Qualitative analysis of sandstone reservoirs predicted by Rune Inversion in the North Sea Megagrid

Introduction

During and after the 2015-2018 oil industry downturn, which limited exploration budgets, there was a need to develop a technology that could utilise the available post-stack seismic data on the Norwegian Continental Shelf for more than just structural interpretation. Usually, the extraction of rock properties from post-stack seismic data is limited to seismic attributes and P-impedance, with the potential future derivation of additional properties through regression. However, this does not provide the rock properties required for exploration and prospect delineation, such as porosity, density, saturation, and mineral composition, as can be obtained from pre-stack inversion. The availability of modern computational resources and the development of mathematical methods allowed us to develop an algorithm that can invert post-stack seismic data into independent P-wave velocity and density cubes, which are subsequently converted into porosity and the Best Reservoir Quality attribute.

Our approach was implemented on merged true-amplitude post-stack seismic data from the Norwegian Sea, the North Sea (Norway), and many other areas worldwide. Here, we demonstrate the results for 69,660 km² of the North Sea using the PSS-GEO Elephant database, which was created from publicly available well data and true-amplitude post-stack seismic data in September 2022. The Norwegian Offshore Directorate (NOD) publishes announcements of drilled wells in the form of short reports and tables. We used the NOD website to track the locations of new wells and analyse the available information, including whether sandstone was encountered, its quality (good, moderate, or poor), and whether hydrocarbons were found. We then analysed the corresponding locations in our predicted cubes to determine whether the predictions matched the drilling results. The results of our predictions have been made publicly available; see the Data and Materials Availability section.

We performed a qualitative analysis and concluded that the reservoirs were predicted very well. Most mismatches occurred in thin layers, below the seismic resolution, or in differentiating between moderate- and good-quality sandstone. Overall, the reservoir prediction accuracy reached 87%.

Methods

Data preparation and inversion

Publicly available post-stack seismic data between 58°N and 63°N, covering approximately 78,450 km², were merged. Areas with AGC applied or unknown amplitude scaling were excluded, leaving approximately 69,660 km² suitable for inversion. Before inversion, all 3D surveys were regridded onto a common grid, denoised, deghosted, and matched in phase, time, and amplitude. Phase corrections were based on well ties. For model building, the quality of the P-wave velocity and density logs was checked for each well. The logs were edited, cleaned, and tied to the merged seismic data over their full-length using Hampson-Russell software. The validated P-wave velocity and density logs were then used to build the initial model. To propagate the P-wave velocity and density away from the wells, we used nine detailed interpreted geological surfaces.

Rune Inversion

The Rune Inversion algorithm allows both post-stack and pre-stack seismic data to be inverted into non-linked velocity and density volumes (Kalashnikova & Øverås, 2026). For post-stack inversion, we used a simplified version of the algorithm, developed in 2019, in which NMO-corrected gathers are generated.

To understand how we can alternatively estimate rock properties from seismic signals (without the concept of generalised linear inversion (Smith and Gidlow, 1987)), let us first consider the case where a gather is available (Øverås et al., 2023). We aim to estimate P-wave velocity and density over the full frequency band as independent parameters. First, we estimate the P-wave velocity. We have a real, true-amplitude migrated and processed non-NMO-corrected seismic gather. First, we apply NMO correction using a very smooth or low-frequency (e.g. 0–3 Hz) initial velocity model. It is expected that the gather reflectors will not be flat or correspond directly to those observed in the real gather, and therefore a proper comparison cannot be made (Figure 1, first iteration). The comparison is made by calculating the first misfit objective function between the real gather and the generated gather. In the second iteration, we randomly update the velocity function by creating artificial kinematic constraints and then generate the second synthetic gather. By comparing it with the real gather, we compute the second misfit objective function. We repeat the same steps in the subsequent iterations. We aim to approximate the global optimum and minimise the cost function associated with seismic trace mismatches using the probabilistic optimisation technique of simulated annealing (Goffe, Ferrier and Rogers, 1994). When the global minimum is approached, we output the P-wave velocity (Vp) from iteration N as the accepted solution.

