Two east–west-trending, north-dipping faults that form the relay ramp were interpreted (Fig. 1b, d). These two faults are termed the western and eastern faults (Fig. 1b and d, blue and red faults, respectively). Two faulted seismic reflectors (top Fuglen and Fruholmen formations; Fig. 1c–d) were also interpreted. These reflectors were chosen because the top Fuglen is a very strong, easily interpreted reflector, while the top Fruholmen is poorly imaged and is more challenging to interpret. Both the top Fuglen and top Fruholmen are peaks (increases in acoustic impedance). The Stø Formation, which falls between the Fuglen and Fruholmen tops, is a prolific petroleum reservoir. Five different seismic interpretation methods (Exps. 1–5) were used with the aim of systematically studying how seismic interpretation techniques (Fig. 2a) influence the fault analysis workflow (Fig. 2b). The first three experiments are manual 2D auto-tracking horizon interpretation techniques with different IL and XL spacing (from every 8 to 32 lines), while the fourth and fifth experiments are a combination of automated (3D auto-tracked horizons) and manual fault interpretations. In all experiments, the faults were interpreted first, followed by the horizons. 2D and 3D auto-tracking of all horizons used a seed confidence of 30 % and a basic 3 × 3 seed expansion value, which pushed the interpretation to the nearest eight seed points of the interpreted seed point on the peak. In 2D auto-tracking, seed expansion only occurs in 2D on the IL or XL being interpreted, while in 3D auto-tracking, the seed points extend in both the x and y directions from the interpreted seed point to the eight nearest seed locations. In both 2D and 3D, if a fault was encountered, the interpretation stopped and needed to be guided to the correct horizon on the other side of the fault. In all experiments, faults were interpreted as simple planar features in the seismic data. Although faults are complex 3D bodies in the subsurface, due to the seismic resolution of the data (i.e. Wood et al., 2015), this detail was not captured and has therefore not been considered further.

Fault length (Fig. 4a) is defined as the maximum horizontal distance of a fault in three dimensions (Peacock et al., 2016; Walsh and Watterson, 1988). An analysis of fault length was conducted on the western and eastern faults (Fig. 1b, d) using the gridded fault surfaces. These data were extracted from the edge of the study area to the fault tipline for both faults. The data were graphically compared to understand the impact of the interpretation method on fault length.

To analyse fault morphology, the horizon surfaces described in Sect. 3.1.6 were used. In creating the surfaces, all horizon interpretations that fall within the fault polygons were removed, leaving behind a gap in the surface where the faults' extent and morphology through that horizon are clear. These fault polygons were generated using patch and trim distances; this is explained in detail in Sect. 3.3. The analysis of morphology considers these voids in the horizon surfaces. The graphical representations of fault throw (Sect. 3.3) can also be used to understand fault length.

Fault throw is defined as the vertical component of dip separation on a fault (Fig. 4a). Fault throw along the length of an isolated fault typically follows a trend whereby the highest throw occurs in the centre of the fault and progressively decreases towards the tiplines (Barnett et al., 1987; Walsh and Watterson, 1990; Fig. 4a, inset). In this study, a separate fault throw analysis was created for each of the five experiments. To calculate throw, hanging wall and footwall cutoff lines were produced for the top Kolje, top Fuglen, and top Fruholmen in each experiment using patch and trim distances on both faults of 150 and 75 m, respectively (Fig. 4b). These deal with the poor seismic image close to the fault: horizon data within the trim distance are rejected, while those within the patch distance are used to extrapolate the horizon onto the fault (e.g. Elliott et al., 2012; Wilson et al., 2009, 2013). The top Kolje (Fig. 1c, d) was used only to help in any lithological projections in the sections to follow. This younger horizon is only partially folded at the western margin of the western fault, so it is not discussed further with respect to deformation. The cutoff lines and their dip separation were then used to calculate the throw across the fault surface (Fig. 4a, bottom left inset). The results were displayed directly on the fault plane, and they were also graphed to understand how fault throw changes across each of the experiments.

