-angle information by including a wider range of
traces into the stack. Arrows show the potential increase in the
effective foldage (see Figures 4 and 5).
Tau-p transformation
There are advantages of carrying out the major steps of data
processing in the plane-wave domain, especially when
densely sampled multi-channel long-offset seismic data is
under consideration. In Figure 3 a super CMP gather is
shown before and after tau-p transform. Merging,
weighting, muting and tapering are applied before the
transformation in order to reduce the effect of spatial
aliasing and the edge effect noise, and also to attenuate
multiples energy from the zone of interest.
An inner trace mute removes part of the multiples as well
as part of the primaries. However, the muted part of
multiples corresponds to a wider range of slownesses than
that of primaries in the tau-p domain. For instance a
primary event and a multiple event arriving at the same
two-way-time are considered in Figure 3. For the primary
reflector, the inner mute removes only the information
corresponding to slownesses up to 10 ms/km, whereas for a
multiple at the similar time, the inner mute removes up to
200 ms/km. In this way, inner mute provides a multiplefree
zone in the low slowness portion of the tau-p gathers.
The parts of multiples not removed by the inner mute are

mapped into a region of the tau-p panel that will not
contribute to the stacking process. For example, the
highlighted primary terminates at 200 ms/km slowness,
which means that the slownesses beyond 200 ms/km that
are occupied by the remaining parts of the multiples will
not contribute to stack. The significant difference in the
slowness ranges of primaries and multiples provides a
natural de-multiple approach as an alternative to the Radon
filter in the t-x domain. This approach, based on an inner
trace mute in t-x and a high-slowness mute in the tau-p
domain, is fast and efficient for long-offset data.
Note that the primaries remaining in the lower slowness
part of the gather have relatively uniform amplitudes. The
effect of spherical divergence is automatically corrected by
plane-wave decomposition. The other amplitude decay
factors, such as attenuation and energy partitioning, still
remain, and therefore a further amplitude recovery process
either before or after stack should be applied.
Figure 3: A super gather after offset optimization including
weighting, tapering and muting (left). After tau-p transformation
(right), linear noise caused by spatial aliasing and edge effect is
controlled. As a result of inner mute, multiples are removed from
the highlighted low slowness zone where primaries are expected to
occur. The remaining high-slowness parts of multiples will not
contribute in the stacking process in the tau-p domain.
Non-stretch higher-order moveout correction
Most of the interesting parts of the reflected events from
deep targets are prone to be enormously stretched and
consequently muted as a result of conventional standard
moveout correction. Enormous stretch happens around the
critical distance, where wide-angle arrivals are strong.
Aiming to exploit long offset information in the data most
fully, we perform non-stretch stacks both in the time-offset
and in the tau-p domains. Figure 4 shows a sample super
CMP gather after multi-window non-elliptical constant
moveout correction using window size of 160 ms with half
window overlap. Event discontinuities at the window edges
Sub-basalt imaging in the Faeroe-Shetland Basin
are controlled by the use of overlapping horizon-consistent
windows picked via the analysis of an animation of
constant moveout stacks. Based on the idea of shiftedhyperbola
introduced by de Bazelaire (1988) and improved
by Castle (1994), we extract equations for fourth-order
non-elliptical and non-hyperbolic moveout correction:
2 2
Δ τ = τ k ( 1 − 1 − p v ) ,
Δt
= k
0 shif
(
2
t 0
2
+ 2 2
k v shift
− t 0
where
k ≅
⎛
⎜ 1 −
⎜
⎝
1 −
2
v dive
2
v int
⎞ ⎛
⎟ /
⎜
1 −
⎟ ⎜
⎠ ⎝
1 −
2
v shift
2
v int
⎞
⎟ .