We have published the concept of finding high-resolution velocity as a basic idea previously using dynamic time warping (Kalashnikova et al., 2020). We showed that this could be achieved kinematically or dynamically. However, the previous approach could lead to parameter overestimation. We then add density to constrain the P-wave velocity estimation. Thus, we repeat the same steps described above but add a density function to the initial low-frequency model.

For the post-stack case, we follow the same procedure as for the pre-stack case. The difference is that we generate a gather by using the real trace as a near-offset trace and duplicating it several times to simulate offsets. In this case, the amplitudes lose the true-amplitude information that normally changes with offset because of stretch, angle and shear-wave effects. Therefore, we expect some errors in the estimation of Vp and density. Also, it will not be possible to restore S-wave information in such autopatch. We then create synthetic gathers at each iteration, stack the gathers, and compare them with the real trace. The process operates trace by trace and does not depend on the window. To generate synthetic traces, we use a full-band wavelet extracted from a long window.

It should be noted that it is not necessary to create NMO-corrected gathers; simulating unlimited offsets and comparing them with the real non-NMO-corrected gathers produces better results because it avoids far-offset stretch and utilises the amplitudes of all offsets (Kalashnikova & Øverås, 2026).

Reservoir attribute

To compute the volume of clay (Vclay) and the volume of sandstone (Vsand = 1 − Vclay), we used the Gyllenhammar formula (Gyllenhammar, 2020), which allows these parameters to be computed from only Vp and density. The computation itself is not described here; instead, we recommend referring to the detailed description presented by Gyllenhammar (2020) and Kalashnikova et al. (2025). In this study, we use a single average equation for the target interval, extending from the Cromer Knoll Group (Middle Cretaceous) to the Middle Triassic. However, at EAGE 2025 we also presented, and subsequently published, formation-specific equations and coefficients for the North Sea region, together with a comparison against petrophysical estimates (Nekrasova et al., 2025). The Gyllenhammar formula performs well for both inversion applications and petrophysical estimation. Once Vclay and density are known, porosity is calculated using the conventional formula (Kalashnikova et al., 2025).

Subsequently, the Best Reservoir Quality attribute (Res_att) was computed as the product of sandstone volume and porosity. This attribute facilitates regional identification of prospective reservoir intervals, with higher values indicating better reservoir quality. Figure 2 illustrates the study area together with examples of the Res_att maps. The Base Cretaceous Unconformity (BCU) divides the target interval into two main zones: the Cretaceous above the BCU and the Jurassic–Triassic below. The Res_att maps delineate the principal sandstone bodies in both intervals, where the predicted reservoir distribution closely matches the known field outline.

Qualitative comparison of drilling results with properties predicted from the inverted cubes

Figure 3 illustrates the inversion quality control. The inversion produces only Vp and density (Figure 3b and c). The algorithm does not ‘know’ the well locations and treats all supplied initial model trends equally. Sandstone volume and porosity are derived from these properties and can be adjusted according to the available well logs and target formations. As we mentioned earlier, in this study, we used an average equation for the entire target interval. Sandstone volume was validated against gamma-ray logs and NOD descriptions, while porosity was evaluated using published data. The comparisons are presented in Figure 3d and e.

The total number of available wells for evaluation between 2019 and June 2024 was 164, including 35 drilled below the BCU. Some of these were sidetrack wells. All available information on the penetrated formations and their rock descriptions was collected and qualitatively rated for each well according to the results from the inverted cubes. The qualitative analysis was rated on a three-point scale (1 = correct prediction, 0 = missed prediction, 0.5 = partially correct prediction). For example, the inverted cubes might indicate a thick, high-porosity sandstone, whereas drilling encountered sandstone of only moderate quality — in this case, the prediction was rated as 0.5. The total score was then calculated to evaluate the qualitative prediction success rate of the proposed technique across all target formations within the Elephant project.

