Geophysical Journal International

SummaryFull-waveform inversion (FWI) is a method that utilizes seismic data to invert the physical parameters of subsurface media by minimizing the difference between simulated and observed waveforms. Due to its ill-posed nature, FWI is susceptible to getting trapped in local minima. Consequently, various research efforts have attempted to combine neural networks with FWI to stabilize the inversion process. This study presents a bidirectional physics-constrained full waveform inversion (BP-FWI) framework that leverages transfer learning by pre-training on simple initial models and utilizing the results. Additionally, it employs FWI gradients to co-optimize both the neural network and the adaptive residual learning module under bidirectional physics constraints. By eliminating the reliance on a large amount of manually constructed synthetic datasets, the proposed training strategy addresses the challenge of data dependency. Furthermore, through the joint optimization strategy guided by bidirectional constraints, the neural network is able to focus on integrating physically-informed prior knowledge into global stratigraphic representations, while the adaptive residual learning module specializes in learning residual mappings from the network’s output, thereby capturing subtle inter-layer velocity variations in local geological structures. Evaluating the method on two benchmark models under various conditions, including absent low-frequency data, noise interference, non-uniform receiver configurations, and differing initial models, along with corresponding ablation experiments, consistently demonstrates the superiority of the proposed approach.

SummaryA comprehensive full-waveform inversion model of the seismic velocity, covering nearly the entire tectonic domain of the western Pacific (FWP24) is developed using an optimized many-core version of SPECFEM3D_GLOBE on the New Generation Sunway supercomputer. Taking the global adjoint tomography model GLAD-M25 as the initial model, the three-component seismograms from 1 228 earthquakes recorded at 3 687 stations are employed in iterative gradient-based inversions for three period bands: 40-100 s, 17-40 s, and 10-60 s. A total of 36 iterations are carried out using the conjugate gradient method to update the velocities of horizontally and vertically polarized P-waves and S-waves (Vph, Vpv, Vsh, and Vsv) in the FWP24 model. This process systematically reduces the phase difference between the synthetic and observed seismograms within the phase measurements. Compared with existing region inversion results, the FWP24 model realizes a wider, more continuous, and higher-resolution inversion range, including all subduction zones in the western Pacific (e.g. Kurile-Japan, Izu-Bonin-Mariana, New-Britain-Solomon, New-Hebrides, and Tonga-Kermadec). Furthermore, compared to the initial model, FWP24 reveals more detailed structures particularly in oceanic regions around the Philippine Sea Plate, the Caroline Sea Plate and the Ontong-Java Plateau by applying more seismic data.

SummaryThe ratio R of shear-wave to compressional-wave velocity variations (dlnVs/dlnVp) is a useful physical parameter to study the thermochemical properties of the Earth’s interior. Several approaches have been employed to estimate R (or its inverse 1/R), but they either assume the same local resolution in models of dlnVs and dlnVp or assume the same ray paths for S- and P-phases, while excluding valuable data and overlooking uncertainties. We overcome these issues by characterizing both dlnVs and dlnVp through the Backus-Gilbert based SOLA method to obtain R including its uncertainties. This approach enables us to ensure that dlnVs and dlnVp share the same local resolution, making it possible to compute their ratio through division. In addition, SOLA provides uncertainties on dlnVs and dlnVp, which we propagate into our estimates of R using the Hinkley distribution for dlnVs/dlnVp. When resembling a Gaussian, the Hinkley distribution provides Gaussian uncertainties for R, enabling us to interpret tomographic features as for instance in terms of slab morphology or partial melt with greater confidence. To illustrate our new approach, we use a data set of P- and S-phase onset-time residuals from ISC to infer the velocity anomalies and the ratio R (or 1/R) in South-East Asia between 100 and 800 km depth. As the SOLA method is driven by data uncertainties, we reassess the provided ISC uncertainties using a statistical approach before developing models of dlnVs and dlnVp with their uncertainties. Based on our quantitative model estimates, we argue that a large velocity anomaly below the Sumatra slab, with a value of R over 2.5, is resolved given our data and their uncertainties. However, in contrast to previous work, we do not find evidence for a slab hole under Java. Our proposed approach to obtain R with uncertainties using the Hinkley distribution can be applied to a large range of tomographic imaging settings.

