Geophysical Journal International

SummarySeismic reflections from mantle discontinuities provide critical constraints on the structure and dynamics of Earth’s interior, but their extraction remains challenging due to low signal-to-noise ratios (SNR), interference from other seismic phases and uneven spatial distributions. In this study, we propose an array-based extraction strategy that integrates data reconstruction with subsequent denoising for enhancing the extraction of weak mantle reflections. This proposed strategy is independently implemented using the Curvelet-, F–K, and Radon transforms, and the performance of the three implementations is systematically evaluated. Compared with the time-space domain, coherent signals and noise are more easily separated in the transformed domain. We apply these methods to synthetic waveforms generated using a modified ak135 Earth model and test their effectiveness in retrieving reflections from the mantle transition zone (SS/PP precursors) and the D″ discontinuity (ScS/PcP precursors), including cases with random noise and missing traces. All three methods effectively isolate weak mantle reflections, with the Curvelet transform demonstrates the highest robustness and SNR improvement, particularly under conditions of sparse or noisy data. Field applications to data sampling the Central Pacific and Central America further confirm the methods’ ability to recover weak mantle reflections and expand the distance range of usable data. These results demonstrate the potential of array-based extraction strategy to advance deep Earth seismic imaging.

SummaryThe Earth’s figure axis is the axis of maximum inertia for the deformed (oblate) Earth, as described by the degree-two, order-one geopotential coefficients C21 and S21. An extended mean-pole model is presented for evaluating solid-Earth and ocean-pole tides. 50-year Satellite Laser Ranging (SLR) data and 24-year GRACE/GRACE-FO data were analyzed to determine variations in Earth’s figure axis, as reflected in changes in the C21 and S21 coefficients. This study reveals that a significant atmosphere-ocean motion induced a variation in C21 that is captured by SLR data but does not appear in the GRACE solution. The current glacial isostatic adjustment (GIA) ICE-6G model requires improvement to account for the observed linear rates of C21 and S21. A significant 30-year and 60-year signal with an amplitude of ∼ 3×10−11 in the Earth’s figure axis is observed using SLR and a ∼10 mas in the PM (polar-motion), which could be predominantly driven by a 0.05-degree tilt of the inner-core figure axis relative to the figure axis of the entire core and is linked to partial electromagnetic core-mantle coupling.

SummaryContinuous, high-density strain and strain-rate distributed acoustic sensing (DAS) recordings are valuable for resolving the shallow Earth’s structure at a low cost, especially in environments that are otherwise difficult to access, such as continental shelves and near-coastal oceanic crust. In this study, we apply seismic ambient-noise methods to extract high-quality empirical Green’s functions (EGFs) from natural noise sources and model the velocity structure along a 30-km-long dark fiber-optic cable connecting the offshore CASTOR gas storage field in the Gulf of Valencia (Spain) with the associated land facility. We extract broadband EGFs containing a rich variety of seismic waves using wavelet phase cross-correlation and time-scale phase-weighted stacking methods. In the common-source EGF gathers, clean fundamental and first-overtone Scholte waves dominate the marine channel pairs, while the fundamental Rayleigh mode appears in the land channel pairs. In addition, weak wavefields reflected from the basin edge follow the main surface waves. We then construct a 2-D Vs model from local phase-velocity observations of the fundamental and first-overtone Scholte waves by solving pointwise depth inversions using Markov chain Monte Carlo methods. The model resolves the marine sedimentary basin from very shallow water-saturated sediments to depths exceeding 1 km, identifying the Amposta Central Fault and the basement bedrock west of this fault at roughly 1 km depth. These results help refine the offshore velocity model along the cable in a region where induced seismic activity has been observed, improving the accuracy of seismic monitoring and seismic hazard characterization.

