The deep seismic reflection line CROP-04 NVR (Mazzotti et al., 2000, 2007a), hereafter CROP-04, runs from the south-western sector of the Campania region (near the town of Agropoli) to the city of Barletta in the Puglia Region (Cippitelli, 2007; https://www.videpi.com/videpi/crop/crop.asp, last accessed 17 August 2022). In this study, we worked on the westernmost portion of this seismic line, the limit of which is located ∼ 6 km northeast of the city of Venosa (CDP 2620, Mazzotti et al., 2000, 2007a) in the Basilicata region (Figure 1). The acquisition of the CROP-04 profile was funded by C.N.R., AGIP (now known as Eni S.p.A.) in collaboration with ENEL, and it was acquired between 1989 and 1990. As in other Italian areas, during the 1980s–1990s, the Southern Apennines were the object of hydrocarbon exploration, and the oil industry collected new data and reprocessed available industrial seismic lines (Cippitelli, 2007). Therefore, the acquisition and analysis of these reflection profiles, collected at a depth range of ∼6,000 m for hydrocarbon exploration, aided in the discovery of important oil fields in Val d’Agri, which are currently still operational. On average, these seismic reflection profiles achieved a higher resolution than the deep CROP-04, since the latter targeted the crustal structure of the mountain range and the foreland (Cippitelli, 2007). This consideration is supported by examination of several of the main acquisition parameters of the CROP-04: this stack line, acquired using a combination of both dynamite and vibroseis sources, shows a fold of 120%, a CDP distance of 40 m, and a time window of 10 s TWT (Figure 2), but Patacca and Scandone (2001) report continuous and well-structured reflections visible until ∼9 s TWT.

The average amplitude spectrum shows a bandwidth ranging from 5 to 45 Hz, with a dominant frequency of ∼11 Hz (Table 1). Assuming an average velocity of 5,000 m/s for the subsurface, a value of ∼110 m is estimated for the vertical resolution. Based on the acquisition parameters, processing difficulties can be anticipated: one is the combination of zero and minimum phase sources (dynamite and vibroseis), each most probably differing in frequency content; spatial aliasing (Steeples and Miller, 1998) is also critical, meaning that spatial sampling might not be sufficient to image sub-vertical structures such as normal faults and/or fractures.

The availability of this deep profile, as well as all the other commercial data that have been released, has favored important improvements in the understanding of the geological frameworks and the evolution of the Southern Apennines region. The seismic interpretation and re-interpretation of such data (Mazzoli et al., 2000; 2001; Menardi Noguera and Rea, 2000; Patacca and Scandone, 2001; Scrocca et al., 2005), including full reprocessing of CROP-04 (Mazzotti et al., 2000; Scrocca et al., 2003; Stucchi et al., 2003; Mazzotti et al., 2007a), has contributed to the definition of a variety of contrasting geological models (.e.g.,Menardi Noguera and Rea, 2000; Butler et al., 2004; Shiner et al., 2004; Finetti, 2005; Cippitelli, 2007; Patacca and Scandone, 2007; Scrocca et al., 2007; Scrocca, 2010; Vezzani et al., 2010; Mazzoli et al., 2013; Savastano and Piana Agostinetti, 2019). Controversies in this domain are nearly always attributable to the relatively low S/N ratio in the data, which hampers the application of more sophisticated processing techniques, as reported by Mazzotti et al. (2000; 2007b), and thus a clear mapping of the fault architecture within the subsurface of the region is still not well defined. For this reason, any achievements in improving noise reduction and increasing the lateral resolution will help with re-interpretation of the data, especially if further information can be extracted from the seismic fabrics and their patterns.

