Geophysical Prospecting - Volume 72, Issue 3, 2024

Due to subsurface viscosity and heterogeneity, the vertical resolution of observed seismic data is decreased after wave propagation, generating nonstationary seismic data with amplitude attenuation and phase distortion. Inverse Q filtering techniques are always used to enhance the vertical resolution of seismic data. However, the majority of inverse Q filtering methods treat attenuation compensation trace by trace, which may produce non‐robust compensation results with poor transverse continuity and amplify noise energy in noisy cases. Thus, we develop a novel sparsity‐promoting inversion‐based multichannel seismic data attenuation compensation approach by introducing a sparse constraint for curvelet coefficients of multichannel compensated data, which takes the transverse continuity of compensated data into account. Besides, the proposed method with a sparse constraint for curvelet coefficients has a better noise‐resistance property, which can attenuate the noise energy in noisy cases during attenuation compensation, improving compensation accuracy and robustness. To improve its computational efficiency, a fast iterative shrinkage–thresholding algorithm is adopted to solve the established lasso problem. Synthetic data examples with different noise levels and two post‐stack field data examples validate the effectiveness of the proposed multichannel method. Its compensation results have superior vertical resolution, transverse continuity and noise robustness in comparison to the conventional single‐channel compensation method using a damped least squares algorithm.

The presence of the air–water interface (or free‐surface) creates two major problems in marine seismic data for conventional seismic processing and imaging: free‐surface multiples and ghost reflections. The attenuation of free‐surface multiples remains one of the most challenging noise attenuation problems in seismic data processing. Current solutions suffer from the removal of the primary events along with the multiple events especially when the primary and multiple events overlap (e.g., adaptive subtraction). The effective attenuation of ghost reflections (or deghosting) requires acquisition‐ and/or processing‐related solutions which generally address the source‐side and receiver‐side ghosts separately. Additionally, an essential requirement for a successful implementation of free‐surface multiple attenuation and seismic dehosting is the requirement of dense seismic data acquisition parameters which is not realistic for two‐dimensional and/or three‐dimensional marine cases. We present a convolutional neural network approach for free‐surface multiple attenuation and seismic deghosting. Unlike the existing solutions, our approach operates on a single trace at a time, and neither relies on the dense acquisition parameters nor requires a subtraction process to eliminate free‐surface multiples, and it removes both the source ghost and receiver ghost simultaneously. We train a network using subsets of the Marmousi and Pluto velocity models and make predictions using subsets of the Sigsbee velocity model. We show that the convolutional neural network predictions give a correlation coefficient of 0.97 on average with the numerically modeled data for the synthetic examples. We illustrate the efficacy of our convolutional neural network–based technique using the Mobil AVO Viking Graben field data set. The application of our algorithm demonstrates that our convolutional neural network–based approach removes different orders of free‐surface multiples (e.g., first and second orders) and recovers the low‐frequency content of the seismic data (which is essential for, for instance, full‐waveform inversion applications and broadband processing) by successfully removing the ghost reflections while preserving and increasing the continuity of the primary reflection.

Recently, we established a blended‐acquisition method: temporally signatured and/or modulated and spatially dispersed source array that jointly uses various signaturing and/or modulation in the time dimension and dispersed source array in the space dimension. We acquired and processed the first pilot programme with our method onshore Abu Dhabi. In this paper, we review the resulting acquisition‐productivity enhancement in the time dimension and discuss it in the space dimension as well. We then review the deblended‐data‐reconstruction processing, followed by imaging processing, and discuss their performance and resulting data quality. We last establish a relationship between the acquisition productivity and the processing performance. We found that this method significantly enhances the acquisition productivity compared to conventional blending methods. For the processing performance, the deblended data can successfully be reconstructed from the blended data; afterwards, the subsurface sections can naturally be imaged from the deblended data. Furthermore, this method owns a relationship: The deblended‐data‐reconstruction performance increases with the acquisition time (i.e. the acquisition effort that is inversely proportional to the acquisition productivity) up to a plateau; the imaging processing improves the data quality as the flow makes progress and eventually reaches a high data quality after poststack processing, regardless of the acquisition effort.