Geology of the region for the target Jurassic and Lower Cretaceous interval and cross-inversion analysis

The Best Quality Sands distribution map summarises the reservoir prediction results across the study area (Figure 4). The map illustrates the combined response from predicted Jurassic reservoirs, mainly ranging from the Lower Jurassic to the Upper Jurassic reservoir groups. Due to the Late Jurassic rifting phase, major block faulting caused uplift and tilting, creating considerable local topography, erosion, and sediment supply. Therefore, the distribution of the Jurassic sediments is not continuous and reflects pre- and syn-depositional fault activity and differential subsidence (Figure 4a).

The largest predicted areas of thick porous sands are located within the Sogn Graben, the northern part of the Viking Graben, and the Viking Graben itself (blue arrows, Figure 4b). The eastern area corresponds to increased net thickness of Jurassic reservoirs, from the Lower Jurassic Statfjord Group to the Upper Jurassic Viking Group. Increased reservoir quality is also predicted in the west, mainly within the Viking and Brent reservoir groups.

Smaller predicted porous sandstone bodies occur across the Viking Graben (black arrows, Figure 4b), corresponding to the Middle Jurassic Brent and Dunlin reservoir groups. The best reservoir quality is concentrated along the eastern flank, with shale content increasing westwards, reflecting the depositional history of both groups. The predicted sandstone distribution confirms the previously interpreted progradation and subsequent retreat of a major deltaic system within the Brent Group (Halland et al., 2013), whereas the Dunlin Group consists predominantly of marine to marginal-marine sandstones.

Less continuous sandstone bodies are also predicted along the eastern and southern flanks of the Southern Viking Graben (orange arrows, Figure 4b). These correspond to the Vestland Group reservoirs formally outlined by the mudstone-dominated Viking Group. The predicted sandstone distribution also delineates the Sleipner and Hugin formations adjacent to the Utsira High, consistent with the known geology of the Vestland Group.

In the Central Graben (light-green arrows, Figure 4b), the map indicates a patchy distribution of porous sandstones within the predominantly mudstone lithology of the Upper to Middle Jurassic sequences (Ula and Bryne formations of the Vestland Group). The depositional environment varies from deltaic coal-bearing silt and shale at the base to more marine-influenced sandstones in the deeper parts of the basin (Halland et al., 2013). The attribute map suggests a low probability of thick sandstone occurrence, consistent with the depositional environment, well results, and previous seismic mapping. The current study demonstrates that the Best Reservoir Quality attribute is an effective tool for reservoir prediction in areas containing coal beds. This is also confirmed by the Mugnetind well (7/11-14 S; light green circle, Figure 4b), where a thin sandstone interval at the BCU and a thick coal bed below were clearly identified by the inversion results prior to drilling. The study also identifies several relatively small porous sandstone bodies along the flanks of syn-sedimentary fault-bounded sub-basins related to halokinesis and around salt diapirs (Finogenova et al., 2025). These areas may represent potential exploration targets, and the application of Rune Inversion may improve exploration and appraisal success where coal beds hinder conventional reservoir mapping. It is also noteworthy that all dry wells (red arrows) are clearly located outside the predicted reservoir quality sandstone distribution shown in Figure 4b.

Figure 5 represents the Lower Cretaceous units deposited during the thermal subsidence that followed Jurassic rifting. The Lower Cretaceous formations are associated with the uppermost part of the Viking Group. The Krossfjord, Fensfjord, and Sognefjord formations represent sandier facies and are generally restricted to the Horda Platform and northwards to approximately 62°N, according to regional seismic mapping and well analysis (Halland, 2013).

The current study’s prediction of porous sandstones is illustrated in Figure 5b, which shows the Best Quality Sands attribute extracted above the BCU. The map outlines previously proven Cretaceous sandstones within the Sogn Graben and suggests additional sandstone bodies basinwards to the west and north of the current regional Cretaceous play outline. In addition, the current study indicates the possible presence of Cretaceous (possibly Barremian) sandstones in the Southern Viking Graben overlying the Jurassic sequences.