Geophysical features in seismic reflection data can be considerably enhanced with the application of seismic attribute analysis, which can be used to extract qualitative and quantitative information from data and to emphasize the display of structural features and relationships. A seismic attribute is a descriptive and quantifiable parameter (e.g., time or dip) that can be calculated from a single seismic trace within 3D cubes (Taner et al., 1979; Barnes, 1996; 1999; Chen and Sidney, 1997; Taner, 2001; Chopra and Marfurt, 2007; Iacopini, 2011; Forte et al., 2016; Dewett et al., 2021; Iacopini and Butler, 2021). Seismic attributes can be computed over pre-stack or post-stack seismic data and, nowadays, over 3D seismic volumes. The latter guarantees more efficient applicability of the proposed analysis by allowing the identification of stratigraphic and tectonic structures along time slices or depth horizons (Chopra and Marfurt, 2005; Marfurt et al., 2011; Hale, 2013; Torvela et al., 2013; Barnes, 2016; Wu and Hale, 2016; Di and AlRegib, 2019). Currently, the extraction of seismic attributes represents a fundamental tool for the exploration industry, as it speeds up the seismic interpretation process, reduces interpretation bias (Alcalde et al., 2019), and improves quantitative results (Chopra and Marfurt, 2007; Marfurt, 2018). Nevertheless, attribute analysis of a 2D seismic transect has some limitations, even when the tools are applied proficiently (e.g., “apparent dip”; Ha et al., 2019).

In the 80s and 90s, the acquisition of sparse 2D seismic reflection data was the operative standard, and nowadays, many such datasets are released by industry for research purposes. Therefore, the re-evaluation of these data can contribute to energy (geothermal) and mineral exploration research, as well as work on characterization of the subsurface for geological storage and seismotectonic studies (Malehmir et al., 2016; Schmelzbach et al., 2016; Ercoli et al., 2020; Malehmir et al., 2021; Pertuz et al., 2022), together with additional novel high-resolution data acquired using modern technologies (Manning et al., 2019; Dieulangard et al., 2021; Strobbia et al., 2022). For this reason, it is worth attempting any available strategy to extract as much information as possible from legacy data, which are typically characterized by both advantages and disadvantages (Ercoli et al., 2020). Conventional seismic interpretation is based on analysis of the amplitudes, geometry, and lateral continuity or discontinuity of reflections; thus, any successful reduction of random noise and increase in signal quality might represent a major step forward for re-evaluation of legacy data. For example, seismic discontinuities interpreted as faults, possibly not visible or unclear in conventionally processed profiles, can be strongly enhanced with relatively quick and cheap data treatment strategies like attribute analysis, without the need to fully reprocess everything from the pre-stack raw data. However, seismic attributes can be extremely sensitive to incoherent noise (Marfurt and Alves, 2015; Ercoli et al., 2020 and references therein), so it is necessary to hamper or reduce the generation of seismic artifacts by suppressing random noise and improving the linear features using pre-conditioning techniques (e.g., Tingdahl and de Groot, 2003; Acuña-Uribe et al., 2021). The latter can be operated via conventional frequency filtering or via structural filters and statistical attributes (Barnes, 2016), based on the extraction of reflections dip and azimuth (Chopra and Marfurt, 2007; Qayyum and de Groot, 2012). After this step, the use of geometric or structural attributes based on the determination of reflector geometry, dip and trace coherence, similarity, variance, and curvature (Bahorich and Farmer, 1995; Gersztenkorn and Marfurt, 1999; Marfurt et al., 1999; Randen et al., 2000; Roberts, 2001; Al-Dossary and Marfurt, 2006; Marfurt, 2006; Chopra and Marfurt, 2008; Dewett and Henza, 2016) might favor the identification and enhancement of the amplitude and phase of seismic events (Acuña-Uribe et al., 2021 and references therein). It must also be noted that during the last few years, the detection of fault structures has been optimized using various techniques (Meldahl et al., 2001; Pedersen et al., 2002; Pepper and Bejarano, 2005; Vasudevan et al., 2005; Aqwari and Boe, 2011; Aqwari et al., 2012; Babangida et al., 2013; Hale, 2013; Di and Gao, 2017; Qi and Marfurt, 2018). On unconditioned legacy profiles, the main faults and related splays can be masked or complicated by several factors, including random noise, dispersion effects, and geological complexities due to intense deformation, which, in unmigrated data, might generate many diffractions.