A transversely isotropic elastic medium with a vertical axis of symmetry is considered. We obtain dispersion equations in real terms for guided Love and Rayleigh waves in a such a medium consisting of horizontal layers sandwiched between two half‐spaces by brief modifications of the available literatures on dispersion equations in elastic layered media through transfer matrix. To illustrate the applicability, dispersion curves of guided waves are computed for three‐ and five‐layered symmetric models with transversely isotropic coal seam in the middle. The Airy phase is marked by a minimum or maximum of group velocity in a dispersion curve, and this phase is important to get seam structure for mining safety. For the first three modes, the effect of Thomsen anisotropy parameters γ, ε and δ of a coal seam on frequency (fA) and group velocity (UA) of the Airy phase is similar in three‐ and five‐layered models. For guided Love waves, fA and UA have nearly a uniform increase with the increase of γ. For guided Rayleigh waves, the increase of ε causes fA and UA to increase; however, the increase of δ causes fA and UA to decrease.

Trace‐wise noise is a type of noise often seen in seismic data, which is characterized by vertical coherency and horizontal incoherency. Using self‐supervised deep learning to attenuate this type of noise, the conventional blind‐trace deep learning trains a network to blindly reconstruct each trace in the data from its surrounding traces; it attenuates isolated trace‐wise noise but causes signal leakage in clean and noisy traces and reconstruction errors next to each noisy trace. To reduce signal leakage and improve denoising, we propose a new loss function and masking procedure in a semi‐blind‐trace deep learning framework. Our hybrid loss function has weighted active zones that cover masked and non‐masked traces. Therefore, the network is not blinded to clean traces during their reconstruction. During training, we dynamically change the masks' characteristics. The goal is to train the network to learn the characteristics of the signal instead of noise. The proposed algorithm enables the designed U‐net to detect and attenuate trace‐wise noise without having prior information about the noise. A new hyperparameter of our method is the relative weight between the masked and non‐masked traces' contribution to the loss function. Numerical experiments show that selecting a small value for this parameter is enough to significantly decrease signal leakage. The proposed algorithm is tested on synthetic and real off‐shore and land data sets with different noises. The results show the superb ability of the method to attenuate trace‐wise noise while preserving other events. An implementation of the proposed algorithm as a Python code is also made available.

Seismic random noise is one of the main factors that degrade the quality of seismic data. Therefore, seismic random noise attenuation should be performed appropriately through several stages during seismic data processing, and this requires sufficient experience and knowledge because the proper hyperparameters need to be determined based on the features of the noise in the target seismic data. Recently, machine learning–based seismic noise attenuation has been widely studied because it provides suitable results by learning noise features from noisy data, unlike conventional physics‐based approaches. There are many important factors in machine learning, and, here, we focus on the loss functions of machine learning in terms of seismic random noise attenuation. The most widely used loss function is l2, but we train a model with various kinds of single and multiple loss functions and attenuate seismic random noise. We analyse the efficiency of loss functions by comparing the noise‐attenuated results of synthetic and field seismic data qualitatively and quantitatively. Our analysis indicates that the multiple loss function with the l1 norm can be a proper choice for random noise suppression of seismic data.

Seismic geometry design is the first step for the seismic prospecting, which would substantially influence the quality of acquired seismic data. Earlier researchers have focused more on tuning systemic parameters and offsetting source points. However, due to the complex geographical factors, the real layout is often different from the pre‐plan design, which is designed by experts in advance. In this paper, we aim to address the source point deviating problem when the pre‐plan design is unsuitable for geographical conditions. Although there exist many research works in this aspect, few of them have been applied in practice and the offset of source points in the field still mainly relies on manual work. In this paper, we propose a method based on dynamic programming to deviate source points automatically while maintaining the smoothness of source lines and taking the uniformity of fold distribution into account. To verify the effectiveness of our method, we also compare the uniformity of offset and azimuth after automatically deviating source points with the pre‐plan design. Experimental results show that our proposed method achieves better uniformity and keeps more smooth source lines which are conducive to the construction. Moreover, on the data set collected in a mountainous area, we observe that our method slightly suppresses the acquisition footprints.

The critical angle plays a crucial role in data processing for refraction seismology. In the context of three‐dimensional data, the critical angle exhibits azimuthal dependence, particularly in the presence of an anisotropic model. In this paper, we propose a method to determine the critical angle (phase angle) and analyse the sensitivity of the critical angle to the model parameters and the available azimuthal range for both transversely isotropic medium with a vertical symmetry axis and orthorhombic models. For more complex orthorhombic models, the critical angle can be computed while considering changes in azimuth. In a numerical example, we apply sensitivity analysis to examine the existence of the critical angle and determine its corresponding values to variations in model parameters and the azimuthal range. Additionally, we conduct computations of the critical angle for the simplified acoustic and elliptical orthorhombic models. This analysis can be extended to encompass all model parameters for different wave modes (pure and converted waves), and they provide generalized predictions for the available range of data in seismic data processing applications.

