Xiaofei Chen

Array‐based surface wave tomography has significantly enhanced the resolution of crustal imaging; however, rigorous quantification of posterior uncertainty in these regularized inversions remains a critical challenge. Conventional statistical methods, such as raw bootstrap resampling, often underestimate uncertainties in data‐sparse regions due to the dominance of regularization terms. This underestimation creates an artificial precision that risks misinterpreting inversion artifacts as robust structural features. To address this, we propose a hybrid uncertainty quantification strategy tailored for regional‐scale dense arrays, integrating data‐driven bootstrap resampling with model‐based resolution matrix correction. By utilizing the resolution matrix to explicitly debias the bootstrap variance, our approach provides a first‐order uncertainty metric for identifying and down‐weighting poorly constrained model estimates in the inversion's null‐space while retaining sensitivity to data inconsistencies in well‐resolved regions. We applied this workflow to ambient noise data from the ChinArray‐Himalaya project, extracting multimodal dispersion curves via the frequency‐Bessel (F‐J) transform. The derived hybrid uncertainty metric serves as a quantitative validity mask, successfully delineating reliable imaging regions for both fundamental and higher‐order modes, thereby facilitating objective, automated quality control for large‐scale data sets. Semi‐synthetic robustness tests further demonstrate that, unlike analytical or raw bootstrap methods, the hybrid strategy reliably distinguishes among data gaps (null‐spaces) and data conflicts. Ultimately, incorporating this resolution‐corrected uncertainty to weight subsequent 1D shear wave velocity inversions significantly suppresses regularization artifacts, laying a robust statistical foundation for constructing physically reliable 3D structural models.

Straight faults are considered prone to large supershear ruptures and seismic gaps to likely host future earthquakes. Both predictions may have been fulfilled in the magnitude ( M ) 7.7 Mandalay earthquake of 28 March 2025 on the Sagaing Fault. Exceptional video footage of the fault rupture, high-rate Global Navigation Satellite System, and seismological and remote sensing data reveal a supershear slip pulse that formed early and ruptured a well-identified seismic gap but propagated well beyond it, resulting in an exceptional rupture length of 510 kilometers. By contrast, the slip pulse observed in the video footage has subshear characteristics. These observations can be reconciled by a preferred conceptual interpretation invoking depth-dependent weakening properties, stress heterogeneity, and possible dynamic triggering. These properties allow for energetic ruptures that stay confined to the depth range locked in the interseismic period but are not confined to seismic gaps along strike.

The near‐surface shear‐wave velocity structure is crucial for geological engineering and near‐surface exploration. Surface wave inversion is a key technique for obtaining the velocity structure, but traditional methods suffer from limitations such as heavy reliance on the initial model, high computational cost, and susceptibility to local minima. This study proposes INVNET, a deep learning model based on Convolutional Neural Networks, for efficient inversion from surface wave dispersion spectra to the Versus structure. We first constructed a large‐scale synthetic data set by randomly generating layered velocity models conforming to geological priors and performing forward modeling using the Generalized Reflection/Transmission Coefficient (GRTC) method. To enhance data realism, we introduced a multi‐depth noise source weighted stacking strategy, generating 20000 sample pairs of noise‐rich dispersion spectra and their corresponding velocity models. INVNET adopts an encoder style architecture that can automatically extract deep features from dispersion spectra and map them to velocity values. Testing on synthetic data shows that INVNET achieves high inversion accuracy, with a single inversion taking less than 1 s. In a practical application to field data from the Qademah area, INVNET rapidly inverted a reasonable three‐layer velocity structure, consistent with results from conventional methods. This result demonstrates the model's potential for real‐world applications. This research demonstrates that deep learning provides a novel solution for surface wave inversion that is independent of initial models, offers high accuracy, and is fast.

