Imaging via pre-stack depth migration (PSDM) of reflection towed-streamer multichannel seismic (MCS) data at the scale of the whole crust is inherently difficult. This is because the depth penetration of the seismic wavefield is controlled, firstly, by the acquisition design, such as streamer length and air-gun source configuration, and secondly by the complexity of the crustal structure. Indeed, the limited length of the streamer makes the estimation of velocities from deep targets challenging due to the velocity–depth ambiguity. This problem is even more pronounced when processing 2-D seismic data due to the lack of multi-azimuthal coverage. Therefore, in order to broaden our knowledge about the deep crust using seismic methods, we present the development of specific imaging workflows that integrate different seismic data. Here we propose the combination of velocity model building using (i) first-arrival tomography (FAT) and full-waveform inversion (FWI) of wide-angle, long-offset data collected by stationary ocean-bottom seismometers (OBSs) and (ii) PSDM of short-spread towed-streamer MCS data for reflectivity imaging, with the former velocity model as a background model. We present an application of such a workflow to seismic data collected by the Japan Agency for Marine-Earth Science and Technology (JAMSTEC) and the Institut Français de Recherche pour l'Exploitation de la Mer (IFREMER) in the eastern Nankai Trough (Tokai area) during the 2000–2001 Seize France Japan (SFJ) experiment. We show that the FWI model, although derived from OBS data, provides an acceptable background velocity field for the PSDM of the MCS data. From the initial PSDM, we refine the FWI background velocity model by minimizing the residual move-outs (RMOs) picked in the pre-stack-migrated volume through slope tomography (ST), from which we generate a better-focused migrated image. Such integration of different seismic datasets and leading-edge imaging techniques led to greatly improved imaging at different scales. That is, large to intermediate crustal units identified in the high-resolution FWI velocity model extensively complement the short-wavelength reflectivity inferred from the MCS data to better constrain the structural factors controlling the geodynamics of the Nankai Trough.
Seismic methods remain the primary source of information about the deep crust. The number of techniques related to seismic data acquisition and processing is continuously increasing, stimulated mainly by the hydrocarbon exploration community. However, along with the impressive technological expansion made by the oil and gas industry in terms of 3-D depth imaging at the reservoir scale, one observes apparent stagnation in the corresponding development of seismic technologies like full-waveform inversion (FWI)
The majority of regional seismic surveys conducted by the academic community still rely on sparse wide-angle 2-D profiles carried out with a 5 to 10 km receiver spacing. While a few attempts at FWI of crustal-scale ocean-bottom seismometer (OBS) datasets have been published
To mitigate this compromise, the development of up-to-date acquisition strategies and processing workflows dedicated to the integration of wide-angle and short-spread reflection seismic data is needed. Indeed, during wide-angle offshore seismic experiments, multichannel seismic (MCS) data can be collected with the aim of further processing reflection arrivals using migration techniques. To obtain high-resolution images of complex subsurface targets with strong lateral velocity variations, one may consider some variant of pre-stack depth migration (PSDM), e.g., Kirchhoff depth migration
Therefore, without doubt, MCS data provide sufficient information to image targets of interest for hydrocarbon exploration. However, the finite length of the streamer hampers the procedure of picking sufficient move-outs in deep reflections, hence preventing a reliable estimation of velocities for deep crustal imaging. Therefore, since the methods routinely applied for reservoir-scale imaging are ineffective when employed at the crustal scale, we shall investigate how to overcome the abovementioned limitations by optimally combining MCS acquisitions with sparse stationary-receiver OBS deployments.