SummaryThe conductive and capacitive properties of rocks are influenced by the type and concentration of the electrolyte present in the pore water. Sodium (Na⁺) and potassium (K⁺) are common pore water cations in saturated sedimentary rocks. Their distinct physicochemical properties are expected to produce different frequency-dependent electrical dispersion when adsorbed onto mineral surfaces. We tested this expectation by using spectral induced polarization (SIP), a method sensitive to interfacial processes. Complex conductivity spectra (10–2 to 105 Hz) were measured on two clayey, opal-A-rich diatomite samples, saturated with either NaCl or KCl solutions. One sample was tested over a stepwise increase in molar concentration (5.4–53 mM), while the other was tested over a stepwise increase in bulk water conductivity (0.050–0.48 S/m). At equivalent molar concentration, the in-phase conductivity of a sample was ~20 per cent higher when KCl saturated than when NaCl saturated, reflecting the greater molar conductivity of K⁺. At matched bulk water conductivity, which required a ~20 per cent higher NaCl molarity than KCl molarity, in-phase conductivity was ~10 per cent higher when NaCl saturated. In both tests, the quadrature conductivity and normalized chargeability followed a lower trend in the KCl-saturated state than in the NaCl-saturated state. This relatively low polarization for the K+ saturated state can be attributed to a weaker hydration and more compact adsorption of K⁺ within the inner layer of the electrical double layer. Additionally, time-lapse monitoring of complex conductivity spectra indicates that chemical equilibration via diffusion is achieved within 72 hours for both electrolyte types. This relatively rapid ionic diffusion is consistent with estimates based on the intrinsic formation factor and probably reflects the high porosity of the diatomite (~0.7). These findings establish that pore-water cation identity (Na⁺ vs. K⁺) is a primary control on SIP-derived polarization parameters, and cation identity must therefore be incorporated into petrophysical models to avoid biased estimates of surface area, permeability, and hydrogeochemical state.

SummaryImages of the Earth’s interior can provide us with insight into the underlying properties of the Earth, such as how seismic activity might emerge and the interplay between seismic and volcanic activity. Understanding these systems requires reliable high-resolution images to understand mechanisms and estimate physical quantities. However, reliable images are often difficult to obtain due to the non-linear nature of seismic wave propagation and the ill-posedness of the related inverse problem. Reconstructions rely on good initial estimates as well as hand-crafted priors, which can ultimately bias solutions. In our work, we present a 3D reconstruction of Kilauea’s magmatic system at a previously unattained resolution. Our eikonal tomography procedure improves upon prior imaging results of Kilauea through increased resolution and per-pixel uncertainties estimated through variational inference. In particular, solving eikonal imaging using variational inference with stochastic gradient descent enables stable inversion and uncertainty quantification in the absence of strong prior knowledge of the velocity structure. Our work makes two key contributions: developing a stochastic eikonal tomography scheme with uncertainty quantification and illuminating the structure and melt quantity of the magmatic system that underlies Kilauea.

SummaryThe global energy transition has created an urgent need for expanded critical mineral supply. Projected production from existing deposits and current discovery rates remains insufficient to meet this demand. More efficient exploration strategies are therefore required, particularly in optimizing costly and low-success-rate data acquisition campaigns. To address this challenge, we introduce the concept of sequential Efficacy of Information (sequential EOI), a decision-making metric that quantifies the uncertainty reduction of target variables under proposed exploration action sequences. Unlike Value of Information (VOI), sequential EOI operates in the domain of uncertainty reduction, removing the need for an economic model that is rarely available in early-stage exploration. We demonstrate the framework using synthetic 2D and more realistic 3D porphyry copper systems, evaluating sequential combinations of exploration plans including ambient noise tomography (ANT) surveys and borehole drilling campaigns. In both cases, sequential EOI identified exploration plans that maximized the uncertainty reduction to the target variables. These results demonstrate that sequential EOI offers a principled framework for multi-physics, multi-step, and uncertainty-driven plan optimization in mineral exploration, providing exploration teams with a practical and scalable decision-analytic tool for rational campaign design without requiring economic assumptions.

