Geophysical Journal International

SummaryNorthern Europe experiences vertical land motion and sea level changes that deviate from the average as a consequence of past changes in ice sheet cover in Fennoscandia and the British Isles. The process, called Glacial Isostatic Adjustment (GIA), is controlled by the subsurface structure. Numerical models of GIA can be compared to observations of uplift or past sea level changes to constrain the subsurface structure, and such models can also be used to correct present-day sea level observations to reveal sea level changes due to climate change. GIA models for northern Europe usually adopt a homogeneous upper mantle viscosity even though seismic studies indicate contrasting elastic lithosphere thickness and upper mantle structure between Northwestern Europe and Eastern Europe. This raises the question whether the effect of lateral variations in structure (3D viscosity) can be detected in observations of GIA and whether including such variations can improve GIA model predictions. In this study we compare model output from a finite element GIA model with 3D viscosity to observations of paleo sea level and current vertical land motion. We use two different methods to derive 3D viscosities, based on seismic velocity anomalies and upper mantle temperature estimates. We use three different reconstructions of the Eurasian ice sheet, one based on an inversion using a 1D model, and two others based on glacial geology and modelling. When we use these two reconstructions, we find that the data are fit better using 3D viscosity models. Models with two separate 1D viscosities for Fennoscandia and for the British Isles cannot replicate a 3D model because a 3D model redistributes GIA-induced stresses differently from a combination of models with separate 2D viscosities. The fit to data across Fennoscandia is improved when, as indicated by seismic models, the upper mantle viscosity is higher than for the rest of Northern Europe. The best fit is obtained with a model with dry olivine rheology, in agreement with other evidence from Fennoscandia.

SummaryIn both onshore and offshore seismic exploration, seismic source localization plays a crucial role in ensuring operational safety and environmental protection. With the continuous advancement of the Marchenko method in the fields of seismic migration and internal multiple elimination, this paper investigates a seismic source localization method based on the Marchenko method, aiming to further extend application domain of this method. The key to this method lies in the data reconstruction based on convolution operations. The conventional Marchenko method is then applied to obtain a seismic profile, which includes the location of the seismic source. In the experiments, this study first uses an anticline model to simulate seismic source localization in onshore seismic exploration. The results show that the proposed method can accurately estimate both the distance to the seismic source and its depth. Furthermore, in large-scale marine model experiments, the method is also able to reliably determine the distance between the seismic source and the observation stations.

SummaryFor the interpretation of Spectral Induced Polarization spectra, the determination of the Relaxation Time Distributions (RTD) can be useful, for instance to extract the grain size distribution. However, this is an ill-posed problem, and retrieving the RTD often requires regularization during the inversion process. In this note, we use Bézier curves and simulated annealing to determine the RTD. The procedure that does not require any regularization nor smoothing, by reducing the number of parameters thanks to Bézier curves which are intrinsically continuous and infinitely derivable. We successfully applied our methodology to three examples (Cole-Cole model, Davidson-Cole model, and an experimental spectrum), demonstrating its interest and efficiency.

SummaryThe extension of direct current resistivity methods to induced polarization methods has enriched the tools available for subsurface exploration. This enrichment involves an increase in the number of parameters used in the models, as well as addressing different physical phenomena than those observed with direct current. Accounting for non-linearities, if they exist, can further enhance the sophistication of our models. Non-linearities are often observed, particularly in laboratory experiments. However, we question their origin, and the experiment described here suggests that the non-linearities observed under typical experimental conditions may be artifacts related to the electrodes, rather than reflecting the actual response of the subsurface. Indeed, we first replaced the polarizable injection electrodes with non-polarizable electrodes. The non-linearities observed due to the presence of harmonics were significantly reduced. Then, we replaced the voltage control with a current control, which completely eliminated the non-linearities still present.We know that it is impossible to prove the non-existence of a phenomenon that does not exist. This fundamental epistemological principle (as pointed out by Russell and Popper) means that we are not claiming that nonlinearity does not exist. We are simply describing an experiment that can raise doubts about its existence.