SummaryWe investigate the effect of statistically non-stationary turbulence in the Earth’s outer core on the effective turbulent electromotive force generated by the convectively driven flow of liquid iron and the evolution characteristics of the geomagnetic field. The non-stationarity means that interactions of distinct waves are crucial, and the effect of beat induces a slow time variation of the large-scale electromotive force. This provides an attractive and fairly simple physical mechanism for the random appearance of short-lived geomagnetic excursions and reversals separating long periods of relatively stable field, through non-synchronized evolution of the amplifying α-effect and turbulent diffusion. This implies rare and random appearance of simultaneous suppression of the α-effect and enhancement of diffusion which leads to a sudden magnetic energy drop, i.e. an excursion. The turbulent field of what is termed MAR waves (Magnetic-Archemedean-Rossby) is analysed. The dispersion relation and structure of such waves involving the joint effect of the Lorentz, buoyancy, and Coriolis forces together with curvature of the core-mantle boundary are obtained and utilized for estimation of the non-stationary electromotive force in the core. The solutions for the large-scale dipole possess an Earth-like behaviour, magnitude, and timescales, and the physical mechanism of the process, including identification of two dynamically important parameters, is discussed. Similar ideas concerning the dynamics of waves within the so-called Stratified Ocean at the top of the Core (SOC) were considered in the recent work Mizerski (2025). The SOC is an important but thin, strongly stratified layer near the core-mantle boundary, and here, the possibility of global non-equilibrium dynamo mechanisms is analysed. It is possible that the surface and bulk mechanisms coexist in the core, both adding to the complexity of the observed picture of reversal occurrences.

SummaryEvaluating sedimentary texture is crucial for managing of aquifers and assessing groundwater quality. Direct methods of aquifer characterization (like pumping tests) are of limited value because of their invasive character and connection to boreholes, which results in scarce data networks. Non-invasive geophysical methods complement the invasive methods. Among geophysical methods, the spectral induced polarization (SIP) method, which allows measuring frequency dependent complex electrical impedance, is especially promising in this context, because it offers insights into the specific surface area (which is the proxy of the clay content), as well as into textural features of porous media (like pore-size or grain-size distributions).However, SIP data concerning sand-clay mixtures remain scarce. This lack of experimental data, in turn, hinders further development of the IP theory of clayey media. To fill this gap, we carried out a set of experiments to study SIP signatures of artificial sand-clay mixtures with varying clay content, clay types, and pore water salinity. The dataset includes nine mixtures with kaolinite or bentonite in various contents. We equilibrated each mixture with six NaCl solutions ranging in salinity from 0.1 to 30 g l−1. Based on these measurements, we first obtained the formation factor and the surface conductivity values of the samples. Then, we interpreted this dataset in terms of the real and imaginary conductivity, the formation factor, the in-phase surface conductivity, and the normalized total chargeability. Finally, based on the Debye decomposition approach, we converted the IP spectra into relaxation time distributions (RTDs) to analyse dominant relaxation times in comparison with micro-computed tomography (μCT) images.We show that the in-phase conductivity of the sand-bentonite mixtures strongly exceeds that of the sand-kaolinite mixtures with the same clay content. We attributed this difference to the higher surface conductivity of the bentonite clays. The quadrature conductivity exhibits a clear dependence on the clay type, its content, and the conductivity of the pore water. Our observations reveal that both quadrature conductivity and normalized chargeability increase with kaolinite content. However, for the bentonite samples, these parameters show maxima rather than a gradual trend. We explained this behaviour when comparing RTDs with μCT images. This comparison allowed us to identify elements of texture (sand grains coated with a thin film made of clay, clay aggregates of different sizes, clay “bridges” connecting two grains or multiple grains, etc.), which are responsible for IP of various intensities and different time constants. The size and morphology of these elements depend on the clay content, mineralogy, the clay phase topology, and pore water salinity.Ultimately, the combined application of the SIP and μCT methods to a variety of sand-clay mixtures enabled us to differentiate the samples with different clay types, contents, and water salinities. We believe that these petrophysical results can serve as the basis for SIP application to detect remotely different clay types and content, and to monitor the water salinity in clayey rocks, soils, and sediments.