AbstractAccurate and rapid magnitude prediction is critical for earthquake early warning systems, directly affecting emergency response decisions and public safety. With global seismic monitoring networks expanding to over 15,000 stations and the emergence of crowdsourcing-based IoT device monitoring systems, daily seismic data has reached petabyte scales, posing enormous challenges for real-time processing under the typical 3-10 second warning window constraint. Existing deep learning methods predominantly adopt single-modal information processing strategies, focusing either solely on temporal features of time-domain waveforms or spectral information after frequency-domain transformation, failing to fully exploit the joint evolution patterns and complementary information of seismic signals in the time-frequency domain, thereby limiting prediction accuracy and generalization performance. This paper proposes MP-Net, an end-to-end deep learning framework based on multi-scale time-frequency fusion for local magnitude (ML) prediction. The method employs a dual-branch architecture that simultaneously processes raw three-component waveforms and spectrograms: the time-domain branch captures features from microscopic waveform details to macroscopic energy evolution through parallel multi-scale convolutions; the frequency-domain branch combines hierarchical 2D convolutional networks with adaptive spectral attention mechanisms to automatically identify magnitude-related frequency components while suppressing noise; a cross-attention based fusion module achieves deep integration of complementary information from both modalities. To preserve the absolute amplitude information physically consistent with the ML definition, logarithmic amplitude features are extracted prior to waveform normalization and provided as auxiliary inputs to the fusion layer. Comprehensive experiments on the large-scale STEAD dataset demonstrate substantial improvements over baseline models: mean absolute error decreased to 0.28, coefficient of determination R2 reached 0.872, with 82.5% of predictions achieving acceptable precision (error≤0.5). The proposed approach provides an efficient and accurate solution for real-time single-station magnitude prediction, applicable to earthquake early warning systems operating in both centralized and distributed computing environments.

SummaryIn marine seismic exploration, various types of seismic sources are employed to visualize geological structures beneath the seafloor, depending on survey objectives. Airgun sources, which generate large amounts of energy by releasing compressed air underwater, are typically used for imaging deep area; however, they have limited vertical resolution due to their low peak frequencies. In contrast, sparker sources generate wavelets with high peak frequencies using bubbles produced by discharging electrical energy to vaporize water, resulting in high vertical resolution. Sparker sources are useful for the detailed imaging of shallow strata but have a shallow penetration depth due to their low source energy. This paper proposes a method to integrate airgun and sparker data to broaden the frequency bandwidth and thus achieve more accurate geological interpretations. The study used small-scale airgun data and sparker data acquired in Yeongil Bay, Pohang, South Korea. A machine-learning-based shaping filter model was developed along with synthetic training data representing the airgun and sparker source wavelet characteristics, and the trained models were applied to regularize these source wavelets. Subsequently, time-variant spectral whitening (TVSW) and weighted integration were performed to yield the flattened broadband frequency spectrum. The integrated data have enhanced penetration depth and vertical resolution compared with the original single-source datasets, thus overcoming the interpretational limitations imposed by their limited frequency bandwidth and penetration depth and enhancing the reliability of associated geological interpretations.

SummarySite amplification in the Kumamoto area, Japan, is analyzed using 985 high-quality horizontal strong-motion records from 45 aftershocks (Mj = 2.7–4.9) recorded within 24 hours following the 2016 Kumamoto Mj 7.3 earthquake, as observed by 51 K-NET and KiK-net stations. For the generalized inversion technique (GIT), a reference station is required as a standard. In the GIT process, the number of events available for analysis is limited to those recorded by the reference station, and the stations whose site effects can be estimated are restricted to those that record common events with the reference station. To overcome the limitation of the GIT, the ‘transfer-station generalized inversion method (TSGI),’ a modified GIT, is introduced to increase the number of analyzed events and stations. The site responses obtained from GIT and TSGI for the same stations exhibit a high degree of consistency, thereby demonstrating the effectiveness of the TSGI. The discrepancies between the ${{Q}_S}$ estimates of GIT and TSGI can be attributed to the gradual expansion of the region represented by ${{Q}_S}$ as more events and stations are included in the inversion. However, the results of GIT and TSGI are relative to the reference station that may itself exhibit site effects. Thus, a reference-independent technique, i.e. genetic algorithm (GA), is also introduced to obtain the absolute site amplifications. The results show that at frequencies greater than about 1 Hz, the site response of the reference station is significantly lower than the theoretical amplification factor of 2, resulting in an overestimation of the site responses at other stations. When the results of GIT are corrected with the site response of the reference station obtained from GA, these two results agree very well for most of the stations. This indicates that the results of GIT are reliable if the reference station is an ideal surface rock station, and that the GA produces accurate absolute site amplification factors for the stations investigated in this study. In addition, we analyze the high-frequency attenuation characteristics of S-waves in the Kumamoto area, and establish $\kappa $ models for different site conditions and an empirical ${{\kappa }_0}$-${{V}_{S30}}$ relationship.