The dip separation gradient and the longitudinal and shear strains are useful tools for QC seismic interpretations (Freeman et al., 2010). The dip separation gradient was calculated using the top Kolje, top Fuglen, and top Fruholmen cutoff lines. The longitudinal strain (also known as the vertical gradient) is the dip separation gradient in the direction of fault dip, while shear strain (horizontal gradient) is the dip separation gradient along the strike of the fault (Freeman et al., 2010; Walsh and Watterson, 1989). In this study, we use the principles introduced in Freeman et al. (2010) to analyse these measurements. This can help us to understand how the different seismic interpretations produce results that differ from what is considered geologically realistic and to compare how the different methods affect the value of these properties.

Juxtaposed lithology (a.k.a. an Allan diagram) is a representation of the hanging wall and footwall lithologies and their juxtaposition on the fault plane (Allan, 1989; Knipe, 1997). To calculate juxtaposed lithology (JL), horizons, faults, and a well (NO 7120/6-1, Fig. 1b, d) containing lithological information were used. JL was calculated using the resulting horizon and fault surfaces from the five experiments. The key lithological units were defined in the well using a combination of logs, core photographs, information from Norwegian Petroleum Directorate (NPD) fact pages, and post-well reports. Sonic and density logs were used to generate a well synthetic seismogram, which was tied to the seismology. Using the same hanging wall and footwall cutoff lines as in the fault throw analysis and the interpreted horizons as guiding surfaces, the well lithologies were projected onto the faults and used to generate a JL (Allan) diagram.

The geological modelling and volume calculations were conducted on the least and most densely interpreted experiments (Exps. 1 and 4). This analysis was completed using a combination of structural and property modelling workflows in Petrel™, and the 5 × 5 km study area was considered to represent the limits of the hydrocarbon field. Firstly, fault and horizon surfaces from Sect. 3.1 were used to create a structural model for each experiment (Fig. 5a). A 3D corner-point grid was generated, and the cells were then populated between the top Fuglen and top Fruholmen horizons using a grid cell size of 12.5 × 12.5 × 1 m (i, j, k direction), matching the resolution of the original horizon surfaces (Fig. 5b). These two horizons define the main reservoir interval (Fig. 5e; Linjordet and Olsen, 1992; Ostanin et al., 2012). In the depth (k) direction, the cells were divided using the proportional method with an approximate thickness of 1 m (∼ 250 cells in total between the Fuglen and Fruholmen top surfaces). The grid follows the shape of the interpreted horizons precisely and the grid pillars align with the fault dip, making an accurate geological representation (Fig. 5b). The faults were included into the grid as zigzag faults, meaning they were not precisely represented in i and j, but the detailed grid resolution cancelled out most of this effect. Facies and porosity data (Fig. 5c) were upscaled from the logs of a single well (NO 7120/6-1) to the grid cells at the well locations, and then they were populated across the structural models for each experiment. The facies were extrapolated using the sequential indicator simulation method (Fig. 5d). For simplicity, all sands were considered to be a net reservoir. A constant oil saturation of 0.9 was used over the whole model for cells located inside the oil leg. Finally, an area-wide oil–water contact (OWC) was placed at a depth of 2420 m, the deepest point of the top Fuglen surface within the model area, to simulate a spill point with a footwall trap. Volumes were calculated, including gross rock volume, pore volume, and in-place hydrocarbon volume (STOIIP), for both Exps. 1 and 4 (Fig. 5f). This simplified modelling was used to quantify the effects of the interpretation methodology on the hydrocarbon-related volume calculations.