SummaryDirect Current Electrical Resistivity Tomography (ERT) is a widely used geophysical method for near-surface investigations, offering high-resolution imaging for geological, engineering, and environmental applications. While traditional ERT surveys typically target depths of 0–200 m, technological advancements have enabled deeper investigations, commonly referred to as Deep ERT. In this study, we explore the practical challenges and methodological improvements associated with Deep ERT, particularly when combined with Induced Polarization (IP) measurements. Rather than other electromagnetic methods, ERT offers a more straightforward framework for analyzing IP effects, which can potentially correlate with the volume fraction of ore minerals. Nevertheless, deep IP investigations are often challenged by weak signal strength and various sources of electromagnetic interference. To address these challenges, we evaluated key strategies including survey planning, high-power current injection, unconventional electrode configurations, and advanced signal processing techniques. The adoption of nodal geophysical recording systems eliminates the logistical constraints of cabled multi-electrode setups, improving flexibility and data acquisition efficiency. Additionally, continuous full time-series recording allows for enhanced noise filtering and signal stacking, ultimately increasing the signal-to-noise ratio and extending the effective exploration depth. We demonstrate this methodology through a comprehensive case study conducted at the Koillismaa Linear Intrusion Complex in Finland, where a 3D Deep ERT-IP survey successfully delineated conductive and chargeable anomalies at depths exceeding 1.5 km. These anomalies closely align with independent gravity and borehole logging data, consistent with the mafic-ultramafic intrusion structures. Our results emphasize the importance of balancing data quality, survey efficiency, and spatial resolution in survey design. This work not only provides a robust workflow for the implementation of Deep ERT-IP surveys but also represents the first documented successful acquisition of high-quality IP data at these substantial depths, significantly advancing the state of deep geoelectrical exploration.

SummaryDistributed Acoustic Sensing (DAS) technology has gained widespread attention in seismic exploration due to its high spatial resolution and low deployment cost. However, the presence of coupling noise in DAS data significantly affects the accurate extraction and interpretation of seismic signals. Coupling noise typically appears as narrowband stripe-like or zigzag-like interference and shares similar characteristics with seismic signals in the time-space (T-S) domain, making it challenging for traditional denoising methods to achieve effective signal-noise separation without residual noise or signal leakage. To address these challenges, this paper proposes a deep learning-based dual-domain fusion approach that integrates both T-S and frequency-wavenumber (F-K) domain information to enhance the accuracy of coupling noise separation. The method leverages the narrowband characteristics of coupling noise in the F-K domain while incorporating spatiotemporal information from the T-S domain to achieve cross-domain feature fusion, thereby improving the separability between signals and coupling noise. Experimental results demonstrate that the proposed method significantly improves coupling noise suppression performance on both synthetic and field DAS vertical seismic profile (VSP) data while minimizing signal leakage. Furthermore, in corridor stacking experiments, the method effectively reduces the impact of coupling noise on seismic interpretation, improving the reliability of subsurface formation analysis. Compared to conventional F-K filtering, single-domain network and denoising diffusion model, the proposed approach achieves superior performance in terms of coupling noise suppression and signal amplitude preservation.

SummaryElectrical properties of porous media consisting of solid and fluid phases have been continuously investigated in the study of geomaterials given their strong link to pore space characteristics (e.g. pore size and connectivity). Over the past decades, numerous theoretical models have been developed to determine the electrical conductivity of the porous media as a function of conductivities of their constituents and their relative proportions. In this paper, we present a new theoretical model to calculate the electrical conductivity of a weakly transversely isotropic (TI) porous medium with two conducting phases. We first use the well-established series and parallel electrical connections, together with a newly introduced coefficient g, to construct a conductive cell that represents the porous medium’s microscopic structure. We then obtain the macroscopic weakly TI conductivity stacking these cells in one dimension. We innovatively introduce a normal probability distribution to simulate the distribution of porosities across cells. Good agreement between our theoretical predictions and literature data validates the model for weakly TI porous media. We also show that series and parallel connections of the solid and liquid phases provide reliable building blocks for more advanced models, and that using a normal distribution to simulate electrical anisotropy in quasi-isotropic or weakly TI porous media is viable. Finally, we use the model to study the effects of key variables on weakly TI conductivity. We find that increasing the coefficient g reduces electrical conductivity and that the Electrical Anisotropy Coefficient (EAC) attains its maximum at 50 per cent porosity in weakly TI porous media with two conducting phases.