SummaryHydraulic fracturing in unconventional gas development has intensified concerns over induced seismicity, generating significant seismic hazards with potential risk implications for surrounding environments and communities. Accurate prediction and transparent interpretation of such hazards remain open challenges in seismology and engineering practice. This study addresses these challenges by developing a One-Step Prediction Random Forest framework to model the spatiotemporal relationships among hydraulic fracturing well deployment, geological factors, historical seismicity, and the likelihood of future seismic occurrences. A seismic energy labeling scheme based on one-step prediction enables the framework to estimate potential seismic energy release and identify dominant controlling factors through feature attribution. Building on these results, the conventional traffic-light risk management system is conceptually extended to a traffic-light hazard management scheme, which incorporates theoretical insights from risk analysis to improve interpretability and operational relevance. Predictive performance across low-, medium-, and high-hazard scenarios is assessed using confusion matrix analysis, while SHAP-based interpretability confirms that the framework preserves physical consistency by linking geological and operational drivers with seismic energy release. The findings advance methodological innovation in induced seismicity research by combining hazard prediction with a hazard-management framework inspired by risk theory, providing both theoretical insights and practical tools for shale gas development.

SummaryThe 23 April 2025 Mw 6.2 Marmara earthquake near İstanbul in northwestern Türkiye provided a rare opportunity to evaluate the capability of distributed acoustic sensing (DAS) for near-fault seismology in a densely populated and tectonically active region. The DAS monitoring complements the borehole-based Geophysical Observatory at the North Anatolian Fault (GONAF) expanding near-fault observatory towards high-resolution monitoring of deformation along the Main Marmara Fault (MMF) in the vicinity of Istanbul. In this experiment, an existing telecommunication dark fiber within the national optical network was repurposed as a long-offset submarine seismic array without any dedicated underwater installation. This fiber-optic cable, which extends underwater along the southern coast of Istanbul and follows the northern boundary of the Princes Islands, successfully recorded the Mw 6.2 Marmara earthquake. In this study, we analyzed the resulting dataset, which provides an aperture of approximately 34 km with 8 m channel spacing and a 250 Hz sampling rate, and includes sections extending both parallel and perpendicular to the fault. The DAS strain rate recordings of the Mw 6.2 mainshock exhibit clear P- and S- wave arrivals across thousands of channels with high signal coherence. After amplitude and phase calibration, the recordings were converted into equivalent particle velocity using a gauge length based transfer function and compared with nearby GONAF borehole seismic waveform data. Spectral analyses show that the DAS array preserves seismic energy up to ~25 Hz, with spectral shape and amplitude distributions consistent with those of the borehole recordings. Curvelet, f–k, and time domain slowness based conversion methods were employed for the transformation process. The submarine DAS array interrogated by an OptoDAS system enables high resolution spatial sampling of strain derived from the recorded strain-rate data. Systematic variations in apparent velocity (Vp ≈ 2.38 km/s, Vs ≈ 2.30 km/s) and amplitude decay are observed, which could possibly represent shallow sediment heterogeneity along the fiber. Short-wavelength fluctuations in strain are present, which may reflect local structural complexity as well as cable-related factors such as orientation and burial conditions. Overall, these results demonstrate that existing dark-fiber infrastructure can be used as long-aperture seismic arrays for capturing local seismic wavefields providing high-density measurements useful for monitoring near-fault strain variations. This study shows the long-offset DAS recording of a moderate size earthquake in Türkiye and one of the few submarine DAS observations worldwide. It demonstrates that dense strain measurements derived from fiber-optic infrastructure can complement the monitoring of offshore active fault segments, offering unprecedented spatial resolution for monitoring dynamic strain, and fault zone deformation in regions of high seismic hazard such as the Sea of Marmara.