The 2023 Mw 6.0 Jishishan earthquake in China caused severe damage and fatalities. Waveform analyses indicate that using a 1 Hz cutoff frequency underestimates broadband ground motions by about 70% for peak ground velocity (PGV) and up to 90% for peak ground acceleration. Physics-based simulations up to 3.5 Hz using an inverted finite-fault rupture model still underestimate PGV due to missing high-frequency components, emphasizing the need for broadband rupture models. In this study, we develop a broadband hybrid source model incorporating deterministic features of the inverted finite-fault model. Simulations with this hybrid source model produce PGV values that better match observations. Further analyses demonstrate that, for the 2023 Jishishan earthquake, ground motions simulated using the hybrid source model are jointly controlled by large-scale rupture characteristics (e.g., slip distribution and rupture velocity), which are generally well constrained by finite-fault inversion, and rise-time distributions, which are typically poorly constrained. These findings emphasize the importance of broadband hybrid source models for improving physics-based ground-motion simulations and in turn, seismic hazard assessment.

Despite the growing wealth of seismic observations, the physics of earthquake initiation remains elusive. And foreshocks offer one of the few direct observational constraints on this process. Here we exploit Japan’s dense seismic networks to analyze immediate foreshock–mainshock pairs with inter‑event times ≤ 60 s. The foreshock–mainshock interval depends only weakly on their magnitude difference. Spectral estimates show that some immediate foreshocks share apparent corner frequencies with their mainshocks despite magnitude contrasts of 1–2, implying departures from constant stress drop scaling. Although relocations indicate that immediate foreshocks nearly coincide with the mainshock hypocentre, empirical Green’s function analysis of representative sequences shows their stress drops are lower by a factor of about 3–10 relative to nearby events of similar magnitude. Guided by these findings, we develop a paused‑rupture model, implemented via dynamic‑rupture simulations with rate‑and‑state friction, that reproduces the short foreshock–mainshock lags, spectral similarity and reduced foreshock stress drops. Our results suggest a lagged reinitiation mechanism whereby the foreshock and mainshock constitute successive phases of a single dynamic rupture.

With increasing computational power, global seismic wavefield simulations have become feasible, with the spectral-element method (SEM) demonstrating outstanding performance. Despite these advances, extending other numerical methods to the global scale remains nontrivial. In particular, the traditional finite-difference method (FDM) encounters notable challenges in global-scale applications due to its limited grid flexibility. To address this issue, we propose a curvilinear grid FDM (CGFDM) tailored for 2D global seismic-wave modeling. The method integrates several key techniques: (1) the dispersion-relation–preserving optimized MacCormack operators on collocated grids to preserve dispersion and accuracy, (2) the traction-image method for free-surface boundary conditions, (3) the split-node technique for solid–fluid coupling, and (4) multiblock structured grids to overcome polar singularities and enable nonuniform meshing adapted to Earth’s heterogeneous structure. These innovations substantially enhance both the geometric adaptability and numerical stability of FDM for global-scale problems. This article presents the theoretical formulation and numerical implementation of CGFDM, introduces a multiblock mesh design suitable for global simulations, and verifies the method against the SEM benchmark. The results demonstrate that CGFDM achieves high accuracy and computational efficiency, establishing it as a robust and promising alternative for large-scale seismic wavefield modeling.