AbstractSeismic observations reveal significant anisotropy in the D″ region, providing direct constraints on mantle flow and deformation. However, the global anisotropy pattern and its relationship with subduction history, mineral deformation, and rheology in the lower mantle remain unclear. We analyze published regional shear-wave splitting and null measurements, along with waveform inversions, which reveal rapid lateral variations in anisotropy near the edges of large low shear velocity provinces (LLSVPs). We combine mineral physics results of temperature- and pressure-dependent elastic tensors, slip systems, and phase transition mechanisms to explore potential deformation scenarios. We set up models that begin with dynamic thermochemical convection, tracking the deformation history driven by the subduction, evolving crystal fabrics, and cumulative seismic anisotropy. Models show that post-perovskite (pPv) with a (001)-dominant slip system, combined with viscosity changes and texture inheritance during the bridgmanite-post-perovskite (Br-pPv) phase transition and the reverse transition, best reproduces the distinct anisotropy patterns observed in upwelling regions such as plume roots and LLSVP edges. The nominal model is time-dependent, showing strong seismic anisotropy when slabs impinge on the CMB that diminishes toward the LLSVP, followed by plume development at the LLSVP edge with significant anisotropy. Within LLSVPs, internal convective upwellings and downwellings can explain the intermittent, spatially clustered anisotropy. We further demonstrate the potential for constraining LLSVP composition through the observed weaker anisotropy within these structures compared to the surrounding mantle, with our results favoring a Br-rich composition. Computations indicate that the bulk of the lower mantle remains nearly isotropic despite significant texture accumulation through dislocation glide, and that seismic anisotropy can extend several hundred kilometers above the core–mantle boundary.

AbstractReverse time migration based on geometric mean or cross-correlation is a powerful passive-source imaging technique that can produce high-resolution source images even under low signal-to-noise ratio conditions. When the velocity model is inaccurate, a hybrid method combining geometric-mean and arithmetic-mean reverse time migration is typically used to reduce sensitivity to model errors. Conventional hybrid methods usually employ a grouping strategy, in which receivers are divided into groups and multiplicative operations are performed between these groups. However, this strategy essentially utilizes only a subset of receiver combinations, which may compromise imaging quality when the number of receivers is insufficient. To overcome this limitation, a novel combinational autocorrelation reverse time migration imaging condition is proposed. Our method forms multiple combinations of receivers and conducts zero-lag autocorrelation on the extrapolated wavefields of these combinations. The cross terms generated by the autocorrelation operation correspond to all possible receiver combinations. Finally, these autocorrelation results are linearly stacked in order to eliminate interference terms while preserving the cross terms. By including more receiver combinations, the proposed method can provide improved imaging performance. Furthermore, due to the adoption of the autocorrelation algorithm, the new method achieves minimal memory usage among methods based on reverse time migration, which makes it especially suitable for three-dimensional problems. Acoustic numerical simulations verify the effectiveness and advantages of the new method in both two-dimensional and three-dimensional scenarios. Additionally, the mathematical relationship between our method and conventional methods is discussed, clarifying the applicable scope of the new method.

SummaryIn the previous paper of this series, a petrophysical model named the Dynamic Stern Layer (DSL) model was extended to describe induced polarization phenomena associated with permafrost by capturing direct and indirect effects associated with the presence of ice in porous media. In the present paper, time-domain induced polarization data obtained in field conditions are interpreted thanks to this updated DSL model. We selected three different test sites in order to apply the DSL model to very different conditions of low and high ice contents to see how ice content directly and indirectly affects geoelectrical measurements. A first survey is performed along a cross-section of a ridge in the Kangerlussuaq mountains of Greenland (Site I). In this area, the rock corresponds to a Precambrian granite characterized by a rather low (< 5%) porosity and therefore a low ice volumetric content on the North face of the ridge. We do not see any direct ice polarization contribution in the data obtained with a current injection period of 1 s. We also performed a field survey close to Col des Vés (2846 m a.s.l., Tignes, French Alps, Site II). This site corresponds to a complex ground ice body overlying a substratum made of a low-porosity marble, both having high resistivity values. The front of this body is characterized by a small amount of residual ice while the roots are ice-rich. Therefore the porosity at this site is high and the ice content highly variable. This case study showcases the role of ice in the induced polarization data in terms of high chargeability values (close to 1 as predicted by the theory in which ice behaves as a surfacic protonic semi-conductor) at the roots of the complex ground ice body. A third site (Site III) corresponds to a profile crossing the Aiguille du Midi (3842 m a.s.l., Chamonix), also in the French Alps in a low porosity granitic environment. Laboratory experiments are used to interpret the tomograms of the electrical conductivity and normalized chargeability using the DSL model and water content and Cation Exchange Capacity tomograms are reconstructed at these sites. This study demonstrates the ability of induced polarization to be an efficient tool to characterize permafrost in very different field conditions.

