Simulated Seismic Data Research Articles

Bayesian rock-physics inversion using a localized ensemble-based approach — With an application to the Alvheim field

We describe, implement, and show the results of a localized ensemble-based approach for seismic amplitude-variation-with-offset (AVO) inversion with uncertainty quantification. Ensembles are simulated from prior probability distributions for fluid saturations and clay content. Starting with continuous saturations and clay content variables, we use depth-varying models for cementation and grain contact theory, Gassmann fluid substitution with mixed saturations, and approximations to the Zoeppritz equations for the AVO attributes at the top-reservoir. The local conditioning to seismic AVO observations relies on (1) the misfit between ensemble simulated seismic AVO data and the field observations in a local partition of the grid/local patch, of inlines/crosslines around the locations where we aim to predict, (2) correlations between the simulated reservoir properties and the data in local patches, and (3) local assessment to avoid unrealistic updates based on spurious correlations in the ensembles. Data from the Alvheim field in the North Sea are used to demonstrate the approach. The influence of the prior information from the well logs in combination with the seismic reflection data indicates the presence of higher oil and gas saturation in the lobe structures of the field and increased clay content at their edges.

Semisupervised seismic impedance inversion with data augmentation and uncertainty analysis

Seismic inversion for impedance is an important task for reservoir characterization. Supervised-learning-based methods improve inversion accuracy for impedance. However, they require a large number of training labels. To reduce the reliance on training labels and further improve the accuracy of impedance inversion, we propose a new data augmentation strategy to simulate layer thickness variation during semisupervised learning which uses unlabeled seismic data properly. Using the Marmousi-II model, we first compare the impedance estimated using the open-loop (OL) method and that obtained using the physics-constrained closed-loop (PC-CL) method at different data augmentation rates to demonstrate the effectiveness of the proposed inversion method. In the case of applying the proposed method to a field data set, comparisons between the estimated impedance results using the OL and PC-CL methods and those obtained using the model-based inversion method are shown for verification. Detailed comparisons between the inversion results of the impedance and well-log data are displayed to emphasize the superiority of the proposed PC-CL method. The residuals between the observed field data and simulated seismic data generated using the estimated impedance are shown to further verify the accuracy of the estimated impedance of the proposed PC-CL method. Aside from the impedance inversion results, epistemic uncertainties are calculated for further performance evaluations of the impedance inversion.

Corrosion-related parameter estimation for RC structures using UKF-based Bayesian nonlinear finite element model updating with seismic data

Model updating based on field measurements (e.g., from ambient vibration) has proven to be an efficient approach for assessing the condition of reinforced concrete (RC) structures. Despite existing studies in the literature, no effort has been devoted to corrosion-related parameter estimation for RC structures based on Bayesian nonlinear finite element model updating (FEMU) with seismic data. To bridge this gap, this research work extends the application of nonlinear FEMU to corroded RC structures and provides a methodology to estimate the corrosion-related parameters from seismic measurement of RC structures with high nonlinearity. The essential idea of this approach is to integrate recursive Bayesian inference (i.e., unscented Kalman filter) with advanced finite element (FE) modeling of corroded RC structures to estimate corrosion-affected properties. The advanced FE modeling approach used in this study is an efficient modeling strategy that accounts for different aspects of the corrosion, particularly corroded bond-slip in RC structures. The capability of the approach in evaluating the corrosion-affected model parameters is examined by 13 case studies using an example RC column with simulated seismic data due to lack of real-word measurements. These study cases consider different corrosion levels, measurement noise levels, corroded features, seismic intensity levels, and types of measurement. The successful estimation of the corrosion-affected properties for the corroded column under various scenarios indicates that the proposed methodology can be employed to identify the unknown corroded properties of RC structures using recorded seismic response data. The updated FE model of a corroded RC structure can potentially be utilized for reliable seismic performance assessment, which assists in future decision-making for repair or retrofit of existing structures.

Full-waveform inversion with adversarial losses via deep learning

Full-waveform inversion (FWI) is a powerful technology for determining geophysical properties, nevertheless, with high computational cost and difficulty in convergence. Motivated by the powerful tools available in modern deep learning system to deal with complex non-linear inverse problems, we implement the FWI workflow with deep neural networks and propose the use of two different neural networks with adversarial losses to provide fast and accurate solutions. Specifically, a deep neural network is used to generate simulated seismic data based on the finite-difference method, and another fully connected neural network is used to fit the Wasserstein distance between simulated and observed seismic data. The Wasserstein distance and the L2 metric together form an adversarial loss function. The parallel-layer model and the SEAM model are used to verify the benefits of the proposed method and its performance in the face of seismic data lack low-frequency components and noise interference. Numerical experiments demonstrate that, compared with other algorithms, the adaptive gradient (Adagrad) algorithm performs best. In addition, the Adagrad algorithm with adversarial losses has faster convergence speed and more inverting details than with L2 losses. The proposed method provides reliable inversion results, even when seismic data without low-frequency components. Moreover, the adversarial losses are less sensitive to noise, although still difficult to converge.

