Seismic Structural Interpretation Research Articles

Distilling Knowledge From an Ensemble of Convolutional Neural Networks for Seismic Fault Detection

Fault detection is a crucial task in seismic structure interpretation. Convolutional neural network (CNN)-based methods, in general, require large amount of labeled data for network training. One way to build the labeled data is to create synthetic seismic images with corresponding fault labels. However, it is hard to ensure that the synthetic data have the same fault feature distributions as the field data, which may lead to inaccurate and unreliable prediction results. Another way is to manually label the faults, which is time-consuming and subjective. In this letter, we propose that using knowledge distillation (KD) to improve the performance of fault detection by integrating the features from large number of synthetic samples and a small number of field samples. We distill knowledge from an ensemble of two teacher CNNs to train a student CNN (applied to final target) for seismic fault detection. In our work, one segmentation teacher CNN is trained on synthetic samples with known ground truth fault labels and another classification teacher CNN is trained on field samples with manually picked labels. Then, a classification student network is trained on samples generated by voting the results from two teacher models. The student CNN learns not only the general fault characteristics in the synthetic data but also the specific fault features of the target field data. Test on the field data shows that the student CNN highlights seismic fault more accurately with higher resolution than the teacher CNNs.

Common-azimuth seismic data fault analysis using residual UNet

Seismic fault interpretation is one of the key steps for seismic structure interpretation, which is a time-consuming task and strongly depends on the experience of the interpreter. Aiming to automate fault interpretation, we have considered it as an image segmentation issue and adopt a solution using a residual UNet (ResUNet), which introduces residual units to UNet. Using the ResUNet model, we develop a fault-versus-azimuth analysis based on offset vector tile data, which, as common-azimuth seismic data, provide more detailed and useful information for interpreting seismic faults. To avoid manual efforts for picking training labels and the inaccuracy introduced by different interpreters, we use synthetic seismic data with a random number of faults with different locations and throws as the training and validation data sets. ResUNet is finally trained using only synthetic data and tested on field data. Field data applications show that the proposed fault-detection algorithm using ResUNet can predict seismic faults more accurately than coherence- and UNet-based approaches. Moreover, geologic fault interpretation results computed using common-azimuth data exhibit higher lateral resolution than those computed using poststack seismic data.

Deep learning for relative geologic time and seismic horizons

Constructing a relative geologic time (RGT) image from a seismic image is crucial for seismic structural and stratigraphic interpretation. In conventional methods, automatic RGT estimation from a seismic image is typically based on only local image features, which makes it challenging to cope with discontinuous structures (e.g., faults and unconformities). We have considered the estimation of 2D RGT images as a regression problem, where we design a deep convolutional neural network (CNN) to directly and automatically compute an RGT image from a 2D seismic image. This CNN consists of three parts: an encoder, a decoder, and a refinement module. We train this CNN by using 2080 pairs of synthetic input seismic images and target RGT images, and then we test it on 960 testing seismic images. Although trained with only synthetic images, the network can generate accurate results on real seismic images. Multiple field examples show that our CNN-based method is significantly superior to conventional methods, especially in dealing with complex structures such as crossing faults and complicatedly folded horizons, without the need of any manual picking.

An improved local phase variation attribute and its application in channels detection

Instantaneous phase is a commonly used attribute for structural and stratigraphic feature characterization. The conventional calculation method is to construct the complex-valued seismic trace, then get the ratio of the imaginary part to the real part and finally compute the antitangent of the ratio as the instantaneous phase attribute. In this way, the phase result at one time sample point is the total phase rotation from the beginning of the trace to this point, which means the traditional instantaneous phase is cumulative. Furthermore, the phase obtained by arctangent is usually entangled, which makes it more difficult to apply to seismic interpretation. To address the two issues above, we proposed a new way to calculate the improved local phase variation attributes. Firstly, we calculate traditional instantaneous phase and unwrap it. Then we set a time window on the unwrapped phase to compute the local phase variation by using some difference methods. Finally, we slide the time window on the whole trace to obtain the final phase variation attributes. This strategy turns the whole cumulative value into local variational value, which makes the obtained local phase variation nearly zero in the continuous region but changed greatly at the interface or the abnormal structure areas. Tested by the numerical model and the real data, the proposed attributes have a good application effect in channel detection, which provides a train of thought to seismic structure interpretation with phase attributes.

