Partial Surface Related Multiple Elimination: Pierre Hugonnet, CGG Paris

Partial Surface Related Multiple Elimination
Pierre Hugonnet*, CGG Paris
Summary Iterating is however not always necessary. At the first

iteration, D is used as an operator instead of P. The
kinematics of the modeled multiples is still correct, but the
The surface-related multiple elimination (SRME) method,
amplitudes of the high order multiples are overestimated:
described by Verschuur et al. (1992), represented a
technological breakthrough in the world of multiple − s −1 * D ⊗ P = M 1 + M 2 + M 3 + K
attenuation. However, some points have been identified as
problematic in practice. These include the convergence of − s −1 * D ⊗ D = M 1 + 2 M 2 + 3 M 3 + K (E 4)
the iterative process, the 3D effects that alter the modeling, Where Mi are the multiples of the ith order.
and the adaptive subtraction step. We propose here a slight
variation of the method. It is non-iterative, thanks to a more For a given order, the error is a global gain, which can be
accurate modeling at the very first "iteration". The easily compensated by the adaptive subtraction. The
drawback is that only part of the surface-related multiples troubles arise when different orders interfere within a
are modeled. As a side effect, the multiple model is also subtraction window, so that they cannot be adapted
split into several independent models: the subtraction stage simultaneously. But if the different orders are mostly
can however take advantage of this decoupling to enhance separated, or if a given order is dominant, then the
the attenuation. subtraction can do the job. This is typically the case with
deep water data, but also for less deep water, depending on
Introduction the size of the subtraction windows (the smaller the
windows, the less interference between orders).
In the SRME method, the surface multiples M are modeled
by applying to the recorded data D an operator that But for some data, like shallow/medium water, the
involves the primary wavefield P itself. This result must be iterations are still necessary, or at least useful. A known
deconvolved by the source wavelet s (including ghosts). A solution to reduce their number is to input a partially
minus sign accounts for the free surface reflection: demultipled dataset (for example by a parabolic Radon
M = − s −1 * D ⊗ P (E 1) filtering, Berkhout and Verschuur, 1997) as a first guess of
P. This however doesn’t totally suppress the iterations, and
The symbol ⊗ consists of a 2D pre-stack convolution, and the cost of the preliminary demultiple must be considered.
it includes various corrections (see Verschuur et al. (1992) And in fact this suggests another technique.
for more details). Subtracting the modeled multiples from
the original data gives the primaries: Partial, single-pass modelling
P = D + s −1 * D ⊗ P (E 2)
With marine data, if we exclude the ultra-shallow water
In practice, this equation is close to useless, since both the cases (less than 100ms), it is always possible to isolate a
primaries and the source wavelet are unknown beforehand. shallow window Psh that is free of surface multiples, the
It is usually solved iteratively with a two-step first multiple arriving at twice the water-bottom time
modeling/subtraction loop (Verschuur and Berkhout, (Figure 1). These windowed data can serve as an operator:
1997). The estimation of P is progressively refined, starting M ′ = − s −1 * D ⊗ Psh (E 5)
from a first guess that can be the data D themselves. The
inverse source wavelet is estimated statistically by an
adaptive subtraction, based on the minimum energy
assumption: D P s h
P ( 0) = D
 M ( n ) = − D ⊗ P (n −1) (E 3)

 P (n ) = D − f ( n ) * M ( n )
The iterations in (E 3) are not only costly, but the

convergence is not always satisfactorily controlled (Nekut
and Verschuur, 1998). Figure 1: Primaries are in black and blue. Multiples are in (dashed)
red. The shallow window Psh contains only primaries (in blue).
SEG International Exposition and 72nd Annual Meeting Main Menu

