Xiaokai Wang

High-speed trains (HSTs) generate strong ground vibrations when running on railways, which excite stable seismic waves. As these seismic waves propagate through the subsurface medium, they carry valuable information about geologic structures, which can be effectively used for imaging and inversion. To fully use HST seismic data, it is crucial to understand and characterize the wavefield characteristics of the HST source. Using the theory of a uniform linear array, the directional characteristics of a seismic wavefield excited by HST sources were analyzed. Furthermore, using the ABAQUS finite element analysis software, a model depicting an HST traversing a tunnel was established. The simulation results of the HST seismic wavefield were used for theoretical verification. By comparing the theoretical radiation patterns of the HST seismic wavefield with those of the simulated wavefield, strong consistency between the main lobes of the interference patterns was observed. Notably, this conclusion remained valid even when the HST speed or medium parameters were changed.

Seismic erratic noise, characterized by large isolated events following non-Gaussian distributions, significantly degrades seismic data quality by masking useful signals. Methods based on conventional priors remain essential but face inherent challenges as they struggle to balance noise attenuation and signal preservation. Supervised deep learning approaches are constrained by the scarcity of high-quality labeled training pairs, while existing unsupervised techniques often suffer from suboptimal accuracy and high computational cost. To address these limitations, we propose an unsupervised deep learning framework based on implicit neural representation (INR) for erratic noise suppression in seismic data. The proposed method employs Fourier feature mapping to encode the spatial coordinates of noisy seismic data, which are then processed by a lightweight multilayer perceptron (MLP). The MLP is optimized using a robust Huber loss function to learn a continuous representation of the underlying seismic wavefield, effectively attenuating erratic noise while preserving valuable signal components. The Fourier feature mapping enhances the MLP’s ability to capture high-frequency signal details, while the Huber loss adaptively weights residuals based on amplitude, enabling precise noise suppression. Experimental results on synthetic and field datasets demonstrate its superior performance in suppressing noise while preserving signal fidelity.

Distributed Acoustic Sensing (DAS) systems offer a promising framework for advanced subsurface imaging and monitoring. Despite the great potential, the complex noise characteristics inherent in seismic data, such as environmental, mechanical, and instrumental disturbances, pose significant challenges to data fidelity and stability. Conventional noise suppression methods cannot be adequately adapted to dynamic seismic environments due to the need to design filters for different a priori conditions. To overcome these limitations, we introduce the conditional diffusion probabilistic model (C-DDPM), which is shown to have strong a priori extraction capabilities and can better cope with seismic signal extraction under different noise combinations. In addition, we incorporate a adaptive FK conditioning approach into the diffusion process, allowing C-DDPM to better learn the data distribution. We also use asymmetric dilated convolution (ADConv) to effectively suppress noise. Our approach is rigorously tested on both synthetic and real-world seismic datasets, demonstrating satisfactory improvements in noise reduction and signal clarity. Comparative analyses with existing classical methods reveal that our framework not only achieves higher PSNR but also reveals previously obscured waveform details, outperforming existing methods in challenging geophysical scenarios.

As a new type of seismic source, the high-speed train (HST) seismic source has great application potential in subsurface imaging in terms of high intensity, strong repeatability and wide distribution compared with traditional seismic sources. However, the HST seismic data are severely interfered in the time-space domain due to the complex mobile combination source. Besides, the body waves are almost swamped by strong surface waves. Therefore, it is difficult to directly use this kind of seismic data for subsurface exploration. Considering the characteristic differences between the surface and body waves in the frequency-wavenumber (F-K) domain, it is more conducive to separate the surface and body waves for the HST seismic data in the F-K domain, so as to use different wavefield information in sequence. Therefore, in this study, we attempt to apply the inversion method with F-K spectrum as the objective function to HST seismic data, which aims to more effectively utilize the surface and body waves therein. In addition, to alleviate the coupling effect between P-wave and S-wave in elastic waveform inversion, we introduce the divided-offset strategy to further improve the quality of the inverted P-wave and S-wave velocities. Numerical experiments indicate that the proposed inversion method is beneficial to make better use of the surface and body waves in HST seismic data, and is expected to provide more insights on the use of HST seismic data for subsurface monitoring and detection.

