The Mediterranean Salt Giant (MSG) is a vast and thick evaporitic unit that was deposited during the Messinian Salinity Crisis (MSC), an extreme environmental episode that occurred from ~ 5.97 to ~ 5.33 Ma (Hsü et al., 1973; Krijgsman et al., 1999; CIESM, 2008; Manzi et al., 2013). Since its discovery in 1970, and despite intense multi-disciplinary research, the MSG and the related MSC are still poorly understood and subject to many unresolved controversies (Camerlenghi and Aloisi, 2019). This is in large part due to a lack of data in the deep-water offshore domain, which represents more than 80% of the volume of MSC-related sediments, emphasizing the need for new and improved offshore data to better constrain the MSC (Roveri et al., 2014). Multi-channel seismic reflection profiling is the most common geophysical method applied to image the architecture of offshore basins and to prospect for potential drilling sites. Acquiring new marine seismic data, however, is challenging due to high acquisition costs associated with marine operations. Numerous offshore seismic datasets have been acquired in the last decades in the Mediterranean Sea, providing countless 2-D profiles that could no longer be acquired today because they cover areas that are currently subject to restrictions related to obtaining exploration permits (Diviacco et al., 2015). Many of these datasets are currently poorly exploited due to lack of public data access or poor-quality seismic processing. Reprocessing legacy data, therefore, is a potential source of new geological information that could be extracted from these dormant datasets, directly contributing to a better understanding of the MSG and the MSC.
The SALTFLU (‘Salt deformation and sub-salt fluid circulation’) 2-D multi-channel seismic reflection dataset was acquired in June-July 2012 by the R/V OGS-Explora, Eurofleets Cruise No. E12 (acquisition parameters listed in Table 1). The survey was planned to study the influence of the MSG on pore fluid circulation during basin evolution since the post-Messinian. The legacy SALTFLU processing followed a ‘narrowband’ approach, without deghosting, using narrow bandpass filters coupled with a source designature based on statistical deconvolution and no zero-phasing of the target wavelet. The filtering eliminated much of the low frequency signal (below around 6 Hz), whilst the source designature boosted the high frequency noise and produced a mixed-phase wavelet that has inconsistent phase across the survey. The wavelet also contains strong residual energy (likely from the bubble pulse) that overprints and obscures the primary signal, particularly the shallow geology close to the water bottom (Jovanovich et al., 1983; Sheriff and Geldart, 1995; Yilmaz, 2001).
Common goals of reprocessing are to improve the data bandwidth, spatial resolution, signal-to-noise ratio, reflector continuity and, where relevant, seismic-borehole correlations (Sadhu et al., 2008; Lille et al., 2017). In this study we aim to better image the salt structures, particularly the base salt, previously interpreted from the legacy processing as a flat surface lying around 2.7 km below the seafloor, at a water depth of 3.2 km (Dal Cin et al., 2016). Due to the complex overburden geology and the short far-offset (3.1 km) with respect to the target depth, the main imaging challenges include accurately resolving velocity variations, eliminating multiples and improving the signal-to-noise ratio at depth. We confront these challenges by outlining three key stages that should be systematically included in modern processing flows for similar marine seismic datasets: i) ‘broadband’ processing, ii) multi-domain denoising and demultiple, and iii) geologically guided velocity model building using iterative pre-stack migration and travel-time tomography.
Compared to traditional ‘narrowband’ seismic processing, ‘broadband’ processing aims to improve the spectral accuracy by expanding the data bandwidth and restoring frequency content attenuated by the source- and receiver-side ‘ghost’ effect and by seismic absorption (Masoomzadeh et al., 2013; Lille et al., 2017). High frequencies are most strongly affected by seismic absorption, which preferentially attenuates the highest frequency parts of the spectrum (Futterman, 1962; Sams et al., 1997). Recovering this information improves the resolution and results in more accurate true amplitudes, improving the performance of other data processing steps such as velocity model estimation and migration, and quantitative interpretation such as amplitude variation with offset (AVO) analysis, impedance inversion, and attribute analysis (Chopra and Marfurt, 2007; Mavko et al., 2009; Amundsen and Zhou, 2013; Lille et al., 2017). Conversely, lack of low frequency signal commonly results in poor focusing in the deep part of the section, as low frequencies suffer less from scattering and absorption, so they penetrate deeper and display a better trace-to-trace moveout coherence, allowing us to build a more accurate velocity model (ten Kroode et al., 2013).