Seismic integrated study for offshore deep- and thin-stratum characterization: A case study from the Enping sag within the Pearl River Mouth Basin

In the Enping 17 sag within the Pearl River Mouth Basin in the South China Sea, one wildcat well has been drilled to the Lower Paleogene Enping Formation (FM EP) and partially into the Wenchang Formation (FM WC) for deep formation hydrocarbon exploration. However, no commercial play was discovered. The reasons for this are clear if the petroleum systems modeling is examined. In FM EP, the main reason for failure is due to poor sealing. In FM WC, the failure is due to the lack of a good reservoir for hydrocarbon accumulation. Encountering a 9 m thick reservoir at a depth of 4650 m indicates that braided fluvial delta and lowstand turbidite sandstone may develop in FM WC. With the objective of establishing cap rock in FM EP and reservoir rock in FM WC, and in the absence of sufficient well data, an integrated framework for 3D seismic reservoir characterization of offshore deep and thin layers was developed. The workflow includes seismic data reprocessing, well-log-based rock-physics analysis, seismic structure interpretation, simultaneous amplitude variation with offset (AVO) inversion, 3D lithology prediction, and geologic integrated analysis. We present four key solutions to address four specific challenges in this case study: (1) the application of adaptive deghosting techniques to remove the source and streamer depth-related ghost notches in the seismic data bandwidth and the relative amplitude-preserved bandwidth extension technique to improve the seismic data resolution; (2) a practical rock-physics modeling approach to consider the formation overpressure for pseudoshear sonic log prediction; (3) interactive and synchronized workflow between prestack 3D AVO inversion and seismic processing to predict a 9 m thick layer in FM WC through more than 60 rounds of cyclic tests; and (4) cross validation between seismic qualitative attributes and quantitative inversion results to verify the lithology prediction result under the condition of insufficient well data.

Building realistic structure models to train convolutional neural networks for seismic structural interpretation

Seismic structural interpretation involves highlighting and extracting faults and horizons that are apparent as geometric features in a seismic image. Although seismic image processing methods have been proposed to automate fault and horizon interpretation, each of which today still requires significant human effort. We improve automatic structural interpretation in seismic images by using convolutional neural networks (CNNs) that recently have shown excellent performances in detecting and extracting useful image features and objects. The main limitation of applying CNNs in seismic interpretation is the preparation of many training data sets and especially the corresponding geologic labels. Manually labeling geologic features in a seismic image is highly time-consuming and subjective, which often results in incompletely or inaccurately labeled training images. To solve this problem, we have developed a workflow to automatically build diverse structure models with realistic folding and faulting features. In this workflow, with some assumptions about typical folding and faulting patterns, we simulate structural features in a 3D model by using a set of parameters. By randomly choosing the parameters from some predefined ranges, we are able to automatically generate numerous structure models with realistic and diverse structural features. Based on these structure models with known structural information, we further automatically create numerous synthetic seismic images and the corresponding ground truth of structural labels to train CNNs for structural interpretation in field seismic images. Accurate results of structural interpretation in multiple field seismic images indicate that our workflow simulates realistic and generalized structure models from which the CNNs effectively learn to recognize real structures in field images.

Structural Inversion Magnitude and its Impacts on the Hydrocarbon Accumulation

Abstract West Wadi El-Rayan is located in the Western Desert at about 140 km SE of Cairo. Also, it lies between Gindi basin to the east and Abu Gharadig basin to the west. In order to construct a 3D structural model and to delineate the subsurface structure styles of the area, seismic structural interpretation and structural restoration are used. The structural geometry within the area is inverted half-graben, since the area was controlled by reactivation of older faults. The magnitude of the inversion-related shortening in the study area was estimated and was suggested to be strong. The result of the strong inversion magnitude occurred toward northeast of the study area can be concluded that, the area suffered shortening and part of the Jurassic / Early Cretaceous normal faults are reactivated as reverse faults. Also the cap, the main reservoirs and the source rock sections are brought to the surface and thus breached, as well any previous mature source rock becoming non-generative where the dry wells are located. However, any less severe inversion structure in this case where producing wells are located that remain buried and will have a better chance or preserving the structure geometry and therefore top and lateral seal.

