Processing seismic data in the presence of residual statics

Aaron Stanton*, Nasser Kazemi, and Mauricio D. Sacchi, Department of Physics, University of Alberta

SUMMARY each trace as an estimate of the residual statics. A problem

with this method is that it is sensitive to the velocity used to
Many common processing steps are degraded by static shifts NMO correct the input gathers (Eriksen and Willen, 1990).
in the data. The effects of static shifts are analogous to noise or Traonmilin and Gulunay (2011) tackle the problem by simulta-
missing samples in the data, and therefore can be treated using neously estimating the statics during projection filtering for the
constraints on sparsity or simplicity. In this paper we show that purpose of denoising seismic data. Recently Gholami (2013)
random static shifts decrease sparsity in the Fourier and Radon showed that sparsity maximization can be used as a criterion
transforms, as well as increase the rank of seismic data. We for short period statics correction. Building off the work of
also show that the concepts of sparsity promotion and rank re- Stanton and Sacchi (2012) we propose two algorithms that al-
duction can be used to solve for static shifts as well as to carry low for statics to be estimated during 5D trace interpolation
out conventional processes in the presence of statics. The first and denoising as well as for demultiple using the sparse Radon
algorithm presented is a modification to the reinsertion step of transform in the presence of statics.
Projection Onto Convex Sets (POCS) and Tensor Completion
(TCOM) that allows for the compensation of residual statics
during 5D denoising and interpolation of seismic data. The THEORY
method allows preserving residual statics during denoising, or
correction of residual statics in the case of simultaneous de- 5D Reconstruction
noising and interpolation. An example is shown for a 5D re- Our method begins with the observation that statics appear
construction of synthetic data with added noise, missing traces to have the same character as noise or missing traces in the
and random static shifts, as well as for a 2D stacked section Fourier domain. Figure 1 shows a 2D synthetic gather (a) with
with missing traces and static shifts. While standard recon- added noise (b), missing traces (c), static shifts (d), and their
struction struggles in the presence of even small static shifts, respective f-k amplitude spectra (e)-(h). The destruction of the
reconstruction with simultaneous estimation of statics is able signal observed in the f-k amplitude spectra for each of these
to accurately reconstruct the data. The second algorithm pre- three cases are remarkably similar.
sented is a Statics Preserving Sparse Radon transform (SPSR).
The assumption of any Fourier reconstruction method is that
This algorithm includes statics in the Radon bases functions,
the desired noise-free, fully-sampled signal can be sparsely
allowing for a sparse representation of statics-contaminated
represented in the Fourier domain. In this paper we show that
data in the Radon domain.
this same sparsity relation can be used for the removal of static
shifts within the data.

INTRODUCTION In the case of 5D Projection Onto Convex Sets (POCS) recon-

struction a Fourier estimate of the data is found by iteratively
Many processing steps can fail in the presence of even small thresholding the amplitude spectrum of the data (Abma and
static shifts (≤ ±10ms). Creating robust methods that can han- Kabir, 2006). For a given temporal frequency, ω, the data in
dle residual static shifts can benefit early stages of processing the ω − mx − my − hx − hy domain at the kth iteration of POCS
workflows, such as denoising, interpolation, and multiple at- are given by
tenuation. This paper is divided into two parts. In the first
Dk = α1 Dobs + (1 − α1 S)FD−1 T FD Dk−1 , k = 1, ..., N, (1)
part a modification to the reinsertion step of Tensor Comple-
tion (TCOM) and Projection Onto Convex Sets (POCS) based where mx , my , hx , hy are midpoints and offsets in the x and y
reconstruction is provided. This step inserts the estimated data directions respectively, Dobs are the original data with missing
into the original data and can be used for both interpolation traces, and FD and FD−1 are the forward and reverse 4D Fourier
and denoising. We propose a modification to the reinsertion transforms in the spatial dimensions respectively. In this nota-
that allows for the compensation of residual static shifts dur- tion Dk (ω, kmx , kmy , khx , khy ) = FD Dk (ω, mx , my , hx , hy ), and T
ing this step. In the case of POCS the property being exploited is an iteration dependent threshold operator that is designed us-
is sparsity in the frequency-wavenumber domain, while in the ing the amplitudes of the input data (Gao and Sacchi, 2011). S
case of TCOM the property exploited is rank in the frequency- is the sampling operator and is equal to one for points with ex-
space domain. isting traces and zero for points with unrecorded observations.
The scaling factor α1 ≤ 1 can be used to simultaneously de-
Residual statics are typically attributed to near surface lateral
noise the data. A choice of α1 = 1 reinserts the noisy original
velocity and topographical variations (Ronen and Claerbout,
data at each iteration, whereas a lower value of α1 will denoise
1984). Because these effects are considered to be surface con-
the volume by averaging the original and reconstructed data.
sistent their correction involves a single static shift for each
trace. A common practice for residual static estimation is to The modification we propose to allow for statics to be com-
use the cross-correlation of unstacked data within a CMP gather pensated for during the reconstruction is to derive a static shift
compared with the stacked trace, taking the maximum lags for between the thresholded data and the data from the previous
 Processing seismic data in the presence of residual statics