Table 1
Survey parameters for seismic survey Salt deformation and sub-salt Fluid circulation (SALTFLU) 2012
SALTFLU 2012
|
Source
|
2 x 4 GI guns 2940 cubic inch air-gun array (2000psi)
|
Shot Interval
|
25 m
|
Source Depth
|
3 m ± 0.5 m
|
Streamer Length
|
3000 m
|
Near offset
|
100 m
|
Number of Channels
|
240
|
Group Interval
|
12.5 m
|
Streamer Type
|
Sercel Sentinel 428
|
Streamer Depth
|
4 m ± 0.5 m (20 5010/5011 birds for depth control)
|
Record Length
|
8000 ms
|
Sample Interval
|
2 ms
|
Field filters
|
3 Hz – 200 Hz
|
Nominal CMP interval
|
6.25 m
|
Nominal Fold
|
60 fold
|
Coordinate Reference System
|
WGS84 UTM Zone 31N
|
Acquired By
|
OGS and Fugro
|
Vessel
|
OGS Explora
|
Acquisition Date
|
2012
|
In marine seismic data acquired with an airgun source, the source signature is a combination of a relatively broad impulsive signal (approximately a minimum-phase wavelet), periodic oscillations caused by the so-called ‘bubble pulse’ and an inverted polarity ‘ghost’ multiple, caused by the time-delayed reflection of the signal from the sea surface (Ziolkowski, 1970; Sheriff and Geldart, 1995; Hegna and Parkes, 2011; Watson et al., 2019). GI-guns are designed to increase the primary-to-bubble ratio by injecting air inside the bubble generated by the primary pulse so that it collapses quickly (Pascouet, 1989). Here, the GI-guns were operated in harmonic mode, where generator and injector volumes are equal, thereby reducing the bubble oscillation, without completely suppressing it (Landrø, 1992; Dondurur, 2018). The data bandwidth is principally widened by performing a source designature, whereby the primary airgun impulse and the bubble pulse are collapsed into a sharp, zero-phase wavelet (Sheriff and Geldart, 1995; Amundsen and Zhou, 2013; ten Kroode et al., 2013; Baldock et al., 2013; O’Driscoll et al., 2013). Deterministic deconvolution, where the operator is designed using an estimated source wavelet, can yield geological imaging superior to traditional statistical deconvolution methods, particularly for recovery of low frequencies and the preservation of amplitude information (Yilmaz, 2001; Sargent et al., 2011; Scholtz et al., 2015; Davison and Poole, 2015). Deghosting, instead, aims to deconvolve both the source- and receiver-side ghosts from the wavefield, further sharpening the wavelet and removing the frequency ‘notches’ associated with the ghost effect (e.g., Sargent et al., 2011; Chuan et al., 2014; Davison and Poole, 2015; Tyagi et al., 2016; Willis et al., 2018).
The quality of the bandwidth enhancement depends on the signal-to-noise ratio of the input data (Amundsen and Zhou, 2013). It is therefore essential to attenuate as much as possible the low frequency noise (e.g., reverberation from the direct arrival, ‘swell’ noise caused by pressure fluctuation near the sea-surface) beforehand, because any remaining low frequency noise not correlated with the source pulse may be artificially boosted by the source deconvolution and deghosting filters (Yilmaz and Baysal, 2015). Thanks to lower computational costs in recent decades, multi-channel filtering and analysis in transform domains has become routine for noise reduction (Schultz, 1985). For example, ‘swell’ noise can often be better attenuated by predictive filters in the frequency-space (F-X) domain than by a simple time domain low-cut filter that results in loss of the low frequency signal along with the attenuated noise (Liu and Goulty, 1999; Schonewille et al., 2008). Multiple attenuation is also better tackled by move-out discrimination techniques in the parabolic Radon domain rather than by traditional statistical deconvolution based methods, for example (Basak et al., 2012; Verschuur, 2013). A ‘broadband’ processing flow combined with an efficient multi-domain noise separation can improve the signal-to-noise ratio in the deep part of the section (in our case below the MSG), allowing considerable improvements in velocity model building (Chuan et al., 2014).
Seismic imaging restores the correct geometry of seismic reflectors and requires an accurate velocity model of the subsurface (Jones and Davison, 2014; Jones, 2015). The legacy SALTFLU data were imaged using a Kirchhoff pre-stack time migration. Time domain migrations are relatively robust to errors in the velocity model but are only well-suited to imaging geology containing weak lateral velocity variation (i.e., approximately ‘layer cake’ geology), as they do not properly account for ray path refraction. This can lead to degradation in image quality in time domain images of complex geology such as salt diapirs. Depth domain migrations, instead, can more accurately reproduce the ray paths of reflections in the subsurface, but the image quality is more sensitive to velocity errors (Sheriff and Geldart, 1995; Yilmaz, 2001; Jones and Davison, 2014). Migration velocities are generally estimated based on the flatness of reflections on common midpoint gathers after migration (Tsvankin and Thomsen, 1994; Jones, 2015). In depth domain, the process of picking reflectors is often automated and used as input to travel-time tomography. Iterative rounds of analysis of the residual curvature of reflectors on depth migrated gathers followed by travel-time tomography to calculate model updates (Jones, 2015).
In this study, we aim to showcase a ‘broadband’ reprocessing strategy designed to improve imaging of the MSG for the SALTFLU dataset. We demonstrate multi-domain de-noising, deghosting and a source designature using a seabed reflection derived operator. We then perform multiple attenuation and geologically driven iterative migration velocity analysis. Our results include pre-stack time and depth migrated images, which we compare to the legacy ‘narrowband’ processing. These reprocessed images highlight some new geological insights that these new results provide on the salt system of the Algerian basin, the seismic expression of the MSC and the basement structure of the study area.