SummaryThe Fuyang Depression, located in the Southern North China Basin, exhibits promising gas generation potential and favorable hydrocarbon accumulation conditions. The hydrocarbon resource survey in this region has primarily involved active-source seismic exploration and drilling operations. Dense-array ambient noise imaging has proven to be an efficient and low-cost detection technique, providing background information on the reservoir accumulation conditions in sedimentary basins. This study utilizes continuous ambient noise records from two linear dense arrays in the Fuyang depression to build S-wave velocity structures and radial anisotropy models down to 3 km beneath the arrays. A modified cross-correlation beamforming method is applied to sub-arrays of the linear array to extract dispersion curves of the fundamental mode and first overtone Rayleigh waves, as well as fundamental mode Love waves, from the ambient noise. The phase-velocity cross sections of different surface wave modes beneath two linear arrays are thereby obtained directly without tomographic inversion. Depth inversion is then performed to derive SV- and SH-wave velocity structures and radial anisotropy for the two linear arrays. The sediment thickness of Quaternary and Neogene (Q+N) sedimentary sequences are delineated by the iso-velocity of S-wave at 1.3 km/s, calibrated with the borehole data. The widespread negative radial anisotropy layers are observed at the shallow subsurface, which are interpreted as the water-saturated open fractures. The S-wave velocity and radial anisotropy models provide valuable constraints for both seismic hazard assessment and hydrocarbon exploration within sedimentary basins.

SummarySolving wave equations using machine learning methods, such as physics-informed neural networks (PINNs) and neural operator approaches, has become an active area of research in the computational seismology community. However, a significant challenge associated with these methods is the degradation of long-term prediction accuracy, which arises from the inherently nonlinear dynamics of wavefields governed by partial differential equations (PDEs). Koopman theory provides a promising framework by enabling the transformation of nonlinear dynamical systems into linear ones, thus allowing linear analysis tools to be applied to complex, nonlinear problems. In this study, we introduce a data-driven operator learning method, the Koopman Neural Operator (KNO), for solving the two-dimensional acoustic wave equation in the time domain. Within the KNO framework, the time-domain wavefield is treated as the state variables, while the velocity model acts as the control variables. These form a nonlinear dynamical system, which is mapped into a linear latent space using a convolutional encoder. The Koopman operator is then approximated by parameterizing the integral kernel in the wavenumber domain, facilitating linear time evolution of the encoded variables. A convolutional decoder subsequently transforms the evolved latent variables back into the original wavefield domain to obtain the predicted time-domain wavefields. To evaluate the performance of KNO, we first conducted numerical experiments on the three most complex datasets from the OpenFWI benchmark and compared the results with those of the current state-of-the-art Fourier Neural Operator (FNO). The results demonstrate that KNO outperforms FNO in terms of prediction accuracy, computational efficiency, memory consumption, and convergence speed. Additionally, KNO exhibits notably strong stability in seismic wavefield extrapolation on the Marmousi model. Finally, we comprehensively evaluate the parallel scalability of the proposed KNO model, and compared KNO with the finite difference method (FDM) (based on both CPU and GPU) in terms of computational speed. Collectively, these experiments indicate that KNO provides a promising new approach for long-term and relatively high-precision wavefield extrapolation in seismic modeling.

SummarySpectral Induced Polarization (SIP) has gained recognition as an advanced geophysical method for monitoring soil water content. SIP’s ability to simultaneously assess soil texture and water content makes it particularly valuable for studying soil dynamics under varying environmental conditions. However, its application in complex field environments has been hindered by issues such as capacitive and inductive coupling, which affect the quality of measurements. In this study, we combined water monitoring in soil and plant (field and lab SIP measurements, sap flow and soil moisture monitoring) to characterize soil heterogeneity and evaluate vine water availability in a Médoc vineyard during the summer drought of 2023. Different SIP field acquisition strategies relying on multiwire cables, fully coaxial cables array or hybrid coaxial/multiwire were tested. The acquisition setup was shown to strongly affect data quality depending on soil moisture conditions. Lab and field SIP measurements confirmed a strong correlation between the quadrature conductivity at 0.25 Hz and soil volumetric water content (VWC) as well as a linear relationship between phase shift at 0.25 Hz and VWC. The real and imaginary parts of the conductivity was used to infer VWC dynamics based on empirical petrophysical relationships established in situ. A mechanistic model based on the Dynamic Stern Layer model was also applied to high-quality SIP data for the same purpose. We found that imaginary conductivity was much less sensitive to soil water conductivity than real conductivity. Thus, in vineyard soils subject to soil amendments and resulting variations in soil water salinity, we hypothesized that SIP monitoring provide more reliable estimates of changes in soil moisture content than standard electrical resistivity tomography. We showed that SIP monitoring effectively captured soil drying dynamics down to a depth of 1 m during the growing season. The SIP method combined with soil moisture probes could thus provide simultaneous access to both soil moisture dynamics and the spatial distribution of soil texture, opening up new perspectives for mapping soil moisture dynamics in the field, even in case of potentially large soil water salinity fluctuations. In our case, SIP indicated a decrease in soil water storage from 150 to 50 mmH20 during the summer drought of 2023. By combining SIP and vine sapflow monitoring, Vine water availability, defined as total transpirable soil water could also be estimated at 98±8 mm H20 for the vines equipped with sapflow sensor, which is of great interest for culture water management. Finally, the distinct responses of the real and imaginary conductivity components underscore the value of SIP for soil moisture assessment in viticultural environments subject to variable salinity inputs. This work is the first to attempt a quantitative estimation of soil water storage in commercial vineyards using SIP methods. It extends previous applications limited to other agricultural settings and broaden the applicability of mechanistic models (Dynamic Stern Layer model) for predicting volumetric water content based on multi-frequency complex conductivity measurements under field conditions.