The detection of faults in seismic data can be considerably improved and refined through use of pre-conditioning filters such as a dip-steered median filter (DSMF; Tingdahl, 1999; Tingdahl and de Groot, 2003), a dip-steered fault-enhancement filter like the structure-oriented filter (SOF; Fehmers and Höcker, 2003), edge-preserving smoothing (EPS; Luo et al., 2002; AlBinHassan et al., 2006), and structural smoothing (SS; Iske and Randen, 2005; D'Argenio et al., 1992). Some seismic attributes are typically sensitive to sharp variations in wave geometry and amplitude among traces or sectors of traces. The attribute response across faults is also quite sensitive to the dip of traces. The dip-steering technique allows seismic reflections to be followed by auto-tracking of the pre-calculated dip-field from a given starting position (Tingdahl and de Rooij, 2005), so that a sample-to-sample analysis clearly improves the lateral correlation of events across seismic traces and along tracks. Attributes such as (geological) dip, azimuth, and curvature are computed directly from steering data; auto-tracker tools, conventional amplitude, and similarity trackers, which can include dip-field, are used to constrain horizon-tracking and fill in any gaps. Steering cubes are typically calculated over 3D seismic volumes, and these require dip values in both the inline and the crossline directions at each seismic sample position. Dip-steering can also be calculated over 2D seismic profiles by storing the dip at every sample position only in the line direction (Ha et al., 2019). Conventionally, at least two steering outputs are generated. The “Detailed-Steering” (DS) form of output includes dips as calculated by the dip-computation algorithm (central-steering or full-steering, as in Figure 3; dGB Earth Sciences, 2021), and this is used to preserve details in the data (e.g., detection of fractures through computation of curvature attributes). The “Background Steering” (BS) output is a smoothed version of DS obtained through the application of a median filter. As BS is less noisy and includes dips relating to larger (regional) structural trends, it is suitable for dip-steered filtering operations. Dip-steering filters can dramatically improve the output of attributes such as similarity, coherence, texture, and volume statistics, and are also critical in neural network-based “probability” cubes (e.g., chimney cubes and fault cubes; dGB Earth Sciences, 2021).

To define an efficient pre-conditioning strategy for application of the seismic attribute analysis, we tested a workflow, including DSMF, on a synthetic seismic profile (Botter et al., 2014; Iacopini et al., 2016), to which we artificially added statistical noise to perturb the reflections resulting from the Vp velocity contrasts. The reflectors built in the model were dislocated by a set of simulated high-angle faults (70° dip angle), including a master fault on the left side and two progressively closer secondary splays (at a distance of ∼400 and 800 m); the three displaced reflectors showed a few hundred meters of vertical offset, which increased with depth and decreased across the splays (Figure 4). The synthetic profile was obtained through forward modeling of the seismic wavefield using finite-difference time-domain (FDTD) numerical simulations, with two main goals: 1) to verify the capacity to image and resolve subtle tectonic structures using a dominant frequency wavelet of 11 Hz, comparable to that of CROP-04, which is lower than that of wavelets typically observed in vintage commercial seismic lines (Ercoli et al., 2020); and 2) to test the efficiency of the proposed workflow on this synthetic image. The aim was to reveal subtle seismic features in order to improve data interpretability through attribute analysis of pre-conditioned data (i.e., with noise removed). The model in Figure 4 was designed using the modeling package of the Reflex-Win software tool (v. 9.1.3; Sandmaier, 2022). A model was designed based on a multilayer sequence, with thicknesses increasing with depth. The assigned media properties (Vp) generated impedance contrast, which also increased with depth (Figure 4A). The simulation was executed using a finite-difference acoustic-wave propagator and the exploding reflector source approximation. The main fault was designed to be sub-vertical (e.g., 70° dip angle), inspired by preliminary interpretations of the CROP-04 profile (e.g., Brozzetti, 2011). Similarly, Vp reference value ranges and parameters (Table 2) were derived from the literature and from borehole stratigraphic information across the area (e.g., Improta et al., 2000; Adigun and Ayolabi, 2013). The synthetic profile obtained, illustrated in Figure 4B, showed criss-crossing diffraction patterns across the faults and several displaced events of variable amplitude. The synthetic profile was migrated using a 2D diffraction (post-stack) algorithm based on the input 2D velocity model. The migration operator collapsed the hyperbolic diffractions and restored a reliable geometry of the events (Figure 4C). The synthetic stack profile was migrated and overlaid over the depth-to-time conversion of the velocity model (Figure 4D). Subsequently, 10% high-frequency noise was added to the data. This noise was characterized by an exponential distribution and by a dominant frequency centered at 30 Hz (Figure 4E). The image illustrates the fact that imaging of the faulted zones was hampered by the random noise, particularly in the case of the shallower section characterized by smaller fault throws.