SummaryModelling induced polarization (IP) effects in electromagnetic (EM) data is increasingly becoming a standard tool in mineral exploration, but the industry standard is still based on one-dimensional (1D) forward and Jacobian modelling. We have developed a three-dimensional (3D) electromagnetic forward and inversion method within the EEMverter modelling platform, incorporating IP effects. The 3D computations are performed in the frequency domain using the vector finite element method and then transformed into the time domain via Hankel transformation. This approach enables modeling of any IP parameterization, ranging from the simple constant phase angle model to a full Debye decomposition. Furthermore, 3D forward modeling mesh and inversion mesh are built independently: an Octree forward mesh is designed for efficient spatial segmentation for single or multiple soundings, while the inversion parameters are defined on a structured model mesh, which is linked to the forward meshes via interpolation. In conjunction with the development of a full 3D EM-IP inversion, we introduce a novel 3D inversion workflow. This workflow allows for hybrid 1D-3D computations, both sequentially and spatially, enabling 3D modeling exclusively in the most significant and interesting areas of the survey. We tested the hybrid 1D-3D inversion workflow using airborne electromagnetic (AEM) data acquired by Xcalibur with the HeliTEM system in the Staré Ransko area (Czech Republic), known for its gabbro-peridotite rocks hosting nickel-copper±cobalt, platinum group element (Ni-Cu±Co, PGE) mineralization. The results demonstrate that the hybrid inversion effectively addresses the challenges of 3D modeling on large-scale datasets. It enhances interpretation reliability in regions with strong 3D effects and shows a significant spatial correlation between resistivity and chargeability phase anomalies and known mineral deposits. Moreover, both synthetic and field data indicate that the resistivity parameter is more sensitive to 3D effects than the chargeability phase parameter.

SummarySeismic reflection and transmission provide essential insights into the composition of reservoir solids and fluids. Reservoir media often consist of layered structures that contain solids, fluids, pores, and cracks. In such complex layered media, the stable and accurate modeling of seismic wave propagation is crucial for effective reservoir evaluation using seismic waves. By solving the cracked porous medium wave equation for layered structures using the propagator matrix method, we calculate the frequency-dependent oblique incident P-SV and SH wave reflection and transmission for the layered poroelastic media containing cracks. This approach accounts for the combined effects of impedance contrast and crack squirt flow on wave reflection and transmission. The newly developed model includes interlayer fluid flow, crack squirt flow, and global fluid flow. Among these mechanisms, interlayer fluid flow and crack squirt flow can both be prominent in the seismic frequency band. Then, the model was applied to simulate seismic reflection and transmission in cracked interlayer and interface geological structures. The results show that the pore-crack squirt flow mechanism plays a significant role in determining seismic reflection and transmission. Increased crack density and gas saturation significantly enhance P-wave reflection and generate seismic reflection bright spots, while for the S-wave reflection, the effect is largely controlled by crack density, and, when crack density is high, is moderately affected by fluid saturation. This fluid sensitivity results from the crack squirt flow mechanism, which is absent from the classical Biot-Gassmann theory. In all known limiting cases, the model predictions agree with those from the Biot-Gassmann theory.

