Processing variable-depth streamer acquisition has recently become possible with a new joint deconvolution algorithm. In this particular acquisition, using variable-depth broadband streamers, the receiver depth regularly increases with offset, which allows a wide diversity of receiver ghosts and dramatically increases the possible frequency bandwidth in both low and high frequencies from 2.5 Hz to source notch. Compared to conventional flat streamer data, processing broadband data requires a major change since the receiver ghosts are rigorously taken into account.

In conventional processing, both source and receiver ghosts are included in a wavelet that is assumed to be consistent from offset to offset. With a broadband dataset, the receiver ghosts change from near offsets to far offsets and cannot be included in a wavelet. This breaks the implicit assumption of many processing steps such as surface-related multiple elimination (SRME). These receiver ghosts will then be removed from the final image with a prestack or post-stack joint deconvolution. Of course, the receiver ghost preservation is a constraint for some programs developed for conventional processing.

One of the key challenges is how to deal with demultiple techniques and variable-depth streamer data in both deepwater and shallow-water environments.

Demultiple techniques in shallow water

Figure 1 image

FIGURE 1. An autocorrelation of a BroadSeis shot gather on a window shows the ghosts varying along the offset. (Images courtesy of CGGVeritas)

In shallow-water environments, the SRME method is not well adapted for short-period multiples reflections; due to the lack of near offsets, the recorded water-bottom reflections used by SRME are often not good enough or are missing. Other common demultiple methods such as Tau-P deconvolution and shallow-water demultiple (SWD) have been tested on broadband data. The predictive deconvolution in Tau-P domain is frequently used for attenuating short-period multiples, mainly from a relatively flat and shallow-water bottom. For broadband data, this method also could be applied in both shot and receiver domain, but this could affect receiver ghosts with a periodicity close to that of the water layer. The key point is to keep a gap long enough to preserve the receiver ghosts (Figure 1).

The SWD method uses the water-layer-related multiples from the data to reconstruct the missing water-bottom primary reflections. The prediction operators, used to compute a short-period multiples model, are derived from the nearest offsets where the wavelets are close to those of conventional zero-phase wavelets. In practice, SWD allows the processor to efficiently remove the short-period multiples, with results equivalent to those expected on conventional data.

These demultiple methods were tested on a 2-D line in the central North Sea. Different trials were done by combining different tools, and the best result was achieved by combining SWD, predictive deconvolution in a Tau-P domain, and SRME (Figure 2). Using this technique, the water-layer multiples are handled by SWD and Tau-P predictive deconvolution, and SRME tackles the free-surface multiples that have longer periods. In this case, the water bottom has to be muted prior to generating the SRME model.

image- central North Sea 2-D dataset

FIGURE 2. Different demultiple results on a central North Sea 2-D dataset (prestack time migrations with post-stack joint deconvolution) demonstrate (a) predictive deconvolution in both shots and receivers, (b) a shallow-water demultiple followed by the previous predictive deconvolution in shots and receivers, and (c) shallow-water demultiple with predictive deconvolution in shots and receivers followed by 2-D SRME.

Demultiple techniques in deep water

The demultiple technique commonly used in deepwater environments is 2-D or 3-D SRME. By applying SRME on conventional data, where both source and receiver ghosts already have been included in a wavelet, the modeled multiples are close to the input data multiples. A key issue appears with broadband data because of the receiver ghosts. Variable receiver depth creates visible differences in wavelets from near to far offsets. By convolving traces with different wavelets, the standard SRME method produces multiple models with mismatched wavelets that are actually different from input data, and the differences vary from offset to offset.

Even if this particular problem can be partially solved through a wavelet adjustment, this method was developed for conventional data and cannot properly handle the multiples wavelet variations. Indeed, the standard SRME method leaves a lot of residual multiples, and the low-frequency multiples provided by the variable-depth streamer acquisition cannot be properly addressed. That is why some algorithmic modifications were introduced to improve the model prediction by normalizing the receiver ghosts. This new SRME technique allows processors to create a multiples model with correct wavelets on the full-frequency bandwidth. Once the multiples model matches perfectly with the input data, the multiples model adjustment could be even more accurate and efficient.

This new technique was successfully applied on 2-D and 3-D datasets and has consistently produced better results than standard SRME. Figure 3, from a dataset in the west of Shetlands area, illustrates the results of standard SRME and new SRME after post-stack deghosting.

image- results of different processing methodologies

FIGURE 3. These results of different processing methodologies (post-stack deghosting) show (a) a deghosted stack without SRME, (b) a deghosted stack with standard SRME, and (c) a deghosted stack with new SRME.

The receiver ghost normalization method could theoretically be extended to any demultiple technique producing a multiples model that could be adapted and subtracted to the input data. This method is currently being tested with other standard demultiple techniques such as SWD, convolutive interbed demultiple, and radon demultiple.

Acknowledgements

The authors would like to thank CGGVeritas for permission to publish this paper and Salvador Rodriguez and Robert Dowle for coordinating the BroadSeis test campaigns. This paper was originally presented at the 2012 European Association of Geoscientists and Engineers annual meeting and has been reprinted with permission from the authors.