Seismic data deconvolution is vital for enhancing resolution and accurate subsurface interpretation. Traditional methods heavily rely on predefined assumptions that limit their robustness to noisy data. As state-of-the-art generative models, diffusion models excel in capturing accurate prior distributions, which are beneficial to inversion. Moreover, diffusion models inherently resist noise due to their training on reverse noisy processes. Building on this foundation, we introduce an unsupervised diffusion model for seismic deconvolution, leveraging Diffusion Posterior Sampling (DPS) to incorporate observed seismic data into the sampling process to guide high-accuracy reflectivity generation. Unlike traditional single-trace approaches, our method performs deconvolution across entire 2D profiles, effectively capturing spatial continuity. Though solely trained on synthetic data, our method exhibits satisfactory performance when applied to synthetic and field datasets, demonstrating strong noise resistance and remarkable generalization capabilities.

Blended acquisition enables near-simultaneous firing of multiple shots, facilitating faster and denser seismic data collection, which reduces acquisition time, lowers costs, and potentially enhances subsurface illumination. However, these benefits require an additional deblending process to recover the seismic data one would have acquired via a conventional survey. Numerous data-driven deblending algorithms have emerged, predominantly relying on supervised deep learning. Nevertheless, constructing training pairs that are representative and transferable to complex field data remains challenging. Some unsupervised methods have been developed, but they often focus on local similarities within individual seismic data, neglecting the continuity of seismic wavefields and relationships between different data across the entire data set. Implicit neural representation (INR) provides a versatile framework for modeling continuous seismic signals using neural network parameterized functions, capturing the intrinsic relationships throughout the entire data set. Building on this advantage, this study introduces INR for self-supervised seismic deblending. Specifically, to capitalize on the prior knowledge that desired unblended signals in nearby unblended common-receiver gathers (CRGs) vary smoothly due to the continuity of seismic waveforms, a deep neural network is used as an implicit function to parameterize desired unblended CRGs with receiver indices as inputs. By leveraging the blending operator, a physics-informed training strategy is implemented to enforce measurement consistency, enabling the trained network to recover corresponding deblended CRGs accurately. Numerical experiments conducted on synthetic and field data sets demonstrate the efficacy of our approach. Compared with sparsity-promoting inversion with a patched 2D Fourier transform, a widely adopted method in the industry, and the plug-and-play method with blind-spot networks, one of the most widely accepted self-supervised approaches, our method shows superior performance.

The Loess Plateau in China presents a formidable challenge for seismic exploration due to its thick, porous surface loess layers that severely attenuate high-frequency seismic waves, degrading the resolution of conventional 3D acquisition. However, the region’s unique topography, crisscrossed by deep gullies formed through consistent rainfall erosion, provides a natural solution to acquire high-resolution (HR) data. With thin or absent loess cover, these gullies delineate natural pathways ideal for 2D crooked-line seismic surveys, where reduced loess interference preserves high-frequency content. Accordingly, these 2D surveys deliver better resolution than traditional 3D acquisition in the loess-covered areas. Their seismic response distributions are expected to closely resemble those of a hypothetical HR 3D data set unaffected by loess attenuation. Although these localized 2D surveys capture geologically representative HR features, existing methods struggle to extrapolate their high-frequency characteristics to broader 3D volumes, limiting their ability to mitigate loess-induced resolution loss. To bridge this gap, we use a cycle-generative adversarial network under weak supervision to enhance 3D data resolution by leveraging unpaired 2D HR crooked-line data. Specifically, our approach transfers high-frequency features from 2D profiles to 3D volumes processed by conventional swath techniques through a bidirectional cycle structure, enforcing cross-distribution consistency while preserving geologic integrity. Custom loss functions and data augmentation further address spectral mismatches and stabilize training under loess-induced complexity. Synthetic and field experiments demonstrate that our method effectively captures HR characteristics of 2D data and recovers high-frequency content attenuated by loess in 3D data. Our approach achieves improved fidelity and noise robustness compared with traditional spectral whitening and zero-phase spiking deconvolution. This work underscores the untapped potential of integrating spatially sparse but information-rich 2D surveys with modern deep-learning methods to overcome persistent resolution limitations in seismic exploration.