SummaryWe present and validate an efficient GPU-accelerated solver for seismic wave propagation in three-dimensional elastic media. The solver achieves up to a 372× speedup relative to a CPU implementation and supports forward simulations on grids ranging from 100 million to 1 billion cells. It is based on a velocity-stress, first-order formulation of the elastodynamic wave equation and supports kilometer-scale models with layered isotropic and anisotropic structure. We validate the solver by comparing synthetic seismograms to analytical solutions from a propagator matrix method in axisymmetric media. Simulations include moment-tensor sources for a 2017 nuclear explosion and collapse in North Korea, and a magnitude ∼4 earthquake near Linthal, Switzerland (6 March 2017). Anisotropic effects for the Swiss event are modeled using rotated orthorhombic stiffness tensors derived from laboratory measurements of gneiss. Projection onto orthorhombic symmetry enables solver compatibility. We find that anisotropy changes waveform polarity, amplitude, and phase at near-source stations. Unscaled laboratory values produce polarity reversals, while velocity-rescaled tensors correct them. These results demonstrate the impact of anisotropy on waveform modeling and indicate that simplified 1D isotropic models may be insufficient for complex crustal settings. We review how structural effects, including anisotropy and 3D heterogeneity, contribute to transverse-component energy in the 2017 DPRK explosion and discuss implications for seismic source classification.

AbstractTransient stress changes from seismic waves of distant earthquakes can promote local fault slip, a phenomenon referred to as remote dynamic triggering. This study examines the remote triggering susceptibility of faults in the Lower Rhine Embayment (LRE) in the Weisweiler region, Germany, a proposed site for geothermal energy production. Assessment of remote triggering can guide industrial operations to assess seismic hazard and mitigate risks associated with fault reactivation caused by small stress perturbations. We select a set of 23 candidate mainshocks from global earthquake catalogs that produce peak ground velocities (PGVs) that exceed 0.02 cm/s in the LRE. The magnitude of these mainshocks ranges from 5.4 to 9.1, epicentral distances range from 50 to 12,300 km, and back azimuth ranges from 16○ to 350○ with a maximum azimuthal gap of 91○. The candidate mainshocks generated PGVs locally from 0.02 to 0.28 cm/s (compared to typical threshold values ranging from 0.02 to 6 cm/s), corresponding to dynamic stress (σpd) values of 1.4 to 26 kPa. We use P-statistics and waveform data from local seismic stations to identify seismicity rate changes and uncatalogued earthquakes that were potentially triggered by the passing mainshock waves. The analysis reveals a statistically significant increase in seismicity rates following four mainshocks: the 1992 Mw5.4 Roermond, Netherlands, 2021 Mw8.2 Chignik, Alsaka, USA, 2023 Mw7.6 Kahramanmaraş, Republic of Türkiye, and 2025 Mw8.8 Kamchatka, Russia earthquakes. The 1992 Roermond mainshock triggered earthquakes within 50 km of its epicenter that were clustered between the Feldbiss and Sandgewand faults. The same area experienced a triggered earthquake sequence following the Chignik mainshock, suggesting that future detailed monitoring in this area may be warranted. The Roermond aftershock distribution can be divided two groups of events, including 61 that occur on the fault and in the near-field, which can be explained by static-stress increase and fluid diffusion. Another 32 remote aftershocks occurred that are consistent with secondary triggering promoted by aseismic slip propagation. The alignment of triggering mainshock back azimuths with the dominant strike direction of regional faults suggests that the orientation of incoming seismic waves is an important factor influencing susceptibility. Despite evidence of triggering, the majority of mainshocks (19 out of 23) were not followed by detectable seismicity-rate changes in the LRE, highlighting the complexity of conditions that lead to remote dynamic triggering. The study area does not respond to a triggering stress threshold, suggesting that non-linear, or a combination of linear and non-linear effects, dominate possible triggering mechanisms. Although the LRE does not respond to a clear triggering threshold, this study demonstrates that peak dynamic stress perturbations of approximately 1.4 kPa or greater can still trigger earthquakes. But, susceptibility is modulated by additional factors such as fault orientation, earthquake fault-zone properties, their state in the seismic cycle, and pre-existing stress state.