The synthetic profile was exported to the SEGY format and imported into the OpendTect software package, with the aim of setting up and testing the proposed workflow based on data pre-conditioning and seismic attribute analysis. After working on the seismic lines using a combination of a dip-steered median filter and a convolve low-pass filter (CLP) to clean up most of the random noise, several attributes were tested on the synthetic profile and on the experimental CROP-04 profile. Attributes ranging from instantaneous ones, like the cosine of instantaneous phase (Taner et al., 1979), up to multi-trace attributes (e.g., similarity), were applied over CROP-04 and displayed using multi-attribute co-rendering (Chopra and Marfurt, 2005; 2011). Among the attributes tested, we selected those that were most effective in enhancing the lateral discontinuities, namely the following.

1) Cosine of instantaneous phase (CIP). This is a frequency-based instantaneous attribute, representing the sample-by-sample instantaneous variation in phase (derivative of the instantaneous amplitude or “envelope”), determined from complex traces (the imaginary part of the complex trace is computed by Hilbert transform; Taner et al., 1979). This is an amplitude-independent algorithm that emphasizes the spatial continuity or discontinuity of reflections. It is continually smoothed, and therefore, it is not plagued by the 180° discontinuity characterizing the instantaneous phase. Another benefit is that amplitude peaks and troughs maintain the same position, but both strong and weak events are displayed at equal strength, thus enhancing thin events and subtle seismic fabrics (which can, for example, correspond to subtle faults) (Forte et al., 2012; Ercoli et al., 2015).

The pre-conditioning procedure on the synthetic profile is based on a first generation of a dip profile and on the application of a median filter (MF). Then, a DSMF was applied on the noisy synthetic line to obtain an improved steered profile (SP) with a reduced amount of random noise. This output was again filtered using a CLP by further smoothing the random noise (filtered steered profile, FSP). We have finally computed the selected attributes (CIP, EN, PR, SM, and workflow summarized in Figure 5, whose parametrization can be found in Table 3), using both the unfiltered and pre-conditioned data for comparison (Figure 6).

A similar workflow based on a DSMF and a series of post-stack attributes was customized and applied, after an extensive phase of parametrization, to the vintage deep seismic profile CROP-04 (Figure 7, parameters in Table 3).

The pre-conditioning data treatment included a structural filter operated via a DSMF and a BG algorithm [1:1], used compute a raw-steering (RS) profile, filtered using an MF [5:9] to obtain a background-steering output (BS). The DSMF noise-reduced profile obtained was then used as the input data, following application of an additional CLP filter, for computation of the same seismic attributes as performed for the synthetic line, using a co-rendered display. A pure methodological comparison between the two results (based on conditioned and un-preconditioned outputs) is presented in Figure 8.

This procedure was customized to maximize the structural information derivable from the data through advanced seismic interpretation, which benefited from enhanced display of the subtle fabric pattern. These fabrics can be attributed to subtle faults. The results are discussed in detail in the next section.

The synthetic profile (Figure 4E) was imported into the OpendTect software package (displayed on a color scale as illustrated in Figure 6A), along with the DSMF and the CLF shown in Figure 6B, as inputs for computation of the PR attribute (Figures 6C, D). The result illustrates the benefits provided by the pre-conditioned profile, which produced a higher-quality image with a reduced number of artifacts. The pre-conditioned PR attribute also showed better resolution and an overall improved display of the thin, weak reflections located at shallow depth. This workflow effectively reduced random noise and improved the performance of seismic attribute computation to reveal main (deeper) and weak/subtle (shallower) lateral discontinuities, interpreted as faults. Figure 6C clearly shows the random noise present in the shallow sector hampering the detection of subtle fabrics/discontinuities/faults, while in Figure 6D, the profile is cleaner, meaning that the vertical displacements across the first three layers can be better interpreted in comparison to the unfiltered PR result.

The same operations were repeated for computation of the EN and CIP attributes, for which we report only the pre-conditioned results in Figure 6E (red “Energy” palette, co-rendered in transparency with the images in Figures 6B, D) and Figure 6F, respectively.

