Fault interpretation tasks become more and more difficult as the complexity of seismic exploration increases, especially for ultra-deep seismic data. Recently, numerous researchers have utilized automatic interpretation techniques based on deep learning to improve the efficiency and accuracy of fault prediction. Although deep learning methods have powerful data information processing capabilities, the applicability of deep neural networks may still be limited by the range of learned information. Therefore, we develop a new technique called structural data augmentation to enhance the diversity of the datasets. Concretely, we utilize different geological structure theories to incorporate virtual folds and faults in the field seismic data to improve the diversity and generalization ability of the training datasets. To cope with the multi-stage and multi-scale complex structures developed in ultra-deep strata, the proposed augmentation workflow increases data diversity by generating various virtual structures containing multi-scale folds, listric faults, oblique-slip displacement fields, and multi-directional fault drags. Tests on the field seismic data show that our method not only outperforms conventional seismic attributes but also has advantages over other machine learning methods.

Three-dimensional magnetic inversion, based on the least-square and regularization algorithm in the space domain, is an important tool for quantitative interpretation of magnetic data. However, the common 3D inversion approaches usually require great numbers of forward and inversion calculations and cause low efficiency for inverting large-scale data. Three-dimensional imaging is an alternate rapid tool for qualitative and quantitative interpretation of magnetic data. In this paper, we present a wavenumber-domain iterative approach for 3D imaging of magnetic anomalies and gradients, which could increase imaging efficiency and is suitable for rapidly imaging large-scale data. The wavenumber-domain formulas for forward modeling and imaging of total magnetic anomaly, three magnetic components, magnetic gradients and magnetic full-tensor gradients are deduced and provided. A depth-scale factor and the constraints of magnetic interface are included into the imaging formulas to enhance depth resolution. An iterative algorithm is adopted for the imaging to reduce the fitting error and improve the imaging accuracy. Tests on synthetic and real data from the Sichuan basin, China, verified the feasibility of the presented approaches.

The remanent magnetization effects pose a great challenge to the application of magnetic exploration in fields such as metallic mineral prospecting, igneous rock detection, and tectonic studies. Magnetic data inversion is one important tool to produce the magnetism structure in the subsurface, but the traditional inversion techniques for large-scale data suffer not only from the complicated influence of remanent magnetization but also from calculational difficulties in solving large-scale non-linear inverse problems. Aiming to address these difficulties under the remanence, we propose a three-dimensional (3-D) frequency-domain imaging approach of magnetic amplitude anomaly (MAA) to derive a 3-D model of susceptibility or magnetization intensity in the subsurface. The MAA, insensitive to the remanence, is adopted to weaken the influence of remanent magnetization effectively on imaging. The frequency-domain formulas for forward calculation and imaging of MAA are derived without the involvement of traditional non-linear inverse equations and are utilized for 3-D imaging, both of which make the imaging fast, stable and suitable for large-scale data. In addition, the iterative optimization algorithm is employed to improve the fitting between the real MAA and the predicted one caused by the imaging model, making the imaging high-precision. Tests on synthetic data with complicated remanent magnetization prove the effectiveness of our approach, and demonstrate that it is easy to conduct, rapid, stable, noise-resistant and produces much better models than the traditional space-domain or frequency-domain inversion approaches. Test on the real aeromagnetic data from the Datong area in North China also verifies the effectiveness of our approach with the resultant crustal 3-D susceptibility model better than the traditional space-domain inversion approach. Our model reconstructs the architecture of Precambrian basement within the crust in the study area, and reveals the existence of a northeast-trending high-magnetism body beneath the Datong volcanic group, extending from the deep crust upward to near the surface, which is interpreted as the cooled mafic magma conduit. Our model will provide significant constraints for understanding volcanic genesis of the Datong volcanic group.

Density interface inversion is an important approach for quantitative interpretation of gravity data, used to reveal the morphology of structural interfaces such as the basement, Moho boundary, and the Lithosphere-Asthenosphere Boundary (LAB). However, common density interface forward modeling and inversion algorithms in Cartesian coordinate system are unsuitable for large regional scales, while conventional algorithms in spherical coordinate system are not suitable for ground observation plane and rarely account for the density variations underground. In this paper, we develop algorithms for gravity forward modeling and iterative inversion of variable-density interface in spherical coordinate system. The gravity forward modeling formula of variable-density tesseroid unit in spherical coordinate system is derived based on the Gauss-Legendre quadrature integration algorithm, and the improved adaptive subdivision strategy is utilized to make the forward modeling algorithm suitable for ground observation plane with high precision. Then, the iterative inversion algorithm for variable-density interface in spherical coordinate system is proposed based on our gravity forward modeling algorithm and Bott’s iteration algorithm. We validate the effectiveness and reliability of the proposed algorithms by using a spherical shell theoretical model and a variable-density interface model. Finally, we apply the proposed algorithms to the gravity data on ground observation plane to derive the depth distribution of the LAB beneath Chinese mainland. The inversion result indicates that the lithospheric thickness generally exhibits a trend of being thicker in the west and thinner in the east, with 110°E as the boundary. The thickest region exceeds 250 km, located in the central Tibetan Plateau, while the thinnest area is approximately 60 km, found from the eastern Northeast China to the eastern South China.