E009
Simultaneous Time-lapse Binning and
Regularization of 4D Data
P.J. Smith* (WesternGeco), I. Scott (WesternGeco) & T. Traylen
(WesternGeco)
SUMMARY
The time-lapse seismic method compares repeated 3D reflection seismic cubes acquired at different dates.
Time-lapse binning is routinely used to ensure that these 3D cubes are created from recorded seismic
traces that are as similar as possible, with poorly repeated traces being discarded. Another important
component of time-lapse seismic processing flows is the conversion of the irregularly and differently
sampled seismic surveys into collocated measurements that can be accurately compared. This
regularization procedure is usually performed independently of time-lapse binning, and the binning may
well discard traces that would have benefited the regularization and subsequent imaging. We have
developed a procedure that combines time-lapse binning and regularization into a single algorithm,
thereby achieving an optimum compromise between inter-survey trace repetition accuracy and image
quality. Initial results on real time-lapse seismic data sets are promising.
74th EAGE Conference & Exhibition incorporating SPE EUROPEC 2012
Copenhagen, Denmark, 4-7 June 2012
Introduction
The time-lapse seismic method compares repeated 3D reflection seismic cubes acquired at different
dates. Time-lapse binning (Li et al. 2003) is routinely used to ensure that these 3D cubes are created
from recorded seismic traces that are as similar as possible, poorly repeated traces being discarded.
Another important component of time-lapse seismic processing flows is the conversion of the
irregularly and differently sampled seismic surveys into collocated measurements that can be
accurately compared. While Kirchhoff imaging algorithms can perform this task, they tend to generate
artefacts when the input sampling is irregular, and a dedicated regularization procedure is usually
applied prior to imaging (Poole and Lecerf 2006).
Time-lapse binning and regularization procedures are usually performed independently, and time-
lapse binning may well discard traces that would have benefited the regularization. This can be
avoided by combining the two processes in a single algorithm.
Time-lapse binning
The term “binning” refers to the process of dividing the area of a 3D survey into a 2D grid of cells, or
“bins”, and assigning traces to those cells based on the location of the midpoint between source and
receiver. Common-offset bins are also specified, assigning traces by the distance between source and
receiver, so as to define a suite of common-offset 3D cubes. Time-lapse binning extends the concept
to multiple repeated surveys by using, for each cell of a common-offset bin, a similarity criterion to
select the most similar traces from each vintage. A criterion that minimizes ΔS+ΔR (Figure 1) is often
used (Calvert 2005). Other geometric or seismic attributes, or combinations of these, may also be
employed.
Figure 1 – Schematic diagrams showing the time-lapse binning selection procedure for a given cell.
The following description assumes, for ease of discussion, a pair of time-lapse surveys, although
multiple vintages must often be handled. A typical time-lapse binning workflow is as follows:
1) Bin the prestack 3D seismic data sets into common-offset 3D cubes.
2) For each cell of equivalent common-offset cubes from the two surveys:
a. Identify the traces from each survey that have midpoints that fall within the current
3D cell and define all possible survey to survey trace pairs.
b. Compute the time-lapse binning criterion between the available trace pairs and select
the pair with the optimum criterion value. The remaining traces are discarded.
74th EAGE Conference & Exhibition incorporating SPE EUROPEC 2012
Copenhagen, Denmark, 4-7 June 2012
The left panel of Figure 1 illustrates this procedure schematically for a given cell. Blue ellipses mark
the traces that are chosen when we use a criterion that minimizes ΔS+ΔR. In the right panel, one
member of the best match trace pair lies just outside the current cell. Conventional time-lapse binning
would choose the pair marked by the red ellipses. If we allow the binning procedure to search into
adjacent cells, indicated by the light blue shading, and into adjacent offset cubes, we can select the
most appropriate trace pair, indicated in the right panel of Figure 1 by blue ellipses.
Figure 2 – Schematic diagram showing the nominal acquisition layout of a real time-lapse survey and
the relative crossline positions of sources and receivers for a single pair of base and monitor lines.
Expanded bin time-lapse binning is not guaranteed to output one pair of traces per cell. Some cells of
a given vintage will contain multiple traces and some may be empty. In fact, for some combinations
of acquisition configurations, the outputs can be quite sparsely sampled relative to the input. Figure 2
shows schematically the acquisition configurations of a real time-lapse survey pair. The baseline
survey used two sources (S1 and S2) and four cables (C1 to C4). The monitor survey used one source
(S1) and eight cables (C1 to C8), with the source tracks repeating those of S1 on the baseline survey.
Each sail line acquires eight subsurface lines, seven of which are repeated by coincident pairs of base
and monitor lines. The upper part of the table (shaded orange) shows the source/cable combinations
chosen by conventional time-lapse binning for a hypothetical zero inline offset data set, together with
ΔS and ΔR values. The lower part (shaded blue) shows the combinations chosen by expanded-bin
time-lapse binning, which will discard all traces acquired by Source 2 of the baseline survey unless
constrained in some way. Parameterization of these constraints can be difficult and time-consuming.
Regularization
Accurate regularization is fundamental to time-lapse processing. It transforms all vintages from their
recorded locations to a regular and consistent inline, crossline, offset, and azimuth sampling, enabling
accurate comparisons. It can also reduce Kirchhoff imaging artefacts associated with irregular trace
distributions, which further improves the similarity of the repeated cubes.
A common regularization procedure uses a multidimensional interpolation to transform the irregularly
sampled input data sets onto a regular grid. 2D operators are often applied to individual common-
offset 3D cubes to achieve regular inline and crossline sampling. Alternatively, 3D operators can be
applied to the full prestack data volume to interpolate to regular inline, crossline, and offset sampling.