For the volume calculations, there was a concern that any differences between Exps. 1 and 4 might be caused, or at least exaggerated, by the stochastic facies and porosity modelling. Different facies and porosity realizations will result in different volumes. We needed to be certain that any variations in volume were caused by the different interpretation methods and not by the stochastic property modelling. Several options were examined to negate this possibility. As the grids are identical in their i, j, and k dimensions, it was expected that Petrel™ would produce the same realization in the two grids when the same seed number was selected; this proved to be an incorrect assumption. The method selected to make sure that the same realizations were being used, and to ensure that an extreme case was not being selected, was to (1) generate 100 realizations on the Exp. 1 grid, (2) copy all 100 realizations to the Exp. 4 grid, and (3) run the volumetric analysis on all realizations for both grids. Once the volumes had been calculated for 100 realizations on each grid, they were analysed to determine the average volumes. This negated the possibility of selecting an extreme case. Using the same set of realizations in the two experiments meant that the differences in volumes could be assigned, with certainty, to the differences in interpretation methods used.

Five seismic interpretation experiments (Fig. 3) were analysed to understand the effect that the interpretation methodology has on the resulting fault and horizon surfaces.

Firstly, it is important to consider the areal coverage and visible patterns contained in the interpretation before it is converted into surfaces (Fig. 3). When analysing the interpretation of the top Fuglen and top Fruholmen, Exps. 1–4 have an increase in interpretation density (the horizon interpretation of Exps. 4 and 5 is the same; Fig. 3a–d). All the horizon interpretations show the same general trends in topography, but as expected, the topography is more detailed and most sharply defined in the most densely interpreted data (Exps. 4 and 5; Fig. 3d, e). The top Fuglen is the most clearly imaged reflector, which resulted in complete interpretation coverage in all experiments (i.e. no gaps in the interpreted lines; Fig. 3). The clear imaging of this reflector is especially evident in the auto-tracked horizon in Exps. 4 and 5 (Fig. 3, top Fuglen). The top Fruholmen is a poorly imaged reflector, which consequently resulted in gaps in the interpreted lines (Fig. 3, top Fruholmen). The areas lacking interpretation of this reflector are evident in all experiments, but they are most clear in the auto-tracked horizon (Fig. 3d, e; top Fruholmen). The fault polygons for the two horizons do appear to have the same general trends, but this will be discussed in detail in the next section.

The horizon and fault interpretations were converted into surfaces. The horizon surfaces show the same general patterns with respect to topography in all the experiments (Fig. 6). Generally, all top Fruholmen structure maps show a topographic low on the north (hanging wall) side of each fault. The footwall blocks are uplifted relative to the hanging walls, and the points of highest elevation are located adjacent to the faults (Fig. 6, top Fruholmen). In the top Fuglen surface, the same overall topographic patterns are evident, but the amount of footwall uplift and depth of topographic lows on the hanging wall are less than on the top Fruholmen surface (Fig. 6a). The greatest differences between the experiments occur in areas where the lateral continuity of the interpretations were disrupted due to the presence of a fault, where horizon interpretations do not continue across the fault plane, and when the interpretation density was low (Exps. 1–3; Fig. 6a–c). In these cases, it is possible to identify topographic features near the faults, which are clearly artefacts (Fig. 6a–b; Exps. 1–2, white arrows).

To better visualize the surface anomalies, thickness difference maps were generated between the surfaces of Exps. 1–3 and the most dense surfaces of Exps. 4–5. Visual inspection indicates that surfaces 1–3 all contain interpretation anomalies. The difference maps show a decrease in thickness difference with increasing interpretation density (Exps. 1–3). The maps also show that the top Fuglen surfaces are a closer match to the auto-tracked horizon than the top Fruholmen (Fig. 7).

Exp. 1 shows the most significant differences from the 3D auto-tracked horizons due to a sparse interpretation grid and the introduction of gridding anomalies (Fig. 7a). The thickness anomalies in both the top Fuglen and top Fruholmen can measure ± 30 m from the 3D auto-tracked surface, and the anomalous areas are up to 400 m wide and long (i.e. comparable to the interpretation spacing; Fig. 7a). The top Fuglen from Exp. 1 correlates moderately well in unfaulted areas, and all the major anomalies occur close to the faults (Fig. 7a, top Fuglen). On the hanging wall side of the faults the anomalies are predominantly depressions (i.e. a sparse interpretation grid generates a surface that is too deep), while on the footwall side the anomalies trend upward (i.e. the surface from the sparse grid is too shallow). The top Fruholmen from Exp. 1 is more anomalous across the entire surface; there is no clear correlation between the tendencies of the anomalies on the hanging wall and footwall (Fig. 7a, top Fruholmen). The areas of divergence occur in the gaps between interpreted ILs and XLs.