How does a source rupture propagate in a geometrically complex multifault system? Reconstructing the spatiotemporal evolution process of such complex geometric ruptures is a crucial way to advance understanding of earthquake source physics, but sufficient observations of such events are difficult to obtain due to their rare occurrence. On 6 February 2023, two destructive earthquakes successively struck Kahramanmaraş, Türkiye, within 9 hr, forming the largest continental earthquake doublet ever recorded by modern seismological instruments. Here, we integrate the teleseismic backprojection imaging and the joint finite-fault inversion to investigate the rupture kinematics of the two earthquakes. We find that the two earthquakes exhibit distinct rupture propagation patterns, with transient and persistent supershear ruptures occurring in the first Mw 7.8 event and the subsequent Mw 7.7 one, respectively. The Mw 7.8 earthquake originates on a splay fault branching from the east Anatolian fault (EAF), and propagates to the main strand of the EAF at a subshear velocity. After passing through fault bends, the rupture generates three transient transitions from subshear to supershear, with two located in the Amanos segment and one in the Erkenek segment. In contrast, the rupture of the Mw 7.7 earthquake initiates in the middle section of the northern strand of the EAF, and propagates bilaterally at a persistent supershear velocity. Based on the analysis of a series of potential causes contributing to the supershear rupture, we argue that the difference in the geometric complexity of the seismogenic faults directly controls the variation in rupture patterns between the two earthquakes. This highlights that the geometric complexity is a key factor capable of modulating the kinematic characteristics of large-scale source ruptures.

The Cenozoic collision and convergence between the Indian and Eurasian plates drove north-south crustal shortening, uplift, and outward growth of the Tibetan Plateau. Since the Late Miocene-Early Pliocene, the plateau has expanded eastward and southeastward, while lateral extrusion, crustal thickening, and surface uplift have jointly shaped its eastern and southeastern margins. Here we collect continuous waveform data from dense seismic arrays in the eastern and southeastern Tibetan Plateau to construct three-dimensional S-wave velocity and radial anisotropy models. We first apply the frequency-Bessel transform (F-J) method to extract multimodal surface wave dispersion curves from ambient noise data, and then invert these measurements using multimodal ambient noise tomography.Our models delineate the detailed three-dimensional geometry of the mid- lower crustal low-velocity zone (LVZ). The spatial distribution of the LVZ shows a close correspondence with regions of pronounced lateral extrusion and crustal deformation. The LVZ also exhibits strong positive radial anisotropy, which provides seismic constraints on possible ductile crustal flow pathways. In addition, variations in LVZ thickness show clear spatial correlations with surface elevation, Moho depth, and present-day GPS velocities. These observations suggest that weak mid-lower crustal materials may play an important role in surface uplift, crustal thickening, and ongoing deformation across the eastern and southeastern Tibetan Plateau. Based on these results, we propose a three-dimensional geodynamic model that explains two contrasting geomorphic end members: the steep eastern margin and the gently sloping southeastern margin of the Tibetan Plateau. This model provides a framework for understanding the tectonic evolution and outward growth of eastern and southeastern Tibet since the late Cenozoic.

:The collision and convergence of the Indian and Eurasian plates during the Cenozoic caused the Tibetan plateau to shorten in the north-south direction. As the plateau grew, lateral extrusion contributed to its rapid uplift and further crustal thickening. The Tibetan plateau has experienced eastward to southeastward growth from the Late Miocene to the Early Pliocene. We collected the continuous waveform data by dense arrays in the eastern and southeastern Tibetan plateau to obtain high-resolution 3-D S-wave velocity and radial anisotropy model by the Frequency-Bessel transform method. The high resolution 3D S wave velocity model reveals the detailed top and bottom boundaries of the Low Velocity Zone (LVZ) within the mid-lower crust (depth range 12~36 km). LVZ exhibits strong positive radial anisotropy, providing direct evidence for the ductile crustal flow paths. The LVZ distribution coincides with regions of pronounced lateral extrusion and deformation in eastern and southeastern Tibetan plateau.The high resolution model reveals strong correlations between variations in LVZ thickness, with surface topography, the depth of Moho boundary. and even with the current GPS velocities. These evidences indicate that ductile crustal flow acts as an active mechanical layer dominating uplift, crustal thickening, and ongoing deformation in the Tibetan plateau. Based on the ductile crustal flow model, this study explains two distinct geomorphic end-member cases, the steep eastern margin and the gently sloping southeastern margin of the Tibetan plateau. These findings establish the dynamic framework for the tectonic evolution in the eastern and southeastern Tibetan plateau since the late Cenozoic.