Active-source seismic surveys are the main geophysical method for the exploration of subsurface geological structures and reservoir characteristics. Current seismic surveys are often designed to acquire the reflection waves, which are the seismic energy bounced back at the interfaces between rock layers. Since oil/gas reservoirs are mainly found in sedimentary rocks dominated by the flat layer feature, the seismic surveys set the maximum offset to 1–1.5 times the depth of targets (Yilmaz, 2001). As a result, the surveys are adequate for recording reflection data.

The conventional seismic data processes aim at utilizing reflections, especially primary reflections. Other types of events, including direct arrivals and refractions, are treated as noise; therefore, they are suppressed during data processing (Yilmaz, 2001). The main steps of the data processing include velocity model building and migration. The data processing can be performed either in the time domain or the depth domain. The time-domain data process is still popular because of its high efficiency and robustness. However, it suffers from low accuracy. Its migration profiles only show the travel time from the surface to the interfaces. Thus, the migration profiles do not provide the depth of the interfaces. In addition, time-domain migration assumes that the rocks are homogenous above an interface in the aperture of one common-middle-point (CMP) gather. Consequently, time-domain migration could produce artificial geological structures, especially in the region where the velocity possesses strong variation. Thus, time-domain migration results are inappropriate for tasks requiring high accuracy imaging, for example, bore well-drilling design.

By contrast, the depth-domain migration provides the image of geological structures matching the real world (Zhang and Sun, 2008; Liu et al., 2011; Li et al., 2021). Its migration profiles, therefore, are ideal for subsequent processes and interpretation. However, depth migration needs an accurate depth velocity. The migration velocity does not need to contain fine details but must provide an accurate background velocity, which gives the correct travel time and wave paths of reflections. Therefore, building the accurate migration velocity is crucial for depth migration. Limited by the offset, reflections are the only information source for building velocity reaching the depth of targets. Reflection-waveform inversion (RWI) (Xu et al., 2012; Wu and Alkhalifah, 2015; Zhou et al., 2015; Yao and Wu, 2017; Yao et al., 2020) and reflection-traveltime tomography (Sherwood et al., 1986; Kosloff et al., 1996) can be used to achieve this goal. Currently, reflection-traveltime tomography is the mainstream tool for velocity building in the industry. The principle is to utilize the move out residual to update the velocity above the selected reflection interfaces.

Two factors hamper the success of velocity building based on reflections: first, the trade-off between the depth of reflectors and velocity above reflectors. This phenomenon is illustrated in Figure 1: if a reflector is moved up/down by reducing/increasing the velocity above the reflector, the travel time keeps the same as that for the true depth and velocity. One symptom in practical applications is that the inverted velocity has strong fluctuation along the strata, which leads to artificial undulated reflectors in the migration image. This undulation becomes more severe in deep parts of the model than its shallow parts due to fewer data constraints as depth increases, for example, Figure 7A of Yao et al. (2019). The fundamental reason for this phenomenon is that inadequate information is used in the inversion. In other words, the inversion system is under-determined. To mitigate this issue, it is necessary to incorporate more information into the system. For instance, structural smoothing is a common constraint used for this purpose (Lewis et al., 2014; Yao et al., 2019).

The second factor is the low S/N ratio. This is common in land seismic surveys. Figure 2 shows one typical shot gather with a low S/N ratio. The reflection events are barely seen in the profile. The low S/N ratio is caused by the near-surface complexity and complex structures. Consequently, it is almost impossible to build a correct velocity model using reflections in such a scenario.

These two factors lead to a paradox: time migration gives even better migration images than depth migration in this type of scenario, but the migration theory tells that depth migration is more accurate than time migration. One reason why time migration works better here is that it can build an optimal velocity by scanning the semblance of stacking velocity, resulting in plausible migration images.

The analysis above indicates that building an accurate velocity model is the key to reliable depth migration. To achieve this goal, the fundamental solution is to acquire adequate and reliable data. The company BP set such a pioneering example for full-waveform inversion (FWI). BP created a marine vibrator, called WolfSpar, which can generate reliable signals as low as about 1.6 Hz at an offset of 30 km (Dellinger et al., 2016). By combining with ocean bottom nodes (OBN), the acquired dataset successfully delivered much more accurate velocity models with FWI than with reflection travel-time tomography, resulting in stunning migration images (Shen et al., 2018; Zhang et al., 2018).