Accordingly, this paper aims to illustrate an up-to-date workflow combining different types of seismic data with a case study devoted to the imaging of the eastern Nankai Trough subduction zone (Tokai area) offshore of Japan. Our approach combines Kirchhoff PSDM of the MCS data with the independent procedure for velocity model building. We start our seismic imaging with two velocity models derived from VA of the MCS data and FWI of the OBS data, respectively. The inaccuracy of the VA model with respect to crustal-scale PSDM relies on the lateral-homogeneity assumption of classical time-domain VA, as well as the limited depth penetration of the recorded wavefield. On the other hand, the velocities in the FWI model are derived from the different seismic data, which due to the wide-angle propagation regime may not have the same meaning as those found by the MCS data (which are more suitable for PSDM), in particular in the presence of anisotropy. Therefore, based on residual move-outs (RMOs) picked after initial PSDM inferred from the VA and FWI models, we update both models using slope tomography (ST). Consequently we perform PSDM using updated models and compare the final results with their initial counterparts. We show that, thanks to the long-distance propagation paths traveled by the wide-aperture wavefields, we are able to build a reliable velocity model of the whole crust using FAT and FWI. Taking into account the relatively small improvements introduced by ST in this model, we conclude that it provides a good approximation of the velocity field for PSDM of the MCS data. Compared to the workflow in which the velocity model is obtained directly from the MCS data (namely via VA and VA, followed by ST update), the approach employing FWI of the OBS data provides superior migrated sections and better flattening of the CIGs. Moreover, structural consistency between the PSDM section and the high-resolution FWI model further validates the geological reliability of the crustal images.
This paper is organized as follows. We start with a brief introduction of the study area and the seismic acquisition. After that, we describe our integrated processing workflow, followed by the presentation of the results. We compare the two velocity models inferred from the OBS (FAT+FWI processing) and the MCS data (standard VA), as well as the CIGs and the migrated sections built from these two velocity models. Consequently, we present an analogous comparison of the PSDM results derived from the VA and FWI models updated by the ST. This comparison highlights how the integrated approach combining MCS and OBS data leads to more accurate and better-resolved images of the crustal structure than those inferred from MCS data only. We further check the relevance of our imaging results by pointing out the correlations between the well-documented geological structures reported in the area and the geological features interpreted in both the migrated section and the FWI velocity model. We finalize the article with a discussion, followed by a summary of the study in the Conclusion.
The Nankai Trough, offshore of Japan, is one of the most complex subduction zone settings around the world. In this area the Philippine Sea Plate subducts below the Eurasian Plate towards the northwest, developing a large, sediment-dominated accretionary prism (Fig.
In order to gain new insight into the crustal structure of the eastern Nankai Trough, 2-D MCS and dense OBS datasets were acquired in 2000 and 2001 as part of the Seize France Japan (SFJ) project. The profiles were roughly perpendicular to the trench axis, resulting in significant variations in bathymetry between approximately 500 and 3750 m (Fig.
During the 2001 SFJ OBS leg, the wide-angle dataset was collected by 100 OBSs spaced 1 km apart and equipped with 4.5 Hz three-component geophones and hydrophones
A coincident towed-streamer profile was also acquired during the SFJ MCS survey
Examples of MCS and OBS gathers are shown in Fig.
Integration of migration- and tomography-like techniques has been a common practice in seismic imaging for decades
Velocity models derived from the OBS data using FAT
To summarize the motivation behind our seismic processing workflow for deep crustal imaging, the application of FWI to OBS data have done the following: (i) led to a reliable velocity field for PSDM; (ii) provided a high-resolution velocity model reducing the wavenumber gap; (iii) significantly contributed to the geological interpretation of the imaging results; and (iv) integrated wide-angle seismic data into the depth imaging workflow of reflection seismic data.
The aim of processing the SFJ OBS dataset was to obtain a crustal-scale P-wave velocity model using FWI such that it could provide insight into deep crustal targets with wavelength-scale resolution. The dataset was recently reprocessed by
The first step consisted of building an accurate starting model using FAT. In the case of regional-scale datasets, long offsets imply that a large number of wavelengths are propagated, during which kinematic errors accumulate, hence making FWI prone to cycle skipping
In the next step, acoustic Laplace–Fourier FWI
Comparison of the final FAT and FWI models in Fig.
