Field Seismic Data Research Articles

Fault3DNnet: A lightweight 3D seismic fault detection network with bidirectional decoding

Fault detection from seismic data is a critical component of hydrocarbon exploration and production. In recent years, deep-learning techniques have made tremendous achievements and breakthroughs in seismic fault detection. Especially U-nets, represented by FaultSeg3D and its variants, are a dominant method for 3D deep-learning-based seismic fault interpretation. The incorporation of state-of-the-art network modules or training strategies with U-nets continuously improves the performance of fault detection for field seismic data, usually at the cost of increased computational intensity and hardware consumption. We design a lightweight bidirectional decoding network, Fault3DNnet, for fault detection from 3D seismic data volume. To make it simple and concise, our Fault3DNnet is equipped with basic conventional convolution and regular operation modules. The two decoding branches in Fault3DNnet could provide complementary information and improve fault detection performance. The superiority of Fault3DNnet is verified using publicly available synthetic and four field seismic data sets. The experimental results indicate that the method predicts fault structures with improved completeness and continuity compared with FaultSeg3D. Meanwhile, the parameter and computational quantity of the network are only 1/7 and 1/5 of FaultSeg3D, respectively. With the same graphics processing unit memory consumption, Fault3DNnet can handle the size of approximately six times the volume of 3D seismic data during inferencing as that of FaultSeg3D. It helps to improve prediction stability and decrease discontinuous artifacts at the stitching boundaries when predicting the large field data in chunks. In general, our Fault3DNnet delivers superior fault detection results while greatly reducing computational intensity and memory usage.

Time-lapse difference seismic inversion based on scattering theory

Time-lapse seismic monitoring technology is an effective tool for quantitatively estimating subsurface medium variations and plays a crucial role in resource and engineering explorations. Among existing time-lapse inversion methods, the difference inversion technique offers higher accuracy but requires linear inversion operators. During linearization, the background is often approximated as a homogeneous medium, resulting in the loss of effective information. Additionally, merely using reflection wave information is not conducive to improving inversion accuracy. However, the true structure of the subsurface is often complex and inhomogeneous, which can easily generate complex wavefields, such as scattering waves, thereby hindering the resolution of differences between time-lapse models. Therefore, incorporating more comprehensive wavefield information is essential for enhancing the accuracy of time-lapse seismic inversion. To address this issue, a wave-equation-based time-lapse difference inversion method is proposed to retrieve the production-induced property perturbation. In the proposed method, the forward operator is formulated through the scattering wave equation, the operator is linearized through the Born approximation, and the WKBJ method is used to approximate Green's function with high accuracy for the inhomogeneous background. In this study, time-lapse seismic parallel inversion and difference inversion are conducted using the proposed forward operator. Both numerical experiments and application of field seismic data indicate that the proposed time-lapse difference inversion method can realize a more accurate quantitative characterization of time-lapse variations.

Addressing cycle‐skipping in full‐waveform inversion using acoustic wave energy

AbstractLimitations in acquisition technologies lead to insufficient low‐frequency signals in field seismic data. Local optimization methods are the common approaches for full‐waveform inversion. Inaccurate initial velocity models and lack of low‐frequency signals in seismic data typically cause the local‐gradient‐based full‐waveform inversion to converge to a local minimum due to cycle‐skipping. The existing energy‐based objective functions can generate artificial low‐frequency signals successfully by squaring the pressure but overlook the law of energy conservation, which may mislead model updates. To overcome this issue, we combine acoustic wave potential energy and kinetic energy to develop a new objective function that fits the acoustic wave energy. The new acoustic‐wave‐energy‐based full‐waveform inversion considers the law of energy conservation. The new system creates low‐frequency signals to avoid cycle‐skipping and produce an accurate smooth background velocity model, which provides a sufficient starting model for conventional full‐waveform inversion. Numerical examples demonstrate that the combination of acoustic‐wave‐energy‐based full‐waveform inversion and conventional full‐waveform inversion can deliver more faithful and accurate final results than conventional full‐waveform inversion alone.