⎟
⎠
Empirical relations are used to estimate vdive and
from vrms and vavg or v int .
v shift
Time (sec)
2
3
4
5
6
Slowness (ms/km)
100 200 300
Downward continuation
The downward continuation method using the interval
velocities in a layer-stripping basis provides exact moveout
correction in the tau-p domain (Clayton and McMechan,
1981). In a practical processing work-flow we used the
depth conversion tool to implement the moveout correction
process. We convert the interval velocity to a conversion
velocity field using the equation:
v con
=
x
2000
2
1 − p v
2
int
in which 2000 (m/s) is the velocity that converts a trace in
TWT (ms) to an identical trace in depth (m). We import
this conversion velocity field into a depth conversion tool
to convert the tau-p traces into their zero-slowness
equivalent status. The dynamic modification performed by
Time (sec)
2
3
4
5
6
)
,
Slowness (ms/km)
100 200 300
Figure 4: Sample gather after elliptical moveout correction in the
tau-p domain (left), and after non-stretch fourth order (shiftedelliptical)
moveout correction (right).

such conversion is equal to the alteration required by the
summation of elliptical shifts for a number of individual
layers in order to perform exact moveout correction in the
tau-p domain.
When the downward continuation approach is considered
for moveout correction in the tau-p domain, the non-stretch
moveout correction can be performed by considering an
imaginary thin layer with interval velocity of zero, just
beneath the targeted interface. The corresponding RMS
velocity function then follows the local iso-moveout curve
(see Masoomzadeh et al, 2004).
The result of the application of the strategy in the tau-p
domain is shown in Figure 5; it clearly shows the top of the
basalt, structures within the basalt, sediments below the
basalt and basement. On the conventional stack (not shown
here), we were only able to image the top of the basalt.
Conclusions
We have developed a strategy for the advanced processing
of long-offset profiles from the Faeroe-Shetland Basin that
provides an excellent quality of the seismic images. The
specialized processing sequence was applied to the dataset
to tackle a number of specific problems such as frequency
content, water bottom multiples, moveout stretching, nonhyperbolic
travel time, turning ray contamination etc.
Animations of non-stretch stacks in the tau-p domain, and
in the t-x domain, allows the extraction of horizon
consistent iso-moveout strips which are combined to
perform the non-stretch stacked section. We also applied
higher order moveout correction based on the ideas of
‘downward continuation’ and ‘shifted-ellipse’.
The advantages of long-offset data processing in the plane-
Sub-basalt imaging in the Faeroe-Shetland Basin
wave domain includes the inherent spherical divergence
correction, the possibility of source and receiver directivity
compensation, the feasibility of layer stripping downward
continuation and the efficiency of filtering multiples and
turning rays. By stacking in the plane-wave domain we
avoid the side effects of the inverse tau-p transform.
Although we have obtained convincing stack images for the
area, there are still potential avenues to be explored.
Forward modelling, inversion, pre-stack migration and
converted wave processing are promising fields of further
work. Performing a scan of constant-velocity pre-stack
migrated sections can potentially provide precise velocity
information as well as non-stretch pre-stack migrated
section.
References
Castle, R.J., 1994, A theory of normal moveout,
Geophysics, 59, 983-999.
Clayton, R.W. and G.A. McMechan 1981, Inversion of
refraction data by wave field continuation. Geophysics, 46,
860-868.
de Bazelaire, E., 1988, Normal moveout revisited:
Inhomogeneous media and curved interfaces, Geophysics,
53, 143-157.
Masoomzadeh, H., Barton P. J., Singh, S. C., 2004, Nonstretch
stacking in the tau-p domain: exploiting long offset
arrivals, 74 th Ann. Internat. Mtg: Soc. of Expl. Geophys.,
SP-P2.
Spitzer, R., and White R.S., 2003, Enhancing sub-basalt
reflections using parabolic τ-p transformation. The Leading
Edge, 22, 1184-1201.
Top basalt
Figure 5: Non-stretch stack in the tau-p domain. Benefiting from the processing strategy shown in Figure 1 (route B), the intra-basalt and subbasalt
structures in the Faeroe-Shetland Basin are imaged more clearly than the previous attempts based on conventional processing.