The data‐driven Marchenko method is able to redatum wavefields to arbitrary locations in the subsurface, and can, therefore, be used to isolate zones of specific interest. This creates a new reflection response of the target zone without interference from over‐ or underburden reflectors. Consequently, the method is well suited to obtain a clear response of a subsurface reservoir, which can be advantageous in time‐lapse studies. The isolated responses of a baseline and monitor survey can be more effectively compared; hence, the retrieval of time‐lapse characteristics is improved. This research aims to apply Marchenko‐based isolation to a time‐lapse marine data set of the Troll field in Norway in order to acquire an unobstructed image of the primary reflections and retrieve small time‐lapse traveltime difference in the reservoir. It is found that the method not only isolates the primary reflections but can also estimate internal multiples outside the recording time. Both the primaries and the multiples can then be utilized to find time‐lapse traveltime differences. More accurate ways of time‐lapse monitoring will allow for a better understanding of dynamic processes in the subsurface, such as observing saturation and pressure changes in a reservoir or monitoring underground storage of hydrogen and CO2.

The presence of sets of vertical, parallel fractures is very common in the Earth's upper crust. Notably, open fractures exert critical control on the mechanical and hydraulic properties of the host formation. There is great interest in understanding how fractures interact with seismic waves, as this knowledge could be used to detect and characterize fractures from seismic data. When a seismic wave travels through a fractured formation, it induces oscillatory fluid pressure diffusion between the fractures and the embedding porous background, a physical process that produces attenuation and dispersion of the seismic wave. Although there are numerous studies on this topic, the case of parallel fractures saturated with different immiscible fluids, such as brine and CO2, remains rather unexplored. With these motivations, in this work, we propose an analytical approach to compute the phase velocity and attenuation of P‐waves travelling perpendicularly to a set of planar, parallel fractures. While we consider that the background is saturated with brine, the fractures can be saturated with brine or gas. Our numerical analysis shows for the first time that two manifestations of fluid pressure diffusion arise in these cases. One of them, associated with relatively high levels of attenuation, is due to fluid pressure diffusion occurring between consecutive fractures saturated with different fluids. In this case, the fluid pressure diffusion process is initiated at a fracture saturated with brine and reaches a consecutive fracture saturated with gas. The other fluid pressure diffusion manifestation, on the other hand, arises at higher frequencies, is characterized by lower levels of attenuation and results from the interaction in the background of fluid pressure diffusion processes initiated at consecutive fractures, irrespective of the fluid content. Finally, by considering a stochastic distribution of the two fluids and a classical frequency of interest in seismic experiments, we obtain P‐wave attenuation and phase velocity as functions of the fracture gas saturation. This approach could be of interest for the remote detection and quantification of pore fluids using seismic waves.

The formations above a producing reservoir can exhibit large mechanical changes, creating a risk of significant subsidence and loss of rock integrity. These changes can be monitored by time‐lapse seismic acquisition, which measures the corresponding velocity changes via time‐shifts. Third‐order elastic theory can be used to connect subsurface strains and stress changes to these seismic attribute changes. Existing models assume isotropic strain dependence of the dynamic stiffness in shales. It is important to re‐evaluate this isotropic assumption considering the inherent anisotropy of shales and their abundance in the overburden. Thus, we instead propose a third‐order elastic model with a transversely isotropic strain dependence of the dynamic stiffness. When calibrated, this new model satisfactorily predicted P‐wave velocity changes determined in undrained laboratory experiments conducted on overburden field shales, covering a wide range of propagation directions and stress variations. The shales exhibit anisotropic dynamic strain sensitivity, resulting in a significantly higher strain sensitivity predicted for Thomsen's anisotropy parameters epsilon and delta subjected to a uniaxial strain parallel to the horizontal bedding plane compared to the vertical direction. Geomechanical modelling, considering a depleting disk‐shaped reservoir surrounded by shales, was employed to predict the dynamic stiffness changes of the overburden using the laboratory‐calibrated third‐order elastic model. The overburden time‐shifts increased with offset angle, peaking at about 45°, suggesting a strong influence of shear strains on the time‐shifts. In contrast, a corresponding model with an isotropic third‐order elastic tensor, calibrated to the same data, exhibited a significantly lower sensitivity to the shear strains. These results underscore the importance of considering the anisotropic strain dependence of the dynamic stiffness when studying shales. Interpreting offset‐dependent trends in pre‐stack time‐lapse seismic data, along with geomechanical modelling and an appropriate strain‐dependent rock physics model, can assist in quantifying subsurface strains and stress changes.