Figures 6 and 7 illustrate two lines through the seismic data cube, the inverted velocity and density cubes, and the computed properties: sandstone volume, porosity, and Res att. The porosity values shown on the porosity sections (Figures 6e and 7e) are taken from published sources. The final sections (Figures 6f and 7f) display the primary attribute of interest for detecting the highest-quality sandstones (Res_att). The known discoveries are indicated on the hydrocarbon-bearing structures. Interestingly, the fault-block sandstone reservoirs to the east are well represented on the Res_att sections. For example, the Fram Field contains hydrocarbons in three Jurassic reservoir levels, from the Middle to the Upper Jurassic (white arrow, Figure 6e and f), making the sandstone reservoirs readily identifiable. The Gullfaks, Tordis, Statfjord, and Kvitebjørn fields are also clearly visible. Greater difficulty is encountered at the Mulder Field, where the reservoirs are thinner and require a more detailed view.

During the qualitative analysis, the velocity and density sections were scaled to the target formations to reflect property variations, as shown in Figure 8. A decrease in density within the predicted sandstone bodies may indicate the presence of hydrocarbons. Figure 8a shows a density section through two wells that penetrated sandstone within the pre-Devonian basement. Well 16/1-4 encountered a 4 m Lower Cretaceous interval consisting of claystone or marl and immature sandstones containing igneous rock fragments similar to the underlying basement. Gas condensate was found in the upper part of the basement section. In contrast, well 16/4-5 contains no Cretaceous or Late Jurassic sandstones but encountered Middle Jurassic–Triassic conglomerates with hydrocarbon shows only. This difference is clearly illustrated in Figure 8b and is supported by the predicted density section.

Figure 9 shows an example of the qualitative analysis in detail for wells drilled between 2018 and July 2024. Two wells, 35/6-2 S (DRY) and 35/9-3 (OIL/GAS, Hamlet), which penetrated the basement, were included in the initial inversion models, providing a reliable model for the region. The primary target of the latest well was the Early Cretaceous Agat Formation sandstones (Figure 9c). The operator reported that the seismic data did not exhibit clear anomalies, making hydrocarbon-bearing sandstones difficult to detect. The predicted sandstone distribution is clearly visible on the Res_att section (Figure 9b and d), supported by density decreases within the Agat Formation sandstones and increased porosity (Figure 9e). For the latest wells drilled in 2024, the operator had not released the seismic data or the interpreted reflectors corresponding to the Cerisa discovery. Therefore, the presented section represents the authors’ interpretation at the time of drilling. Considering that the interpretation was carried out using only post-stack seismic data, the prediction matched the drilling results with remarkably good accuracy.

Results

Out of the 46 available wells drilled in the Late-Jurrasic to Sub-Triassic interval between 2019 and 2024, 35 underwent detailed analysis. The remaining wells were either abandoned because of drilling hazards or were sidetracks within the same formations and were therefore excluded from the analysis. Most wells were predicted correctly and assigned a score of 1. The remaining wells were assigned partial or zero scores depending on the degree of mismatch. Overall, the property prediction achieved a success rate of 87%.

In the Early Cretaceous interval, a total of 114 wells were drilled during the same period, of which 63 underwent detailed analysis. Among these, 50 wells were predicted correctly, three were missed because of similar seismic resolution limitations, and ten showed discrepancies in reservoir thickness or sandstone quality. This corresponds to a prediction success rate of 87%.

Conclusion

We applied an unconventional seismic inversion workflow to post-stack seismic data. The inversion produced P-wave velocity and density cubes, from which sandstone volume, porosity, and the Best Reservoir Quality attribute were computed. Reservoir prediction was validated against newly drilled wells targeting primarily Cretaceous and Jurassic sandstones on the Norwegian Continental Shelf. The analysis showed good agreement between areas predicted to contain poor-to-no reservoirs and wells that encountered no sandstone or only poor-quality reservoirs. The main deviations occurred in distinguishing between moderate- and good-quality reservoirs and in predicting thin reservoirs below seismic resolution. The average success rate for predicting quality sandstone reservoirs was 87%.