The linear canonical W-transform

The time-frequency spectrum of seismic data is commonly used as an attribute in the analysis and interpretation of nonstationary seismic signals. To improve the capability of the W-transform to analyze nonstationary data, we introduce the linear canonical W-transform (LCWT) by implementing an affine matrix along with scaling parameters to generate high resolution and energy-concentrated time-frequency spectra. More features of nonstationary signals can be explored in the time-linear canonical frequency domain by transforming the time-frequency plane, which makes the algorithm more applicable to seismic signals. For the completeness of the LCWT, we also develop the inverse LCWT to extend its applications. Examples of synthetic and field seismic data show that a highly resolved and energy-concentrated time-frequency spectrum can be estimated with the LCWT with negligible additional computational burden. In addition, our algorithm is promising for several seismic data processing and interpretation techniques, such as reservoir characterization and thin layer identification.

Evolution of the transtensional Barreirinhas pull-apart system in the Brazilian Equatorial margin and its correlation with the African conjugate counterpart

The Barreirinhas pull-apart system encompasses marginal basins in divergent and transform margin segments in the central sector of the Brazilian Equatorial Margin and its African conjugate counterpart. This ancient pull-apart system evolved through transtensional strike-slip motion within a highly heterogeneous crystalline basement affected by multiple rift phases. The geometry and development of pull-apart structural elements during the final rifting phase before continental breakup and the mechanisms and extent to which they were influenced by preexisting crustal heterogeneities are comprehensively addressed using an extensive database of potential field (magnetic and gravity) and 2D seismic reflection data. We also assess the lithospheric thermomechanical conditions and their influence on transtensional extension throughout Curie Point Depth, Heat Flow, and Moho depth, derived from potential field data and published seismological models. Plate reconstruction of Brazilian and African equatorial margins based on gravity patterns and comparison with sandbox analog models allow a 3D synoptic model to reveal the Barreirinhas pull-apart system evolution during the Equatorial Atlantic opening. During the rift phase I, the location of major grabens was controlled by favorably oriented Neoproterozoic shear zones, while the cooler, stronger, and thicker crust beneath cratonic areas formed the western barrier to strike-slip rift activity during rift phase II. This same geological domain anchored the onset of the pull-apart system in the last rift phase III, whose principal displacement zones developed along the extensive oceanic fracture zones linked by sigmoidal fault systems. Toward the end of the rift phase, a large asymmetric lozangle to a lazy-Z-shaped, pull-apart basin developed above low overlapping ∼90°, releasing stepover in oblique transtensional strike-slip motion.

Automatic seismic first‐break picking based on multi‐view feature fusion network

AbstractAutomatic first‐break picking is a basic step in seismic data processing, so much so that the quality of the picking largely determines the effect of subsequent processing. To a certain extent, artificial intelligence technology has solved the shortcomings of traditional first‐break picking algorithms, such as poor applicability and low efficiency. However, some problems still remain for seismic data, with a low signal‐to‐noise ratio and large first‐break change leading to inaccurate picking and poor generalization of the network. In order to improve the accuracy of the automatic first‐break picking results of the above seismic data, we propose a multi‐view automatic first‐break picking method driven by multi‐network. First, we analysed the single‐trace boundary characteristics and the two‐dimensional boundary characteristics of the first break. Based on these two characteristics of the first break, we used the Long Short‐Term Memory and the ResNet attention gate UNet (resudual attention gate UNet) networks to extract the characteristics of the first arrival and its location from the seismic data, respectively. Then, we introduced the idea of multi‐network learning in the first‐break picking work and designed a feature fusion network. Finally, the multi‐view first‐break features extracted by the Long Short‐Term Memory and resudual attention gate UNet networks are fused, which effectively improves the picking accuracy. The results obtained after applying the method to field seismic data show that the accuracy of the first break detected by a feature fusion network is higher than that given by the above two networks alone and has good applicability and resistance to noise.

Assessment of the seismic failure of reinforced concrete structures considering the directional effects of ground motions