4D operators can comprehend an azimuth dimension as well. WesternGeco interpolates with compact
regularization operators (CRO) (Moore and Ferber 2008). Each interpolated trace is an optimally
weighted sum of all input traces that fall within a multidimensional window about the desired trace
74th EAGE Conference & Exhibition incorporating SPE EUROPEC 2012
Copenhagen, Denmark, 4-7 June 2012
location. The algorithm provides, for each output trace, an estimate of the interpolation accuracy,
which is a function of the distribution of the input traces within the multidimensional window.
Simultaneous time-lapse binning and regularization
Time-lapse binning aims to maximise the similarity of the time-lapse seismic cubes by discarding
traces from the input data, but, by doing this, can induce interpolation errors that reduce the similarity
of the regularized time-lapse seismic cubes. We have implemented an algorithm that simultaneously
regularizes multiple time-lapse seismic data sets and incorporates the functionality of time-lapse
binning. It can automatically determine the best compromise between trace similarity and
regularization accuracy, and works as follows (again we assume two surveys for ease of explanation):
1. Prestack common-offset sorted data for the two surveys are input to the program.
2. For a given output location (inline, crossline and, optionally, offset), the following is done:
a. Define a 2D (inline, crossline) or 3D (inline, crossline, offset) window, centred on the
current cell.
b. Collect the traces that fall within the multidimensional window.
c. Compute the time-lapse binning criterion for all possible trace pairs.
d. The CRO algorithm is used to create, for each survey, an interpolated trace at the
desired output location, together with its interpolation quality measure.
e. An optimization procedure then progressively discards the worst repeated trace pairs,
repeating the interpolation, until an interpolation quality threshold is reached. The
NRMS difference (Kragh and Christie 2001) between the interpolated traces for each
survey can be used as an additional optimization criterion. Being computed between
accurately collocated traces, the NRMS difference can be an effective discriminator.
f. The final interpolated traces, one for each survey, are output.
Data examples
We demonstrate the algorithm on a two-vintage time-lapse survey, acquired with the parameters
described in Table 1, from the Sleipner Vest field, offshore Norway. The highly redundant nature of
the monitor survey was intended to maximize the likelihood of accurately repeating the source and
receiver locations of the baseline survey, with the intention that, during processing, time-lapse binning
would be used to select these accurately repeated traces and discard the redundant ones.
Baseline Monitor
Number of sources / source separation 2 / 50 m 2 / 50 m
Number of streamers / streamer separation 3 / 100 m 9 / 50 m
Crossline distance between sail lines 150 150
Table 1 – The acquisition configuration for the data sets shown in Figure 3
We compared the new algorithm with both conventional and expanded bin time-lapse binning,
regularizing the time-lapse binned data with CRO, and processing each result through prestack time
migration and stack. Figure 3 shows NRMS difference maps for the three tests. Assuming that lower
NRMS difference corresponds to better time-lapse seismic quality, we see that the simultaneous
approach gives the best results and the conventional approach, the worst. This conclusion is borne out
by visual inspection of the time-lapse seismic difference sections in Figure 3. The 4D signal stands
out clearly on all three comparisons, but the residual background energy is strongest on the
conventional time-lapse binning result and weakest on the simultaneous approach. This is largely
because the simultaneous approach selects more traces than the other methods. Standard time-lapse
binning is generally parameterized to select only one trace per bin, which results in potentially useful
traces being discarded. The simultaneous approach is able to automatically identify and preserve these
useful traces, without the need for extensive testing and user input.
74th EAGE Conference & Exhibition incorporating SPE EUROPEC 2012
Copenhagen, Denmark, 4-7 June 2012
Figure 3 – NRMS maps and inline 4D seismic difference sections showing the effect of different time-
lapse binning algorithms (the position of the seismic section is marked by a dashed line on the maps).
Conclusions
Conventional time-lapse binning does not always choose the traces that optimize survey repeatability.
Expanded-bin time-lapse binning can overcome this issue, but often requires careful parameterization
to achieve adequate spatial sampling. We have implemented a simultaneous time-lapse binning and
regularization scheme that can automatically select traces that give the best compromise between
survey repeatability and image quality. The results of application to real time-lapse seismic data sets
are promising.
Acknowledgements
We thank the PL046 licence partners (Statoil AS, ExxonMobil Exploration and Production Norway
AS, and Total E&P Norge AS) for permission to show the above data examples.
References
Calvert, R. [2005] 4D technology: where are we, and where are we going? Geophysical Prospecting,
53, 161-171.
Kragh, E., and Christie, P. [2001] Seismic Repeatability, Normalized RMS and Predictability. 71st
SEG Annual Meeting, Expanded Abstracts, 20, 1, 1656-1659.
Li, X.P., Brittan, J., Harwood, S., and Widmaier, M. [2003] Azimuth Preserved Trace Binning of 4D
Seismic Data for Improved Repeatability. 65th EAGE Conference and Exhibition, Expanded
Abstracts, A27.
Moore, I., and Ferber, R. [2008] Bandwidth Optimization for Compact Fourier Interpolation. 70th
EAGE Conference and Exhibition, Expanded Abstracts, G026.
Poole, G., and Lecerf, D [2006] Effect of regularization in the migration of time-lapse data. First
Break, 24 (4), 25-31.
74th EAGE Conference & Exhibition incorporating SPE EUROPEC 2012
Copenhagen, Denmark, 4-7 June 2012