iteration. The data in the ω − mx − my − hx − hy domain at the and regularization term. The cost function of equation (5) is
kth iteration of POCS with statics compensation are given by complete if data has no statics. As discussed earlier statics in-
k
troduce artifacts and smears the Radon model. Considering
Dk = α1 Dobs e−iω(1−α2 )τ +(1−α1 S)FD−1 T FD Dk−1 , k = 1, ..., N, statics changes equation (5) to
(2)
where τ k (mx , my , hx , hy ) are the estimated static shifts at the kth 1
argmin J = argmin kQd − Lmk2 + τ kmk1 , (6)
iteration and are constant for all frequencies, ω. The scaling m,Q m,Q 2
factor α2 ≤ 1 can be used to control the level of static cor-
where Q is shifting operator that corrects statics in data. Equa-
rection. A choice of α2 = 1 can be used for data with little
tion (6) can also be written as
to no residual statics, whereas a value of α2 = 0 will remove
statics more aggressively by fully applying the estimated static 1
shifts at a given iteration. The time shifts τ k (mx , my , hx , hy ) are argmin J = argmin kd − QT Lmk2 + τ kmk1 , (7)
m,Q m,Q 2
the lags given by the maximum values of the cross correlation
of the static corrected input data from the previous iteration, where QT is the adjoint operator of Q and puts statics back in
k−1
Dobs e−iω(1−α2 )τ (where τ k−1 = k−1 n
P
n=1 τ ), with the thresh- to the Radon predicted data. The only advantage in using equa-
olded data from the current iteration, FD−1 T FD Dk−1 . This al- tion (7) instead of equation (6) is that equation (7) preserves
lows for the iterative application of noise attenuation, missing statics in the predicted data. The cost function of equation (7)
trace interpolation, and static correction. can be minimized by alternatively solving the following sub-
problems:
The reinsertion step of POCS is identical to that used during
5D Tensor Completion (TCOM) (Stanton et al., 2012). This 1 b T Lmk2 + τ kmk1 , (8)
m- step b = argmin
m kd − Q
implies that we can replace the Fourier estimate of the data, m 2
FD−1 T FD Dk−1 , at a given iteration, k, with a rank-reduced
1
version of the data R(Dk−1 ). This formulation has the advan- Q- step b = argmin
Q kd − QT Lmk
b 2. (9)
Q 2
tage that it can deal with the reconstruction of curved events
(Kreimer and Sacchi, 2011). In the case of noise attenuation
By considering the initial shifting operator as the identity ma-
given data that is fully spatially sampled one may wish to de-
trix, the m- step can be solved using Fast Iterative Shrinkage-
noise the data while preserving residual statics. The total resid-
Thresholding Algorithm (FISTA) (Beck 2009). The Q- step
ual static corrections applied during the denoising are con-
has a closed form solution but it could be considered a full
tained in τ N (mx , my , hx , hy ) allowing for them to be removed
matrix and without physical meaning. So, instead of solving
from the data. In the case of interpolation it is preferable to
directly the Q- step we update the shifting operator by simply
leave the static corrections applied to avoid static shifts be-
cross correlating the d and Lmb vectors. Note that this is also a
tween interpolated and original traces. The fact that the algo-
valid solution by defining the Q operator space as a combina-
rithm produces a tensor of time-shifts τ(mx , my , hx , hy ) could
tion of some shifting basis functions.
offer an advantage for other processing steps. Converting this
tensor to shot and receiver coordinates, τ(sx , sy , gx , gy ), the
time shifts could then be used for further processing or to gain
EXAMPLES
a better understanding of the near surface.
Sparse Radon Transform To test the reconstruction algorithm we apply the algorithm to
The Sparse Radon Transform can be written as the linear op- a 5D synthetic dataset with dimension 100x12x12x12x12. The
eration. Data, d can be generated from a model m under the data has hyperbolic moveout in all four of the spatial directions
action of the Radon operator L as follows: as seen in figure 2. This figure shows one central bin location
out of a total of 144 bins that comprise the complete data. The
d = Lm + n, (3) complete noise free data is shown in figure 2 (a). Before recon-
struction random noise was added to the data giving a signal to
where d is data, m is the Radon model and n is the noise con-
noise ratio of 2. Random traces were then decimated from the
tent. Thorson and Claerbout (1985) showed that given the data,
data leaving 50% of the original traces. Random static shifts
equation (3) can be solved via damped least squares approach.
between ±10ms were then applied to the data producing the
In other words, the objective function is
data seen in figure 2 (b). Standard 5D POCS reconstruction
minimize ||m||2 s.t. ||d − Lm||2 < ε (4) was applied to the data resulting in the data shown in figure 2
(c). The static shifts cause a very low quality reconstruction
where ε is some estimate of noise level in the data. To increase that smears the signal. 5D POCS Reconstruction with static
the resolution of the Radon model, one can adopt l2 norm for compensation gives a much higher quality result as seen in fig-
data misfit and l1 norm for the model ure 2 (d). Figure 2 shows reconstruction results for a stacked
inline of data that has been corrupted with random static shifts
1
b = argmin
m kd − Lmk2 + τ kmk1 , (5) (a). The result after simultaneous statics computation and re-
m 2 construction (c) is of higher quality compared to the result after
where m is desired sparse model and τ is a regularization pa- standard reconstruction (b). To test the Radon demultiple al-
rameter that balances the importance of the misfit functional gorithm we show an NMO corrected CMP gather from a Gulf
 Processing seismic data in the presence of residual statics
(a) (b) (c) (d)
0 0 0 0

