The separation of upgoing and downgoing waves in vertical seismic profiling (VSP) data is crucial for subsequent imaging, interpretation, and inversion. The interweaving of upgoing and downgoing waves and the presence of noise complicate the entire wave field, making it difficult to separate upgoing and downgoing waves. Various model-driven separation methods including low-rank approximation (LRA) and data-driven methods have achieved promising results. However, the performance of pure model-driven methods, such as F-K filtering and Radon transform, rely on domain transformation sparse representation for both upgoing and downgoing wavefield. Additionally, pure data-driven methods require a large number of precisely separated signals as training samples, which is a nontrivial task. To overcome these difficulties, this article proposes a model-data-driven VSP wavefield separation framework that iteratively completes the task of wavefield separation in an unsupervised manner. The key to this model is to use the powerful feature representation capability of the deep convolutional autoencoder (DCAE) to model the downgoing waves and use LRA to model the upgoing waves. By accurately modeling the upgoing and downgoing waves, we integrate the model-driven and data-driven methods together, while protecting both the upgoing and downgoing waves, and inheriting the advantages of the DCAE and LRA methods. Subsequently, we also proposed an alternating minimization optimization strategy to optimize the parameters of the model and iteratively obtain high-quality solutions. Comparative experiments on synthetic data and real data show that our method can achieve effective separation results while suppressing Gaussian noise.

Full-waveform inversion (FWI) is increasingly used in land seismic exploration to achieve high-resolution near-surface models. In complex near-surface environments, however, FWI is challenged by noisy data and inaccurate initial models, raising the need for an effective regularization strategy to mitigate the inherent ill-posedness of FWI. Incorporating geological information, such as structural dips, into the regularization operator, has proven effective in stabilizing FWI and promoting geologically meaningful results. Obtaining clear seismic images that provide structural insights is, nonetheless often hindered in complex near-surface surveys due to noisy reflection data and gradient weathering velocity. Structural regularization treats imaged reflectors as velocity contours and penalizes velocity variations along the dips. We propose a novel approach involving deformable-layer tomography (DLT) to directly invert for velocity contour distributions. DLT is well suited for near-surface applications as it accommodates both gradient velocity variations and abrupt velocity contrasts. In order to avoid erroneous structural regularization, we evaluate the accuracy of the DLT-estimated dips and assign weights accordingly. The weighted dip field is used to construct directional total variation (TV) in regularizing FWI, using the edge-preserving smoothing of TV regularization as well as the dip constraint. Using a realistic near-surface model, we demonstrate that FWI with the DLT-guided directional TV regularization outperforms conventional TV regularization. Our findings underscore the advantages of incorporating geologic structural constraint into FWI under complex near-surface conditions, highlighting the improved fidelity and resolution in near-surface velocity model building.

Vertical seismic profiling (VSP) is widely applied in the field of seismic exploration to deliver high-quality subsurface images and enable quantitative characterization around the wellbore region. The separation of VSP upgoing and downgoing wavefields is a practical step in the wavefield processing, which sets the foundation for the final imaging quality. With recent advances in deep learning (DL), a number of new attempts in seismic signal processing have proven effective. Inspired by that, we propose an unsupervised VSP wavefield separation framework via dual convolutional autoencoders (dualCAEs). Our method is based on two characteristics of up- and downgoing wavefields: one is directional continuity, and the other is the zero-mean feature. Two regularizers are proposed to constrain these two features. Thanks to this, our method does not require any training data other than the input data itself. Ablation and comparison studies on synthetic data validate the effectiveness of our method. Generalization tests on SEAM open data and field VSP records in the Dong area show the superiority of robustness and fidelity over traditional$F$–$K$and median filtering methods.

