Seismic Data Cube Research Articles

Automatic geologic fault identification from seismic data using 2.5D channel attention U-net

Fault identification plays a vital role in geologic structure interpretation, reservoir characterization, and reservoir evaluation. The traditional process for fault identification is label intensive. Deep learning-based automatic fault detection from seismic data has attracted extensive attention in recent years. However, using a 2D neural network (i.e., a single seismic data slice) leads to the loss of correlation among seismic slices, and using a 3D neural network (i.e., a 3D seismic data cube) increases the workload of fault labeling and the difficulty of training. To overcome the drawbacks of 2D and 3D neural networks, we have introduced a channel attention mechanism in the original U-net architecture. The new model, which is referred to as the 2.5D channel attention U-net (2.5D CAU-net), takes four adjacent seismic slices as inputs instead of the entire 3D seismic data cube. The efficacy of 2.5D CAU-net for fault detection is demonstrated using publicly available synthetic seismic data sets and real seismic field data sets. We explored the influence of cropping the original seismic data into various sizes on fault identification. Experimental results find that the 2.5D CAU-net can effectively use the correlation information among adjacent seismic slices to improve fault detection performance. For different seismic data cropping strategies, results indicate that the more extensive size cropping method allows the network to obtain more contextual information and more complete fault features, improving the efficacy of fault picking. However, the cropping approach with a larger size leads to a smaller number of labeled examples. When the amount of training data is not large enough, the larger size cropping approach tends to lead to model overfitting, which will decrease the performance of the fault recognition, especially in a 3D neural network. Overall, 2.5D CAU-net can improve the efficiency of fault recognition with less annotation effort and computational resources.

Robust vector median filtering with a structure-adaptive implementation

The 3D seismic data in the prestack domain are contaminated by impulse noise. We have adopted a robust vector median filter (VMF) for attenuating the impulse noise from 3D seismic data cubes. The proposed filter has two attractive features. First, it is robust; the vector median that is the output of the filter not only has a minimum distance to all input data vectors, but it also has a high similarity to the original data vector. Second, it is structure adaptive; the filter is implemented following the local structure of coherent seismic events. The application of the robust and structure-adaptive VMF is demonstrated using an example data set acquired from an area with strong sedimentary rhythmites composed of steep-dipping thin layers. This robust filter significantly improves the signal-to-noise ratio of seismic data while preserving any discontinuity of reflections and maintaining the fidelity of amplitudes, which will facilitate the reservoir characterization that follows.

Probabilistic falsification of prior geologic uncertainty with seismic amplitude data: Application to a turbidite reservoir case

Geophysical subsurface modeling is often highly uncertain due to limited data resolution. At the same time, a wealth of geologic information, often from databases and outcrop studies, is available to state prior uncertainty on key geologic modeling parameters. In most inversion procedures, these uncertainties are ignored and only a limited number of geologic scenarios are considered. We have developed a workflow, inspired by a Popper-Bayes philosophy that aims at probabilistically updating prior geologic parameters uncertainty with data without the need for inversion or iterative data conditioning. Our approach consists of stating a wide prior model of uncertainty (in rock physics and facies), and generating a set of facies models and their corresponding geophysical responses (here the seismic response) by forward simulation. Then, a global measure of similarity is defined between any two generated seismic responses and the field data. The proposed measure, based on the wavelet transform decomposition, analyzes differences in patterns found in the seismic data and aims at comparing seismic images as a whole, avoiding any local trace-by-trace comparison. Based on this global measure of similarity, our procedure identifies the geologic interpretations that are unlikely given the observed seismic data. The proposed approach was applied and validated using seismic data from a synthetic example and an offshore reservoir in west Africa. The uncertainty included geologic parameters, such as infill channel size, proportion, and sinuosity, and geophysical parameters, such as the rock physics model and the wavelet in the forward seismic model. The synthetic example allowed for the validation of the methodology using rejection sampling. For the offshore reservoir, updated probabilities were obtained for each uncertain parameter, and some of the prior geologic interpretations were found to be inconsistent with the available seismic amplitude data cube (e.g., narrow channel width).