Exp. 2 exhibits much less significant changes in thickness with respect to the auto-tracked horizons on both the top Fuglen and top Fruholmen (Fig. 7b). For the top Fuglen, a pattern like Exp. 1 is observed; most thickness anomalies occur near the faults and correspond to gaps in the interpretation (Fig. 7b, top Fuglen). The top Fruholmen is more chaotic, but in this case the anomalies are smaller (up to 200 × 200 m) and exhibit smaller thickness differences (± 15 m) than in Exp. 1. Like in Exp. 1, the thickness differences in both the top Fuglen and Fruholmen correlate with gaps in the interpretation.

Finally, the thickness anomalies for Exp. 3 show the same trends as in Exps. 1 and 2, but again they are smaller in area (up to 100 × 100 m) and magnitude (± 5 m; Fig. 7c). The anomalies occur at points of gaps in the interpretation. The thickness anomalies in the top Fuglen are almost always observed near the faults, while those on the top Fruholmen are more widespread across the whole surface (Fig. 7c). It is important to keep in mind that the top Fuglen has complete areal coverage in the study area, while the top Fruholmen does not. In Exps. 1–3, the thickness anomalies in the top Fruholmen structure maps are in some instances linked to inconsistencies in the auto-tracked horizon.

Fault polygons are displayed on structure maps (Fig. 6) and plotted graphically (Figs. 8, 9) to show how fault length and morphology change with the interpretation method. Generally, fault length on the interpreted horizons increases with interpretation density from Exp. 1 (shortest faults) to Exps. 4 and 5 (longest faults). These observations are clear for both the top Fuglen (Fig. 8a, b) and the top Fruholmen (Fig. 8c, d). In Exp. 5 (horizontal fault sticks), the eastern fault is longer than the fault interpreted by vertical fault sticks in Exp. 4, while the western fault is shorter than in Exp. 4 (Fig. 8).

The morphology of the faults also changes with interpretation. In Exp. 1, there is a minimal amount of interaction between the two very straight faults forming the relay (Fig. 6a). In Exp. 2, the faults are also straight and do not appear to interact (Fig. 6b). In experiments 3 to 5, the northward curvature and lengthening of the eastern faults towards the western fault increase, which suggests that the relay is close to breaching or may even be breached (Fig. 6c–e). This near-breach relay is evident in the top Fuglen for Exps. 4 and 5, but it is less prominent in the top Fruholmen (Fig. 6d, e).

The effect of the interpretation method on fault length is clearly seen in the graph of fault trace distance versus fault throw (Fig. 9). The data in these graphs were sampled from the interpreted fault sticks and show that in Exp. 1 there is minimal overlap between the two faults, and the amount of overlap increases towards Exp. 4 (Fig. 9a–d). For Exp. 5, fault trace distance versus throw shows that the eastern fault is longer, while the western fault is shorter than Exp. 4 (Fig. 9e), which confirms our observations from Fig. 8.

