Geophysical Prospecting - Volume 71, Issue 1, 2022

Besides classical imaging techniques, full‐waveform inversion is an increasingly popular method to derive elastic subsurface properties from seismic data. High‐resolution velocity models can be obtained, and spatial sampling criteria are less strict than for imaging methods, because the entire information content of the seismic waveforms is used. As high operational costs arise from seismic surveys, the acquirable data volume is often limited by economic criteria. By selecting optimal locations for seismic sources, the information content of the data can be maximized, and the number of sources and thus the acquisition costs can be reduced compared with standard acquisition designs. The computation of such optimized designs for large‐size 3D inverse problems at affordable computational cost is challenging. By using a sequential receiver‐wise optimization strategy, we substantially reduce the computational requirements of the optimization process. We prove the applicability of this method by means of numerical 3D acoustic examples. Optimized source designs for different receiver patterns are computed for a realistic subsurface model, and the value of the designs is evaluated by comparing checkerboard inversion tests with different acquisition designs. Our examples show that inversion results with higher accuracy can be obtained with the optimized designs, regardless of the number of sources, the number of receivers, or the receiver distribution. Larger benefits of the optimized designs are visible when a sparse receiver geometry is used.

Pore‐pressure depletion causes changes in the triaxial stress state. Pore‐pressure depletion in a flat reservoir, for example, can be reasonably approximated as uniaxial compaction, in which the horizontal effective stress change is smaller than the vertical effective stress. Furthermore, the stress sensitivity of velocities can be angle‐dependent. Therefore, time‐lapse changes in reservoir elastic anisotropy are expected as a consequence of production, which can complicate the interpretation of the 4D seismic response. The anisotropic 4D seismic response caused by pore‐pressure depletion was investigated using existing core velocity measurements. To make a direct comparison between the anisotropic 4D seismic response and the isotropic response based only on vertical velocities, pseudoisotropic elastic properties were utilized, and the two responses were compared in terms of a dynamic rock physics template. A comparison of the dynamic rock physics templates indicates that time‐lapse changes in reservoir elastic anisotropy have a noticeable impact on the interpretation of 4D seismic data. Changes in anisotropy as a result of pore‐pressure depletion cause a time‐lapse amplitude variation with offset response as if there is a reduction in VP/VS (i.e., pseudoisotropic VP/VS decreases), although the vertical VP/VS increases. The impact of time‐lapse changes in anisotropy on the amplitude variation with offset gradient was also investigated, and the time‐lapse anisotropy was found to enhance changes in the amplitude variation with offset gradient for a given case.

Multiples have longer propagation paths and smaller reflection angles than primaries for the same source–receiver combination, so they cover a larger illumination area. Therefore, multiples can be used to image shadow zones of primaries. Least‐squares reverse‐time migration of multiples can produce high‐quality images with fewer artefacts, high resolution and balanced amplitudes. However, viscoelasticity exists widely in the earth, especially in the deep‐sea environment, and the influence of Q attenuation on multiples is much more serious than primaries due to multiples have longer paths. To compensate for Q attenuation of multiples, Q‐compensated least‐squares reverse‐time migration of different‐order multiples is proposed by deriving viscoacoustic Born modelling operators, adjoint operators and demigration operators for different‐order multiples. Based on inversion theory, this method compensates for Q attenuation along all the propagation paths of multiples. Examples of a simple four‐layer model, a modified attenuating Sigsbee2B model and a field data set suggest that the proposed method can produce better imaging results than Q‐compensated least‐squares reverse‐time migration of primaries and regular least‐squares reverse‐time migration of multiples.

The quality factor Q is a vital parameter for quantitatively describing the attenuation information of underground reservoirs, which is of great significance for hydrocarbon detection and reservoir characterization. A frequency‐weighted‐exponential (FWE) method utilizing the symmetry index and the characteristic frequency can obtain this parameter. Unfortunately, the constant symmetry index assumption of it reduces the accuracy of Q values under the non‐standard FWE shape. Selecting an optimal symmetry index is also a problem for this method. Hence, the basic idea of a novel Q estimation method is to substitute the symmetry index with the standard deviation of the source and attenuated wavelet spectrums. The added standard deviation varies with the degree of attenuation under different wavelet shapes, which can reduce the effects of the constant assumption. Meanwhile, it is simple to calculate this parameter directly from the spectrums. In this way, the proposed method has wider applicability to various wavelets. The synthetic records show the better performance of the novel method in improving the accuracy of Q values and the resistance to random noise than the FWE method. Furthermore, the results of matching wavelets exhibit the applicability of the proposed method in real data, and the quadratic spectrum simulation method improves the stability of the spectrums. Finally, real data experiments indicate the effectiveness of the proposed method after the above two processing means.