0.2 0.2 0.2 0.2

0.4 0.4 0.4 0.4

time (s)

time (s)

time (s)

time (s)
0.6 0.6 0.6 0.6

0.8 0.8 0.8 0.8

1 1 1 1
0 500 1000 1500 0 500 1000 1500 0 500 1000 1500 0 500 1000 1500
offset (m) offset (m) offset (m) offset (m)
(e) (f) (g) (h)
0 0 0 0

20 20 20 20
frequency (Hz)

frequency (Hz)

frequency (Hz)

frequency (Hz)
40 40 40 40

60 60 60 60

80 80 80 80

100 100 100 100

−0.02 0 0.02 0.04 −0.02 0 0.02 0.04 −0.02 0 0.02 0.04 −0.02 0 0.02 0.04
wavenumber (cycles/m) wavenumber (cycles/m) wavenumber (cycles/m) wavenumber (cycles/m)

Figure 1: a) Original 2D synthetic data. b) Data with random noise. c) Data with random missing traces (50%). d) Data with
random ±10ms statics shifts. e-h) Are the f-k amplitude spectra of a-d.
(a) (b)
0 0
0.05 0.05
0.1 0.1
0.15 0.15
time (s)

time (s)

0.2 0.2
0.25 0.25
0.3 0.3
0.35 0.35

0 100 200 300 400 500 600 700 800 900 1000 1100 1200 0 100 200 300 400 500 600 700 800 900 1000 1100 1200
y−offset (m) y−offset (m)
(c) (d)
0 0
0.05 0.05
0.1 0.1
0.15 0.15
time (s)

time (s)

0.2 0.2
0.25 0.25
0.3 0.3
0.35 0.35

0 100 200 300 400 500 600 700 800 900 1000 1100 1200 0 100 200 300 400 500 600 700 800 900 1000 1100 1200
y−offset (m) y−offset (m)

Figure 2: a) A portion of noise-free, static-free, fully sampled 5-D synthetic data. b) Data after adding random noise (SNR =
2), random ±10ms static shifts, and randomly removing traces (50%). c) Data after standard 5D reconstruction d) Data after
simultaneous 5D reconstruction and statics computation.
Input: with statics: 50% of traces decimated Output: with statics: Standard POCS Output: with statics: POCS with statics
0 0 0