In many subsurface studies, facies and petrophysical properties are two important reservoir parameters that are closely correlated. They are routinely used in well interpretation, hydrocarbon reserve calculation and production profile prediction. These two parameters have commonly been determined in two separate tasks because of their mathematical differences (facies are discrete, and petrophysical properties are continuous). However, this is incorrect because facies and petrophysical properties are often strongly correlated. Therefore, we propose a new joint inversion method of facies and petrophysical properties based on a bi‐level optimization model. In the bi‐level optimization model, the upper‐level problem is the petrophysical property inversion while the lower‐level problem can identify the facies and add a facies‐related constraint for the upper‐level optimization. We also develop a new genetic algorithm for the discrete‐continuous inversion problem based on the bi‐level optimization model because the inversion problem usually has multiple local solutions. In addition, rock physics and statistics are combined in the inversion process. A rock physics model is used to establish the basic relationships between the petrophysical and elastic parameters, and the statistical approach is used to describe the intrinsic connection among the multiple reservoir parameters based on well log data. The numerical experiments indicate that the traditional separate prediction method and the new joint inversion method can quickly obtain more accurate results. In the application examples of real data, the inversion results can be matched to the well log data within the limits of the input data resolution, which further verifies the reliability and application potential of this new method.

The carbonate fault–karst reservoir is a special and significant reservoir in the Shunbei area, and the development of the cave has been controlled by strike‐slip faults. Due to the complex subsurface structures, fault–karst reservoir characterization is generally divided into fault and cave detection tasks. The potential spatial relationships between faults and caves might be neglected by using the separate detection scheme. The multi‐task learning network can perform multiple tasks simultaneously and exploit the potential features of training data by using a deep neural network. In this study, we built fault–karst models based on the geological background of the Shunbei area and synthesized fault–karst training data using three‐dimensional point spread function convolution. Then, we developed a multi‐task learning network to learn fault–karst features and detect faults and caves simultaneously. The test result demonstrates that the multi‐task learning network trained by synthetic fault–karst data can effectively identify the faults and caves in field seismic data. The comparisons of the multi‐task learning network, single‐task learning networks and conventional methods demonstrate the importance of spatial relationships between faults and caves and show the superiority of the multi‐task learning network. This technique could significantly assist in the exploration, development and well deployment for an ultra‐deep carbonate reservoir.

Full waveform inversion for land seismic data requires the development of specific strategies for modelling the complex response associated to the near surface. Seismic wave propagation is distorted by several effects, such as topographic relief, wavefield scattering, attenuation (often frequency‐dependent) and anisotropy. The modelling of such shallow complexities is often unmanageable by parametric inversions such that data preconditioning and surface‐related corrections are required before tackling the full waveform inversion velocity modelling problem. We developed a unified framework addressing both surface‐consistent corrections and full waveform inversion by using the transmitted portion of the wavefield. The problem of surface‐consistent decomposition is recast in terms of the transmitted wavefield leading to kinematic and dynamic corrections to account for sub‐resolution wave propagation distortions occurring in the near surface weathering layer. Seismic data are deconvolved from the effects of the near surface to better represent the deeper subsurface wavefield propagation. Signal‐to‐noise enhancement is obtained after the surface‐consistent transmission preconditioning by the generation of virtual super gathers reconstructed in the midpoint‐offset sorting domain. An original scheme of 1.5D Laplace–Fourier full waveform inversion, involving 3D radiation and 1D velocity inversion, is then applied for the velocity reconstruction where the amplitude information of the seismic data is fully preserved and utilized. The unified approach of surface‐consistent transmission preconditioning and velocity modelling is demonstrated on the Society of Exploration Geophysicists Advanced Modelling arid model dataset as well as on two complex field datasets containing near surface complexities typical of arid regions. The approach provides the solution of the near surface distortions by performing data preconditioning and velocity inversions in a unified scheme. The surface‐consistent full waveform inversion approach has been utilized for large land seismic projects and represents a robust tool for supporting land seismic imaging.