Fault throw contours from all five interpretation experiments exhibit generally consistent patterns (Fig. 4a) on the eastern and western faults but also some bullseye patterns (Fig. 10a). The western fault has similar throw magnitudes across all experiments. The lowest throws occur on the eastern margin and the highest throws (up to 100 m) on the western side. With increasing interpretation density, the throw results for this fault appear smoother and more laterally extensive. For example, in Exp. 1 the western fault shows three separate bullseye patterns, while in Exps. 2–4 it shows a progressively smoother throw distribution (Fig. 10a). For the eastern fault, the throw patterns are similar between experiments, but the throw magnitudes increase with increasing interpretation density (Fig. 10a). In Exp. 1, fault throw reaches a maximum of ∼ 150 m on the eastern side of the fault. For Exps. 2 and 3, the results have slightly higher maximum throw (∼ 175 m), but they are segmented into geologically unrealistic bullseye patterns (Fig. 10a). In Exp. 4, the maximum throw of the eastern fault is up to 200 m, and the results are more concentric, smoother, and more geologically realistic than in Exps. 1–3. Exp. 5 (fault sticks interpreted on depth slices) shows similar patterns as those observed in Exp. 4 but with more irregularities.

Fault trace distance versus throw also illustrates how fault displacement is influenced by the interpretation method (Fig. 9). As discussed before, the fault throw of all experiments is greater on the edges of the study area than near the relay (centre of the graphs in Fig. 9). For the western fault, the top Fruholmen is always displaced more than the top Fuglen. For the eastern fault, the top Fruholmen is displaced more than the top Fuglen in Exps. 4 and 5 (Fig. 9d, e), but it exhibits similar throws as the top Fuglen in Exps. 1–3 (Fig. 9a–c). In all experiments, the throw distributions for the top Fuglen are smoother than those for the top Fruholmen. This smoothness is also observable in the throw fault plane projections where the bullseye patterns occur on the top Fruholmen level. The highest throw values for the eastern fault at the top Fruholmen in Exps. 1–5 are ∼ 147, 155, 161, 189, and 187 m, respectively. These values occur near the eastern margin of the study area (Fig. 9). For the western fault, the top Fruholmen peak throw values in Exps. 1–5 are ∼ 91, 87, 90, 97, and 92 m, respectively. However, these peaks do not always fall near the western edge of the study area, as the western fault is relatively constant in throw outside the relay (Fig. 9). The top Fuglen throw on the eastern and western faults has a similar distribution as observed for the top Fruholmen (Fig. 9). At the top Fuglen level, the eastern fault has maximum throws of ∼ 150, 155, 154, 155, and 149 m, and the western fault has maximum throws of ∼ 77, 72, 72, 78, and 77 m for Exps. 1–5, respectively. Figure 9 clearly shows that the trends of throw for Exp. 1 are overly smooth, while those of Exps. 2–4 are similar. Exp. 5 shows more or less the same result as Exp. 4, with slight changes due to the extent of the faults.

Lithology data projected onto the fault planes can help us to understand how interpretation methods influence the evaluation of reservoir juxtaposition and the potential for fault sealing. All experiments were populated with the same lithological data from well NO 7120/6-1 (Fig. 1b, yellow dot); the only variation is the interpretation method. On a broad scale, the juxtaposition diagrams for the five experiments look very similar on both the eastern and western faults (Fig. 10b). The uppermost section of the faults is characterized by shale–shale juxtaposition (dark grey, western fault), or it has not been characterized due to a lack of conformable top Kolje distribution on the eastern side of the study area (light grey, eastern fault). The next unit down is a homogenous sand–sand interval, followed by a shale–shale section at the fault centres, which is segmented by thin sand–sand units. Finally, the deepest lithology juxtaposition is another homogeneous sand–sand unit. On closer examination, however, comparison of the different experiments reveals that the lateral extent and definition of the intra-shale sand overlaps improve with increasing interpretation density (Fig. 10b). This is especially true when comparing the least dense seismic interpretation (Exp. 1) to the densest one (Exp. 4). Exp. 5 follows the same pattern as Exp. 4 in areas where the juxtaposed lithology ran smoothly, but there are some issues with the juxtaposition (light grey triangle at the base of the eastern fault, Fig. 10b). This anomaly is caused by a limitation in the software, whereby the horizontal interpretation of the fault on depth slices results in some sections of the fault having vertical dips. It is not possible to generate juxtaposition diagrams in these vertical fault areas.