Processing of the SFJ MCS data was focused on the best possible reflection imaging of the subduction system. The difficulty of the task is not only due to the complexity of the underlying structure but also due to the low ratio between the streamer length and the deepest targeted structure as well as two-dimensional acquisition setting. The processing sequence is summarized in Table
At the preprocessing stage, we mainly focus on the noise attenuation, wavelet corrections and multiple elimination. To improve the signal-to-noise ratio (SNR), filters designed to remove different kinds of coherent noise (e.g., swell noise, seismic interference noise, linear noise, double shots) were applied. The bubble effect was removed, followed by the zero phasing of the wavelet and reduction of the ghost on the source and receiver side.
Processing sequence of SFJ MCS data.
Next, we focused on the attenuation of multiples. The lack of a general demultiple approach imposes the need to tune the processing workflow according to the limitations caused by the geological setting and the acquisition parameters. In our case, the deeply underlying complex structure causes a lot of triplications and discontinuities of the wavefield, especially at later times. Also, due to the relatively short streamer, the ability to track the move-outs is limited. This in turn confines the robustness of move-out-discrimination-based methods
CIGs converted to time domain
In our processing scheme, we chose first to apply 2-D SRME, which is unlikely to affect the primaries. After SRME the vast majority of the free-surface multiples were attenuated. However, in certain areas of the accretionary wedge (e.g., between 55 and 80 km of the model distance in Fig.
In this section, we will examine the issue of building a velocity model from short-streamer MCS data. In the classical VA the move-out curve is (in a typical case) approximated by the hyperbolic relationship between zero-offset time and velocity. Stacking velocities are often picked at multiple locations along a line and are further interpolated between analysis points. This procedure is usually repeated several times with the simultaneous refining of the sampling interval leading to more detailed velocity estimation
Results of semblance-based velocity analysis performed on the gathers presented in Fig.
In Fig.
The results confirm that meaningful information from hyperbolic move-out can only be retrieved in the shallow part in which the semblance amplitudes are sufficiently well focused and allow the increasing velocity trend with time to be tracked. However, at intermediate and later times the VA panels become more chaotic and no distinct maximum semblance focusing points can be picked. This is of course a consequence of the increasing depth of the target to be imaged. Moreover, it is also worth mentioning that the final FWI model (Fig.
To further improve the velocity model obtained via VA, one may consider application of some variant of reflection travel time tomography utilizing information about the RMOs of reflections tracked in the CIGs after initial PSDM. Of course, the robustness of the tomographic approach relies on the precision and redundancy of the picked RMOs. In the simplest case, the velocity model is built through the minimization of the difference between travel times generated in the model with ray tracing and those measured from the data
Local coherent events picked for ST. Each local coherent event is described by the source and receiver positions (s, r), the slopes (
In the data domain, the input data (picks) are defined by their source-to-receiver two-way travel time and two slopes, namely the horizontal component of the slowness vectors at the shot and receiver positions (Fig.
One of the advantageous assumptions about the input data for ST is that the picked events need to be only locally coherent (see Fig.
From Fig.
PSDM sections inferred from the
It is worth mentioning here that in Fig.
To express in a more quantitative way the relative sampling of the model by the picks we calculate their hit rate inside square grid cells of size 0.5 km. Consequently, we clip 10 % of the highest values, normalize the results between 0 and 1, and divide them into three uniform intervals. Figure
To conclude we can make the statement that the robustness of the ST method is reflected by the distribution of the scattering points demonstrating high consistency with the dipping structures observed in the PSDM section. This in turn ensures regular and redundant information that might be exploited for velocity model building. However, the number of picks and their robustness (especially during 2-D processing) are significantly reduced with depth as the reflectivity of the structure decreases. This is more pronounced on the landward part (backstop) of the model. Therefore, the constraints on the velocity model at larger depths, although significantly better than those provided by classical VA, are weaker than in the shallow section.
During the PSDM step, we tested different background velocity models, targeting the optimal flattening of the CIGs. Here we consider only four of them. The first one was built from MCS data by classical VA, while the second one is the FWI model inferred from the OBS data (Fig.
