With the use of seismic flowlines as a guide, we here present a workflow for automatic identification and 3D extraction of prominent seismic horizons and sequence boundaries. Seismic flowlines represent trajectories extracted from the local dip across the seismic volume. These seismic flowlines highlight apparent unconformities as they often appear as areas of convergence from the sequence above and the sequence below. This makes seismic flowlines suitable for identifying and mapping sequence boundaries. In addition, by combining seismic flowlines, which provides spatial relationship and first-order geometries, together with a series of seismic attributes. We are able to identify and map prominent reflectors in 3D. With this work we aim towards a framework for automatic interpretation that both identifies and extracts prominent reflections and sequence boundaries in 3D, informed by the seismic data itself, and without the need for extensive manual input.

Misalignment of reflections is a common problem in migrated prestack seismic data, which occurs due to moveout correction with incomplete knowledge of the velocity model. This issue persists in partial angle stack data often used for amplitude-variation-with offset (AVO) analyses and is traditionally treated with residual moveout (RMO) and trim static corrections. However, these methods require time-consuming velocity selection and parameter tuning and commonly lead to spurious alignment of reflections with noise. Therefore, we train a convolutional neural network (CNN) on synthetic examples to automatically perform conditioning and alignment of angle stack data. First, 2D synthetic common-depth point gathers are created using a convolutional modeling method and then made into input-target pairs, where the inputs are poorly moveout corrected (using inaccurate normal-moveout velocities) and the targets are accurately moveout corrected. These input-target examples are split into angle stacks for the near, mid, and far offsets and then used to train a CNN to transform the misaligned angle stacks into their more-aligned counterparts. In testing, the trained network increases the alignment of unseen synthetic data, improving the correlation with the target far-offset trace from 0.76 to 0.85 on average. The network is applied on two field examples from offshore Norway with class 3 and class 2P AVO responses, respectively. In both cases, the angle stack data predicted by the network indicate improved alignment, greater resolution in the far-offset stacked section, and increased correspondence to a well-tie synthetic seismogram, all while retaining the AVO responses. The CNN predictions are compared with angle stack data after traditional conditioning (RMO and trim static corrections), with the results being of similar or greater quality. This method enables high-quality conditioning of angle stack data at a fraction of the time required for conventional methods and is easily adaptable to different data sets.

Net-to-gross ratio and net pay are essential properties for characterizing turbidite reservoirs. We present a Bayesian inversion that estimates the probability density distributions of the reservoir properties from the amplitude-variation-with-offset (AVO) attributes intercept and gradient, which are measured at the top of the reservoir. The method is adapted to the region-specific characteristics of the sand-shale interbedding as observed from well data. The likelihood function is estimated by a Monte Carlo simulation, which involves generating pseudo-wells, seismic modeling using the reflectivity method, picking the amplitudes at the top of the reservoir, and estimating the AVO intercept and gradient. In a North Sea oil field case example, the AVO gradient is most sensitive to variations in the net-to-gross ratio, while the AVO intercept is most sensitive to the type of pore fluid. The inversion was successfully tested on pseudo-wells and synthetic seismic AVO from well data. We show that the inversion can be applied to AVO maps to produce maps of the most likely estimates of the net-to-gross ratio and the net pay-to-net ratio, the resulting net pay, and the uncertainty.