SummaryDespite generally low seismicity typical of intraplate regions, magnitudes larger than 6 have occurred in the Amazon craton. The installation of permanent stations of the Brazilian Seismic Network in the Amazon around 2014, the Vale 5-station network in the Carajás mineral province in 2019, and a 30-station temporary deployment (October2021-March/2024) significantly improved earthquake detectability in the eastern part of the Amazon craton. A review of the seismotectonic characteristics of the oldest, Eastern part of the Amazon craton is presented here. Seismicity is not uniform, and areas of higher seismicity are identified, such as in the northern part of the Tapajós-Parima province and along the eastern border of the craton. No clear correlation with the main trends of faults was observed. Contrary to Central and Eastern Brazil, seismicity is not directly correlated with lithospheric thin spots in the Amazon Craton but tends to occur in the flanks of thick keels (“craton edge” effect). A possible influence of free-air gravity anomalies was noticed suggesting that flexural stresses contribute to control seismicity. Five new focal mechanisms are presented for the eastern edge of the Amazon craton, indicating a stress field with NW-SE compression and NE-SW extension. An updated map of the stress field for mid-plate South America shows that the maximum horizontal stresses vary from E-W in SE Brazil, NW-SE in central Brazil and SW-NE in the north. This pattern can potentially be explained by upper mantle flow, provided more detailed convection models are used.

SummaryElectrokinetic signals generated by coupled stress–fluid processes are increasingly recognized as indicators of fault-zone dynamics prior to earthquakes. However, their interpretation is often limited by the common reliance on point-source approximations, which neglect the inherently distributed nature of stress accumulation and fluid migration along fault planes. Here, we develop a quasi-static finite-fault electrokinetic framework in which coupled stress and fluid perturbations are represented as spatially distributed, time-evolving sources. The approach combines an extended Luco–Apsel–Chen generalized reflection and transmission method with a point-source superposition scheme, enabling efficient simulation of electrokinetic responses to area sources in layered porous media. Numerical results reveal that the horizontal components of geoelectric fields in the coupled stress–fluid system are highly sensitive to fluid-source geometry, whereas vertical components primarily reflect stress loading. Spatial variability in initiation time, arising from finite-rate fluid migration, further introduces waveform complexity, amplitude modulation, and multi-stage temporal evolution in surface signals. Notably, we find that the directional variations of the geoelectric field provide a robust diagnostic for distinguishing fluid-driven from stress-induced signals, with angular misalignments reaching up to 16.6°. These results establish a quantitative framework for interpreting near-fault electrokinetic signals and for guiding monitoring strategies aimed at constraining fault-zone fluid pathways and stress evolution.