Seismic intensity measures at different levels significantly impact the seismic damage of building clusters. Reinforced concrete (RC) structures are a type of building widely used in different regions worldwide. A large amount of field seismic damage observation data indicates that within the same seismic intensity zone, the directional effects of varying ground motions lead to significant differences in damage to RC structures on different floors. However, relatively little research has been conducted on estimating the seismic fragility and vulnerability of RC structures considering the directional effects of seismic action. To further study the directional effect of seismic motion on the damage of different floors, this study conducted dynamic time history and spectral analyses on 1472,379 acceleration records in multiple directions obtained from monitoring at nine actual seismic stations during the 2008 Wenchuan earthquake in China. This paper innovatively reports on the actual damage features of RC structures during the Mw6.2 magnitude Jishishan earthquake in Gansu Province, China, on December 18, 2023 (the most severe destructive earthquake in China in 2023) and analyses the influence of earthquake directionality on the damage to different floors of RC structures. A three-dimensional numerical model of a four-story RC frame structure was established using numerical simulation and the finite element method. Different intensity measures were taken to input seismic excitations of different intensities into the RC structure in the 0°, 30°, 60°, and 90° directions, and damage simulation analysis was conducted. The nonlinear dynamic time history curves of structures under multiple directions of seismic excitation in different intensity zones have been developed. Using the interstory stiffness and interstory displacement angle as engineering demand parameters, damage assessment curves and stress clouds for RC structures with 1–4 floors considering different seismic directional effects were generated. The analysis results and major findings indicate that the seismic sequence in the 0° direction has a relatively heavy impact on the first floor of the RC structure in zone IX, which is consistent with the actual field observation results.

Synchrosqueezing voices through deep neural networks for horizon interpretation

Horizon picking is a crucial element in reservoir characterization, but it remains a labor-intensive process. Manually interpreting horizons across thousands of vertical seismic slices in a 3D seismic survey significantly increases the time and labor required for this task. Although several automated methods for extracting the horizons in seismic images have been developed, their effectiveness can be hampered by interruptions in lateral continuity, such as faults and noise. In addition, closely spaced horizons pose a challenge, making it even more difficult to recognize their exact position. To track the horizons through a 3D seismic volume, other seismic attributes extracted from the 3D seismic data must be used. We suggest using spectral voice components together with the original seismic amplitudes to track the surfaces of the target horizons. We generate the time-frequency spectrum using a high-resolution method, namely the synchrosqueezing wavelet transform (SWT), and the real part of the complex SWT spectrum is the voice component. We import the spectral voice component and the seismic amplitude into a neural network. A deep convolutional neural network is used to track the horizon surfaces within a 3D seismic volume. We evaluate this application on a field seismic data set where closely spaced thin layers are located within a complex faulted formation with noisy seismic data with a low signal-to-noise ratio. The integration of the amplitude and phase within the attribute of the voice component indicates that it improves the quality of the generated horizons, especially compared with using the seismic amplitude only for this task. An example from field data demonstrates the ability of our method to accurately delineate the horizons across fault surfaces and in the immediate vicinity of unconformities. This exceeds the current limits of the existing methods for horizon detection.

Re‐visible blind block network: An unsupervised seismic data random noise attenuation method

AbstractNoise is inevitable when acquiring seismic data, and effective random noise attenuation is crucial for seismic data processing and interpretation. Training and inferencing two‐stage deep learning‐based denoising methods typically require massive noisy–clean or noisy–noisy pairs to train the network. In this paper, we propose an unsupervised seismic data denoising framework called a re‐visible blind block network. It is a training‐as‐inferencing one‐stage method and utilizes only single noisy data for denoising, thereby eliminating the effort to prepare training data pairs. First, we introduce a global masker and a corresponding mask mapper to obtain the denoised result containing all blind block information, enabling simultaneous optimization of all blind blocks via the loss function. The global masker consists of two complementary block‐wise masks. It is utilized to mask noisy data to obtain two corrupted data, which are then input into the denoising network for noise attenuation. The mask mapper samples the value of blind blocks in the denoised data and projects it onto the same channel to gather the denoised results of all blind blocks together. Second, the original noisy data are incorporated into the network training process to prevent information loss, and a hybrid loss function is employed for updating the network parameters. Synthetic and field seismic data experiments demonstrate that our proposed method can protect seismic signals while suppressing random noise compared with traditional methods and several state‐of‐the‐art unsupervised deep learning denoising techniques.

