
Abstract Data-driven methods provide an end-to-end framework for seismic noise attenuation, but the limited receptive field of conventional convolutional neural networks (CNNs) restricts their ability to model long-range dependencies in seismic data. Their generalization is further limited by the distribution mismatch between synthetic training data and field records. To address these challenges, we propose a local-orthogonality-constrained Swin Transformer convolutional network (LOSTCG). The core STCG module combines convolution-based local feature extraction with attention-based adaptive feature weighting, enabling effective multiscale representation of seismic events. In addition, a local orthogonality term is incorporated into the loss function to reduce the leakage of coherent signals into the estimated noise and improve robustness to complex field noise. A diverse synthetic training dataset is also constructed and augmented with field noise to narrow the distribution gap between synthetic and field data. Experiments on synthetic data and low-signal-to-noise-ratio field records demonstrate that LOSTCG achieves better overall performance in noise attenuation and signal preservation than the f–k method, singular value decomposition (SVD), ResUNet, and SCRN. These results indicate that the proposed method generalizes well to field data and has considerable potential for practical seismic data processing.
Abstract Differential compaction and diagenesis produce depth-dependent changes in the petrophysical properties of deep tight sandstone and adjacent mudstone. Overlapping P-wave impedance distributions can hinder conventional reservoir prediction, creating an impedance trap. Using well logs from three wells in the Songliao Basin, China, we compare the ability of P-wave impedance and density to distinguish sandstone from mudstone at depths of 3000–4700 m. Two metrics quantify their separation: the probability-density overlap area (Soa) and the separation coefficient (Dsc). P-wave impedance increases in both lithologies, but its discrimination capability improves only slightly with depth. Density shows a contrasting trend: mudstone becomes denser, whereas sandstone density changes little. The density overlap area decreases from approximately 0.8 to 0.3, and its separation coefficient increases from approximately 0.5 to 1.8. The deep intervals exhibit three density endmembers: high-density mudstone, medium-density tight sandstone, and low-density high-quality sandstone. We then apply noise-resistant density inversion (NRDI) to extract a density-gradient AVO attribute from field seismic gathers. In the studied dataset, this attribute delineates deep sand bodies and minor faults more clearly than conventional post-stack attributes. These results support the use of density contrasts to complement impedance-based interpretation in deep tight sandstone reservoirs, subject to adequate amplitude fidelity, angle coverage, and lithological contrast.
Abstract As seismic exploration increasingly focuses on complex areas, such as subsalt regions, steeply dipping structures, and fault zones, complex wavefield phenomena (including caustics and multipathing) make accurate migration more challenging. Conventional Kirchhoff migration typically images subsurface positions through diffraction stacking based on single-valued traveltimes, which cannot handle multipathing well in complex structures, especially after long-distance propagation. This paper proposes a recursive Kirchhoff migration method accelerated by the butterfly algorithm. Recursive Kirchhoff migration divides the velocity model into several local depth intervals and recursively extrapolates seismic records to the datums within these intervals. Then, the conventional Kirchhoff migration is used within each local interval. This process helps maintain the validity of the first-arrival traveltimes while implicitly accounting for multipathing effects through recursive wavefield extrapolation. However, recursive wavefield extrapolation involves a large number of repeated Kirchhoff integral summations, and the computational cost increases rapidly with grid size. We therefore introduce the butterfly algorithm to accelerate the wavefield extrapolations. The butterfly algorithm uses low-rank approximations of the oscillatory integral kernel over box pairs and transforms the dense summations into the recursive transfer of equivalent source coefficients on a tree structure, which reduces the computational complexity of wavefield extrapolation. Numerical examples on the Marmousi 2 model and the 2D SEG/EAGE salt model demonstrate that the proposed method effectively reduces the computational cost while preserving imaging accuracy. Its application to an offshore field dataset from the deepwater Gulf of Mexico further confirms its practical applicability.
Abstract In land seismic data, strong coherent scattering noise generated by small-scale, low-velocity near-surface heterogeneities can severely degrade reflection continuity and waveform fidelity. Unlike primary reflections, which remain approximately locally planar, such scattering noise violates the local plane-wave assumption and is characterized by irregular wavefronts and rapidly varying apparent slopes. Motivated by this physical distinction, we propose a physics-guided, self-supervised denoising framework that jointly reconstructs seismic data and predicts local slopes, enabling effective separation of scattering noise from reflection signals without requiring clean training data. The proposed framework jointly trains two neural networks: an implicit neural representation (INR) network for local slope estimation and a convolutional neural network (CNN) for seismic data reconstruction. A partial differential equation (PDE)-based loss is introduced to constrain the reconstructed data to satisfy the local plane-wave equation, while the local slopes are adaptively updated to minimize signal leakage. To improve scalability and efficiency, the INR is conditioned using shot-dependent information, while the CNN incorporates slope-based modulation derived from the estimated local slopes. This design allows the framework to learn from multiple shot gathers simultaneously while still generating shot-specific predictions, thereby avoiding redundant per-shot training. Synthetic and field data examples demonstrate that the proposed method effectively attenuates coherent scattering noise, enhances reflection continuity, and provides robust local slope estimation.
Abstract Cross-hole radar is a high-resolution geophysical technique for subsurface characterization. In practical engineering, velocity tomography alone is often insufficient to distinguish different types of geological defects or identify critical features such as water or clay fillings, whereas attenuation is more sensitive to moisture, conductive materials, and fracture fillings. However, attenuation inversion is generally less stable than velocity tomography in strongly heterogeneous media because the measured amplitudes are affected by geometric spreading, scattering, antenna response, and inversion non-uniqueness. To address these issues, this study proposes a sequential velocity-guided attenuation inversion framework for cross-hole radar. Curved-ray traveltime tomography is first used to obtain a stable velocity structure and propagation paths. The gradient of the converged velocity model is then transformed into spatially varying regularization weights for attenuation inversion, allowing the velocity information to act as a soft structural prior rather than being directly imposed on the attenuation model. First-arrival times are picked using the First-arrival Picking through Mathematical Morphology and Edge detection (FPME) method, followed by direct-wave amplitude extraction within a first-arrival-centered time window. Synthetic experiments show that the proposed structural constraint suppresses background artifacts and improves the spatial delineation of attenuation anomalies, particularly in heterogeneous models. Sensitivity and spatially decoupled tests further indicate that moderate structural coupling improves inversion stability while retaining attenuation-data sensitivity. Field application in a karst area of Hechi, Guangxi, shows clearer attenuation anomaly boundaries and reasonable correspondence with the clay-silt overburden and cave- or clay-filled intervals identified from borehole cores. These results demonstrate that velocity-guided attenuation tomography can provide complementary information to velocity imaging for interpreting complex karst structures.