In this work, different stages of gas production in shale reservoirs are modelled. First, hydraulic fracturing is considered by injecting water at high pressures to crack the formation and increase the flow capacity of the reservoir. During the fluid injection, rock properties are modified and water appears in the stimulated area. Then, these changes can be detected through seismic monitoring. Finally, once the fracking stage is completed, the simulation of gas production begins. The simultaneous gas–water flow in the injection and production stages is modelled using the Black‐Oil formulation. Furthermore, a fracture criterion is applied under the hypothesis of constant temperature and constant stress field in this first analysis of the problem. The numerical simulations allow us to analyse the propagation of the fracture and the behaviour of the pore pressure and water saturation in the stimulated area. The advance of the fracturing fluid is delayed in relation to the breakdown of the rock. Besides, the presence of new fractures is detected by applying a poroviscoelastic wave propagation simulator that considers mesoscopic losses induced by heterogeneities in rock and fluids. After the fracture network is created, the injection well becomes a producer, allowing the extraction of gas and the flowback of the injected fluid. The simulated gas flow rates are compared with those obtained by a simplified single‐phase analytical solution used for practical applications, achieving optimum matching results.

Subsurface energy resources are often found in three‐dimensional and non‐spatially continuous rock formations that exhibit electrical anisotropy. Deep‐sensing tri‐axial borehole electromagnetic measurements are currently being used to detect three‐dimensional fluid‐bearing subsurface formations, but borehole environmental, geometrical and instrument‐design factors, together with measurement noise, constrain their practical range of detection and spatial resolution. By understanding the interplay of the above factors on the detectability and sensitivity of borehole deep electromagnetic measurements, one can potentially quantify the uncertainty of both target spatial location (relative to the well trajectory) and electrical resistivity contrast, thereby improving the certainty of three‐dimensional well navigation in real time. We implement a finite‐volume method to numerically solve Maxwell's equations for three‐dimensional electrically anisotropic heterogeneous rock formations in the calculation of magnetic fields measured with deep‐sensing tri‐axial borehole electromagnetic instruments. The calculated magnetic fields at measurement locations are described as the per cent difference between measurements acquired in rock formations with and without conductive or resistive three‐dimensional targets. We quantify the maximum radial distance of detection from the well trajectory and the spatial sensitivity of a commercially available deep‐sensing electromagnetic instrument with respect to environmental factors and measurement acquisition parameters by assuming that the borehole electromagnetic instrument can reliably detect offset three‐dimensional targets if the corresponding per cent measurement difference exceeds the threshold for measurement noise. Results indicate that commercially available tri‐axial deep‐sensing borehole electromagnetic instruments can achieve maximum detection distances between a quarter of and a full transmitter–receiver spacing. In addition, we show that radial detection distances vary from 0.3 to 2 skin depths depending on the geological environment, that is, target conductivity contrast with respect to the embedding background, electrical anisotropy of the background formation and targets and measurement acquisition parameters, that is, frequency and transmitter–receiver spacings. The above findings are not only important for instrument design and measurement‐acquisition planning but also for the effective implementation of real‐time inversion‐based measurement interpretation procedures.

Many different magnetotelluric response functions such as the impedance and tipper which are converted from measured electromagnetic fields to remove the effects of natural sources have been proposed and studied. To obtain good inversion results, it is important to know that the inversion results depend on the sensitivity patterns of the magnetotelluric response functions used as the input, indicating parts to which the magnetotelluric responses are sensitive. We introduce how to use the sensitivity patterns to determine optimal input magnetotelluric response functions for interpreting field data. For magnetotelluric field data, we use the data acquired at the Utah Frontier Observatory for Research in Geothermal Energy survey area. The target structure in the Utah Frontier Observatory for Research in Geothermal Energy survey area is associated with geothermal energy, and most of the sites are located to the left of the target structure. Examining the sensitivity patterns of the major magnetotelluric response functions, that is the components of the impedance tensor and tipper vector, for the target anomaly, the pattern of y‐component of the tipper vector () covers the areas where most of the sites are distributed. This means information about the target structure is mainly contained in . To investigate the impact of on inversion of the target anomalous body, we compare the inversion results for four cases that (1) the xy‐ and yx‐components of the impedance tensor ( and ) are used; (2) , and are used; (3) , and x‐component of the tipper vector () are applied; and (4) the all components of the impedance tensor (, , and ) are used as the input. This comparison shows that selected as the optimal input from the sensitivity patterns for this case study contributes to retrieving the target anomaly. Through the case study for the Utah Frontier Observatory for Research in Geothermal Energy site, we show how to design a strategy of selecting optimal magnetotelluric response functions based on the sensitivity patterns to effectively image target structures. Our experiment supports that the sensitivity patterns can be used to increase the reliability of magnetotelluric inversion.