Marine seismic data is often missing near offset information due to separation between the source and receiver cables. To solve this problem, a convolutional neural network is trained on synthetic seismic data to reconstruct the near offset gap. The synthetic data is created using a two-dimensional finite difference method within a heterogeneous velocity model. These synthetics are generated with a source-over-receiver acquisition geometry so that they contain complete near offset data. The convolutional neural network is then trained on input-target synthetic pairs where the inputs are common midpoint gathers with the near offset section removed, and the targets are the same gathers with the near offset section retained. Following training, the robustness of the method is investigated with regards to common midpoint data sorting, normal moveout correction and changes in the velocity model. It is found that training on common midpoint-sorted data results in 2.8 times lower error than training on shot gathers, that normal moveout correction of the training data makes no significant difference in error levels, and that the model can reconstruct realistic near offsets on synthetic data generated 10 km away within the heterogeneous velocity model. In field data testing, first a dataset with source-over-cable acquisition geometry from the Barents Sea is used to compare the reconstructed wavefields to ground truth values. Although the reconstructed amplitudes require minor scaling to match the true values, predictions on this dataset yield 2.5 times lower near offset reconstruction error compared to a simple Radon transform interpolation method. Furthermore, amplitude versus offset gradient and intercept sections from the Barents Sea dataset are estimated with half the error when including the convolutional neural network-predicted near offset data, compared to only using the conventionally-acquirable portion of the data (beyond 112.5 m of offset). In a secondary field data test, a conventional northern North Sea dataset is used to demonstrate how the method may be applied in practice. Here, the convolutional neural network generates more realistic predictions than the Radon method, and the gradient and intercept sections calculated using the convolutional neural network-predicted traces have higher signal-to-noise ratios than the sections calculated using only the original data. The combination of high-quality synthetic training data and interpolation in the common midpoint domain enables near offset reconstruction at significant depth (1 s of two-way traveltime or more), which is demonstrated in both synthetic and field examples.

ABSTRACTRegularization and interpolation of 3D offset classes prior to imaging are an important and challenging step in the marine seismic data processing flow. Here we describe how to perform this task using a deep neural network, and we explain how to overcome the challenge of creating a suitable training data set. The training data set is generated by de‐migrating stacked pre‐stack depth migration images. For each offset class volume, we de‐migrate the pre‐stack depth migrated stacked image into two configurations: (i) the original survey configuration consisting of the recorded source/receiver positions and (ii) an ‘Ideal’ survey configuration with constant offset and azimuth for each 3D offset class. The training creates a 3D convolutional encoder–decoder model that will regularize and interpolate seismic data. The convolutional encoder–decoder is trained on 3D sliding windows in each 3D offset cube to map from (i) to (ii), i.e. to map the original survey configuration with irregular and sparse sampling into the fully sampled regular offset cubes suitable for offset‐based migration, such as Kirchhoff migration. Such migration algorithms rely on regular and sufficiently dense sampling to achieve constructive interference to image the structures and destructive interference to suppress migration noise. We test the new method on one synthetic and one field data example and show that it performs better than a standard regularization/interpolation method based on anti‐leakage Fourier transform, especially for the smallest offset classes. On the synthetic data, we also demonstrate that the convolutional encoder–decoder method preserves the amplitude versus offset as well as the standard method.

Summary We present a workflow to condition seismic angle stack data using a convolutional neural network. First, 2-D synthetic CDP gathers are generated with a convolutional modelling method, which utilizes randomized reflectivities, wavelets, and North Sea-inspired velocity trends. From each generated gather we form an input/target pair: the target is moveout corrected with the accurate velocity trend, while the input is moveout corrected with a perturbed, inaccurate version of this trend. The result is bending, distorted reflections for the input gathers, and flattened, less-distorted reflections for the targets. These gathers are muted and stacked to form near, mid, and far angle stacks. A 1-D CNN is trained on the pairs of triplet angle stack traces, to transform the misaligned and distorted inputs into improved outputs. The synthetic-trained CNN is then tested on North Sea field data. This dataset includes traditionally-conditioned angle stacks (RMO and trim statics correction), and well data allowing synthetic seismogram generation, showcasing a Class 3 AVO response. Applying the CNN results in alignment improvement comparable to the traditional conditioning, while retaining AVO responses and becoming more similar in character to the well synthetic seismogram. Additionally, the CNN-conditioned far stack section increases in resolution compared to the original data.