AbstractPhysics-informed neural networks (PINNs) face significant challenges in modeling multi-frequency wavefields in complex velocity models due to their slow convergence, difficulty in representing high-frequency details, and lack of generalization to varying frequencies and velocity scenarios. To address these issues, we propose Meta-LRPINN, a novel framework that combines low-rank parameterization using singular value decomposition (SVD) with meta-learning and frequency embedding. Specifically, we decompose the weights of PINN’s hidden layers using SVD and introduce an innovative frequency embedding hypernetwork (FEH) that links input frequencies with the singular values, enabling efficient and frequency-adaptive wavefield representation. Meta-learning is employed to provide robust initialization, improving optimization stability and reducing training time. Additionally, we implement adaptive rank reduction and FEH pruning during the meta-testing phase to further enhance efficiency. Numerical experiments, which are presented on multi-frequency scattered wavefields for different velocity models, demonstrate that Meta-LRPINN achieves much faster convergence speed and much higher accuracy compared to baseline methods such as Meta-PINN and vanilla PINN. Also, the proposed framework shows strong generalization to out-of-distribution frequencies while maintaining computational efficiency. These results highlight the potential of our Meta-LRPINN for scalable and adaptable seismic wavefield modeling.

AbstractP-wave reflections from the 410- and 660-km mantle discontinuities are visible in stacks of ambient noise cross-correlation functions of USArray stations spanning the contiguous United States. The reflections are most visible on the vertical components at frequencies between 0.1 and 0.3 Hz during low-noise periods, which generally occur during the summer months in the northern hemisphere. Common reflection point stacking can be used to resolve apparent lateral differences in discontinuity structure across the continent and suggests the possible existence of sporadic reflectors at other depths. Visibility of the 660-km reflector is correlated with faster P-wave velocities at similar depth in a tomographic model for North America. However, the lack of clear agreement between these P-wave ambient noise features and prior mantle-transition-zone imaging studies using other methods suggests caution should be applied in their interpretation. Ambient noise sources from the southern oceans may not be distributed uniformly enough for cross-correlation stacks to provide unbiased estimates of the true station-to-station P-wave Green’s functions. However, the clear presence of 410- and 660-km reflections in the ambient noise data suggests that it should be possible to unravel the complexities associated with varying noise source locations to produce reliable P-wave reflection profiles, providing new insights into mantle structure under the contiguous United States.

AbstractSeismic wave propagation in poro-viscoelastic anisotropic media is of practical importance for exploration geophysics and global seismology. Existing theories generally utilize homogeneous plane wave theory, which considers only velocity anisotropy but neglects attenuation anisotropy and wave inhomogeneity arising from attenuation. As a result, it poses significant challenges to accurately analyze seismic wave dispersion and attenuation in poro-viscoelastic anisotropic media. In this paper, we investigate the propagation of inhomogeneous plane waves in poro-viscoelastic media, explicitly incorporating both velocity and attenuation anisotropy. Starting from classical Biot theory, we present a fractional differential equation describing wave propagation in attenuative anisotropic porous media that accommodates arbitrary anisotropy in both velocity and attenuation. Then, instead of relying on the traditional complex wave vector approach, we derive new Christoffel and energy balance equations for general inhomogeneous waves by employing an alternative formulation based on the complex slowness vector. The phase velocities and complex slownesses of inhomogeneous fast and slow quasi-compressional (qP1 and qP2) and quasi-shear (qS1 and qS2) waves are determined by solving an eighth-degree algebraic equation. By invoking the derived energy balance equation along with the computed complex slowness, we present explicit and concise expressions for energy velocities. Additionally, we analyze dissipation factors defined by two alternative measures: the ratio of average dissipated energy density to either average strain energy density or average stored energy density. We clarify and discuss the implications of these definitional differences in the context of general poro-viscoelastic anisotropic media. Finally, our expressions are reduced to give their counterparts of the homogeneous waves as a special case, and the reduced forms are identical to those presented by the existing poro-viscoelastic theory. Several examples are provided to illustrate the propagation characteristics of inhomogeneous plane waves in unbounded attenuative vertical transversely isotropic porous media.