High-resolution (HR) seismic data are crucial for accurately identifying subsurface geologic formations and reservoir properties. However, as widely observed in prestack and poststack data, seismic wave attenuation often leads to a gradual decrease in resolution with depth. This degradation leads to resolution differences between data from shallow and deep layers, characterized by high-frequency energy loss in deeper target layers. The latter could complicate the extraction of crucial deep-layer structure information, leading to increased uncertainty in reservoir modeling and significant economic losses due to our inability to image deep targets with sufficient resolution. Traditional methods with simplified assumptions about this nonlinear attenuation hinder modeling geologic complexity and variability. Deep-learning methods excel at capturing complex, nonlinear relationships but often rely heavily on scarce and costly paired labels for supervised learning. As a result, deep learning without requiring paired labels for seismic resolution enhancement has become a key research focus. This study explores a method to enhance 3D seismic data resolution using weakly supervised learning, leveraging the inherent similarity of sedimentary structures across depths. Specifically, our approach uses a relatively shallow, HR data window to transfer and extrapolate learned high-frequency information to a deeper window containing attenuated responses from deeper layers. Thereby, we achieve enhanced resolution in the target region. A tailored 3D convolutional neural network with a bidirectional cycle structure, custom-designed loss functions, and data preprocessing techniques specifically addresses the challenge of resolution differences caused by seismic wave attenuation. The effectiveness of our method is validated with synthetic and real 3D poststack migration data, demonstrating its robustness for regions near the target layer. Compared with conventional spectral whitening, our approach leverages intrinsic data characteristics more adaptively and robustly, making geologic structures more discernible.

Coherent noise separation stands as a crucial step in seismic data processing. The morphological component analysis (MCA)-based separation method, which treats coherent noise and signal as distinct components and represents them sparsely with dictionaries, has been widely adopted for noise suppression. Typically, constructing effective fixed dictionaries for MCA-based methods involves necessary expert knowledge to meticulously select appropriate transform basis functions from an extensive dictionary library and fine-tune their parameters. To reduce time consumption and ensure optimal dictionary construction, we introduce an adaptive framework for identifying optimal dictionaries used in MCA-based coherent noise separation. Initially, we define a fixed dictionary library comprising dictionaries constructed using various transform basis functions with their corresponding parameters. Subsequently, we formulate a relative sparsity minimization problem (RSMP) to identify the optimal fixed dictionaries that minimize relative sparsity within this predefined library. Finally, we design a genetic algorithm to solve RSMP. The identified dictionaries are then applied to MCA-based coherent noise separation. Synthetic and field data examples demonstrate the effectiveness of our method.

Due to challenging field operations and resource constraints, seismic data acquisition often requires coping with missing traces. Interpolation algorithms are crucial for reconstructing these missing traces to enable improved subsurface analysis and interpretation. Although deep learning has made exciting advances in seismic reconstruction, its focus has predominantly been on 2D and 3D data sets with relatively low rates of missing data. Reconstruction of 5D seismic data entails considering simultaneous sources and receivers deployed in areal arrays to solve the reconstruction problem. The latter offers greater data redundancy, which can be leveraged to enhance interpolation quality. Traditional 5D deep-learning interpolation methods rely heavily on synthetic training pairs, posing challenges when applied to real-world data. This necessitates transfer learning techniques, which can be cumbersome. To address this, we introduce a self-supervised, coordinate-based deep interpolation algorithm that mitigates the need for labeled data. Using a multilayer perceptron (MLP) network can effectively encode the continuous seismic 5D wavefield. Once trained, the MLP can infer missing trace amplitudes from their coordinates. We contribute to boosting the MLP, enabling it to generate seismic profiles rather than single-point predictions. This enhancement significantly strengthens the model’s performance and efficiency. Moreover, we apply nuclear norm regularization to the output profiles, improving the reconstruction quality. The effectiveness of our algorithm is illustrated with synthetic and field data experiments.