Elastic full-waveform inversion (EFWI) has received significant attention in the industry for many years. Numerous studies have shown that parameter optimization methods based on physics-informed neural networks (PINNs) are more promising in full-waveform inversion (FWI) compared to traditional methods. Compared with surface seismic techniques, vertical seismic profiling (VSP) offers advantages such as lower acquisition costs, higher signal-to-noise ratios, and higher resolution. To address the problem of FWI for VSP, this study proposed a physics-informed recurrent neural network (RNN) method with prior knowledge constraint for elastic FWI. Our main idea is threefold. First, we mined and characterized the prior knowledge for VSP FWI, which primarily includes geological knowledge, the 1-D velocity profile around the wellbore, and the empirical relationships between compressional and shear wave velocities. Second, we implemented a PIRNN with prior knowledge constraint, using the RNN framework to optimize trainable parameters. Finally, using the PIRNN framework with prior knowledge constraint, we achieved elastic FWI for VSP. Numerical examples demonstrate that the inversion results for the salt and Marmousi models indicate that incorporating prior knowledge increases the accuracy of FWI. Compared to data-driven machine learning methods, this architecture eliminates the impact of sample quality on parameter optimization.

The separation of P- and S-waves is pivotal in the processing of multicomponent seismic data. The complexity of geological structures often leads to intricate P- and S-wavefields, which poses challenges for identifying and separating waves using conventional signal features, such as the F-K domain or$\tau $-p domain distributions, while data-driven machine learning methods overly rely on the quality of samples. This article proposes a novel approach for P- and S-wave separation in complex geological structures based on a knowledge-guided autoencoder network. First, the existing full waveform inversion (FWI) can obtain relatively accurate knowledge representations of P- and S-waves. A recurrent neural network (RNN) was employed for elastic wave FWI to acquire a knowledge representation of intricate P- and S-waves in complex structures. Subsequently, a dual-branch autoencoder network was constructed based on the obtained knowledge representation of the complex P- and S-waves. One branch was guided by the knowledge representation of P-waves for P-wave separation, whereas the other branch was guided by the knowledge representation of S-waves for S-wave separation. Finally, a comprehensive autoencoder network architecture was devised that incorporates waveform reconstruction loss, P-wave knowledge guidance loss, and S-wave knowledge guidance loss for effective P- and S-wave separation. Theoretical analyses and numerical simulations were performed, and they demonstrated the effectiveness of the proposed method for achieving P- and S-wave separation in complex geological structures.

Due to the poorly coupled geophones present in boreholes, vertical seismic profiling (VSP) data are known to suffer from intense coupling noise, which causes severe VSP image deterioration and significantly hinders subsequent processing. Thus, diverse denoising approaches are indispensable preprocessing steps for suppressing this kind of noise to achieve good results. Among them, robust principal component analysis (RPCA) is a common signal and noise separation model that is generally considered a highly promising method for removing intense coupling noise; however, handcrafted priors have limited denoising ability, especially for the low-rank assumption of useful signals. As an alternative, following the RPCA framework, this article proposes an unsupervised iterative robust deep convolutional autoencoder (IRDCAE) model to suppress intense VSP coupling noise without any assumptions regarding valuable signals. The key to the IRDCAE approach is the use of weighted column sparsity (WCS) to characterize the behavior of the intense coupling noise, where the weight prior is derived from the pure noise component before the first break. By adding a WCS regularization term to the conventional deep convolutional autoencoder (DCAE), our IRDCAE method transforms the model from an entirely data-driven model to a model+data driven approach. Thus, the IRDCAE approach has the advantages of both RPCA and DCAE, resulting in the ability to separate intense coupling noise from useful signals in an unsupervised manner by optimizing the IRDCAE model via an alternating minimization algorithm. The exceptional performance of the IRDCAE model is exhibited with synthetic and field VSP data.