In this article, we propose an improved acquisition system for building more reliable velocity models for deep subsurface exploration. The basic idea is to add extra spare sources or receivers along the extension line of the normal geometry. As a result, this acquisition system includes ultra-large offsets, which are the foundation to acquire diving waves from the deep subsurface. The diving waves are refraction waves that bend back to the surface due to the velocity gradient of the Earth (Sheriff, 2002). Their waveform and first-arrival time are particularly useful for recovering deeper velocity, which has paramount significance for the exploration of deep and ultra-deep hydrocarbon reservoirs.

The rest of this article will be organized as follows: the improved acquisition geometry will be introduced first; examples then will be demonstrated; and finally, the discussion and conclusion will be given.

Currently, reflection waves are the main objective for active seismic surveys. Thus, the maximum offset is usually about 1–1.5 times the depth of targets. Due to the two factors, which are elaborated in the section of introduction, it is still difficult for reflection-based travel time tomography to build an acceptable velocity model for migration imaging. By contrast, the diving waves, which are the transmitted waves turning back to the surface due to the velocity increase with depth, are the more robust signals for velocity building than the reflection waves. The reasons are two folds: first, it does not suffer from the trade-off between the depth of reflectors and velocity above reflectors because changing the depth of reflectors does not alter the travel time of diving waves significantly; second, the S/N ratio of diving waves is higher than reflection waves. There are also two other reasons. Diving waves are concentrated on the early arrivals in each trace so that they are not interfered with ground roll noise, which is strong and usually masks reflection signals (Yilmaz, 2001). In addition, reflection coefficients are much smaller than transmission coefficients (Berkhout, 1980). Consequently, the diving waves are robust for velocity building.

There are two ways to utilize the diving waves for velocity building. First-arrival travel time tomography is a robust way to invert the first arrivals of diving waves. The rest of the diving waves is the result of interference of multiple events and multi-path events. Due to the interference, it is impossible to distinguish the travel time of each event. Consequently, travel time tomography cannot be used for these data. One solution for the interfered diving waves is full-waveform inversion (FWI) (Tarantola, 1984; Virieux and Operto, 2009; Warner et al., 2013). In this article, we use travel time tomography (Zhou, 2003) to verify the concept of the improved acquisition geometry, so that the travel time of first arrivals is the information for velocity building.

In this article, we only analyzed the modified geometries for a 2D seismic survey. However, it is straightforward to extend the two types of modified geometries to a 3D seismic survey. Extra sources and extra receivers were deployed around the 2D receiver arrays. For Type-I geometry, extra sources were deployed not only along inlines but also in crosslines; consequently, the number of extra sources increases significantly compared to that for the 2D seismic survey. According to the analysis before, Type-I geometry is equivalent to the densest receiver interval in Type-II geometry. Thus, two means can mitigate this issue: the first one is to reduce the extra source number; the second one is to use a cost-effective seismic source, for example seismic vibrators. For Type-II geometry, the most cost-effective way is to use nodal seismometers for recording the large-offset diving waves. As demonstrated by the field data example, an interval of 3 km still delivered a good velocity model. Thus, one nodal seismometer can cover a 9 km² area. The modified geometries, therefore, are practical for a large 3D seismic survey.

Building accurate seismic velocity models is the key for seismic migration imaging. Current seismic surveys are mainly designed for acquiring reflections. Due to limited offsets, the recorded diving waves in a normal geometry have limited penetration depth. Consequently, the velocity model building relies on reflections. The low S/N ratio and incomplete information of reflection data prevent successful velocity model building. Building velocity models with diving waves can tackle this problem. We, therefore, propose two types of modified geometries to record the diving waves at a large offset, which can penetrate to a large depth. Type-I geometry adds extra sources along the extension line of the receiver array, whereas Type-II geometry deploys extra receivers along the extension line. Analysis shows that the two geometries have different characteristics. Seismic vibrators are a cost-effective choice for Type-I geometry, whereas nodal seismometers are a convenient choice for Type-II geometry. We also provide the workflow for designing the geometries. Both synthetic and field data examples demonstrate that the proposed improved acquisition geometries can record the diving waves from the deep subsurface. Their first arrivals are used to successfully build velocity models that are enough to imaging deep targets. By contrast, the diving wave from conventional geometry only recovers the shallow velocity.