Dip separation gradient (DSG) as well as longitudinal and shear strain (Freeman et al., 2010) were calculated to understand variations in interpretation confidence between the experiments. The results for the dip separation gradient are similar across all five experiments (Fig. 11a). In general, the largest DSG (> 0.2) occurs at the top Fruholmen level. The western fault has a larger distribution of high DSG values in the western top part (0.125 gradient) and a main bullseye on the eastern side (Exps. 1, 3–5; Fig. 11a). The eastern fault has the same three to four bullseyes occurring in all experiments, but Exp. 1 has the lowest DSG values.

The longitudinal strain (LS) patterns are similar to those observed in the DSG results (Fig. 11b). The colour bar for longitudinal strain is set so any values outside a geologically realistic threshold (Freeman et al., 2010) occur as red (LS >0.1) or purple (LS <-0.1). The results for LS for all experiments are similar and exhibit values that are within the defined threshold. In the western fault for Exp. 1, unrealistic LS values at the top Fruholmen level on the eastern side suggest a problem with the interpretation (Fig. 10b, top row). This problem is not present in the other experiments. High (green) LS values in the western upper half of the western fault in Exps. 1–4 are within the acceptable threshold (Fig. 10b). These high values coincide with the area between the top Kolje and top Fuglen. The eastern fault has the same LS bullseyes across its centre as observed in DSG, but they are mostly within the established threshold. In Exps. 4 and 5, there are two areas above the high LS thresholds (red) at the top Fruholmen level (Fig. 11b, black asterisks). All areas above threshold LS values (red, pink) are less than 250 m across.

For the shear strain (SS), the colour bar is also set to display geologically unrealistic values (± 0.05; red and pink, Fig. 10c) (Freeman et al., 2010). Although SS highlights more problematic areas and places more stringent constraints on the interpretation, it indicates extreme highs and lows of SS at the overlap of the western and eastern faults, respectively (Fig. 10c, black arrows). The overlapping sections of the faults are more laterally extensive from Exps. 1 through 5, which is reflected in the lateral extent of extreme SS. Localized (> 250 m) SS bullseyes highlight some slight interpretation problems discussed before in relation to LS (Fig. 11c, black asterisks). Due to the high degree of similarity between the experiments, no attempt has been made to analyse SS variations any further.

In order to test the implications of interpretation techniques for hydrocarbon volume calculations, the least and most densely populated experiments (Exps. 1 and 4) were input through a geological modelling workflow (Fig. 5). A 5 × 5 km geological model was generated for each experiment (Fig. 12a, b) and used to calculate the bulk rock volume, pore volume, and STOIIP (Fig. 9c, d).

There are significant differences in fault morphology, horizon resolution, and lithology distribution between the two geological models. In Exp. 1, the surface anomalies observed in the structure maps (Sect. 4.1, Fig. 6 arrows) are also evident in the 3D grid at the top and base of the gridded interval (Fig. 12a, inset a, label 1). Since the top Fuglen and Fruholmen are used as the input to define the top and base of the gridded interval and the cells within, the surface anomalies also greatly impact the facies distribution in Exp. 1, which undulates to match these anomalies. These facies undulations can be observed on the exposed footwall of the eastern fault and on the eastern geological model boundary, as the facies pull upwards towards the footwall (Fig. 12a, inset a, label 1). In Exp. 1, there are also some problems with respect to the exposed fault planes where some shale cells have bled up and down the fault planes, creating unrealistic peaks (Fig. 12a, inset label 2). This results in poor modelling of the relay ramp structure, although the exposed footwall and hanging wall blocks appear relatively smooth (Fig. 12a, inset a, label 3).