Multitask learning for local seismic image processing: fault detection, structure-oriented smoothing with edge-preserving, and seismic normal estimation by using a single convolutional neural network

SummaryFault detection in a seismic image is a key step of structural interpretation. Structure-oriented smoothing with edge-preserving removes noise while enhancing seismic structures and sharpening structural edges in a seismic image, which, therefore, facilitates and accelerates the seismic structural interpretation. Estimating seismic normal vectors or reflection slopes is a basic step for many other seismic data processing tasks. All the three seismic image processing tasks are related to each other as they all involve the analysis of seismic structural features. In conventional seismic image processing schemes, however, these three tasks are often independently performed by different algorithms and challenges remain in each of them. We propose to simultaneously perform all the three tasks by using a single convolutional neural network (CNN). To train the network, we automatically create thousands of 3-D noisy synthetic seismic images and corresponding ground truth of fault images, clean seismic images and seismic normal vectors. Although trained with only the synthetic data sets, the network automatically learns to accurately perform all the three image processing tasks in a general seismic image. Multiple field examples show that the network is significantly superior to the conventional methods in all the three tasks of computing a more accurate and sharper fault detection, a smoothed seismic volume with better enhanced structures and structural edges, and more accurate seismic normal vectors or reflection slopes. Using a Titan Xp GPU, the training processing takes about 8 hr and the trained model takes only half a second to process a seismic volume with $128\, \times \, 128\, \times \, 128$ image samples.

Seismic structure interpretation based on machine learning: A case study in coal mining

Interpretation of geologic structures entails ambiguity and uncertainties. It usually requires interpreter judgment and is time consuming. Deep exploitation of resources challenges the accuracy and efficiency of geologic structure interpretation. The application of machine-learning algorithms to seismic interpretation can effectively solve these problems. We analyzed the theory and applicability of five machine-learning algorithms. Seismic forward modeling is a key connection between the model and seismic response, and it can obtain seismic data of known geologic structures. Based on the modeling data, we first optimized the seismic attributes sensitive to the target geologic structure and then we verified the accuracy of the five machine-learning algorithms by the cross-checking method. In this case, the random forest algorithm had the highest accuracy. So we examined the structural interpretation method based on a random forest using the 3D seismic reflection data from coalfield exploration. The prediction effect of this interpretation workflow is verified by comparison with known geologic structures on the plane and profile. The results suggest that the random forest algorithm is feasible to indicate geologic structure interpretations in the case of collapsed column and fault structures and it can effectively improve the efficiency of seismic interpretation and its accuracy. The machine-learning-based workflow provides a new technique for seismic structure interpretation in coal mining.

Fault image enhancement using a forward and backward diffusion method

Fault interpretation is a key step for seismic structural interpretation and reservoir characterization. In conventional approach, faults are identified as reflection discontinuity or abruption in post-stack seismic data. In order to highlight fault, several fault attributes by measuring waveform similarity or diversity have been proposed. However, noise and stratigraphic features can also be highlighted in a fault attribute image. We propose a forward and backward diffusion method to enhance the fault image so that the noisy features which unrelated to faults are suppressed and the fault features are sharpened. We first estimate the planarity (linearity in 2D) and fault orientation (strike and dip) of every sample in fault image by using a PCA-based algorithm. Combining the planarity and fault orientation, we then construct fault probability map based on the assumption that faults can be seemed as relatively steep planes in a small spatial window. Thanks to the fault probability map, we can adaptively switch the diffusion process from a forward mode to a backward mode. Forward diffusion force is used to suppress the non-fault features within the low fault probability area and backward diffusion force is used to enhance the fault features within the high fault probability area. A real 3D example is included to demonstrate that this method can effectively suppress the noise and sharpen the faults. Furthermore, automatic fault tracking on an enhanced fault image can get more accurate result than on an original fault image.