0.2 0.2 0.2

0.4 0.4 0.4

0.6 0.6 0.6

0.8 0.8 0.8

Time (s)

Time (s)

Time (s)

1 1 1

1.2 1.2 1.2

1.4 1.4 1.4

1.6 1.6 1.6

1.8 1.8 1.8

1100 1200 1300 1400 1500 1600 1700 1100 1200 1300 1400 1500 1600 1700 1100 1200 1300 1400 1500 1600 1700
CDP CDP CDP

(a) (b) (c)

Figure 3: Stacked section (a) with missing traces and +/-5ms static shifts, (b) after standard reconstruction , and (c) after simulta-
neous reconstruction with statics computation.
 Processing seismic data in the presence of residual statics
Distance (m) Far-offset Residual Moveout (s) Far-offset Residual Moveout (s)
1000 2000 3000 4000 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8

2 2 2

Time (s)

Time (s)

Time (s)
4 4 4

6 6 6

(a) (b) (c)

Figure 4: Marine CMP gather with random +/- 10ms static shifts, (a) input data, (b) sparse Radon Transform with no statics
computation, (c) sparse Radon transform with statics computation.

of Mexico towed-streamer dataset in figure 4 (a). The gather is

corrupted with random +/-10ms static shifts. The sparse Radon
Distance (m)
transform is shown in (b), while the sparse Radon transform 1000 2000 3000 4000

with simultaneous statics computation is shown in (c). The

sparsity of the Radon panel is improved through the use of the
new algorithm. Figure 5 shows the estimated data using the
2
conventional sparse Radon transform (a), compared with us-
ing SPSR. Notice the dimming of amplitudes in the estimate
produced using the conventional sparse Radon transform. The Time (s)

SPSR result preserves the statics on the data, allowing for mul- 4

tiple suppression of statics-contaminated data.

6
CONCLUSION

We presented methods to process seismic data in the presence

of residual statics. The methods are able to preserve the static (a)
shifts in the case of noise and multiple removal, or to compen- Distance (m)
1000 2000 3000 4000
sate for the shifts in the case of simultaneous denoising and
trace interpolation. The methods make use of sparsity or sim-
plicity of the static-free seismic data. A topic of future research
is to incorporate surface consistency into the methodology to 2

characterize anomalies in the near surface.

Time (s)

4
ACKNOWLEDGEMENTS

The authors would like to thank the sponsors of the Signal

Analysis and Imaging Group (SAIG) at the University of Al- 6

berta, and the USGS for use of example seismic data.

(b)

Figure 5: Marine CMP gather with random +/- 10ms static

shifts, (a) data estimated using conventional sparse Radon
transform, (b) data estimated using sparse Radon transform
with statics computation.
 Processing seismic data in the presence of residual statics

REFERENCES

Abma, R., and N. Kabir, 2006, 3D interpolation of irregular

data with a POCS algorithm: Geophysics, 71, E91–E97.
Eriksen, E. A., and D. E. Willen, 1990, Interaction of stacking
velocity errors and residual static corrections: SEG Techni-
cal Program Expanded Abstracts, 9, 1730–1733.
Gao, J., and M. Sacchi, 2011, Convergence improvement and
noise attenuation considerations for POCS reconstruction:
Presented at the 73rd EAGE conference and exhibition.
Gholami, A., 2013, Residual statics estimation by sparsity
maximization: Geophysics, 78, V11–V19.
Kreimer, N., and M. D. Sacchi, 2011, A tensor higher-order
singular value decomposition (hosvd) for pre-stack simul-
taneous noise-reduction and interpolation: SEG Technical
Program Expanded Abstracts, 30, 3069–3074.
Ronen, J., and J. F. Claerbout, 1984, Surface-consistent resid-
ual statics estimation by stack optimization: SEG Technical
Program Expanded Abstracts, 3, 420–422.
Stanton, A., N. Kreimer, D. Bonar, M. Naghizadeh, and M.
Sacchi, 2012, A comparison of 5d reconstruction methods:
SEG Technical Program Expanded Abstracts, 1–5.
Stanton, A., and M. D. Sacchi, 2012, 5D reconstruction in the
presence of residual statics: Presented at the CSEG Annual
Meeting.
Traonmilin, Y., and N. Gulunay, 2011, Statics preserving pro-
jection filtering: SEG Technical Program Expanded Ab-
stracts, 30, 3638–3642.