In Exp. 4, the facies distributions do not have the same undulations that are observed in Exp. 1. This result is more or less expected since these anomalies were not evident in the top Fuglen and Fruholmen, which define the grid. Flat, more geologically representative facies distributions are clear on the uplifted footwall of the eastern and western faults, as well as on the exposed eastern boundary of the model (Fig. 12b, inset b, label 1). A “bleeding of facies” occurs on the margins of the model and slightly on the edges of the faults (Fig. 12b). The relay ramp is much more clearly defined in this experiment than in Exp. 1 (Fig. 12b, yellow arrow). The faults are better defined with respect to length and morphology, but the high density of interpreted fault sticks means that the fault planes have vertical jumps between grid cells in the 3D grid (Fig. 12b, unset b, label 3).

Bulk rock volume, pore volume, and oil (STOIIP) were calculated for both geological models using an oil–water contact of 2420 m (Fig. 12a, b; OWC). This contact was chosen to mimic a spill point at the lowest point of the top Fuglen. The volumetric analysis was run on each of the 100 realizations, and the results presented are given as their average. The stochastic facies and porosity realizations used in these calculations were identical for the two experiments, which allowed any volume differences to be assigned to the impact of the resolution of the interpretation. The volumetric calculations for Exp. 4 were always slightly larger than Exp. 1. The bulk volumes for Exps. 1 and 4 are 1548.7 and 1554.2 × 106 m3, respectively (a difference of 0.36 %). For pore volume, values of 136.8 and 137.4 × 106 m3 were calculated from Exps. 1 and 4, respectively, which is a difference of 0.46 %. Finally, the calculation of oil in place (STOIIP) resulted in 123.1 × 106 m3 for Exp. 1 and 123.7 × 106 m3 for Exp. 4 (a difference of 0.46 %).

The volumes in Exp. 4 are slightly larger than in Exp. 1, with the increase in the bulk rock volume carried through the pore volume and STOIIP calculations. However, the percentage differences are very small: less than 0.5 % for all metrics.

The seismic interpretation method had a significant impact on all aspects of the fault analysis workflow. We found that both Exps. 4 and 5 provided the most geologically accurate representation of the morphologies of horizons, faults, and their intersections. The eastern fault was longest in Exp. 5, while the western fault was longest in Exp. 4, which suggests that a combination of the methods (i.e. vertical and horizontal interpretation) would be the most rigorous approach to fault interpretation. The horizons in Exps. 4 and 5 were quick to interpret because of 3D auto-tracking, and they were also the most detailed. When interpreting the top Fuglen there was no need for a QC process since the imaging of this reflector was clear and the final surface did not contain any artefacts in the interpretation (Fig. 7, top Fuglen columns). The top Fruholmen needed some manual guidance and QC, and it did have some interpretation artefacts, but this was unavoidable due to the poor seismic quality (Fig. 7, top Fruholmen columns). The interpretation of faults was slightly more time-consuming for Exp. 5 relative to 4, but the attribute volume increased the understanding of fault morphology and length compared to Exp. 4 (see Fig. 8, fault lengths). Exp. 1 is considered to be a failure with respect to observed geological morphologies, and this methodology cannot be recommended as a method for fault interpretation, even though it was very time-efficient. The sparsity of the horizons and fault interpretations led to inaccuracies and gridding anomalies proportional to the spacing of the interpreted inlines (400 m), reduced fault length (Fig. 8, up to 400 m difference between Exps. 1 and 4, western fault), and incomplete understanding of the relay morphology. Exps. 2 and 3 were an improvement on Exp. 1, as expected. They captured some important information but not as much as Exps. 4 and 5. The differences between Exps. 2 and 3 were much less significant than those between Exps. 1 and 2. As such, if manual interpretation of faults is required, Exp. 2 should be considered the minimum acceptable interpretation density for performing a detailed fault analysis workflow.

The two aspects of the fault analysis workflow that were the most affected by the interpretation method were fault length and throw. Both the length and throw of the faults differed dramatically depending on interpretation density, which in turn had a large influence on the apparent morphologies of the faults and of the relay ramp (Figs. 8–10). The knock-on effects of these are because the fault lengths and throws impact all other aspects of the workflow. Overall, comparison of the most and least densely interpreted datasets (Exps. 4–5 and 1, respectively) shows that the length, morphology, and throws were different at both the top Fuglen and Fruholmen level (Figs. 7–10).