FaultSeg3D: Using synthetic data sets to train an end-to-end convolutional neural network for 3D seismic fault segmentation

Delineating faults from seismic images is a key step for seismic structural interpretation, reservoir characterization, and well placement. In conventional methods, faults are considered as seismic reflection discontinuities and are detected by calculating attributes that estimate reflection continuities or discontinuities. We consider fault detection as a binary image segmentation problem of labeling a 3D seismic image with ones on faults and zeros elsewhere. We have performed an efficient image-to-image fault segmentation using a supervised fully convolutional neural network. To train the network, we automatically create 200 3D synthetic seismic images and corresponding binary fault labeling images, which are shown to be sufficient to train a good fault segmentation network. Because a binary fault image is highly imbalanced between zeros (nonfault) and ones (fault), we use a class-balanced binary cross-entropy loss function to adjust the imbalance so that the network is not trained or converged to predict only zeros. After training with only the synthetic data sets, the network automatically learns to calculate rich and proper features that are important for fault detection. Multiple field examples indicate that the neural network (trained by only synthetic data sets) can predict faults from 3D seismic images much more accurately and efficiently than conventional methods. With a TITAN Xp GPU, the training processing takes approximately 2 h and predicting faults in a [Formula: see text] seismic volume takes only milliseconds.

Superimposed characteristics and genetic mechanism of strike-slip faults in the Bohai Sea, China

Abstract Based on the 3D seismic structure interpretation of Bohai Sea, combined with physical modeling of structure, structural style analysis and apatite fission track simulation, the structural characteristics and genetic mechanism of the Cenozoic strike-slip faults in Bohai Sea were investigated. The results show that Tanlu strike-slip fault experienced three stages of strike-slip activities in the Cenozoic, and the transition from left-lateral strike to right-lateral strike-slip was completed at the end of the fourth member of the Shahejie Formation. The strike-slip faults in the Bohai Sea have the characteristics of multi-stage and multi-strength stress superposition. According to the superimposed forms of different strengths, different properties and different ratios, they can be divided into three major genetic types, extension and strike-slip superimposition, extension and extrusion superimposition, extrusion and strike-slip superimposition, and fifteen typical structure patterns. Affected by multiple changes in the direction and rate of subduction of the Cenozoic Pacific plate, the difference between the Cenozoic extension and the strike-slip in the Bohai Sea area leads to the diversity of the fault system and the zoning of the depression structure. According to superimposition features of faults, the Bohai Sea area can be divided into the Liaoxi S-type weak strike-slip zone, Liaodong braided strong strike-slip zone, Boxi conjugated medium strike-slip zone, Bodong brush structure medium strike-slip zone and Bonan parallel strong strike-slip zone. These zones differ in oil and gas accumulation features.

Integrated analysis of fault sealing capabilities when studying and mapping hydrocarbon traps in the Dnieper-Donets Depression

This paper demonstrates how a set of advanced technologies provided a better understanding of the geological structure of the Solokha gas condensate field and underground gas storage located in onshore Ukraine. In particular, these technologies helped us improve the process of hydrocarbon trap delineation and subsequent exploration well planning in this area. The study area is structurally complex because it is located in the central part of the axial zone of the Dnieper-Donets Depression (DDD). This part of the DDD is characterized by active halokinesis that has produced morphologically developed salt bodies and associated with them plicative and disjunctive deformations of the host rocks. The structure of the area has to a large extent been formed under the impact of one of such bodies, the closely located and actively developed Bakeyski diapir. The workflow used consists in synergic integration of high-quality 3D seismic data processing, detailed structural and stratigraphic interpretation and seismic inversion combined with pore pressure measurements in wells, lithotype clustering using Kohonen self-organizing maps, and fault seal analysis. A ramified network of faults, which plays an important role in the study area, formed independent tectonic traps in the Middle Jurassic sediments of the Bayosian age and led to the compartmentalization of one of the stratigraphic traps in the Lower Carboniferous sediments of the Visean age. To analyze sealing capabilities of these faults, Allan diagrams and Shale Gouge Ratio are used. The first of these methods is based on the juxtaposition of permeable and impermeable rocks on both sides of the fault, and the second, more advanced, is based on the calculation of the percentage of the clay component of all rocks that have slipped along the fault past a given point.