The EN envelope (Figure 6E) is useful to highlight the main (deeper) reflective contrasts as higher amplitude reflections, although it does not enhance the minor contrasts located at shallow depth. However, the attribute aids the visualization and detection of the main lateral amplitude discontinuities originating from the simulated faults. This output is particularly effective when co-rendered together with the PR attribute in the background (optionally, the conditioned seismic line FSP can be overlapped in slight transparency, as shown in Figure 6E). Finally, the CIP also enhances the detection of faults in comparison to the source data shown in Figure 6B.

The customized workflow applied to the CROP-04 NVR profile is reported and illustrated in extensive detail in Figures 7, 8, 9, and 10. The pre-conditioning steps produced several outputs, illustrated in Figure 7. Starting from the original, noisy CROP-04 profile (Figure 7A), we computed the median filtered profile, shown in Figure 7B, which was then used as input for the DSMF. This filtering operation removed a consistent amount of random noise, illustrated in Figure 7C (“net” profile). Following this process, the steered profile (SP) presented in Figure 7D clearly shows the benefits obtained after application of the DSMF, which reduced the noise and improved the lateral continuity of the events that could be associated with geologic structures. Figure 7E represents our final output (FSP) after an additional application of the convolve LP filter, which further smoothed the random noise, significantly improving the S/N ratio of the profile. Following this phase, we computed the above-described seismic attributes, over both the original CROP-04 profile and the novel noise-reduced version, as shown in Figure 8. Here, we report only one example for comparison (analogous to the synthetic data comparison of Figures 6C, D), showing the results obtained by computing the PR attribute over the entire seismic line (co-rendered with the FSP of Figure 7E). Figure 8A shows the unconditioned result, in which the visualization of the continuity of the reflections is greatly reduced by numerous random noise-generated artifacts, which particularly hamper the extraction of the signal (structural information). A pure methodological comparison with the cleaner profile of Figure 8B shows that the latter offers a much clearer display of the regional events along the CROP-04, and a significantly larger number of details can be locally extracted. An example is displayed in the magnified area on the easternmost side of the seismic line (without introducing, in this work, any geological considerations relating to the area), delimited by the blue and green boxes of Figures 8C, D (representing the unconditioned and pre-conditioned versions, respectively), where the difference between the two images is apparent.

In Figure 9, from top to bottom, the computed seismic attributes represent consistent improvement in the visualization of the continuity of the reflection and, at the same time, visualization of the discontinuity patterns. The PR attribute is shown in grayscale in Figure 9A. The co-rendered display in Figure 9B consists of PR (again in transparency with the FSP of Figure 7E) plus the EN attribute, which effectively enhances the main reflective areas and, at the same time, the lateral signal discontinuities. The attribute analysis also encompassed other two outputs, namely, the cosine phase (CP, in grayscale) and the similarity (SI) attributes presented in Figures 9C, D (again, both with the FSP of Figure 7E in transparency). These attributes represent another two alternative outputs supporting advanced seismic interpretation of the network of fractures by strongly enhancing the lateral discontinuities.

A sector of CROP-04, on which we focus our structural interpretation and discussion, is magnified in Figure 10 to show the considerable improvements obtained through the combination of pre-conditioning filters with seismic attributes. Figures 10A, B show the PR + EN attributes and the PR + FSP of Figure 7E, respectively. Figures 10C, D show the CP + FSP and the PR + FSP + SL + EN attributes, respectively. All these images strongly enhance the visualization and continuity of shallow and deeper reflections, such as W-dipping regional events (indicated by blue arrows) as well as the identification of several major and minor discontinuities disrupting them (indicated by red arrows from the surface). An interpretation of these geophysical features is presented in the following section, in light of the regional tectonic framework of the study area, with a focus on the extensional structures.