Summary A possible new model for the formation of the large sand mounds, or jack-ups, in the Paleogene-Neogene of the Northern North Sea is in the process of being published by Jan Erik Rudjord (AkerBP) and Mads Huuse (University of Manchester). This novel model utilizes density inversion, or Rayleigh-Taylor instabilities in granular media, to explain the mounds as a result of liquefied, granular, overburden sand sinking down fractures in lighter consolidated bio-silicious mudrock (ooze), causing an upwards directed buoyant force which eventually detach the ooze blocks and make them float as rafts in a column of "quick" sand. The term Sinkite to describe the sand mound was originally used by the Lundin team drilling the well in 2015, and some of them are co-authors to this presentation. The presentation/abstract utilizes the new perspectives to explain the formation of the 26/10-1, Zulu East, gas discovery structure, up-dip, east of the Johan Sverdrup Field, and why the reservoir and the seismic DHI is segmented. Zulu East is the only HC discovery in the Upper-Miocene-Pliocene Utsira Fm. in the NNS and the presentation suggests explanations for the lack of success and possible consequences for using the Utsira and Skade aquifers for storing CO2.

Working with seismic data in the flowline domain offers a promising approach to improving geologic interpretation, particularly in identifying unconformities and sequence boundaries. This method utilizes the seismic differential dip field as a fluid velocity field to extract geometric information and generate flowlines. These flowlines, treated as distinct objects, represent the paths of hypothetical particles moving along local velocity vectors within the seismic data. By applying a scoring system based on overlapping paths, major unconformities and sequence boundaries can be efficiently identified and extracted. An additional advantage of the flowline representation is its ability to capture the lateral regional context of seismic geometries, enabling straightforward grouping into stratigraphic sequences through simple clustering techniques, which simplifies the categorization of first-order stratigraphic units. Furthermore, refining the flowline paths based on relative amplitude changes improves their alignment with conformable reflections, supporting a flowline-based approach to effectively track these reflections. The flowline approach can also be extended into three dimensions, providing a straightforward tool for preliminary geologic analysis in 3D seismic data. The flowline-based workflows are demonstrated through application to two seismic sections, one of which addresses challenges posed by faulted areas and showcases the strategies used to resolve them.

Conventional marine seismic surveys lack near offset information, resulting in lower-quality imaging in the shallow subsurface. We present a deep learning workflow to reconstruct near offsets, which utilizes field data collected with source-over-cable acquisition geometry as training data. First, a convolutional neural network (CNN) is trained to reconstruct the near offset gap of source-over-cable common depth point gathers acquired in the North Sea, yielding low reconstruction error on unseen gathers. The trained network is then tested on a source-over-cable dataset acquired in the Barents Sea, resulting in similarly low error. AVO intercept and gradient sections are then calculated using the ground truth near offsets, the CNN-predicted near offsets, and the conventional offsets (beyond 112.5 m). The CNN-predicted sections are about two times lower in error than their conventional-offset counterparts, as compared to ground truth. This workflow could be applied to restore near offsets in existing conventional data that suffer from limited offsets (for example in other areas of the Barents Sea), thus leading to improved imaging and AVO quantity estimation.

Summary We present a workflow for reconstructing missing near offset traces in marine seismic data using a convolutional neural network. We first generate two-dimensional synthetic shot gathers with a finite difference method, and sort them into CMP gathers. From these CMP gathers we form training data of input and target pairs – with the inputs having their near offset traces zeroed out, and the targets retaining their near offset traces. The CNN is trained to transform the inputs to the targets, and reconstruct the missing near offset traces. We test the trained network on synthetic testing data, and compare the results to another network trained on shot gathers, with the CMP-trained network yielding better results. We also demonstrate the robustness of the model to new testing data, which is generated 10 kilometers away from the training data within a heterogeneous velocity model. Finally, the synthetic-trained network is tested on on field data, and we find it yields accurate results compared to a traditional Radon transform method.

Summary In this work we explore deep metric learning as a data-driven approach for segmentation of first-order stratigraphic units in 3D seismic, based on similarities in reflection patterns. We attempt to learn similarity in seismic reflection patterns from a set of general synthetic seismic classes defined in our modelling pipeline. We use synthetics to 1) avoid using manual labels, with their limitations and 2) by using synthetics rather than samples from a specific seismic survey in training, we hope to achieve a more generalizable model that can be applied on multiple surveys without the need for fine-tuning or retraining. Using the N-pair multiclass loss we train a 3D CNN to embed the input such that the distance between positive pairs sampled from the same synthetic class in the features space is minimized, while the distance to the negative pairs is maximized. To introduce the positional context of each prediction we use a pseudo-RGT volume calculated from the unwrapped phase image in 3D of the input seismic, which is added as a weighted feature to the encoded feature vectors prior to clustering. Finally, the number of clusters must be determined before applying agglomerative clustering algorithm with Ward's criterion and a connectivity constraint.