Spectral analysis of seismic data based on a combined generalized S transform

The S transform is widely used to delineate seismic spectral anomalies associated with hydrocarbons during reservoir characterization. To overcome the relatively poor time-frequency resolution due to the fixed varying pattern of the window used in the S transform, we present a combined generalized S transform (CGST) for the spectral analysis of seismic data. First, the CGST decomposes the signal into a series of intrinsic mode functions (IMFs) using a complete ensemble empirical mode decomposition (CEEMD). An objective function is defined to quantitatively determine the spectral energy concentration criteria for optimal parameter selection of the improved Gaussian windows. Then, the time-frequency spectra for all the IMFs within narrower frequency bands are calculated using the respective Gaussian windows with normalized energy and individual optimal controlling parameters. We applied the CGST to spectral analyses of both synthetic signals and seismic field data. The results indicate that the CGST exhibits good performance with a higher time-frequency resolution. Field examples illustrate that the CGST can be used to characterize the spectral behavior of thin beds and indicate low-frequency shadows associated with hydrocarbon reservoirs.

A self‐supervised scheme for ground roll suppression

AbstractIn recent years, self‐supervised procedures have advanced the field of seismic noise attenuation, due to not requiring a massive amount of clean labelled data in the training stage, an unobtainable requirement for seismic data. However, current self‐supervised methods usually suppress simple noise types, such as random and trace‐wise noise, instead of the complicated, aliased ground roll. Here, we propose an adaptation of a self‐supervised procedure, namely, blind‐fan networks, to remove aliased ground roll within seismic shot gathers without any requirement for clean data. The self‐supervised denoising procedure is implemented by designing a noise mask with a predefined direction to avoid the coherency of the ground roll being learned by the network while predicting one pixel's value. Numerical experiments on synthetic and field seismic data demonstrate that our method can effectively attenuate aliased ground roll.

Seismic data AVO analysis in time frequency domain using synchroextracting transform

Time-frequency analysis is a crucial tool for processing and interpreting seismic data in hydrocarbon exploration. Accurately determining the temporal and spectral characteristics is essential for applications such as reservoir characterization and fluid identification. High-resolution representation of seismic data time-frequency properties improves reliability in these applications. Recent advancements in post-processing techniques aim to enhance conventional methods. The synchroextracting transform (SET) is an innovative approach that combines a post-processing procedure with the short-time Fourier transform. It retains time-frequency information associated with spectral components while eliminating excessive energy dispersed in the time-frequency domain. This study evaluates the effectiveness of SET for seismic data time-frequency analysis, comparing it with contemporary post-processing techniques using synthetic signals and field seismic data. Results show that SET provides higher resolution for distinct signal components and effectively reduces energy dispersion. SET exhibits advantages such as robustness against random noise, reasonable computational cost, and offering more interpretable time-frequency information regarding attenuated amplitudes. Building on synthetic experiments, we demonstrate the capability of SET to generate high-resolution single frequency sections of seismic data, facilitating accurate analysis of amplitude variation with offset (AVO) in the time-frequency domain.

3D seismic Fault Detection via Contrastive-Reconstruction Representation Learning

Fault detection is a critical step in structural modeling and characterizing reservoirs. 3D seismic fault labeling is almost impossible to obtain, while networks trained on synthetic fault data have limited generalization to real data. We use self-supervised representation learning to enable networks to learn seismic field data features during pre-training, improving generalization. However, widely popular ViT/SwinViT-based methods cannot capture low-level fault features well. CNN-based have limited capability in transferring to downstream tasks. Tackle this, we designed the Tiny Self-Attention and extensively embedded it into the HRNet. It merges the advantages of convolutional and self-attention, allowing for in-depth learning of seismic representation information during the representation learning phase, thus achieving better inter-class separation in downstream tasks. We crafted two proxy tasks. The contrastive task aims to minimize the projected feature distance of overlapping areas from two distinct views. To tackle the issue of memory overflow and training suspension caused by the high spatial and temporal complexity of 3D high-resolution feature matching, we introduced sparse distance matching and an adaptive feature aggregation. In reconstruction task, we designed a masking strategy tailored to the unique attributes of 3D seismic data. It process named “Fault Detection via Contrast-Reconstruction Representation Learning” (FaultCRL). Experimentally, we compared the latest fault detection methods, and representation learning techniques like MAE and SwinUNETR, showcasing the notable advantage of FaultCRL in fault detection tasks. The appendix provides all qualitative results from mainstream geophysical journals regarding The Netherlands-F3 data in recent years, indicating that our approach has reached the current state-of-the-art level.