Successful leveraging of image processing and machine learning in seismic structural interpretation: A review

As a process that identifies geologic structures of interest such as faults, salt domes, or elements of petroleum systems in general, seismic structural interpretation depends heavily on the domain knowledge and experience of interpreters as well as visual cues of geologic structures, such as texture and geometry. With the dramatic increase in size of seismic data acquired for hydrocarbon exploration, structural interpretation has become more time consuming and labor intensive. By treating seismic data as images rather than signal traces, researchers have been able to utilize advanced image-processing and machine-learning techniques to assist interpretation directly. In this paper, we mainly focus on the interpretation of two important geologic structures, faults and salt domes, and summarize interpretation workflows based on typical or advanced image-processing and machine-learning algorithms. In recent years, increasing computational power and the massive amount of available data have led to the rise of deep learning. Deep-learning models that simulate the human brain's biological neural networks can achieve state-of-the-art accuracy and even exceed human-level performance on numerous applications. The convolutional neural network — a form of deep-learning model that is effective in analyzing visual imagery — has been applied in fault and salt dome interpretation. At the end of this review, we provide insight and discussion on the future of structural interpretation.

Nonlinear anisotropic diffusion techniques for seismic signal enhancing - Carpathian Foredeep study

The use of nonlinear anisotropic diffusion algorithm for advanced seismic signal processing in the complicated geological region of Carpathian Foredeep was examined. This technique allows for an improvement of seismic data quality and for more accurate interpretation by the recovery of a significant amount of structural information in the form of a correlating seismic reflections and by preserving true DHI indicators. It also allows searching for more subtle geological structures. Anisotropic diffusion is an iterative image processing algorithm that removes noise by modifying the data by solving partial differential equations. Moreover, it can reduce image noise without blurring the edges between regions of different chrominance or brightness. This filter preserves edges, lines, or other features relevant to the seismic structural and stratigraphic interpretation. The algorithm also enables noise reduction without removing significant information from a seismic section even for high dips values. For a better estimation of anisotropic diffusion structure tensor, the parameterization is done using the depth field and the calculations in the two-way travel time field. The presented research shows the results of using an anisotropic diffusion algorithm for post-stack and migration processing of seismic 3D data collected in Carpathian reservoir rocks of southern Poland.

The impact of multifocusing in the processing of land 3D seismic data in a fold and thrust belt setting: Ranquil Norte Block, Neuquén Basin, Argentina

A total of 270 km2 of 3D seismic data was acquired in the southeast corner of the Ranquil Norte Block in the north of the fold and thrust belt area of the Neuquén Basin, west of Argentina. The survey was acquired in an area with hard topography, disturbing anthropomorphic conditions, and very complex subsurface geology. All of these situations negatively contributed to the signal-to-noise ratio of the resulting data. Therefore, and considering a structural target only, the main objective of the seismic processing sequence was the mitigation of the impact of such geologic and anthropomorphic noise on the data. Initially, a tailored processing sequence was performed, including noise suppression in the cross-spread domain. This first processing approach included Kirchhoff prestack time migration, structural seismic interpretation, velocity modeling, tomography, and Kirchhoff prestack depth migration. Given that multifocusing can deliver high-resolution subsurface images in areas with similar challenging conditions, this processing technique was then included in the processing workflow. Naturally, this second approach necessarily involved a reinterpretation of the structural model, a rebuilding of the velocity model, and a new tomographic computation. A significant improvement of the results was achieved in this manner.

Effects of Partial Melting on Seismic Velocity and Attenuation: A New Insight from Experiments

The effects of partial melting on seismic velocity and attenuation have long been studied by focusing on the direct effects of melt, such as the poroelastic effect. The direct effects are generally very small for a very small melt fraction. Because geochemical studies have shown that the melt fraction during partial melting is very small (∼0.1%), it is difficult to explain upper-mantle low-velocity regions by the direct effects of melt. Recent experimental studies, by using a rock analog, have captured a significant enhancement of polycrystal anelasticity just before partial melting in the absence of melt. This newly recognized effect enables us to interpret seismological and geochemical observations consistently. The new anelasticity model significantly changes the interpretation of upper-mantle seismic structures. This review summarizes the recent progress in the understanding of polycrystal anelasticity, starting from a basic knowledge of linear anelasticity.