SummaryIntermontane basins in active orogenic regions face significant seismic hazard due to their proximity to sustained tectonic activity. While the sediments deposited in these basins create a relatively flat topography suitable for urban and infrastructure developments, their unconsolidated sedimentary fill locally amplifies earthquake-induced ground motions, thereby increasing the seismic hazard and risk. Documented observations suggest that ground motion estimates in these basins are often poorly constrained due to oversight of surrounding surface topography and insufficient sub-surface information about deeper basin layering, leading to inaccurate hazard assessments. In this study, we systematically evaluate the implications of these two factors on ground motion characteristics up to 4.4 Hz, which is crucial for earthquake engineering practices. We conducted 3D simulations around the Kathmandu catchment area (Nepal) using hypothetical thrust-faulting moment tensor sources at various depths and locations. The results show a significant reduction, by an order of magnitude, in the peak ground velocities (PGV) at the catchment area due to surface topography. However, this effect is prominent only for very shallow earthquakes producing predominant surface waves; for deeper sources, the de-amplification may be negligible or even result in amplification due to scattered body waves converted into surface waves. To incorporate basin-specific material properties, we performed the analysis in a computationally-feasible 2D domain, which shows that the existence of topography can reduce the energy entering the basin, hence resulting in a reduced basin amplification. The deeper layers of the Kathmandu basin play a critical role in controlling the spatial variability of the observed amplification, with significant differences within the basin compared to scenarios that exclude these deeper layers. We conclude that neglecting topography in ground motion predictions may lead to an overestimation of ground motion amplification in the basin. Pronounced topographic features in the surrounding of intermontane basins can result in further scattering of the received energy content from earthquakes occurring outside of the basin, especially for the high-frequency motions. In addition, in order to provide site-specific measures of ground motion in intermontane basins, high spatial resolution of the underlying geological structure is deemed imperative.

SummaryInterpreting geophysical inversion results across diverse applications presents significant challenges, particularly when the resulting images lack distinct, sharp interfaces. Incorporating prior information to constrain the inversion process introduces additional complexity, especially when this prior information itself contains uncertainties. This work explores methods for improving the geometric representation of geologic structures using integrated geophysical and geologic models. While many existing approaches are either data-driven or model-driven techniques, they often fail to fully integrate available data into a dynamic, unified geomodel. We present an approach that integrates geologic models and geophysical data through structure-based inversion. Our approach preserves geological realism through an implicit model while imaging sharp contrasts within the geophysical inversion models. To address the ambiguities of solving for both the geometry and physical parameters, we adopt a sequential inversion process, first resolving shifts of geologic interfaces, then inverting for geophysical parameters using the updated geometry as a structural constraint. The method’s efficacy is demonstrated through cross-hole travel-time tomography using two synthetic and one field data set from the Mont Terri Rock Laboratory (MTRL). The field data results validate the capability of our approach to recover subsurface interface geometries from geophysical data that are comparable to the interpolated interfaces from borehole data. While we demonstrate the method for seismic travel time data in cross-hole geometry, the flexible open-source implementation allows application to 3D scenarios and other geophysical methods.

SummaryWe efficiently extract high-quality Pn wave arrival times from seismograms recorded at recently deployed 120 portable seismic stations of TanluArrays and 317 stations of the Chinese provincial seismic network for 7231 local earthquakes (M >2.0) using the PickNet automatic picking method. Then we use the Pn data to determine 2-D P-wave velocity and anisotropic tomography of the uppermost mantle in and around the Tanlu Fault Zone. Our Pn tomography reveals segmented features of the fault zone, which are well consistent with geological structural features. A continuous low-velocity anomaly parallel to the fault zone is revealed along the Bohai Bay-Weifang, Weifang-Tancheng, and Tancheng-Mingguang segments, whereas the Mingguang-Wuxue segment exhibits a bead-like alternating high and low velocity belt. A similar segmented characteristic also appears in the Pn wave anisotropy in and around the fault zone. A majority of strong earthquakes are located in transitional zones between high and low Pn velocities, suggesting that structural heterogeneities in the uppermost mantle may affect crustal seismogenesis. The low Pn velocities may reflect upwelling of hot and wet upwelling flows in the big mantle wedge due to mantle convection and dehydration of the flat Pacific slab in the mantle transition zone, which cause seismic anisotropy in the upper mantle in and around the Tanlu fault zone.

SummaryEarthquake monitoring plays a critical role in disaster warning and geophysical research, including earthquake phase picking and source parameter estimation. However, traditional methods suffer from cumbersome workflow and challenges in parameter selection during multi-parameter joint estimation. Here, we propose a multitask network for earthquake monitoring that reduces computational complexity by replacing the self-attention mechanism with fast Fourier transform. Through the integration of a fusion module and an enhancement module, the interaction between tasks is strengthened to optimize network performance. Additionally, a dynamic adaptive weight allocation strategy is introduced to achieve a balance among different tasks. The proposed method was trained and tested on the STEAD and INSTANCE datasets and compared with advanced approaches. The results demonstrate that this method outperforms other deep learning methods in earthquake phase picking and source parameter estimation, achieving lower errors and higher evaluation metrics, thus showing potential for practical application in earthquake monitoring.