Rapid Estimation of Earthquake Fatalities in Mainland China Based on Physical Simulation and Empirical Statistics-A Case Study of the 2021 Yangbi Earthquake.

At present, earthquakes cannot be predicted. Scientific decision-making and rescue after an earthquake are the main means of mitigating the immediate consequences of earthquake disasters. If emergency response level and earthquake-related fatalities can be estimated rapidly and quantitatively, this estimation will provide timely, scientific guidance to government organizations and relevant institutions to make decisions on earthquake relief and resource allocation, thereby reducing potential losses. To achieve this goal, a rapid earthquake fatality estimation method for Mainland China is proposed herein, based on a combination of physical simulations and empirical statistics. The numerical approach was based on the three-dimensional (3-D) curved grid finite difference method (CG-FDM), implemented for graphics processing unit (GPU) architecture, to rapidly simulate the entire physical propagation of the seismic wavefield from the source to the surface for a large-scale natural earthquake over a 3-D undulating terrain. Simulated seismic intensity data were used as an input for the fatality estimation model to estimate the fatality and emergency response level. The estimation model was developed by regression analysis of the data on human loss, intensity distribution, and population exposure from the Mainland China Composite Damaging Earthquake Catalog (MCCDE-CAT). We used the 2021 Ms 6.4 Yangbi earthquake as a study case to provide estimated results within 1 h after the earthquake. The number of fatalities estimated by the model was in the range of 0–10 (five expected fatalities). Therefore, Level IV earthquake emergency response plan should have been activated (the government actually overestimated the damage and activated a Level II emergency response plan). The local government finally reported three deaths during this earthquake, which is consistent with the model predictions. We also conducted a case study on a 2013 Ms7.0 earthquake in the discussion, which further proved the effectiveness of the method. The proposed method will play an important role in post-earthquake emergency response and disaster assessment in Mainland China. It can assist decision-makers to undertake scientifically-based actions to mitigate the consequences of earthquakes and could be used as a reference approach for any country or region.

A Method for Denoising Seismic Signals With a CNN Based on an Attention Mechanism

Suppressing random noise in seismic data is a significant problem in seismic data processing. Often, there is serious aliasing between the effective signal and random noise, affecting the identification of weak signals, and even resulting in great difficulties in the suppression of conventional seismic signals. We propose an improved attention-guided convolutional neural network (ADNet) to eliminate seismic interference noise. After a sufficient amount of training, the network removes noise by transferring seismic data features learned from a synthetic dataset to tests with complex field data. Our workflow consists of four parts. First, in the model, we improve the feature enhancement module (FEM) and attention module (AM), increase the convergence speed, and enhance the expressive ability. Second, we use 2-D synthetic data to verify the ability of the model to suppress noise in seismic records. Third, we use 2-D real seismic data to further verify the denoising effect of the improved ADNet. Fourth, we convert the 3-D simulated seismic data and field data into 2-D data for processing and reorganize the 2-D denoising results into 3-D data. By comparing the noise suppression outcomes of several classic denoising methods, simulations and actual experiments show that the improved ADNet effectively maintains the signal amplitude, reduces the network depth, and better suppresses seismic noise. Hence, we believe that our model can be widely applied in the field of seismic data processing.

An unsupervised deep-learning method for porosity estimation based on poststack seismic data

We propose to invert reservoir porosity from poststack seismic data using an innovative approach based on deep-learning methods. We develop an unsupervised approach to circumvent the requirement of large volumes of labeled data sets for a conventional learning process. We apply convolutional neural networks (CNN) on seismic data to predict the relative porosity that is to be added to a low-frequency prior component. We then apply a forward model to synthesize seismic data based on a source wavelet and an acoustic impedance converted from the network-determined porosity. The parameters in the CNN are iteratively updated to minimize the error between recorded and simulated seismic data. We test the capability of our deep-learning approach to estimate reservoir porosity using a synthetic rock-physics model with two different signal-to-noise ratios. We also apply the proposed method to a real case study of seismic data acquired for hydrocarbon exploration of clastic reservoirs in the Vienna Basin. Instead of randomly assigning neural parameters, we use pretrained weights and biases at a previous location as initialization values for the next location, to preserve the geologically lateral continuity of the layers’ physical properties. As shown by these analyses, the unsupervised CNN-based scheme provides more or equally accurate results than standard methods for porosity estimation from seismically inverted acoustic impedance, which makes it a promising tool in seismic reservoir characterization with less user intervention.