In the following sections we present the PSDM results inferred from the four abovementioned background velocity models. We assess them by looking at the flatness of the CIGs as well as through the correlation of the migrated sections with the underlying velocity models. Furthermore, we compare the main observed structures against the results reported by previous geological studies conducted in this area.
The PSDM sections, superimposed on the VA- and FWI-related velocity models with which they were computed, are presented in Fig.
We validate resulting PSDM sections against the corresponding background velocity models. The structure imaged in PSDM sections inferred from VA velocity models is less well focused than in those inferred using the FWI models. This is a consequence of the fact that VA models lack resolution and accuracy due to the estimation method. Comparing panels (a) and (b) in Fig.
Analogous comparison of the FWI and FWI+ST velocity models and related PSDM sections (Fig.
CIGs extracted along the whole profile (every 8.75 km).
To further QC the imaging results in Fig.
The overall trend after ST is the upshifting of the reflectors, suggesting that the velocity refinement by ST mainly lowers the FWI velocities. This likely highlights the imprint of the anisotropy. The velocities estimated by FWI are close to the horizontal velocities, while those constrained by MCS data are closer to the velocities associated with normal move-out. In all cases, RMOs are more pronounced at the side edges of the profile. The effect is caused by the lack of coverage in these areas, which translates to weak constraint during velocity model building.
PSDM section inferred from the FWI+ST velocity model overlaid on the estimated depth shifts. Positive and negative values correspond to upshifts and downshifts introduced in the PSDM section by FWI+ST velocities relative to the FWI velocities, respectively. Note the nonsymmetric color scale.
To quantitatively express the depth changes in the PSDM results introduced by the ST update of the FWI model, we estimate the depth shifts between the PSDM sections inferred from FWI and FWI+ST models. We calculated the local cross-correlation for each pair of corresponding traces from the mentioned sections, which led to consistent map of depth changes between them. Figure
Imaging results and structural interpretation. Migrated section superimposed on the following:
Before interpretation, we filter out (using curvelet transform;
Imaging results and structural interpretation.
The oceanic crust in the region of Tokai is commonly described as being affected by volcanism
In Fig.
Between 40 and 55 km of model distance we clearly identify the negative velocity gradient creating a major LVZ. The top of this LVZ is consistent with the plate interface. In our case, access to the high-resolution velocity model directly indicates a negative velocity gradient, which could alternatively be obtained from the polarization reversal in the reflectivity image addressing, for example, the presence of fluids. However, at this depth we observe several ringing reflections (see arrows in Fig.
On top of the subducting oceanic crust, our seismic imaging results reveal the complex accretion system of the Tokai segment, which can be divided into four domains. The weakly deformed domain is located at the southeast of the profile starting at
The weakly deformed domain is separated form the accretion front – a moderately deformed domain – by the major frontal thrust fault, which generates ocean-floor topography at 85 km and reaches the décollement around 13 km further in the landward direction
The moderately deformed domain extends landward until the Tokai thrust, separating the frontal part of the prism from a heavily deformed domain
We also observe that the thickness of slope sediments in the shallow part of this segment (50–63 km; blue in Fig.
Finally, on the most landward part of the profile, we image the old deformation complex that now acts as a backstop. The deformation of large landward-dipping stacked thrust sheets can be identified by velocity changes correlating with reflectivity visible in the PSDM image. No active deformation can be clearly identified in this area (in contrast to the frontal area of the prism
The ability of the seismic wavefield to penetrate the deep crust is limited by various factors, including acquisition geometry, employed equipment and geological context. Sufficiently dense shot coverage increases the fold and therefore spatial sampling redundancy. The ability to generate and record broadband frequency signals improves resolution and the signal-to-noise ratio. Long streamers, although challenging to operate in the field, provide large enough offsets to track the move-outs of later arrivals originating from deep interfaces. All these factors improve the ability of MCS data to retrieve meaningful information about the deep crust. Our MCS data were recorded using a 4.5 km long streamer, which is quite limited compared to industrial 15 km long streamers. Nevertheless, we have shown that the robust depth imaging of legacy data with short offset and limited fold is possible when supported by velocity model building based on FWI of wide-angle stationary-receiver data. Despite the different regimes of wavefield propagation between wide-angle and reflection data, our FWI model provided an accurate velocity field for PSDM. This accuracy allowed for the picking of slopes for ST and introduced relatively small velocity changes even at greater depths (
CIGs extracted along the whole profile (same as in Fig.
