Three-dimensional (3D) density gravity inversion is a fundamental yet challenging problem in geophysics due to its inherent non-uniqueness and computational complexity, particularly for large scale regions where Earth curvature cannot be neglected. In this study, we propose a deep learning based inversion framework in a spherical coordinate system that integrates tesseroid gravity forward modeling with a modified U-Net architecture incorporating a self-attention bottleneck. The proposed network is designed to capture both local structural details and long range spatial dependencies, while a composite loss function combining weighted mean squared error, structural similarity, and frequency domain constraints enhances reconstruction accuracy and preserves sharp geological boundaries. The method is trained on a diverse synthetic dataset representing a wide range of geological scenarios, including regular and irregular density bodies, to ensure strong generalization capability. Synthetic experiments demonstrate that the network can accurately recover subsurface density distributions from noisy gravity data, preserving the geometry, connectivity, and amplitude of anomalies, although minor smoothing persists near sharp interfaces. Application to the Tibetan Plateau reveals a clear depth dependent evolution of lithospheric structure, with highly heterogeneous shallow crust, increasingly coherent patterns in the middle-lower crust, and relatively uniform density in the upper mantle. A pronounced low density anomaly beneath central Tibet is identified, consistent with crustal thickening and isostatic compensation, while a transition to higher density material toward the eastern region suggests lithospheric modification processes such as thinning or underplating. These results indicate that the proposed framework provides an efficient and robust alternative to conventional inversion approaches, enabling high resolution 3D density imaging with reduced computational cost. Furthermore, the study highlights the potential of combining spherical modeling with deep learning techniques for large scale geophysical inversion problems and offers new insights into the lithospheric structure and tectonic evolution of Tibet.
Deep-focus earthquakes occur under pressure–temperature conditions that inhibit conventional brittle failure, yet their relative abundance varies strongly among subduction systems. Here, I quantify this variability using the deep-seismicity fraction F300, defined as the fraction of slab-associated earthquakes occurring deeper than 300 km. I combine USGS ComCat earthquakes with preferred magnitude M≥5.5 from 1976 to 2025 with three-dimensional Slab2 geometry and independent slab-scale geometric, thermal, and kinematic parameters. The primary sample comprises 18 well-sampled slabs, 17,079 associated earthquakes, and 1342 events deeper than 300 km. F300 ranges from 0 to 0.2647, with nine slabs containing no earthquakes deeper than 300 km. Maximum Slab2 depth is the independent predictor with the most consistent support, showing a Spearman correlation of ρS=0.808 with F300 (95% bias-corrected and accelerated confidence interval 0.591–0.936) and being selected in every nested leave-one-slab-out fold of both binomial and beta-binomial independent-control models. Incoming-plate age provides secondary, data-support-dependent information, whereas convergence rate and other slab-scale controls are less stable across model specifications. Strong extra-binomial variability requires dispersion-adjusted uncertainty, while beta-binomial sensitivity supports structured extra-binomial heterogeneity. The principal slab ordering is robust to magnitude threshold, slab-support criterion, and earthquake-to-slab association distance. Lowering the deep threshold to 250 km preserves the broad slab ranking but changes the relative predictive performance of the model families. These results identify geometric slab depth extent as the most reproducible first-order observational correlate of deep-seismicity persistence without implying a unique causal or microphysical mechanism.
Displacement along the north–south Chaouat–Mornaguia–El Feja strike-slip fault in Tunisia induced localized extension at fault jogs, promoting differential loading between a brittle overburden and an underlying salt layer. The subsurface Ain Krima salt stock in northeastern Tunisia provides a valuable case study for examining diapiric processes and their associated depositional and structural evolution. Field mapping, gravity data, and structural observations from Jbel Ain Krima reveal key fault–salt interaction processes including: fault segmentation, pull-apart basin development, spatial links between salt extrusion and transtensional step-over zones, diapir initiation with associated sedimentary hiatuses, and diagnostic signatures of subsurface diapir. The structural configuration at Jbel Ain Krima exemplifies a releasing step-over zone within a potential N–S shear system, intruded by diapiric activity since the Eocene. During the Tortonian, folding processes enhanced transtensional salt tectonics across linkage zones of sinistral N–S fault segments. From the Pliocene onward, in conjunction with the accumulation of younger, unconsolidated sediments, the aborted diapir intruded the overburden that had been thinned by extensional forces, as indicated by doming with Y-shaped radial extensional fractures and radial drainage network in the roof strata.