A scalable online algorithm for passive seismic tomography in underground mines

Understanding and monitoring the seismic responses of rock masses to massive mining are crucial for safety and economic viability of ever larger and deeper underground operations. Seismic monitoring can be used to detect stress variations and hazardous instabilities, but its effectiveness requires accurate estimations of the nonhomogeneous propagation velocity of microseismic waves. While predetermined velocity models are not accurate enough and might bias hypocenter localization, using active-source seismic tomography methods to estimate the velocity field provides limited spatial coverage. Thus, passive seismic tomography using first-arrival traveltimes of mining-induced microseisms (of unknown hypocenters) constitutes a promising tool. However, available methods solving this high-dimensional statistical inverse problem do not scale well with the data set size and cannot easily refine or update estimations with new data. We have developed a novel passive seismic tomography method able to dynamically learn the nonhomogeneous velocity field from a streaming of noisy first-arrival times, online (in real time) or from catalogs. We have developed a new Bayesian approach that avoids linearizing the forward problem and allows for general 3D velocity models. This is combined with the use of the stochastic gradient descent (SGD) method, which underlies much of the recent progress in machine learning and provides increasing accuracy at a cost scaling linearly with the data set size. Moreover, we introduce an adaptive variant of SGD based on raypath density, which significantly improves the speed of the algorithm, and we implement a parallel version of our method enabling its systematic use in real applications. These include the design of optimal sensor locations, the dynamic update of velocity estimates in production conditions, and the real-time determination of hypocenters and their uncertainty. Our method’s reach and effectiveness are illustrated with simulated seismic data on 3D checkerboards, using synthetic and real acquisition geometries, and on a dense 2D velocity grid.

Near-surface scattered waves enhancement with source-receiver interferometry

Near-surface scattered waves (NSWs) are the main noise in seismic data in areas with a complex near surface and can be divided into surface-to-surface scattered waves and body-to-surface scattered waves. We have developed a method for NSW enhancement that uses modified source-receiver interferometry. The method consists of two parts. First, deconvolutional intersource interferometry is used to cancel the common raypath of seismic waves from a near-surface scatterer to the common receiver and the receiver function. Second, convolutional interreceiver interferometry is used to compensate the common raypath of seismic waves from the common source to the near-surface scatterer and the source function. For an isotropic point scatterer near the earth’s surface in modified source-receiver interferometry, a body-to-surface scattered wave can be reconstructed by constructive interference not only among three body-to-surface scattered waves but also among a body-to-surface scattered wave and two surface-to-surface scattered waves; a surface-to-surface scattered wave can be reconstructed by constructive interference not only among three surface-to-surface scattered waves but also among a surface-to-surface scattered wave and two body-to-surface scattered waves. According to stationary phase analysis based on the superposition principle, we have developed a so-called dual-wheel driving configuration of modified source-receiver interferometry for enhancing NSWs in the data of conventional seismic exploration. The main advantages of the scheme are that (1) it can be used to enhance NSWs without the need for any a priori knowledge of topography and near-surface velocity, (2) it can be used to reconstruct NSWs from real sources to real receivers, including 3D near-surface side-scattered waves, and (3) it can be applied to conventional seismic data with finite-frequency bandwidth, spatially limited and sparse arrays, different source and receiver functions, and static correction. Numerically simulated data and field seismic data are used to demonstrate the feasibility and effectiveness of the scheme.

Least-squares reverse time migration with and without source wavelet estimation

Least-squares reverse time migration (LSRTM) attempts to find the best fit reflectivity model by minimizing the mismatching between the observed and simulated seismic data, where the source wavelet estimation is one of the crucial issues. We divide the frequency-domain observed seismic data by the numerical Green's function at the receiver nodes to estimate the source wavelet for the conventional LSRTM method, and propose the source-independent LSRTM based on a convolution-based objective function. The numerical Green's function can be simulated with a dirac wavelet and the migration velocity in the frequency or time domain. Compared to the conventional method with the additional source estimation procedure, the source-independent LSRTM is insensitive to the source wavelet and can still give full play to the amplitude-preserving ability even using an incorrect wavelet without the source estimation. In order to improve the anti-noise ability, we apply the robust hybrid norm objective function to both the methods and use the synthetic seismic data contaminated by the random Gaussian and spike noises with a signal-to-noise ratio of 5dB to verify their feasibilities. The final migration images show that the source-independent algorithm is more robust and has a higher amplitude-preserving ability than the conventional source-estimated method.