Improving Seismic Fault Recognition with Self-Supervised Pre-Training: A Study of 3D Transformer-Based with Multi-Scale Decoding and Fusion

Seismic fault interpretation holds great significance in the fields of geophysics and geology. However, conventional methods of seismic fault recognition encounter various issues. For example, models trained on synthetic data often exhibit inadequate generalization when applied to field seismic data, and supervised learning is heavily dependent on the quantity and quality of annotated data, being susceptible to the subjectivity of interpreters. To address these challenges, we propose applying self-supervised pre-training methods to seismic fault recognition, exploring the transfer of 3D Transformer-based backbone networks and different pre-training methods on fault recognition tasks, thereby enabling the model to learn more powerful feature representations from extensive unlabeled datasets. Additionally, we propose an innovative pre-training strategy for the entire segmentation network based on the characteristics of seismic data and introduce a multi-scale decoding and fusion module that significantly improves recognition accuracy. Specifically, during the pre-training stage, we compare various self-supervision methods, like MAE, SimMIM, SimCLR, and a joint self-supervised learning approach. We adopt multi-scale decoding step-by-step fitting expansion targets during the fine-tuning stage. Ultimately merging features to refine fault edges, the model displays superior adaptability when handling narrow, elongated, and unevenly distributed fault annotations. Experiments demonstrate that our proposed method achieves state-of-the-art performance on Thebe, the currently largest publicly annotated dataset in this field.

Hypercomplex Processing of Vector Field Seismic Data: Toward vector-valued signal processing [Hypercomplex Signal and Image Processing

Hypercomplex Processing of Vector Field Seismic Data: Toward vector-valued signal processing [Hypercomplex Signal and Image Processing

Vertical bearing capacity of spun precast concrete pile in liquefiable soil: a case study

Spun precast prestressed concrete (SPC) pile has been used in many parts of the world as a viable foundation alternative. This study assesses the load-carrying capacity of SPC piles passing through a deep soft subsoil layer and resting on a dense sand layer. The vertical bearing capacity of the SPC pile has been estimated using various analytical methods and static pile load tests for the Jolshiri area of Dhaka. The liquefaction potential of the site has been assessed using local seismic site conditions, field and laboratory test data. The liquefaction analysis suggests that the topsoil layer is liquefiable to a depth of 4.5 m. The capacity obtained from the load test is compared with those obtained from the different methods, ultimate push-in load, and local design guidelines. Finite Element Analysis (FEA) was performed considering the hardening soil (HS) model using PLAXIS 3D to simulate field conditions. The load settlement response obtained from FEA shows a good agreement with the test results. The capacity examinations primarily suggest that the SPC pile can be a viable foundation solution for the subsoil conditions of Jolshiri Abasion area of Dhaka.

Integrated characterization of deep karsted carbonates in the Tahe Oilfield, Tarim Basin

Abstract As transport channels of oil and gas, fracture networks can greatly improve reservoir seepage, which is of great significance to the hydraulic fracturing and hydrocarbon deposit exploitation in petroleum science and engineering. In this paper, our target reservoirs are deep karsted carbonates at depths of >6000 m and highly heterogeneous, leading to complex seismic responses with weak energy and low resolution. Therefore, it is challenging to predict the spatial distribution of a carbonate fracture-cavern reservoir and to characterize its delicate structure. We present a characterization method for an excellent fracture description by integrating several attribute results on 3D seismic field data. First, we use a noise elimination method to remove the noise interference in seismic data without damaging the fault structure characteristics. Next, we propose a novel spatially windowed 2D Hilbert transform-based operator to perform volumetric edge detection on 3D seismic field data. Then, the volumetric edge results are co-rendered with other seismic geometric attributes to generate multi-attribute fusion results for a comprehensive prediction that can excellently delineate geologic anomalies at different scales in deep carbonates. The results indicate that integrating several scale attributes can obtain more rich geological discontinuity and reveal more subtle fractures than using single attribute. The multi-attribute fusion results can effectively delineate some small-medium-sized faults, and they provide practical support for the exploration and production of Tahe carbonate fracture-cavern reservoirs.