Summary The Nordkapp Basin is a large under-explored salt basin of the Barents Sea. Despite several exploration campaigns over the past decades, no successful drilling was achieved. A new hybrid survey combining streamers and nodes was acquired in 2021 to unlock this new play. Sparse nodes recording continuously during a 3 months period, with a nominal spacing of 1200m in both inline and crossline directions, supplemented a natively dense source over streamer data acquired with 7 simultaneous sources and 18 cables. We present here a fully data-driven FWI flow designed to exploit and combine the different types of data recorded by this survey to obtain optimal velocity model. The flow combines the ultra-low frequencies diving waves obtained from node interferometry and the ultra-wide offsets of node active seismic gathers to obtain a background velocity model for accurately imaging salt flanks. For higher frequency FWI, the streamers dataset with its dense spatial sampling including more near offsets traces complemented the sparse OBN data. The final 200Hz FWI product allows to directly distinguish in the velocity model the Carnian sands target and reveals details in shallow as small as 3 to 4m, which opens up new possibilities for hydrocarbon and shallow hazard detections.

Summary Synthetic modeling of seismic data has many applications. We propose a method of generating detailed pseudo-velocities by inverting legacy data to sharp relative impedance and combining this with a smooth existing velocity field. This is done for both zero-offset data and angle data in order to obtain complex pseudo velocity models for P ans S wave velocities. These models can then be used to model realistic synthetic seismograms with features based on geologically relevant, complex velocity fields. The generated data has potential applications in geophysics and for training machine learning models for seismic data processing.

While machine learning (ML) provides a great tool for image analysis, obtaining accurate fracture segmentation from high-resolution core images is challenging. A major reason is that the segmentation quality of large and detailed objects, such as fractures, is limited by the capacity of the segmentation branch. This challenge can be seen in the Mask Region-based Convolutional Neural Network (Mask R-CNN), which is a common and well-validated instance segmentation model. This study proposes a two-stage segmentation approach using Mask R-CNN to improve fracture segmentation from unwrapped-core images. Two CNN models are used: the first model processes full-size unwrapped-core images to detect and segment fractures; the second model performs a more detailed segmentation by processing smaller regions of the images that include the fractures detected by the first model. In addition, the procedure uses a new architecture of Mask R-CNN with a point-based rendering (PointRend) neural network module that can increase segmentation accuracy. The method is evaluated on approximately 47 m of core from four boreholes and results in an improvement to previous fracture segmentation methods. It achieves an increase in the average intersection over union of approximately 27% from the baseline (one-stage segmentation with standard Mask R-CNN). The enhanced fracture segmentation provides a mean for obtaining an accurate fracture aperture with an average error of less than 1 mm, which represents a reduction of 0.5 mm from the baseline method. This work presents a novel contribution towards developing an ML-based workflow for core-image analysis.

Calcite cement often occurs locally, forming thin layers of calcite-cemented sandstone characterized by high seismic velocities and densities. Because of their strong impedance contrast with the surrounding rock, calcite-cemented intervals produce detectable seismic reflection signals that may interfere with target reflections at the top of a reservoir. In this case, the amplitude-variation-with-offset (AVO) of the effective seismic signature will be altered and may even create a false hydrocarbon indication. From the Monte Carlo simulation, we find that the presence of thin calcite-cemented beds increases the uncertainty of the Bayesian pore-fluid classification based on the AVO attributes intercept and gradient. In the case example of a North Sea turbiditic oil and gas field, the probability of a false-positive hydrocarbon indication increases from 3%–5% to 18%–21% assuming an equal probability of the occurrence of brine, oil, and gas. The results confirm that calcite-cemented beds can create a pitfall in AVO analysis. Realistic estimates of the AVO uncertainty are crucial for the risk assessment of well placement decisions.