Simulated Annealing for Sequential Pattern Detection and Seismic Applications

Sequential pattern detection with simulated annealing (SA) is adopted to estimate parameters and detect lines, ellipses, hyperbolas type by type, and patterns by patterns in each type. The motivation of the sequential detection method is to deal with multiple patterns. The parameters of a pattern are formed as a vector and used as a state, and adjusted in SA. A sequential detection algorithm using SA to detect patterns is proposed. It detects one or a small number of patterns at each step. SA has the capability of the global minimization. The six parameters of patterns are adjusted sequentially step by step. The computation can converge efficiently. In the experiment, the result of sequential detection is better than that of synchronous detection in detecting a large number of patterns. In sequential detection, detection of one pattern at each step can have less computation time and good convergence in total detection than using two or more pattern detections. In simulated seismic data, SA is applied to detect the hyperbolas in the common depth point (CDP) gather. In real one-shot seismogram, SA is applied to detect lines of direct wave and hyperbolas of reflection wave. The results can show that the proposed method is feasible.

Model-reduction techniques for Bayesian finite element model updating using dynamic response data

This work presents a strategy for integrating a class of model reduction techniques into a finite element model updating formulation. In particular a Bayesian model updating approach based on a stochastic simulation method is considered in the present formulation. Stochastic simulation techniques require a large number of finite element model re-analyses to be performed over the space of model parameters during the updating process. Substructure coupling techniques for dynamic analysis are proposed to reduce the computational cost involved in the dynamic re-analyses. The effectiveness of the proposed strategy is demonstrated with identification and model updating applications for finite element building models using simulated seismic response data.

Composition of seismic facies: A case study

In this case study, we used simulated seismic data from outcrops on Svalbard to analyze what seismic facies are composed of, what the dominating factors in forming the facies are, and which consequences this has for the interpretation results. Seismic facies analyses can be used to interpret environmental setting, depositional processes, and lithology. Here, we found that noise is the most important factor in forming the seismic facies. Noise is defined as all reflections that cannot be ascribed directly to the reservoir model. Effects from overburden and processing dominated, and the low-frequency content of the seismic section complicated the seismic facies analyses. The main reason for this is that the analysis relies heavily on identified internal patterns and low-angle terminations. Such patterns and terminations are easily created by the seismic method itself, by overburden effects, and by artifacts generated when processing the data. External form, strong amplitudes, and continuous reflections are robust seismic observations, whereas the internal pattern and terminations are commonly deceptive. Identification of boundaries based on predefined patterns of terminations does not work here, and uncritical use of seismic facies analysis in this interpretation case will create wrong reservoir models. Because of the size of the outcrops, the results from this analysis are relevant for reservoir-scale seismic interpretation and detailed interpretation for prospect evaluation in mature basins. For seismic interpretation at a more regional scale, it is probably less relevant.

Subtraction of Newtonian noise using optimized sensor arrays

Fluctuations in the local Newtonian gravitational field present a limit to high precision measurements, including searches for gravitational waves using laser interferometers. In this work, we present a model of this perturbing gravitational field and evaluate schemes to mitigate the effect by estimating and subtracting it from the interferometer data stream. Information about the Newtonian noise is obtained from simulated seismic data. The method is tested on causal as well as acausal implementations of noise subtraction. In both cases it is demonstrated that broadband mitigation factors close to 10 can be achieved removing Newtonian noise as a dominant noise contribution. The resulting improvement in the detector sensitivity will substantially enhance the detection rate of gravitational radiation from cosmological sources.

Geophysical parameter estimation

Seismic exploration for oil and gas is a parameter estimation problem. Geological properties and fluid content of subsurface reservoirs are sensed from the earth's surface. Seismic data are acquired by generating elastic waves at the surface and recording the reflections off subsurface targets using large receiver arrays. At present, it is not practical to determine earth parameters by iteratively refining an earth model until simulated seismic data match recorded field data. The computational demands of accurately simulating viscoelastic wave propagation in a 3-D heterogeneous earth are prohibitive. Instead, seismic data are processed to remove many complexities of wave propagation in the real environment, like multipath, attenuation, dispersion, shear waves, and interface waves. With these complexities removed, the processed data conform to simpler propagation models which can be solved in reasonable computer time. The cost of each simplification, however, is the loss of some information about the subsu...