The Best Reservoir Quality attribute, combined with well information, proved to be a useful tool for reviewing regional play outlines and assessing reservoir prospectivity. These results demonstrate that the proposed post-stack seismic inversion workflow can provide valuable information for regional reservoir prediction and exploration risk assessment, especially when no pre-stack seismic data are available.

Acknowledgements

The authors would like to thank Pre-Stack Solutions-Geo for permission to publish these materials, and Lime Petroleum for supporting the development of the technology and the work on the database.

Data and materials availability

To facilitate independent verification, seismic sections for 68 drilled wells (2018–2024) are freely available on the PSS-Geo website. The data cubes are available free of charge to universities and other non-commercial organisations for educational and academic purposes.

References

• Finogenova, A., Kalashnikova, V., Klein, B.E., Kaluza, M., Nekrasova, T., Øverås, R., Sopivnik, V., Akhiyarova, E., Ivanova, A., Finstad, E.K. and Kukina, N. [2025] Revealing the salt tectonic puzzle: Mesozoic base of the Norwegian North Sea. First Break, 43(6), 67-73. https://doi.org/10.3997/1365-2397.fb2025045
• Goffe, W.L., Ferrier, G.D. and Rogers, J. [1994] Global optimization of statistical functions with simulated annealing. Journal of Econometrics, 60(1–2), 65-99. https://doi.org/10.1016/0304-4076(94)90038-8
• Gyllenhammar, C.F. [2020] Calculating a synthetic density curve using volume of clay and velocity. First Break, 38(6), 45-49. https://doi. org/10.3997/1365-2397.fb2020041
• Halland, E.K., Johansen, W.T., Riis, F., Magnus, C., Meling, I.M., Pedersen, S., Solbakk, T. and Tappel, I. [2013] CO2 Storage Atlas. Norwegian Petroleum Directorate.
• Kalashnikova, V., Meisingset, I., Øverås, R. and Krasova, D. [2020] High-resolution seismic velocity field estimation techniques and their application to geohazard, lithology and porosity prediction. Near Surface Geophysics, 18(1), 61-72. https://doi.org/10.1002/nsg.12083
• Kalashnikova, V. and Øverås, R. [2026] Robust subsurface velocity and density estimation via seismic inversion with global misfit minimisation and full-offset moveout. Geophysical Prospecting, 74(1). https:// doi.org/10.1111/1365-2478.70124
• Kalashnikova, V., Øverås, R., Nekrasova, T., Gyllenhammar, C.F. and Meisingset, I. [2025]. Sand-Clay Distribution and Best Quality Sand Thickness in the Våle and Lista Formations of Rogaland Group: Comparison of Stratigraphic Reference Maps and AI-based Inversion Results in the Norwegian North Sea Elephant Database. First Break, 43(9), 49-58. https://doi.org/10.3997/1365-2397. fb2025068
• Nekrasova, T., Kalashnikova, V., Finogenova, A., Gyllenhammar, C.F. and Sopivnik, V. [2025] Clay volume calculation from acoustic and density curves using the Gyllenhammar equation. 86^th EAGE Annual Conference & Exhibition, Extended Abstract. https://doi. org/10.3997/2214-4609.202510624
• Norwegian Offshore Directorate. [2024] North Sea plays. Available at: https://www.sodir.no/ (accessed 15 June 2024). 
• Øverås, R. and Kalashnikova, V. [2021] Method of seismic inversion using artificial kinematic constraints. Norwegian Patent NO 346427 B1; U.S. Patent Application No. 18/275,303. Notice of Allowance issued May 2026
• Øverås, R., Kalashnikova, V. and Klein, B.E. [2023] Non-regression pre-stack AI-based seismic inversion algorithm. Abstract. IAMG2023, Trondheim, Norway.
• PSS-Geo. [2024] Predicted rock properties for newly drilled wells in the North Sea (2021–2024). Available at: https://www.pss-geo.com/ elephantfact (accessed 15 June 2024).
• Smith, G.C. and Gidlow, P.M. [1987] Weighted stacking for rock property estimation and detection of gas. Geophysical Prospecting, 35(9), 993-1014. https://doi.org/10.1111/j.1365-2478.1987. tb00856.x