AbstractDeep learning (DL) approach has gained attention for earthquake (EQ) detection. To alleviate the problem of training data shortage, transfer learning (TL) provides a useful framework to adapt pre-trained models, typically through tuning of model parameters. Nonetheless, the current practice still requires considerable data, which hinders its application where only a small number of data is available. Instead of TL, we propose a novel two-stage of model correction as a solution to this important and ubiquitous problem in EQ detection. In the proposed approach, a pre-trained DL model is directly applied to waveform data in the target domain (first stage), and the cases that are classified as an earthquake signal (i.e., positive cases) are further classified as positives and negatives using a non-DL classification method (second stage). Our classification method for the second stage is based on multiple clustering, which characterizes local waveform patterns in terms of amplitude scale in specific time segments that are inferred in a data-driven manner. This characterization captures complex high-dimensional waveform patterns in a low-dimensional space, which leads to the effective classification of true and false positives. Furthermore, the proposed method is useful when only true positive waveforms are labeled (PU classification). Both synthetic and real data analysis clearly demonstrated effectiveness of unsupervised waveform characterization of the proposed method.

AbstractFull-waveform inversion (FWI) is a powerful imaging technique that produces high-resolution subsurface models. In seismology, FWI workflows are traditionally based on seismometer recordings. The development of fibre-optic sensing presents opportunities for harnessing information from new types of measurements. With dense spatial and temporal sampling, fibre-optic sensing captures the seismic wavefield at metre-scale resolution along the cable. Applying FWI to fibre-optic measurements requires the reformulation of the forward and adjoint problems due to two fundamental differences to seismometer data: i) fibre-optic measurements are sensitive to strain rather than translational motion, and ii) they do not represent the motion at a single spatial point, but instead capture the average deformation over a pre-defined cable segment, known as the gauge length. Within this study, we derive the adjoint sources to perform FWI for data from distributed acoustic sensing (DAS) and integrated fibre-optic sensing (IFOS) that are based on moment tensors. Our formulation incorporates gauge-length effects, direction-dependent sensitivity and complex cable layouts. For the numerical simulations, we use a spectral-element solver that allows us to incorporate surface topography, and coupled viscoacoustic and viscoelastic rheologies. In illustrative examples, we present how our theoretical developments can be used in inversions of synthetic fibre-optic data generated for a realistically curved cable placed on irregular topography. As examples, we invert for source parameters, including moment tensor, location, and origin time for noise-free DAS data, noise-contaminated DAS data, and IFOS data. Further, we present the 3-D imaging results for the three data groups and further analyse the effect of scatterers on the FWI based on DAS data. In all example inversions, we compare how close the found model is to the known ground truth. The codes to produce these results are accessible and ready to be applied to real data inversions.