SummarySix-component (6C) seismic observations offer a more comprehensive description of the wavefield than conventional three-component methods. However, current wavefield separation techniques are often constrained by their reliance on dense arrays. This study introduces a novel wavefield separation framework based on single-station 6C polarization analysis, which enables the simultaneous identification and improved separation of major seismic wave types: P-, SV-, Rayleigh, and transversely polarized horizontal waves (SH- and Love waves). Our proposed method models the observed wavefield as a weighted linear combination of theoretical wave models and optimizes the weighting coefficients via the Broyden–Fletcher–Goldfarb–Shanno (BFGS) algorithm, achieving reliable separation. A refined polarization-based filtering strategy incorporating likelihood estimation, degree of polarization, and energy distribution ratios enhances robustness under low signal-to-noise conditions. Furthermore, the inverse Short-Time Fourier Transform (iSTFT) is adopted to mitigate energy leakage issues associated with the inverse S-transform. Validation using synthetic and real teleseismic data suggests the method’s effectiveness and provides insight into its physical limitations. This study demonstrates a polarization-based framework for seismic phase identification and wavefield separation, which may support multi-phase joint inversion in selected seismic applications.

SummaryInjection-induced seismicity poses a major challenge to the safety of Enhanced Geothermal Systems (EGS). We customized a deep learning model to forecast seismicity rate under prescribed injection schedules. The model adopts a two-stage strategy where injection pressure is first forecasted as an intermediate variable and subsequently used to support seismicity rate forecasting. In this way, seismicity rates at both the field-scale Utah Frontier Observatory for Research in Geothermal Energy (Utah FORGE) and the mine-scale EGS Collab projects could be successfully forecasted. While the model without seismogenic index (Σ) could attain low forecast errors, incorporating Σ markedly improves its ability to capture the transient variability of seismicity rate. The forecasts at Utah FORGE and EGS Collab may highlight the importance of integrating key physical parameters calculated from raw observations into data-driven frameworks for forecasting injection-induced seismicity, and may demonstrate the potential of customized deep learning models for cross-stage forecasting in next-generation EGS.

SummaryReconstructing subsurface structures with high resolution is one of the main goals and potentials of full waveform inversion (FWI). However, FWI is a highly nonlinear and ill-posed problem. Conventional physics-based FWI methods, which rely on gradient-based optimization to minimize the difference between observed and synthetic data face cycle-skipping challenge. Although numerous deep-learning inversion approaches have shown promise, they typically focus on latent representations of time-domain seismic data. This often causes an unstable inversion process due to waveform mismatches. To overcome these limitations, we introduce FFT-InversionGAN, an unsupervised seismic inversion framework that integrates physics-based forward modeling with adversarial learning of the frequency-domain data based on Wasserstein generative adversarial network with gradient penalty (WGAN-GP). Fast Fourier transformer (FFT) is employed to transfer the time and phase information of time-domain seismic data into the spectrum and amplitude distributions to modify the feature space and sensitivity of the adversarial loss to different types of mismatches. By leveraging Wasserstein distance constraints, this method can naturally operate on the spectral distributions of seismic data. Compared with L2 norm, Wasserstein distance is far less sensitive to the linear variations in the phase spectrum. And our proposed method eliminates the need for network pre-training while improving stability and flexibility. FFT-InversionGAN demonstrates enhanced accuracy and resilience in numerical experiments on noise-free, noisy and missing low-frequency benchmarks. This was observed when applied to the Marmousi and overthrust models, where it consistently outperformed conventional FWI and FWIGAN. These findings highlight that FFT-InversionGAN has superior inversion effectiveness.

SummaryThe 2020 MS 5.0 Qiaojia earthquake occurred in a tectonically complex region near the Xiaojiang fault in southwestern China. We investigated the rupture process of this moderate-sized earthquake using a multi-empirical Green’s function (EGF) inversion method that integrated waveforms from multiple EGF events. Synthetic tests demonstrated that the multi-EGF inversion method recovered the input model more robustly than any individual EGF inversion. The resolved spatiotemporal rupture model of this earthquake indicated a compact rupture lasting approximately 2.9 s, dominated by a major asperity near the hypocenter and characterized by predominantly eastward rupture propagation. Bootstrap resampling analyses further confirmed the robustness of the resolved major coseismic slip distribution and the overall moment release pattern. We also observed a spatial complementarity between the coseismic slip and aftershock distributions, with most aftershocks clustering around the periphery of the major asperity. This study not only elucidates the source complexity of the 2020 MS 5.0 Qiaojia earthquake, but also validates the robustness and effectiveness of the multi-EGF inversion method in resolving the rupture processes of moderate-sized earthquakes. Our results provide new insights into the rupture kinematics of moderate-sized earthquakes and the heterogeneity of fault strength and stress within the Xiaojiang fault zone and its surrounding regions.