Within the field of seismic data acquisition with active sources, the technique of acquiring simultaneous data, also known as blended data, offers operational advantages. The preferred processing of blended data starts with a step of deblending, that is separation of the data acquired by the different sources, to produce data that mimic data from a conventional seismic acquisition and can be effectively processed by standard methods. Recently, deep learning methods based on the deep neural network have been applied to the deblending task with promising results, in particular using an iterative approach. We propose an enhancement to deblending with an iterative deep neural network, whereby we modify the training stage of the deep neural network in order to achieve better performance through the iterations. We refer to the method that only uses the blended data as the input data as the general training method. Our new multi‐data training method allows the deep neural network to be trained by the data set with the input patches composed of blended data, noisy data with low amplitude crosstalk noise, and unblended data, which can improve the ability of the deep neural network to remove crosstalk noise and protect weak signal. Based on such an extended training data set, the multi‐data training method embedded in the iterative separation framework can result in different outputs at different iterations and converge to the best result in a shorter iteration number. Transfer learning can further improve the generalization and separation efficacy of our proposed method to deblend the simultaneous‐source data. Our proposed method is tested on two synthetic data and two field data to prove the effectiveness and superiority in the deblending of the simultaneous‐source data compared with the general training method, generic noise attenuation network and low‐rank matrix factorization methods.

Separation of diffracted from reflected events in seismic data is still challenging due to the relatively low amplitude of the diffracted wavefield compared to the reflected wavefield as well as the overlap in the kinematics of reflection and diffraction events. A workflow based on deep learning can be a simple and fast alternative, but using training data made by physics‐based modelling is expensive and lacks diversity in terms of noise, amplitude, frequency content and wavelet. This results in poor generalization beyond the training data without retraining and transfer learning. In this paper, we demonstrate successful applications of reflection–diffraction separation using a conventional U‐net architecture. The novelty of our approach is that we do not use synthetic data created from physics‐based modelling, but instead use only synthetic data built from basic geometric shapes. Our domain of application is the pre‐migration common‐offset domain where reflected events resemble local geology and the diffracted wavefield consists of downward convex hyperbolic diffraction patterns. Both patterns were randomly perturbed in many ways while maintaining their intrinsic features. This approach is inspired by the common practice of data augmentation in deep learning for machine vision applications. Since many of the standard data augmentation techniques lack a geophysical motivation, we have instead perturbed our synthetic training data in ways to make more sense from a signal processing perspective or given our ‘domain knowledge’ of the problem at hand. We did not have to retrain the network to show good results on the field data set. The large variety and diversity in examples enabled to trained neural networks to show encouraging results on synthetic and field data sets that were not used in training.

Characterizing the expressions of seismic waves in elastic anisotropic media depends on multiparameters. To reduce the complexity, decomposing the P‐mode wave from elastic seismic data is an effective way to describe the considerably accurate kinematics with fewer parameters. The acoustic approximation for transversely isotropic media is widely used to obtain P‐mode wave by setting the axial S‐wave phase velocity to zero. However, the separated pure P‐wave of this approach is coupled with undesired S‐wave in anisotropic media called S‐wave artefacts. To eliminate the S‐wave artefacts in acoustic waves for anisotropic media, we set the vertical S‐wave phase velocity as a function related to propagation directions. Then, we derive a pure P‐wave equation in transversely isotropic media with a horizontal symmetry axis by introducing the expression of vertical S‐wave phase velocity. The differential form of new expression for pure P‐wave is reduced to second‐order by inserting the expression of S‐wave phase velocity as an auxiliary operator. The results of numerical simulation examples by finite difference illustrate the stability and accuracy of the derived pure P‐wave equation.