The numerical simulation and imaging of seismic scattered wave applied to base metal exploration

The seismic data collected during base metal exploration are often degraded by a variety of non-reflecting signals due to the heterogeneity of ore bodies and host rocks. It is then difficult to generate a high-quality seismic section when conventional reflection seismic processing algorithms are applied to the data. Two seismic records have been simulated from constructed typical ore geological models using the Kirchhoff integration principle. One geological model is the Tongling copper mine in eastern China, and another is the Gejiu (also known as Gegeo) tin mine in the Yunnan Province of south-western China. The simulated seismic data has then been processed using reflection seismic and seismic scatter imaging methods, respectively. The seismic scatter imaging method, which stacked all kinds of reflected/diffracted waves, seems more powerful to image the complexity of the subsurface.

Potential risks of spectrum whitening deconvolution — Compared with well-driven deconvolution

Deconvolution is widely used to increase the resolution of seismic data. To compare the resolution ability of conventional spectrum whitening deconvolution to thin layers with that of well-driven deconvolution, a complex sedimentary geological model was designed, and then the simulated seismic data were processed respectively by each of the two methods. The amplitude spectrum of seismic data was almost white after spectrum whitening, but the wavelet resolution was low. The amplitude spectrum after well-driven deconvolution deviated from white spectrum, but the wavelet resolution was high. Further analysis showed that if an actual reflectivity series could not well satisfy the hypothesis of white spectrum, spectrum whitening deconvolution had a potential risk of wavelet distortion, which might lead to a pitfall in high resolution seismic data interpretation. On the other hand, the wavelet after well-driven deconvolution had higher resolution both in the time and frequency domains. It is favorable for high resolution seismic interpretation and reservoir prediction.

Real-Time Seismic Signal Enhancement Utilizing a Hybrid Rao–Blackwellized Particle Filter and Hidden Markov Model Filter

This letter outlines a novel and robust algorithm for identifying seismic events within low signal-to-noise ratio (SNR) passive seismic data in real time. Since the event detection problem is a continuous, real-time process which has nonlinear mathematical representations, a Rao-Blackwellized particle filter (RBPF) is utilized. In this algorithm, a jump Markov linear Gaussian system (JMLGS) is defined where changes (i.e., jumps) in the state-space system and measurement equations are due to the occurrences and losses of events within the measurement noise. The RBPF obtains optimal estimates of the possible seismic events by individually weighting and subsequently summing a bank of Kalman filters (KFs). These KFs are specified and updated by samples drawn from a Markov chain distribution which defines the probability of the individual dynamical systems which compose the JMLGS. In addition, a hidden Markov model filter is utilized within the RBPF filter formulation so that real-time estimates of the phase of the seismic event can be obtained. The filter is demonstrated to provide up to an 80-fold improvement in the SNR when processing simulated seismic data with Gauss-Markov measurement noise.

Prediction analysis of long-term memory effect for calamity gray series

It is adequate to use the gray theory for modeling and forecasting short-term calamity series. The forecast of calamity gray series is equivalent to predicting an extraordinary event in nature. In order to look for the regularity, the calamity date series, created from the threshold for a fixed time-interval series, are studied. In this paper, the Hurst exponent is applied to defining the long-term memory effect of the simulated calamity series, and is tested for the feasibility of using it as pre-requisite information before the gray modeling and forecasting. Based on the fractional Brownian motion (fBm) model, the time series with a definite length or quantity of data are derived assuming that various kinds of memory effect exist. Different threshold values are defined to yield or to analogize the calamity date series that are required in the prediction of the gray calamity events. After case study, both of the simulated and real seismic data show that the Hurst exponents are greater than 0.5 and, therefore, indicate that the long-term memory ef-fect exists. The correlation between the Hurst exponent and the gray modeling parameter, a, provides criteria for the classification of the forecast.

Impedance inversion experiments with simulated data

Acoustic impedance inversion experiments using simulated seismic data have been presented applying the Gopinath‐Sondhi (GS) method. The purpose is to invert synthetic seismograms to reconstruct impedance profiles. Seismic data have been perturbed by white noise and a band‐limited Butterworth filter. The effect of the dc component of impulse response on the impedance reconstruction has been investigated. The inversion method can be applied for both onshore and offshore explorations due to a similarity between particle velocity and pressure convolution equations. Computer model experiments show that the method is promising for inverting seismograms contaminated with white noise. The study also shows a problem for inversion due to the missing low frequencies including the dc.