The Pampean flat slab is a distinctive segment of the central Andes, where the oceanic Nazca Plate subducts nearly horizontally for several hundred kilometers, undergoing upward and downward bending along its path. This geometry is associated with elevated intraplate seismicity driven by complex stress fields generated during slab flexure. We analyze the finite-source properties of the 2019 M-w 6.7 Coquimbo mainshock to determine its causative fault by combining near-field and teleseismic datasets. We also examine the aftershock activity located using a local array, and moment tensors from global catalogs, including the 1997 M-w 7.1 Punitaqui earthquake. Intraplate yield stresses are estimated using slab curvature and two-dimensional iso-viscous thermal models. Our results indicate that most moderate-magnitude intraplate earthquakes (M-w< 6.0) cannot be solely explained by present-day slab curvature. This implies that plate bending, together with slab pull, suction forces, and interplate friction, significantly contributes to the intraplate stress field. The Punitaqui and Coquimbo earthquakes, located near the onset of flat subduction (50-80 km depth), likely result from tensional stresses induced by abrupt changes in dip and curvature. The flat-slab zone (400-500 km wide) hosts diffuse intraplate seismicity, predominantly tensional and largely independent of slab curvature. Although slab pull-related stresses are reduced in flat-slab zones, regional tension remains dominant and likely intensifies at the leading edge, where the slab resumes steep subduction. Additional suction forces may also contribute to the tensional regime along the flat slab, particularly near the hinge zone.
The Northwestern Ethiopian Plateau (NWEP) formed initially from the uplift and volcanism associated with the Afar mantle plume; however, its present topography is sculpted by the adjacent rifting of the Main Ethiopian Rift (MER) and Afar Rift (AR). This region is known for being covered by the Ethiopian Large Igneous Province and significant volcano-tectonic features, including shield volcanoes (e.g., Choke, Semen, Guna, and Guguftu), a local intraplate basin (e.g., Lake Tana Basin (LTB)), and lithospheric inherited structures (e.g., Yerer Tulu Volcano-Tectonic Lineament (YTVTL). However, the tectonic and geodynamic implications of these features within a broader context with respect to the East African rift system (EARS) are least understood. In this paper, we used remote sensing and geophysical imaging to understand the volcano-tectonic architecture of the LTB and the NWEP and their geodynamic implications within the northern EARS. The Landsat imagery maps reveal three half-graben structures around LTB, including Dengel Ber (NE-SW), Debre Tabor (E-W), and Gondar (N-S). Gravity anomaly maps show positive Bouguer anomalies beneath each graben, suggesting the presence of relatively dense subsurface materials (e.g., intrusive bodies or thinned crust with mantle upwelling), while low-density materials beneath the center of the LTB in the residual anomaly map may represent sedimentary infill and volcanoclastic deposits associated with collapse-related subsidence underlying the basin interior. However, upward continuation maps show that the three grabens merge into a single positive anomaly at a depths, indicating that the LTB represents a common center of subsidence rather than an isolated tectonic depression. In addition, the shield volcanoes and inherited structures show gravity maxima. The 2D forward gravity model reveals that there is magmatic intrusion beneath LTB and associated shield volcanoes. We therefore propose that the LTB represents an intra-plateau volcano-tectonic collapse structure generated by mantle-derived magmatic impingement associated with the broader East African Superswell and Afar plume system. Subsequent extensional deformation related to the evolution of the EARS overprinted this magmatic framework, producing the present graben architecture and fault-controlled basin morphology. This integrated interpretation reconciles the previously competing models by suggesting that both magmatic collapse and later rift-related subsidence contributed to the development of the LTB, with ongoing geothermal and volcanic activity indicating that the underlying magmatic system may remain active today. Further geophysical investigations, including seismic tomography and magnetotelluric surveys, are essential to refine the subsurface distribution of magmatic bodies.