AbstractMeasurements of the propagation of teleseismic fundamental-mode surface waves are essential for studies of Earth structure and earthquake source processes. Understanding sources of noise and error in these measurements can help improve the accuracy and precision of analyses that use these measurements. One prominent source of noise is interference of overtones with the fundamental mode, which is well-studied in the context of surface wave phase observations. In this work, we show that overtone interference also has a substantial impact on group measurements and has uniquely different characteristics when compared with the analogous interference in phase measurements. We illustrate these characteristics using measurements on both synthetic and real data. Importantly, our experiments suggest that group measurements are more vulnerable than phase measurements to interference from overtones; both synthetic data and published datasets show larger and more variable interference in group measurements than in phase measurements. This interference leads to significant errors in group velocity estimates made using regional or array-based approaches. We show that some quality control measures designed to eliminate overtone interference in phase measurements may not be applicable for group measurements. Our results emphasize the need for careful monitoring of group velocity overtone interference in tomographic imaging, as well as the need for accurate uncertainty quantification when group velocity maps are used in further studies.

AbstractCompared to first-arrival traveltime tomography (FATT), first-arrival traveltime and slope tomography (FASTT) integrates both traveltimes and local slopes of first arrivals at sources and/or receivers to construct more accurate subsurface velocity models. Local slopes serve as additional constraints, helping to mitigate the ill-posedness of tomography by better constraining ray directions. This is particularly beneficial in regions with complex topography, where shadow zones arise due to strong velocity contrasts. However, representing complex topography or bathymetry can be less accurate when using classical rectangular grid discretization. To address these complexities with greater versatility and accuracy, we compute traveltimes of locally coherent events using a factored topography-dependent eikonal solver on curvilinear grids. Slopes are then estimated by finite differences in the traveltime maps after a back-and-forth coordinate transform from the curvilinear grid to a rectangular computational grid. Additionally, we solve the inverse problem with the matrix-free approach, where the data misfit gradient is computed with the adjoint-state method. This adjoint—state formulation avoids the explicit construction and storage of large Fréchet derivative matrices and does not require a tedious posterior ray tracing on curvilinear grids. A land synthetic example first illustrates the sensitivity of slopes to topography and the more accurate velocity models with FASTT than with FATT in the presence of topography. We then perform a first application of the topography-dependent FASTT method on a real redatumed ocean-bottom node dataset, where the bathymetry exhibits a steep scarp. We show that the topography-dependent FASTT generates a velocity model that matches more closely a legacy reflection tomography model than conventional FATT. We conclude that the topography-dependent FASTT provides a versatile approach for handling complex surfaces during velocity model building in both marine and land environments.

SummaryMagma transport in dikes is usually modelled by means of lubrication theory, assuming that magma properties are uniform across the dike. We explore the influence of cross-dike temperature heterogeneity on the dynamics of dike propagation using a quasi-2D model, derived from a full 2D model with an assumption of small width to length ratio. The model couples elastic fracture mechanics with multiphase magma flow, solving the governing equations using a hybrid numerical approach that combines the Displacement Discontinuity Method for elasticity with finite volume discretization for fluid flow and heat transfer. The model includes heat exchange with wall rocks, shear heating and latent heat release. It accounts for non-equilibrium magma crystallization, implementing temperature-dependent crystallization kinetics using an Arrhenius formulation for the relaxation timescale. As a case study, we simulate the ascent of a volatile-rich dacite from a source at 30 km depth. The distribution of temperature, crystallinity, and, thus, viscosity across the dike leads to a plug-like velocity profile with magma stagnation near the walls, substantially different from the parabolic Poiseuille flow assumed in classical lubrication theory. With temperature-dependent crystallization rate, rapid cooling of magma near the dike walls can generate a glassy chilled margin. The adjacent magma has higher crystallinity due to intermediate cooling rates, while the hotter core remains depleted in crystals throughout dike propagation. The dike propagates further and is thinner than predicted by (1D) lubrication theory because the low-viscosity core continues to facilitate vertical transport while the wall zones become progressively more viscous due to cooling and crystallization. The latent heat of crystallization can have a substantial impact in slowing down cooling and prolonging propagation. Other important factors include the characteristic crystal growth time, initial magma temperature and water content. Our quasi-2D approach bridges the gap between oversimplified 1D models and computationally expensive 3D simulations, providing a practical framework for investigating magma transport in silicic dikes.