Degel, Milos Cvetkovic, Sarah Spoors, Richard Clarke and. Laurie Geiger for reviewing this paper. The data shown are provided courtesy of Spectrum's ...
How Broadband Processing Can Improve Multiple Attenuation Processes for Conventional Flat Streamer Data Allan Willis*, Steve Ward, Sean Poche, Genmeng Chen and Mike Saunders, Spectrum Geo Inc. Summary Over the past ten years, there has been a significant amount of research and development related to marine broadband technology both to remove the ghost effect and to obtain a broader bandwidth of useable seismic energy. These developments are generally separated into acquisition and processing solutions. In this paper we address the processing challenges and solutions we encountered during multiple attenuation process(es) for conventional towed streamer data after broadband processing. We use a conventional flat streamer dataset from offshore Brazil to show the benefits broadband processing has on various demultiple processes.
certain processing challenges and the implementation of broadband processing can improve multiple attenuation results. Acquisition and Processing Overview The data used in this abstract is a deep-towed 2D dataset from the Sergipe Basin in Brazil. It was acquired with a flat streamer with nominal source and receivers depths of 10 and 15 meters respectively. This acquisition configuration resulted in ghost notches in the amplitude spectra at or near 0Hz (source and receiver ghosts), 50Hz (first order receiver ghost), 75Hz (first order source ghost) and their respective harmonics. In general, the sea state was relatively calm for the 2D line shown here.
Introduction A well-known issue in conventional marine flat streamer data is the interference from the free-surface ghost. These ghost reflections will create notches in the amplitude spectrum of the recorded data and will also distort the phase spectrum of the data which will have implications for subsequent processing steps such as multiple attenuation algorithms. Broadband data processing has been at the forefront of processing research and development for several years, in particular as it provides a solution for removing both source and receiver ghosts, consequently de-ghosting has become a key step in the broadband processing workflow. Since many papers have discussed various de-ghosting methods and processing challenges in broadband processing (Cheriyan et al., 2014; O’Driscoll et al., 2014; Telling et al., 2014; Zhou et al., 2012; and Lin et al., 2011), this paper will focus on the challenges de-ghosted data present before standard multiple attenuation processes. In conventional de-multiple methods like Surface Related Multiple Attenuation (SRME) (Verschuur et al., 1992), both source and receiver ghosts still exist in the input data, consequently the multiple modelling is significantly affected by the interference of both source and receiver ghosts. With the presence of these ghost interferences, SRME may not be adequate to suppress the actual surfacerelated multiples effectively. However, after removal of both source and receiver ghosts, the de-ghosted data should give a better prediction of the multiple model which can then be subtracted adaptively from the input de-ghosted data to produce a record with less multiple contamination. A field data example acquired with a conventional flat marine streamer is used to demonstrate how addressing
The current workflow consists of the following steps: (1) de-swell/de-noise, (2) source and receiver de-ghosting, (3) de-bubbling and zero-phasing, (4) SRME, (5) water velocity Radon de-multiple, and (6) high-resolution parabolic Radon de-multiple. Specific care was given to all pre-processing steps (noise attenuation in particular) so that the de-ghosting process would not be compromised by any noise or time-varying amplitude scaling. A review of the raw data showed a strong presence of low frequency noise (mainly swell noise). Various noise attenuation processes were applied in different domains, with multiple passes of noise modelling and subtraction, to effectively remove the strong swell/impulsive noise and any cable tug noise. De-ghosting and De-bubbling In general, ghosting is an interference effect. The combined effect of both source and receiver ghosts results in both constructive and destructive interference (Figure 1) (Bai et al., 2013), consequently both the amplitude attenuation and the phase distortion introduced by ghosting must be removed prior to any de-multiple process. The ghost response is influenced by many variables including source and receiver depths, water velocity and sea state. The ghost removal process used in our broadband processing is a non-linear inversion process that adapts the least-squares minimization method (based on either minimum energy or minimum absolute amplitude) to estimate both the source and receiver ghost times as well as the reflection coefficients at the air/water interface for both the source and receiver ghosts (Yilmaz et al., 2014). The process derives a recursive filter which is then applied to
Broadband Processing Can Improve De-Multiple
the pre-processed seismic data to create a ghost free output. Both source- and receiver-side de-ghosting were performed simultaneously. Since the process is recursive, it is very sensitive to any non-reflective noise in the seismic data. Consequently, any strong noise contamination (especially low frequency noise) can reduce the reliability of the deghosting filter and must be removed in the data preconditioning step. Since the de-ghosting algorithm requires both the source and receiver depths, it is critical that both the source and receiver delay times and the sea-surface reflection coefficient can be estimated accurately in order to calculate an inverse operator to remove the ghost reflections.