Seismic attributes comprise an effective method for oil and gas reservoir characterization and prediction. Hundreds of seismic attributes have been introduced in the last 30 years. Among the seismic attributes targeting different reservoir features, the autoencoder (AE) receives a significant amount of attention, as it extracts deep attributes of seismic data, providing more details of seismic lateral features than other seismic waveform data and seismic attributes. However, data-driven deep attributes bring new challenges to interpretation as they lack the support of intrinsic physical mechanisms. Hence, a shared AE (S-AE) method is proposed in this article, which can extract seismic deep attributes and fuse traditional seismic attributes simultaneously. An S-AE is a revised version of an AE, which consists of an encoder and decoder. An S-AE takes the seismic waveform as the input of the encoder and obtains the deep attribute, and the decoder then transforms the deep attribute to reconstruct the seismic waveforms and attributes. In an S-AE, the network in front of the decoder is shared, while the networks after the decoder consist of independent layers. Such a network structure ensures the effect of reconstruction and associates seismic attributes with the extracted deep attribute, so as to achieve the purpose of attribute fusion and deep attribute extraction. The proposed S-AE method is compared with conventional seismic data fusion methods, such as RGB and principal component analysis, and the superiority of the S-AE is demonstrated in both synthetic and field applications.

Vertical seismic profiling (VSP) helps to derive high-resolution images around the instrumented borehole and is a cost-effective technique for CO2storage monitoring. In routine VSP data processing, P- and S-wave separation is a crucial step to extract independent single-mode waves for accurate imaging and interpretation. Conventional wave mode separation involves tedious, subjective, and non-reproducible manual interventions, especially when dealing with complex geology. To better automate the process, we propose a data-driven deep learning-based P- and S-wave separation method. Our method adapts a fully convolutional neural network that simultaneously extracts P- and S-potential data from multicomponent VSP measurements. To reduce the enormous computational cost in wave simulation while constructing training datasets with sufficient kinematic and dynamic variations, we introduce virtual wellbores where synthetic VSP data sampling wide variations in seismic kinematics and dynamics are recorded using only a dozen elastic wave simulations on a single velocity model. We qualify the separation results both directly in data space and in image space after reverse time migration (RTM). Generalization tests on various synthetic models and their corresponding RTM images demonstrate that the proposed strategy provides sufficient sampling of the high-dimensional data space and essentially ensures successful applications of the trained neural network to similar yet different geological scenarios.

The pre-/sub-salt fractured networks provide key clues to understanding the tectonic history and the subsurface reservoir evolution. Yet, they are among the most complicated targets for geophysical investigations. We develop a structure-oriented mapping technique to delineate the high-resolution dipping layers across the borehole region using the zero-offset vertical-seismic-profiling (ZVSP) survey, which is conventionally used to provide one-dimensional (1-D) wave propagation information. The key information for the single-shot VSP mapping, which is the structural dip across the borehole, is unreliable from the poor sub-salt surface seismic image. Alternatively, we obtain the structural dips via reflection layer tomography. Taking advantage of the accurate interval velocities and the high-fidelity identifications from the reflection events, we manage to invert for the geometry of the main reflectors identified across the wellbore. Based on such information, we propose an effective processing strategy and achieve a high-resolution image of the dipping structures around the borehole region from the ZVSP. The current result compares reasonably well with the corresponding surface seismic profile but supplies higher-resolution details of the dipping structures and fault networks below the thick evaporite caprock where the surface seismic image degrades sharply. The enhanced subsurface image encourages better structural evaluation, geologic interpretation, and future 2-D/3-D VSP survey design.

Accurate delineation of the salt flank is important for oil and gas exploration in areas with salt intrusion. We describe the application of deformable-layer tomography (DLT) to invert for the geometry of salt flank using travel times of transmitted P- and S-waves from surface sources to downhole receivers. The DLT allows us to take advantage of a common situation that the salt velocity is known and generally invariant, but we need to determine the variable geometry of salt flank. We demonstrate our new method using a physical model experiment mimicking the setup of a walkaway vertical-seismic-profiling (VSP) survey. We first use picked P-wave arrivals to invert for the salt flank geometry; the sinusoid shape of the salt flank is estimated fairly by the DLT due to the uneven P-wave raypath coverage. As an improvement, we further incorporate the S-wave arrivals in the DLT. The picking of S-wave arrivals is assisted with modeled S-wave arrivals based on the P-wave DLT model. The DLT solution using both P- and S-wave arrivals delineates the salt flank geometry more accurately than that using P-wave arrivals alone. The salt flank delineation using DLT can complement with the conventional salt proximity and migration methods. It could also serve as constraints or the initial model for the full-waveform inversion of salt geometry.