Review Paper: An outlook on the future of seismic imaging, Part II: Full‐Wavefield Migration

ABSTRACTThe next‐generation seismic imaging algorithms will consider multiple scattering as indispensable information, being referred to as Full‐Wavefield Migration. In addition, these algorithms will also include autonomous velocity updating in the migration process, being referred to as Joint Migration Inversion. Full‐Wavefield Migration and Joint Migration Inversion address the industrial needs to improve images of very complex reservoirs as well as the industrial ambition to produce these images in a more automatic manner (automation in seismic processing).In this vision paper on seismic imaging, Full‐Wavefield Migration and Joint Migration Inversion are formulated in terms of a closed‐loop estimation algorithm that can be physically explained by an iterative double focusing process (full‐wavefield common‐focus‐point technology). A critical module in this formulation is forward modelling, allowing feedback from migrated output to unmigrated input (closing the loop). For this purpose, a full‐wavefield modelling module has been developed, which utilizes an operator description of complex geology. The full‐wavefield modelling module is pre‐eminently suited to function in the feedback path of a closed‐loop migration algorithm.‘The Future of Seismic Imaging’ is presented as a coherent trilogy, proposing the migration framework of the future in three consecutive parts. In Part I, it was shown that the proposed full‐wavefield modelling module algorithm differs fundamentally from finite‐difference modelling because velocities and densities need not be provided. Instead, an operator description of the subsurface is used. In addition, the concept of reverse modelling was introduced. In Part II, it is shown how the theory of Primary Wavefield Migration can be extended to Full‐Wavefield Migration by correcting for angle‐dependent transmission effects and by utilizing multiple scattering. The potential of the Full‐Wavefield Migration algorithm is illustrated with numerical examples. A multidirectional migration strategy is proposed that navigates the Full‐Wavefield Migration algorithm through the seismic data cube in different directions.

Survival of a submarine canyon during long-term outbuilding of a continental margin

Net-depositional submarine canyons are common in continental slope strata, but how they survive and prograde on constructional margins is poorly understood. In this study we present field evidence for the coevolution of a submarine canyon and the adjacent continental slope. Using a three-dimensional seismic data cube that images the Ebro margin (northwest Mediterranean), we identify a preserved canyon on a middle Pleistocene paleosurface and relate it directly to its expression on the present-day seafloor. A subparallel stacking pattern of seismic reflectors, similar to that seen between prograding clinoforms in intercanyon areas, is observed between the modern and paleocanyon thalwegs. The concavity of the modern long profile differs from the convex-concave long profile on the middle Pleistocene surface, suggesting a long-term change in canyon sedimentation. We interpret this change as a shift to a canyon dominated by turbidity currents from one strongly influenced by the pattern of sedimentation that built the open-slope canyon interfluves. We find support for our interpretation in previous studies of the Ebro margin.

Application of image segmentation to tracking 3D salt boundaries

Delineating salt boundaries is a necessary step in the velocity-model building process. The salt-delineation problem can be thought of as an image-segmentation problem. Normalized cuts image segmentation (NCIS) finds the cut (or cuts) that result in an image being broken into portions which have dissimilar, by some measure, characteristics. We apply a modified version of the NCIS method to partition seismic images along salt boundaries. NCIS can track boundaries that are not continuous, where conventional horizon-tracking algorithms may fail, by calculating a weight connecting each pixel in the image to every other pixel within a local neighborhood. The weights are determined using problem-dependent combinations of attributes, the most important being instantanteous amplitude and dip. The weights for the entire image are used to segment the image via an eigenvector calculation. The weight matrices for 3D seismic data cubes can be quite large and computationally expensive. By imposing bounds and by distributing the algorithm on a parallel cluster, we significantly increase efficiency. This method is demonstrated to be effective on a 3D field seismic data cube.

Dip-scanning coherence algorithm using eigenstructure analysis and supertrace technique

The eigenstructure-based coherence algorithms are robust to noise and able to produce enhanced coherence images. However, the original eigenstructure coherence algorithm does not implement dip scanning; therefore, it produces less satisfactory results in areas with strong structural dips. The supertrace technique also improves the coherence algorithms’ robustness by concatenating multiple seismic traces to form a supertrace. In addition, the supertrace data cube preserves the structural-dip information that is contained in the original seismic data cube; thus, dip scanning can be performed effectively using a number of adjacent supertraces. We combine the eigenstructure analysis and the dip-scanning supertrace technique to obtain a new coherence-estimation algorithm. Application to the real data set shows that the new algorithm provides good coherence estimates in areas with strong structural dips. Furthermore, the algorithm is computationally efficient because of the small covariance matrix [Formula: see text] used for the eigenstructure analysis.

The Marine VHR 2.5-D Seismic Brute Stack Cube as a Feasible Tool for Low Budget Investigation and Research

Between 1997 and 1999 several marine seismic surveys were carried out in Kiel Bay aimed towards the development of a three-dimensional acquisition and interpretation technique for small scale subsurface structures using high-frequency sources and multichannel streamers. The data set was recently revisited by the author and reprocessed to obtain a multichannel stacked seismic data cube. Nominal hydrophone positions are deduced by determining offsets from first arrival times and estimating the hydrophone positions under consideration of the ships track. Processing towards a ‘seismic cube’ mainly comprised CMP sorting, constant velocity NMO correction and stacking. The resulting VHR 3-D seismic ‘brute stack cube’ reveals rich structural details. The fluvial Pleistocene channel system already documented in an earlier publication was tracked further to the north. It is situated below a flat cover of gas-bearing Holocene sediments, which locally constitute the seafloor. This till-horizon is superimposed on a second till layer showing strong topographic variations. Seismic signal phase and shielding effects indicate the possible presence of gas in these formations. This case history demonstrates that the VHR 3-D seismic method is a feasible tool for low budget investigation and research.