In order to illustrate the benefits of the proposed processing flow, an interpretation is presented here of the central sector of the CROP-04 transect within the distance range ∼34–76 km (corresponding to ∼ CDPs 620 - 2620, Figure 1B), for a total length of ∼42 km. This region is characterized by the presence of numerous major normal faults, which are critical in assessing the seismogenesis of the area (Brozzetti, 2011; Ascione et al., 2020 and references therein; Bello et al., 2021). The improved quality of the “revived” CROP-04 elucidates several major stratigraphic and structural features, as shown in Figure 11, which can be summarized as follows: 1) several high-amplitude reflection packages down to a depth of ∼3 s (TWT, Figure 11A), whose lateral continuity is interrupted by a series of small-scale sub-vertical discontinuities interpreted as minor sub-vertical faults (Figure 11B); 2) these bright reflection packages are separated by relatively “transparent” areas, characterized by low-amplitude reflectivity and localized, in particular, where the most relevant normal faults have been recently mapped at the surface with high accuracy (following Bello et al., 2021; indicated with black arrows in Figure 11B); and 3) several continuous or semi-continuous low-angle reflections represent subparallel SW-dipping discontinuities at different depths, among which the basal one is strongly enhanced (laterally continuous up to 10 km eastward of the outcropping Monticello fault, at ∼2 s TWT; Figure 10). Such features, despite not being the focus of this work, may represent, respectively, shallow local thrusts and a deeper regional thrust, the latter rising up to ∼22 km in Figure 11 (corresponding to CDP 1405 in Patacca and Scandone, 2007).

In the following material, the main geophysical features along the profile are described, starting from its western side (profile distance: ∼34 km) and moving toward its eastern end (profile distance: 76 km). In Figure 11, a package of high-amplitude reflections (between ∼34–40 km) are characterized by a general dip toward the SW and are clearly segmented by a series of NE-dipping minor faults/fractures that reach down to ∼3 s (∼9 km using an apparent Vav = 5 km/s). Further to the east, a series of high-amplitude reflections appear to be dislocated by a main NE-dipping structure (marked as AT, at surface) and faint antithetic events interpreted as fractures/faults (between ∼40–46 km). To the NE, the package of reflections is again interrupted by a NE-dipping feature, corresponding to the Inner Irpinia Fault (InF, at surface; Bello et al., 2021). Moving to the east, another area characterized by prominent, laterally discontinuous events and underlying semi-transparent units (corresponding to the Lagonegro basinal deposits and Apulia shallow water carbonates, respectively; Patacca and Scandone, 2007) is visible up to ∼72 km; this appears to be displaced by synthetic and antithetic normal faults (thick red and blue thick in Figure 11B), as well as by a series of subtle minor structures. At ∼51 km, the Irpinia Fault (IF) mapped at the surface (Improta et al, 2003a; b; Brozzetti, 2011; Ascione et al., 2020; Bello et al., 2021, 2022 and all references therein) has been traced to depth, together with parallel structures (e.g., the Paratiello fault - Pa) and some antithetic faults (e.g., the Monticello fault - Mo, Bello et al., 2021), and faint fault splays are particularly readily visible in the easternmost sector of the line (∼60 - 72 km, magnifications of Figures 11C, D). Here, a package of upward convex (antiform-shaped), very bright high-amplitude reflections with opposite dip (∼1 s TWT, Figure 11B) are densely disrupted by several faults, up to the shallower and transparent seismic facies, representing intra-mountain basins characterized by Quaternary fills. In this case, the application of pre-conditioning and attribute analysis was an asset in better visualizing the dense network of subtle, small-scale faults not visible in the original seismic profile. These tectonic structures clearly disrupt the very high-amplitude reflections, although the latter possibly tend to “saturate” the smaller offset associated with the minor faults. Such steep alignments dissect the reflections down to 3 s and are very closely spaced. The average distance between the minor faults is ∼600 m, which is a shorter distance than in the remainder of the profile. The area intersected by this dense network of minor faults extends to almost 8 km in width.

Below ∼3 s (TWT), the seismic image shows low-amplitude reflectivity fabric with limited lateral continuity. However, from ∼2.8 s to 5.5 s (TWT), in the westernmost side of the seismic profile, we identified a narrow package of gently SW-dipping sub-parallel events. The latter are characterized by the extent of their lateral continuity, with only a few main discontinuities, which seems to propagate to depth, and they are enhanced by its complex geometry, which is difficult to interpret. Multiple interpretations of this package of reflections (which, in the original profile (Figure 2), is visible mostly in the westernmost sector) can be found in the literature, such as a possible Moho discontinuity, a regional thrust, very deep deformed sedimentary layers incorporated into the thrust belt, near-bottom Apulian carbonates, and the top of the crystalline basement (Mazzotti et al., 2000; D’Argenio et al., 1973; Cippitelli, 2007; Patacca and Scandone, 2007; Scrocca et al., 2007; Scrocca, 2010).