Figure 2: Near-trace wavelet and its corresponding spectral analysis before (a and b) and after (c and d) de-bubbling.
After de-ghosting, the amplitude spectrum of the full stack section shows improvement in recovering both the low frequencies and the notches that were weakened by the ghosts (Figure 1).
Figure 3: Amplitude spectral comparison from unfiltered brute stack sections before (a and b) and after (c and d) de-bubbling.
Multiple Attenuation Figure 1: Amplitude spectral comparison from unfiltered brute stack sections without (a and b) and with (c and d) source and receiver de-ghosting showing improvements in the interference of the ghosts.
A key issue was noticed with the de-ghosted data. Since the bubble is part of the low frequency component (for the acquisition configuration chosen, it is close to 8Hz), it will be strongly over-boosted after de-ghosting, and this is evident from the spectrum of the de-ghosted stack in Figure 1 (Brookes et al., 2014 and Taner et al., 2014). By applying SRME directly to the de-ghosted data, the multiplemodelling and prediction process in this step will be overly biased by the strong bubble energy.
Multiple attenuation is one of the most challenging steps in seismic data processing, and SRME plays a key initial step in removing multiple contamination, predominantly from the water bottom. Brookes et al. (2014) explain that there are source and receiver ghosts in the first order multiple but there is no ghost on the bounce off the water bottom. Subsequently, when we convolve two traces together to generate the predicted multiple we will have two sets of shot and receiver ghosts (Figure 4), and this extra set of ghosts in the predicted trace makes the adaptive subtraction process more difficult as the modelled multiple is different from the input data (Brookes et al., 2014; Sablon et al., 2012 and Sablon et al., 2011).
This issue can easily be handled by de-bubbling. A zerophase de-bubble filter was derived from the water-bottom of the near trace subsets of the de-ghosted data, and was applied to the de-ghosted data before SRME. A comparison of the wavelets and the brute stacks before and after debubbling is shown in Figures 2 and 3 respectively.
The main challenge with SRME is how to properly model the water-bottom multiples. The standard SRME method leaves a lot of residual multiples, and the low frequency multiples that were “shadowed” by the ghosts cannot be properly addressed. Thus, removing the free-surface ghosts before SRME will more accurately recover the amplitudes
Broadband Processing Can Improve De-Multiple
and correct any phase distortion caused by the ghost reflections, and we can create predicted traces that better match the water-bottom multiples in the data. Figure 5 shows the modeled multiples of a shot gather after deghosting with and without de-bubbling. As evident in Figure 5a the SRME model without de-bubbling looks “ringy” (sinusoidal character), making it difficult for the multiple-modeling process to distinguish what are real multiples.
model with a broad bandwidth wavelet so that the adaptive multiple subtraction can be performed more accurately while preserving primary signal. Figure 6 shows a shot gather before and after SRME with and without applying the broadband-bounded gain deghosting workflow. It is evident that the results from conventional SRME processing were not optimal as there are still some residual low-frequency water bottom multiples evident. Figure 7 compares stack sections showing de-multiple results between conventional and broadband processing. Conclusions
Figure 4: Ray paths of two traces selected to build the predicted multiple.
In this paper, we have shown that broadband processing can improve results of standard multiple attenuation processes such as SRME. We also presented some processing considerations which should be addressed before the application of SRME. Proper de-ghosting can correct both the amplitude and phase variations resulting from the combined source and receiver ghost effects. Recovering both low and notch frequency contents, and restoring a broad and balanced spectrum have significant advantages for subsequent de-multiple processes. Using virtually ghost-free data as the input to SRME allowed us to obtain a more accurate multiple model from the SRME prediction, subsequently making the de-multiple process more reliable and accurate. The data processing workflow used on a conventional deep-towed 2D line from the Sergipe Basin in Brazil with de-ghosting and followed by multiple attenuation processes clearly showed the improvements in frequency recovery and multiple removal. Acknowledgements The authors would like to thank Spectrum Geo Inc. for the permission to publish this paper. We are also very grateful to Mike Ball for his support of this work, and Tomislav Degel, Milos Cvetkovic, Sarah Spoors, Richard Clarke and Laurie Geiger for reviewing this paper. The data shown are provided courtesy of Spectrum’s Multi-Client Division.
Figure 5: SRME model of a shot gather without (a) and with (b) de-ghosting.
In conventional processing, the low frequency multiples were not properly addressed and remain in the data. With broadband processing and recovery of the low frequencies after de-ghosting, the SRME process creates a multiple
Broadband Processing Can Improve De-Multiple
Figure 6: Comparison of a shot gather showing result before (left) and after (right) SRME between conventional (top) and broadband (bottom) processing.
Figure 7: Stack section comparison after SRME with (a) conventional and (b) broadband processing workflow.