A knowledge-embedded close-looped deep-learning framework for intelligent inversion of multisolution problems

Deep learning is prevalent in many fields and attempts have been made to use it in nonbidirectional mapping problems, such as seismic inversion. These nonbidirectional mapping problems have two special issues, that is, insufficient labels and uncertainty of solution. Therefore, current deep-learning structures are not suitable for handling this kind of problem. A distinctive knowledge-embedded close-looped (KECL) deep-learning framework is developed, tuned to the characteristics of the seismic inverse problem. The KECL deep-learning framework is composed of a reservoir parameter generator (RPG) and a reservoir parameter updater (RPU). The former half-loop is RPG, which takes the seismic data as input to generate the initial reservoir parameters. The latter loop is RPU, which takes the initial parameters as input to output synthetic seismic data. Through the training by well data, the difference between field seismic data and synthetic seismic data modeled by the RPU is used to optimize the RPG and RPU. In this deep-learning framework, knowledge of the Robinson convolutional model is embedded to address the problem of insufficient labels. Furthermore, semisupervised learning is used as prior information to reduce the uncertainty of solution. After the training, with the help of prior geologic information data, the RPU is used to update the initial reservoir parameters generated by RPG for final reservoir parameter inversion. Numerical models and field data are used to test the feasibility of our deep-learning framework. We find that intelligent inversion results using data from one well to train the KECL network are consistent with results using multiple well data. Experiments demonstrate that it is adaptable to situations in which insufficient well data are available and is able to achieve reliable intelligent inversion.

Two-dimensional synchrosqueezing transform and its application to tight channel sandstone reservoir characterization

Spectral analyses can effectively highlight the morphological characteristics of tight-channel sandstone reservoirs, which are crucial development targets in the field of unconventional oil and gas. A majority of these methods focus on trace-by-trace analyses of seismic data without lateral constraints. However, they do not depict the morphological characteristics of the channels in detail. Therefore, this study proposes a novel time-frequency (TF) analysis (TFA) method: two-dimensional synchrosqueezing transform (TDSST). The TDSST introduces a space-time window to capture the lateral variation of seismic signals. Then, from a two-dimensional (2D) purely harmonic signal, the instantaneous frequency (IF) and instantaneous wavenumber (IK) estimation formulas were strictly derived in the two-dimensional short-time Fourier transform domain. Furthermore, we define a new two-dimensional synchrosqueezing operator (TDSTO) that can squeeze the TF coefficients into the estimated IF and IK trajectories. Finally, the TDSST can effectively reflect the vertical and horizontal changes, and according to the relationship between the wave number and frequency, the wave number parameter can be determined by maximizing the energy of the TDSST. TDSST can provide a highly energy-concentrated TF representation (TFR) while allowing the reconstruction of signals. Compared to other TFA methods, TDSST can provide a more laterally continuous spectral decomposition that clearly describes tight-channel sandstone reservoirs, even in a strongly noisy environment. The reliability of TDSST is validated through the utilization of both synthetic model and seismic field data.

Time-lapse seismic inversion for CO2 saturation with SeisCO2Net: An application to Frio-II site

Seismic monitoring of geological CO2 storage (GCS) involves highly nonlinear seismic inversion and petrophysical inversion, making it challenging to estimate CO2 volume efficiently and detect possible early CO2 leakages. Deep learning (DL) using convolutional neural networks (CNNs) has shown promise in solving highly nonlinear seismic inversion problems. However, direct estimation of CO2 plume extent/saturation from time-lapse seismic gathers using DL is still underexplored, with no reported field applications to date. The investigation of field data is primarily hindered by scarcity of field data for neural network training. Other obstacles include highly nonlinear seismic-petrophysics inverse relationship, and presence of noise in field seismic data. We introduce SeisCO2Net, a deep CNN that predicts CO2 saturation maps directly from time-lapse full waveform shot gathers. For training, we use site-specific geological information, fluid flow physics, rock physics, and seismic modeling to generate synthetic datasets that closely resemble the CO2 storage site. Synthetic tests show promising results, inspiring us to apply SeisCO2Net's trained weights on field data collected at Frio-II GCS site by leveraging transfer learning principles. As reference, we compare SeisCO2Net's predicted CO2 saturation maps with results obtained from physics-based inversion. Our analyses show both methods display similar CO2 plume shapes, reasonable CO2 plume characteristics, and comparable saturation values. Our results suggest pre-training CNNs on physics-informed synthetic datasets and then applying the learned weights to field data is a viable approach to estimating field CO2 saturation. This method effectively addresses the scarcity of field training data, thus encouraging the feasibility of long-term GCS monitoring.