Summary The Haugaland High, in the Norwegian North Sea, consists of a layered overburden of sub-horizontal sediments almost 2km thick that sits on the chalk basement. The background velocity regime of these top layers has a low vertical gradient down to the chalk interface. This velocity behavior is particularly poorly suited for diving wave FWI, and the strong multiple content present in the data does not allow for an efficient tomographic update. Using all reflections and diving waves recorded, Time-Lag FWI can provide a high-resolution velocity field that explains the complex velocity variation present in the overburden and simplifies the reservoir structure. With the use of narrow-azimuth towed-streamer data covering 2000km2, the velocity was updated up to 40Hz, both helping structural imaging and bringing additional information to better understand the rock properties of the basement over the entire region.

Summary The Nordkapp Basin in the Barents Sea is considered an underexplored basin with limited amount of good quality seismic for exploration. The salt diapirism is prolific with diapirs penetrating all the way up to the seafloor. In order to image the complex salt flanks, accurate 3D velocity models for imaging is required. Experience from TopSeis acquisition both in the Barents Sea and the North Sea has provided sufficient evidence to continue a similar acquisition setup for recording of the seismic wavefield. In addition to a high quality seismic wavefield, a high-resolution 3D velocity model is required to image around and up against the irregular salt bodies. This can be achieved by deploying Ocean Bottom Seismic nodes on the seafloor. Recent Full Waveform Inversion (FWI) technology can obtain very accurate models to high frequencies even from very sparse node geometries. A split-spread source-over-streamer acquisition geometry using six wide-tow sources in addition to a long offset FWI front source has been used to acquire 3700 km2 of high-quality data. Seafloor nodes in a sparse grid of 1200 x 1200 m was deployed to record the necessary low frequencies and long offset full azimuth data required to build an accurate velocity model.

In 3D marine seismic acquisition, the seismic wavefield is not sampled uniformly in the spatial directions. This leads to a seismic wavefield consisting of irregularly and sparsely populated traces with large gaps between consecutive sail lines, especially in the near offsets. The problem of reconstructing the complete seismic wavefield from a subsampled and incomplete wavefield is formulated as an underdetermined inverse problem. We have investigated unsupervised deep learning based on a convolutional neural network for multidimensional wavefield reconstruction of irregularly populated traces defined on a regular grid. Our network is based on an encoder-decoder architecture with an overcomplete latent representation, including appropriate regularization penalties to stabilize the solution. We proposed a combination of penalties, which consists of the [Formula: see text]-norm penalty on the network parameters, and a first- and second-order total-variation penalty on the model. We determined the performance of our method on broadband synthetic data and field data represented by constant-offset gathers from a source-over-cable data set from the Barents Sea. In the field data example, we compare the results to a full production flow from a contractor company, which is based on a 5D Fourier interpolation approach. In this example, our approach displays improved reconstruction of the wavefield with less noise in the sparse near offsets compared with the industry approach, which leads to improved structural definition of the near offsets in the migrated sections.

Summary The 3700km2 Nordkapp Basin area, Barents Sea, was recently acquired with a wide-spread source-over-spread design. With its 6 sources sitting on top of 18 multi-sensor streamers, one sail-line can record a dense carpet of 108 sublines separated by only 6.25m. By thinly sampling the near offset over the full azimuth, this new source-over-spread setup is particularly well suited to image the steeply dipping salt flanks that extend up to the water-bottom. After application of a dedicated processing sequence carefully designed to honour the full resolution of the recorded data, the obtained high-resolution image is able to distinguish even small-scale geological features. This large 3D volume was also compared with an NFH image. Using the full benefits of the hexa-source, a high-end processing sequence was applied to the NFH data to overcome the usual weak signal-to-noise ratio of such records. The comparison between the two final images confirms the high-resolution quality of the source-overspread volume, which includes enhanced lateral resolution, especially along the crossline direction, and access to AVO and RMO information. On the other hand, the very thin vertical sampling of the NFH data extends the recorded bandwidth by two octaves, making it a possible complement to the source-over-spread image.