Geothermal resources have potential for providing cost‐effective and sustainable energy. Monitoring of production‐induced changes in geothermal reservoirs using seismic waves requires understanding of the elastic properties of the rock and how they change due to injection of fluids and opening and closing of natural and hydraulic fractures. P‐ and S‐wave velocities measured in a granitic geothermal reservoir using sonic logging are systematically lower than those predicted using the composition of the rock. Cracks may occur in granitic rocks from tectonic stresses and from the thermal expansion mismatch between differently oriented anisotropic crystals. An isotropic orientation distribution of microcracks causes a significant reduction in both the P‐ and S‐velocities, consistent with the observed sonic P‐ and S‐velocities. Vertical fractures cause a difference in the velocity of vertically propagating shear waves polarized parallel and perpendicular to the fractures. An assumption that the lower measured velocities are caused by the presence of vertical fractures is inconsistent with the sonic data. This is because vertical fractures cause a decrease in slow S‐wave velocity that greatly exceeds the decrease in P‐wave velocity, in contrast to the observed data. The growth of vertical fractures in the geothermal reservoir may be monitored using the difference in velocity of the fast and slow shear waves, while the change in P‐velocity in a crossplot of measured P‐ and slow S‐velocities is useful for estimating the ratio of the normal‐to‐shear compliance of the fractures.

The attenuation of seismic waves propagating in reservoirs can be obtained accurately from the data analysis of vertical seismic profile in terms of the quality‐factor Q. The common methods usually use the downgoing wavefields in vertical seismic profile data. However, the downgoing wavefields consist of more than 90% energy of the spectrum of the vertical seismic profile data, making it difficult to estimate the viscoacoustic parameters accurately. Thus, a joint viscoacoustic waveform inversion of velocity and quality‐factor is proposed based on the multi‐objective functions and analysis of the difference between the results inverted from the separated upgoing and downgoing wavefields. A simple separating step is accomplished by the reflectivity method to obtain the individual wavefields in vertical seismic profile data, and then a joint inversion is carried out to make full use of the information of the individual wavefields and improve the convergence of viscoacoustic full‐waveform inversion. The sensitivity analysis of the different wavefields to the velocity and quality‐factor shows that the upgoing and downgoing wavefields contribute differently to the viscoacoustic parameters. A numerical example validates our method can improve the accuracy of viscoacoustic parameters compared with the direct inversion using full wavefield and the separate inversion using upgoing or downgoing wavefield. The application on real field data indicates our method can recover a reliable viscoacoustic model, which helps reservoir appraisal.

Ambient noise tomography is a powerful tool that has found increasing application in reservoir analysis and imaging. The Cachar fold belt in lower Assam, northeast India encompasses several wells under active hydrocarbon production, along with several dry wells. To overcome the lack of active seismic data over the entire fold belt, a passive seismic study was carried out to image the concealed three‐dimensional sub‐surface structures. The data were recorded from February to November 2011 by a network of 65 wideband seismometers spanning an area of about 40 × 60 km². The data are crosscorrelated in the 2–5 s band, followed by phase‐weighted stacking to estimate noise correlation functions with surface wave signatures. The traveltimes picked from the frequency‐time analysis are utilized in a tomographic inversion for Rayleigh wave group velocities. The group velocity anomalies have a lateral resolution of ~ 3.5 × 5.5 km² and variations of up to for each period. The group velocities are in turn inverted for S‐wave velocity distribution as a function of depth. The three‐dimensional S‐wave velocity tomograms reveal the tight anticlines and broad synclines, with high‐ and low‐velocity zones corresponding to structural highs and lows, respectively. The structural interpretation is supported for the part of the region with producing wells and covered by active seismic data, wherein the post‐stack time migrated seismic section shows anticlinal and synclinal features similar to those obtained from ambient noise tomography. The structures revealed by ambient noise tomography can help identify zones of interest to be targeted by active seismic surveys in the Cachar fold belt.

This note compares different approaches to estimating acoustic plane‐wave components from a set of traces recorded by receivers at known locations. The parameters for these plane waves are estimated following somewhat different approaches:

• 1.  A heuristic approach without reference to any optimization schemes).
• 2.  A frequency‐domain least‐squares approach:
  • With explicit derivation of separate real and imaginary parts of the complex‐valued wave‐field parameters.
  • Using complex partial differential operators.

All approaches give the same solution. Of the three, the first approach gives a better understanding of the physics – the last approach is far more direct and easier. That the results are identical argues for our use of Wirtinger complex partial differential operators for our least‐squares minimization problem. Using these operators allows the complex numbers z and to be considered independent variables, giving .