Simulated migration amplitude for improving amplitude estimates in seismic illumination studies

Ideally, a seismic data cube should contain amplitudes that are proportional to the reflectivity of the subsurface. However, when working with real data, the seismic amplitudes used for interpretation are highly dependent on other factors such as survey geometry, instrumentation, filtering and muting of prestack data, migration artifacts, focusing/defocusing, wave attenuation in the overburden, and several other factors. When an amplitude variation is encountered in the seismic data, determining which factor is responsible is not easy. In this paper we apply a method based on a combination of 3D seismic ray modeling and simulation of migration that is able to distinguish between some factors contributing to the amplitudes. This will reduce the risk of interpreting “false” anomalies and drilling dry wells.

COMBINING GEOSTATISTICS AND MULTI‐ATTRIBUTE TRANSFORMS: A CHANNEL SAND CASE STUDY, BLACKFOOT OILFIELD (ALBERTA)

In this paper, we combine the methods of geostatistics and multi‐attribute reservoir parameter prediction (the multi‐attribute transform) for the integration of seismic and well log data, and illustrate this new procedure with a case study involving the prediction of porosity at the Blackfoot oilfield, central Alberta. The objectives of the survey were to delineate incised, valley‐fill sediments within the Early Cretaceous Glauconitic Formation at this field and to distinguish between sand‐fill and shale‐fill. The input consisted of twelve porosity logs together with a 3D seismic volume and the inversion of this volume. Although an excellent correlation was found between porosity and the initial inverted acoustic impedance volume, the combination of traditional geostatistics and the multi‐attribute transform produced an improved final result.Our approach uses well logs to “train” the multi‐attribute transform algorithm. We first extract average porosity values over the depth zone of interest, and compare these values to average seismic attributes over the same zone. Cross‐validation is used to show which attributes are significant. We then apply the results of the training and cross‐validation to data slices derived from both the seismic data cube and the inverted cube to produce an initial porosity map. Finally, we improve the fit between the well‐log values and the porosity map using cokriging.

Mapping lithofacies and pore‐fluid probabilities in a North Sea reservoir: Seismic inversions and statistical rock physics

Reliably predicting lithologic and saturation heterogeneities is one of the key problems in reservoir characterization. In this study, we show how statistical rock physics techniques combined with seismic information can be used to classify reservoir lithologies and pore fluids. One of the innovations was to use a seismic impedance attribute (related to the [Formula: see text] ratio) that incorporates far‐offset data, but at the same time can be practically obtained using normal incidence inversion algorithms. The methods were applied to a North Sea turbidite system. We incorporated well log measurements with calibration from core data to estimate the near‐offset and far‐offset reflectivity and impedance attributes. Multivariate probability distributions were estimated from the data to identify the attribute clusters and their separability for different facies and fluid saturations. A training data was set up using Monte Carlo simulations based on the well log—derived probability distributions. Fluid substitution by Gassmann’s equation was used to extend the training data, thus accounting for pore fluid conditions not encountered in the well. Seismic inversion of near‐offset and far‐offset stacks gave us two 3‐D cubes of impedance attributes in the interwell region. The near‐offset stack approximates a zero‐offset section, giving an estimate of the normal incidence acoustic impedance. The far offset stack gives an estimate of a [Formula: see text]‐related elastic impedance attribute that is equivalent to the acoustic impedance for non‐normal incidence. These impedance attributes obtained from seismic inversion were then used with the training probability distribution functions to predict the probability of occurrence of the different lithofacies in the interwell region. Statistical classification techniques, as well as geostatistical indicator simulations were applied on the 3‐D seismic data cube. A Markov‐Bayes technique was used to update the probabilities obtained from the seismic data by taking into account the spatial correlation as estimated from the facies indicator variograms. The final results are spatial 3‐D maps of not only the most likely facies and pore fluids, but also their occurrence probabilities. A key ingredient in this study was the exploitation of physically based seismic‐to‐reservoir property transforms optimally combined with statistical techniques.