Evaluation of geological conditions for coalbed methane occurrence based on 3D seismic information: a case study in Fowa region, Xinjing coal mine, China

The research on geological conditions of coalbed methane (CBM) occurrence is of great significance for predicting the high abundance CBM rich region and gas outburst risk area pre-warning. The No. 3 coal seam, in Yangquan coalfield of Qinshui basin, is the research target studied by 3D seismic exploration technique. The geological factors which affect CBM occurrence are interpreted based on the 3D seismic information. First, the geological structure (faults, folds, and collapse columns) is found out by the 3D seismic structural interpretation and the information of buried depth and thickness of the coal seam is calculated by the seismic horizons. Second, 3D elastic impedance (EI) and natural gamma attribute volumes are generated by prestack EI inversion and multi-attribute probabilistic neural network (PNN) inversion techniques which reflect the information of coal structure types and lithology of the roof and floor. Then, the information of metamorphic degree of seam and hydrogeology conditions can be obtained by the geological data. Consequently, geological conditions of CBM occurrence in No. 3 coal seam are evaluated which will provide scientific reference for high abundance CBM rich region prediction and gas outburst risk area pre-warning.

Improvement of seismic structural interpretation of Zagros fold-thrust belt by dip scanning in common diffraction surface imaging method

Geological interpretation and seismic imaging on data acquired from complex media, especially when accompanied by a thick layer of salt and/or anhydrate, is always a controversial task. The aim of this study is to introduce a method for seismic imaging of complex structures and study the folding scenario of the Zagros overthrust in SW Iran based on the improved seismic image processed by the proposed method. A diffraction based version of the common reflection surface stack method, called the common diffraction surface (CDS), was selected to enhance quality of seismic image from the Zagros fold-thrust belt. The 2D CDS method was also developed here to 3D partial CDS. Consequently, three sets of 2D and a set of 3D seismic data were processed with the conventional Kirchhoff prestack time migration and the partial common diffraction surface methods. Geological models released in previous studies of the Zagros overthrust were used as the base model of interpretation. Structural interpretation on seismic sections with new method was in accordance with previous geological models and improved in some parts where the new method could enhance quality of the seismic image. Seismic structural interpretation shows the inner part of the belt was subjected to folding, uplift and erosion. This mechanism constituted a resistant mass in the Dezful Embayment in front of the SW moving thrust and folds. This resistant mass formed tightly folded or thrusted and highly shortened structures in the Izeh zone and High Zagros area. In the oil rich zone of the Dezful Embayment, the thick evaporate—rich Gachsaran Formation, decouples the folding above it from the folding below, making seismic imaging of potential traps beneath the detachment horizons extremely difficult.

Complete wavefield seismic structural, stratigraphic, geomorphic and elastic interpretation in the offshore Browse Basin

The first quantitative interpretation project in the Browse Basin that used complete wavefields from dual-sensor streamer data to characterise a variety of overburden and reservoir seismic structural, sedimentary and geomorphic features is discussed in this extended abstract. Having established that complete wavefield pre-stack angle data are amplitude versus angle (AVA) compliant, the authors analyse the elastic response from thin gas sands in a location where well control predicts a weak gas response in terms of the Vp/Vs ratio and negligible response in terms of the acoustic impedance. Various seismic expressions in the near-surface are better resolved using complete wavefield imaging, including deltaic deposits, potentially prospective patch reefs and intersecting channel systems, and shallow hazards associated with historic drilling failures appear to be resolvable with greater confidence. On the larger scale, the overall data interpretability is substantially enhanced in texture and character by broadband imaging of primary reflections, and then by subsequent reservoir characterisation. Initial (deeper) interpretation focused around mapping of the prospective Early Jurassic Plover Formation and the Early Cretaceous Upper Vulcan Formation, in which evidence of channelisation can be demonstrated on conventional migration volumes and reinforced using complete wavefield migration. Ultra-low frequency signal penetration to the deeper Triassic level has been limited historically, but the first complete seismic wavefield data analysis presented demonstrates that a combination of volume attributes provide substantially improved geologic information from the seafloor to the deepest targets.