The impact of the interpretation method on the length, morphology, and throw profiles in the relay is critical to understand its formation. Fault displacement–throw relationships in relay ramps are dependent on the stage of relay development in question (Fig. 13). In the first stage of relay development, the faults do not overlap and therefore exhibit isolated fault throw profiles (Barnett et al., 1987; Fig. 13a–b). Stage 2 of relay development is defined by the propagation of faults to form a relay ramp (Fig. 13c). Fractures break up the ramp (that in our case are sub-seismic resolution) and accommodate some of the strain of the relay (Larsen, 1988; Peacock and Sanderson, 1994). The throw profiles of the faults interact, and the total throw of the overlapping fault segments is accommodated by the relay ramp (Peacock and Sanderson, 1994; Fig. 13d). The fault extents and throw profiles for Exp. 1 (Figs. 9a, 10a) fall somewhere between Stages 1 and 2, wherein there is a slight overlap of the faults, but a relay is only just starting to form (Fig. 6a). This is because Exp. 1 does not properly capture the full length of the fault. Stage 3 of relay development is defined as when the faults have continued to propagate and fractures have begun to spread through the relay structure, as it is near the maximum amount of strain it can accommodate (Long and Imber, 2012; Peacock and Sanderson, 1994). The propagation of the fault tips toward the relay and increased fault overlap are evident (Fig. 13e–f). Stage 4 of relay development defines the destruction (breaching) of the relay ramp and the formation of branch lines between the two relay-forming faults (Peacock and Sanderson, 1994). The original tiplines of the fault are no longer active, and the faults are now joined along branch lines formed in the weakened and sheared ramp margins (Fig. 13g–h). When analysing Exps. 4 and 5, the morphologies are comparable to those observed in Stage 3 of the relay formation. The northward propagation and curvature of the eastern fault tipline are clear, and there are likely fractures forming in the relay that are below the resolution of the seismic data. The relay in Exps. 4 and 5 has not breached on either the top Fuglen or Fruholmen level, although it is very close to breaching in Exp. 5 at the top Fuglen (Figs. 6d, e, 9d, e, 10a). The potential impact of a relay on a working hydrocarbon system and the implications of misinterpreting the relay are discussed in Sect. 5.2.

A study of longitudinal and shear strain was completed to test the accuracy of the interpretation methods (Freeman et al., 2010). According to Freeman et al. (2010), longitudinal and shear strain values in isolated faults should remain inside their defined threshold values (± 0.1 and ± 0.05, respectively) in order for the interpretation to be deemed accurate. High and low values of longitudinal and shear strain were observed across all experiments, some of which are outside these defined thresholds (Fig. 11b, c). There is a high and low shear strain accumulation in all experiments on the western and eastern faults, respectively, particularly in the parts of the faults exhibiting overlap (Fig. 11c). Freeman et al. (2010) stated that in the event of overlapping faults, higher shear strains (above their defined limit) are to be expected in the overlapping segments of the fault. The shear strain limits in this case are higher than could be expected from an isolated fault (Freeman et al., 2010). These highs and lows appear to change with interpretation density and align with the increased overlapping of the faults (Fig. 11c, double-ended black arrows). There were some bullseye patterns (longitudinal and shear strain plots) which were outside the fault overlap and outside the defined threshold strains; these are interpreted to be artefacts produced by incorrect fault stick interpretations (Fig. 11b, c black asterisks). It is possible that some of the bullseye patterns observed, which did not align with interpretation spacing, are real and linked to the coalescence of faults during their formation (e.g. Lohr et al., 2008), but this is outside the scope of this study and was not investigated further. It is important to note that interpretation accuracy with respect to longitudinal strain and shear strain was not the aim when running the initial interpretations, and therefore it is expected that some inconsistencies are present.