Partial SRME
Since the operator contains only primaries, the model (E 5) Mr = − D ⊗ Psh

has correct amplitudes on all orders of multiples. Thus, the
iterations can be totally discarded. The drawback is that Ms = − Psh ⊗ ( D − Psh ) (E 7)
less multiples are modeled: only the generators present in Msr = − Psh ⊗ D ⊗ Psh
the operator are considered. This is not a serious problem,
since very often the main generators are the water bottom We then subtract simultaneously these three models, by
itself and a few shallow interfaces. searching for three different adaptation filters (fs, fr, fsr),
still under a minimum energy assumption (Lokshtanov,
The fact that Psh contain only a part of the generators has 2000):
another consequence: the operator is no longer
commutative. (E 5) models in fact the receiver-side peg-
P = D − f r * Mr − f s * Ms − f sr * Msr (E 8)
legs only (Bierstaker, 2001). Reversing the application side
of Psh ( Psh ⊗ D ) models the source-side peg-legs. Some Besides being easier to implement, allowing fully
independent filters instead of linking them together is also a
multiples being then modeled twice (by both sides), a way of giving more freedom to the subtraction. On real
corrective term ( Psh ⊗ Psh ) is required. Finally a 2nd order data, this may be an advantage: we know that the modeling
correction term, combining source-side and receiver-side is for example strongly affected by the 3D effects (Ross,
( Psh ⊗ D ⊗ Psh ), is added. The full expression of the 1997). Up to a certain point, the subtraction can
compensate for some small errors. We can easily imagine
partial model is then (Figure 2):
that the errors differ on the different models (different local
M ′ = − s −1 * D ⊗ Psh − s −1 * Psh ⊗ D cross-line dips encountered by the different travel paths):
(E 6) different filters may hence better compensate for them.
+ s −1 * Psh ⊗ Psh − s −2 * Psh ⊗ D ⊗ Psh
Illustration on synthetic data
We use a synthetic model (Figure 3) to illustrate the

different steps of the method.
On the simulated data (2D acoustic finite differences) in

Figure 4 many multiples are observed. On the displayed
offset section, it appears clearly that some multiples have
different arrival times depending on the side of the peg-leg.
A superficial window Psh without multiples is extracted.
Figure 2: the four terms in (E 6) First, we compute the full model (E 4) and subtract it from
the data (Figure 5). The subtraction is performed in the
The model (E 6) is accurate in both terms of kinematics and CMP domain, using a single window. This is the first
amplitudes. However, the inverse source wavelet is still iteration of standard SRME. The attenuation is not perfect
unknown. Estimating it by adaptive subtraction is a non- since several orders of multiples are present in the
linear problem because of the 2nd order term. subtraction window.
Note that this 2nd order term can be neglected if the Then we compute the partial models (E 7) and subtract
reflection coefficients in Psh are small. Then, after summing them with the multi-model adaptation (E 8) (Figure 6). All
the 1st order terms into a single model, a conventional multiples, of any order, are better attenuated than with the
linear adaptive subtraction can be applied. Otherwise, an 1st iteration of the standard SRME. The attenuation is still
alternative for dealing with a linear problem is to search for not perfect but it is now very satisfactory.
different filters for each term of the partial model.
Multi-model subtraction
From (E 6) we consider that we have three models: a

receiver-side model, a modified source-side model, and a
2nd order correction model. Ignoring the inverse source
wavelet, we write:

Partial SRME
Real data example Lokshtanov, D., 2000, Suppression of water-layer

multiples--from deconvolution to wave-equation approach,
The real data chosen here (Figure 7) are not the best 70th Ann. Internat. Mtg: Soc. of Expl. Geophys., 1981-
candidate for SRME. Actually, for such horizontally- 1984.
layered media, where moreover the water-bottom is by far
Nekut, A. G. and Verschuur, D. J., 1998, Minimum energy
the main generator, a tau-p deconvolution algorithm
performs better than SRME. However, horizontally layered adaptive subtraction in surface-related multiple attenuation,
media have the advantage of offering a simple QC tool of 68th Ann. Internat. Mtg: Soc. of Expl. Geophys., 1507-
the demultiple process: the stack autocorrelation. 1510.
Ross, W. S., 1997, Multiple suppression: Beyond 2-D - Part
The water bottom is at 300ms. Its period is very clear on I: Theory, 67th Ann. Internat. Mtg: Soc. of Expl. Geophys.,
the autocorrelation. We applied the standard SRME (with 1387-1390.
only one iteration) and the partial SRME. Both the normal
and multi-model subtraction were applied on CMP gathers, Verschuur D.J., Berkhout A.J., and Wapenaar C.P.A.,
on windows of 40 traces (max fold is 60) by 1000ms. The 1992, Adaptive surface-related multiple elimination:
attenuation is much pronounced after the partial SRME, on Geophysics, 57, 1166-1177
both the stack (see the first water bottom bounce) and the Verschuur D.J. and Berkhout A.J., 1997, Estimation of
autocorrelation. However, the primary preservation seems multiple scattering by iterative inversion, Part II: Practical
to be a problem. Closer observation of the data shows that aspects and examples: Geophysics, 62, 1596-1611
some of them are altered. The explanation probably lies in
the considerable freedom of the multi-model subtraction. Acknowledgments
Perhaps too great as suggested above. A solution is to
increase the window size in the subtraction. In this The author thanks Antonio Pica for the synthetic dataset
example, we tried 60 traces (full fold) by 2000ms. Indeed, used in this paper.
the apparent attenuation is now closer to that of the
standard SRME. The autocorrelation shows a less efficient
multiple attenuation than with the small windows, but still
better than with the standard SRME after one iteration.
On this example it is indeed very difficult to estimate the

primary preservation, but what is certain is that a better
autocorrelation indicates a better primary to multiple ratio.
Figure 3: 2D synthetic model
In this respect, the partial SRME does a better job.
Conclusions
We have proposed a variation of the standard SRME

method, which for a wide range of data can avoid the
iterations of the original method. In addition, it splits the
multiple model into several terms, giving more freedom to
the adaptive subtraction. The multiples are better
attenuated, but the primary preservation is more critical,
and it requires at least more attention in the tuning of
parameters. Introducing more constraints in the subtraction
is probably the way to go.
References
Berkhout A.J. and Verschuur D.J., 1997, Estimation of Figure 4: (left) data D - constant offset section (1000m). Note the
multiple scattering by iterative inversion, Part I: Theoretical doublets on some multiples. (right) superficial window Psh with
considerations: Geophysics, 62, 1586-1595 primaries only (the events are also muted in the shallowest times)
Biersteker J., 2001, MAGIC: Shell’s surface multiple

attenuation technique: 70th Annual Internat. Mtg., Soc.
Expl. Geophys., Expanded Abstracts,

Partial SRME
Figure 5: (top) full model M; (bottom) adaptive subtraction in Figure 6: (from top to bottom and left to right): receiver-side
CMP gathers (single window) model Mr, source-side model Ms, 2 nd order source and receiver-side
model Msr; multi-model adaptive subtraction in CMP gathers
Figure 7 : (top) stack sections; (bottom) autocorrelations.

(from left to right) data with multiples; standard SRME 1 iteration; partial SRME; partial SRME with larger subtraction windows.

Partial Surface Related Multiple Elimination: Pierre Hugonnet, CGG Paris

Uploaded by

Copyright:

Available Formats

Partial Surface Related Multiple Elimination: Pierre Hugonnet, CGG Paris

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Partial Surface Related Multiple Elimination: Pierre Hugonnet, CGG Paris

Uploaded by

Copyright:

Available Formats

Partial Surface Related Multiple Elimination

Pierre Hugonnet*, CGG Paris

Summary Iterating is however not always necessary. At the first

The iterations in (E 3) are not only costly, but the

SEG International Exposition and 72nd Annual Meeting Main Menu

Since the operator contains only primaries, the model (E 5) Mr = − D ⊗ Psh

We use a synthetic model (Figure 3) to illustrate the

On the simulated data (2D acoustic finite differences) in

From (E 6) we consider that we have three models: a

SEG International Exposition and 72nd Annual Meeting Main Menu

Real data example Lokshtanov, D., 2000, Suppression of water-layer

On this example it is indeed very difficult to estimate the

We have proposed a variation of the standard SRME

Biersteker J., 2001, MAGIC: Shell’s surface multiple

SEG International Exposition and 72nd Annual Meeting Main Menu

Figure 7 : (top) stack sections; (bottom) autocorrelations.

SEG International Exposition and 72nd Annual Meeting Main Menu

You might also like