Arid Continental Sand Bodies and Stratal Geometries in the Haynesville Formation as Interpreted from Coherency Processing of a 3-D Seismic Data Cube, Monroe County, Alabama

ABSTRACT In an exploratory effort to extend the Frisco City Trend northward and up-dip of its original limits in central Monroe County, Alabama, a twenty-four square mile 3-D seismic program was undertaken in 1998. Acquisition and processing parameters were established to equal or exceed those of previous data sets. As a result, very good imaging was accomplished for the Upper Jurassic targets in the 8,000 to 10,000 foot depth range. The survey was acquired using deep hole (125) pentalite with a triple-brick 110110 bin design. Throughout the central 80% of the survey 24 to 30 fold data was attained. Following the initial processing, which included DMO pre-stack migration with noise suppression, Coherency Cube Processing was applied to enhance sand body shape, thickness and size recognition. The basement massifs and inselbergs that underlie the Jurassic sediments demonstrate several episodes of incised valley erosion and subsequent onlapping deposition. Through interpretation, alluvial fan, braided wadi, eolian dune, playa and pediment sand flat deposits were delineated through variable amplitude horizon slice displays. The dunes appear to be best developed on the down-wind side of flat intermontane basins. Intermontane basin centers appear to be composed of flat lying, relatively high amplitude reflectors suggestive of playa evaporites. Low amplitude wadi sands, with their associated braided character, occur as packages of inclined valley fills and constructive, on-lapping bajada aprons. Valley fill Norphlet sand deposits and up-dip Smackover strandline facies were identified in addition to the Haynesville sand body geometries. End_of_Record - Last_Page 21--------

Upper Miocene Stevens Sandstone, San Joaquin Basin, California: Reinterpretation of a Petroliferous, Sand-Rich, Deep-Sea Depositional System

Deep-marine sands of the upper Miocene Stevens sandstone, one of the most important hydrocarbon-producing units in the United States, were deposited by sediment-gravity flows in the Bakersfield arch area of the southern San Joaquin basin. The Stevens sandstone historically has been considered to be a thick turbidite succession shed off the southern Sierra Nevada as four fans in a long-lived submarine fan system fed in large part by several large submarine canyons. Access to previously unavailable proprietary two-dimensional and three-dimensional seismic data sets, carefully calibrated by well-log and core data, permits a more complete understanding of the depositional architecture of this highly petroliferous, deep-marine depositional system. We conclude that these units were deposited in a structurally controlled, sand-rich deep-sea system, which was fed by a delta system that lacked major submarine canyons. The uppermost sand unit, the upper Stevens sandstone, developed as a deep-sea braid plain in the area of the present Bakersfield arch. Within the braid plain, curvilinear features detected on horizon slices through a three-dimensional seismic data cube are interpreted as braided channel-form deposits. Hydrocarbon production established along these linear trends may reflect improved reservoir quality localized by channel sedimentary processes.

A Delta Submarine Ramp Alternative to the Canyon-Fed Depositional Model of the Stevens Submarine Fan System, Southeastern San Joaquin Basin, Kern County, California: ABSTRACT

Deep-marine sands of the Upper Miocene Stevens Sandstone, one of the most important hydrocarbon-producing units in the United States, were deposited by sediment-gravity flows in the Bakersfield Arch area of the southern San Joaquin basin. The Stevens Sandstone has historically been considered to be a thick turbidite succession shed off the southern Sierra Nevada as four fans in a long-lived submarine fan system fed by several large submarine canyons. Access to previously unavailable proprietary 2-D and 3-D seismic data sets, carefully calibrated by well-log and core data, permits a more complete understanding of the depositional architecture of this highly petroliferous, deep-marine depositional system. This study concludes that these units were deposited in a delta-fed, line- sourced deep-sea system, whose distribution was structurally-controlled. Seismic lines examined in this study show evidence for a large fault-controlled slump feature in the area that has been referred to as [open quotes]Rosedale Canyon,[close quotes] and no evidence supports the existence of submarine canyons feeding the system. The highly progradational Stevens interval consists of thick siliciclastic units separated by thin, intervening biosiliceous shales. Seismically, the upper bounding surfaces of these biosiliceous shales represent major downlap surfaces. As sands were deposited by high-density turbidity currents, the areamore » of the present Bakersfield Arch developed into a deep-sea braid plain. Smaller-scale linear features detected on horizon slices through the 3-D seismic data cube have been interpreted in this study as braided channelform features deposited on the deep-sea braid plain. Hydrocarbon production along these linear trends may be associated with porosity and permeability variations resulting from channelized versus non-channelized sedimentation.« less

On‐line movies of reflection seismic data with description of a movie machine

On‐line movies are an exciting way to view reflection seismic data. We make a movie by slicing through a three‐dimensional (3-D) seismic data cube in a selected direction (Figure 1). By sweeping through the data fast enough, it is possible to get another perspective of what is going on in the data than by examining still frames. By twiddling various knobs and buttons, we can zoom onto a zone of interest, magnify it, rotate it, and change the color to emphasize a feature of interest.