Two-dimensional seismic data processing is inherently inaccurate because of the three-dimensional wavefield propagation within the lithosphere, which takes place during field experiments. Therefore, the assumption of inline scattering along the shooting profile is unrealistic. Nevertheless, despite the improved accuracy of a 3-D experiment, 2-D surveys remain a practical approach to retrieving geological information about the subsurface. This in turn capitalizes on limited prospects for subsequent processing (for example, the accuracy of retrieved seismic attributes) and the following quality of the results when cross-dipping structures are expected in the vicinity of the 2-D profile. While the imaging of deep crustal targets requires a long time for wavefield propagation, it means that the wavefront is also more prone to travel out of the 2-D profile plane. The complexity of the geological structures that we aimed to reconstruct here combined with their significant depth lead us to believe that the final imaging results are affected by “3-D effects”, reducing the continuity of the deeper interfaces. This was also apparent from the locally occurring arrivals that were stacking into migration artifacts. We decided to filter them out before final interpretation as we were unable to flatten them with any velocity model, even with velocity scaling ranging between 80 % and 120 %. To overcome such issues, one can consider cross-dip processing that takes into account the 3-D character of the data (for example, feathering of the streamer) and the geology
The typical problem from the Tokai region, which can justify the use of 3-D rather than 2-D crustal-scale imaging, is the existence of the basement volcanic ridges and mountains subducting below the accretionary wedge and affecting the geodynamical setting of the region. While their existence is undeniable (even from the bathymetric observations in Fig.
FWI in its classic form is a local optimization problem. In other words, the misfit functional representing the difference between real and synthetic data is iteratively minimized (towards a global minimum), starting from the initial model that is sufficiently close to the true one, such that it fulfills the cycle skipping criterion. If this condition is not met, FWI will be guided towards a local minimum and an inaccurate model. Therefore, the accuracy of the initial FWI model determines the correctness of the final results. In this sense, the robustness of the initial FWI model (in our case derived by FAT) can also be verified through PSDM. Indeed, if the velocity values in the FWI model are searched for in the vicinity of the FAT model, then PSDM utilizing the FAT model should also provide good results. To investigate this issue we performed two additional PSDM tests considering the FAT model and its version updated by ST.
Figure
The presented case study shows that crustal-scale FWI of OBS data can significantly improve depth-migrated images inferred from towed-streamer data. Firstly, the final FWI model provides reliable information about the velocity not only at shallow depths (
The data used in this research are not publicly available. The OBS and MCS data were provided based on the collaboration between Institute of Geophysics, Polish Academy of Sciences and participants of SFJ project.
AG initialized and led the research at all stages, derived FAT and FWI models from the OBS data, performed PSDM of the MCS data, conducted majority of the quality control of the results, derived initial geological interpretation, generated the figures, wrote the initial manuscript and coordinated its revision. SO co-supervised the project and participated in the design of the processing workflow and quality control of the results as well as reviewing the manuscript. LS performed the preliminary processing of SFJ-MCS data and provided the initial velocity model derived form the classical velocity analysis. YY participated in the geological interpretation of the results and participated in writing and reviewing the corresponding part of the manuscript.
The authors declare that they have no conflict of interest.
This article is part of the special issue “Advances in seismic imaging across the scales”. It is not associated with a conference.
This study was partially funded by the SEISCOPE Consortium (
This research has been supported by the Institute of Geophysics, Polish Academy of Sciences (internal grant no. 1A/IGF PAN/2016).
This paper was edited by Charlotte Krawczyk and reviewed by Andrew Calvert and Anke Dannowski.