AbstractHigh-frequency induced polarisation, which measures the complex electrical conductivity in a frequency range up to several hundred kHz, is potentially suitable to detect and quantify ice in the frozen subsurface. In order to estimate ice content from the electrical spectra, a two-component weighted power mean (WPM) model has been suggested and applied to field-scale data. In that model, ice is one of the components, whereas the solid phase, residual liquid water and potentially air form the second component, called “matrix”. Here, we apply the model to laboratory data previously discussed in the literature, with the aim to assess the applicability of the model and to understand the behaviour of the frequency-dependent electrical conductivity. The data were measured on an unconsolidated sediment sample with 20.8% water content from the European Alps, and a consolidated sandstone with 16.6% porosity. Electrical spectra have been measured over a temperature range from approx. - 41 ○C to +20 ○C and a frequency range from 0.01 Hz to 45 kHz. We extend the original WPM model to account for low-frequency polarisation in form of a constant phase angle model. The measured data were fitted with the model by a least-squares inversion algorithm. In order to reduce the ambiguity, we constrained several of the nine underlying parameters by literature values, in particular for the electrical properties of water ice, and the expected ice content according to porosity or water content of the unfrozen sample. Both data sets can be well matched, corroborating the hypothesis that the model is in principle suitable to explain measured data of frozen samples in that frequency range. One important observation is that the mixing parameter, i.e. the power in the WPM model, which is controlled by the geometric arrangement of the two components, depends on temperature. For the unconsolidated sample it even becomes negative at the coldest temperature, which is important because negative shape factors relate to specific geometries. A second observation is that relatively large permittivities of the matrix are required to fit the data, suggesting that processes at the interface between solid/liquid phase and ice, which are not included in the volumetric mixing model, might be relevant and should be considered in future extensions of the model.

SummaryThe dominant forces shaping the unique geometries of flat slabs are still not fully understood. Knowing how the stress field changes with respect to the shape of the slab allows inferences of the dominant forces acting on the slab. In this study we calculated new models of the slab geometry and the intraslab stress field in the Pampean flat slab region of the Chile-Argentina Subduction Zone (latitude ∼25°36°S) where the Nazca Plate subducts together with the aseismic Juan Fernandez Ridge. To build the models, we used a catalog of 1,059 well-located slab earthquakes recorded by the SIEMBRA and ESP temporary seismic arrays and calculated 411 new focal mechanisms that were analyzed together with 407 focal mechanisms from other catalogs. Our results confirmed slab seismicity features such as a reverse dip (i.e. opposite to the subduction direction) of the seismicity band within the flat slab, two bands of descending seismicity, and two regions with an absence of earthquakes. These seismicity patterns express the shape of the slab and its hydration state, with more localized slab dehydration along the inland path of the Juan Fernandez Ridge relative to the surroundings. In one of the regions without earthquakes, the slab is most likely continuous and dry, while in the other one the slab is missing, in agreement with previous works that proposed a hole in the slab visible with other methods. A comparison between the stress field and the local slab dip from both our new model and a previous one (Slab2) indicates that the dominant forces acting on the flat slab are the slab pull and the ridge buoyancy. Finally, the shape of the flat slab is controlled by the geologic migration of the Juan Fernandez Ridge, making the flat slab four times wider than the ridge offshore, and by the competing forces of the slab pull and the ridge buoyancy that creates a notable flexure (bulge) resembling the geometry of the outer rise near the trench.

SummaryIn the analysis of Induced Polarization (IP) data, it is commonly assumed that induction effects (IE) are negligible. However, at higher frequencies, this assumption becomes increasingly invalid, posing challenges for IP measurements. High-frequency induced polarization (HFIP) extends the conventional IP frequency range beyond 100 kHz, allowing estimation of ice content by capturing the characteristic decrease in permittivity of water ice. This study focuses on the interpretation of HFIP data while accounting for IE. We modified an existing one-dimensional simulation code to evaluate HFIP responses over frozen ground with ice, both with and without the influence of IE. Our results demonstrate that IE can distort HFIP measurements in typical permafrost conditions, potentially obscuring the characteristic polarization behavior of water ice. two-dimensional IP inversion codes that account for IE are not routinely available. Even if they were, the simultaneous presence of IE and IP would likely complicate their application. We therefore propose to remove IE from the data, with the aim to use well-established two-dimensional SIP inversion routines. For this purpose, we implemented a one-dimensional inversion routine that includes IE as a frequency-dependent correction factor. The method is based on fitting a layered model to the data. Comparing the responses with and without IE, we calculate a correction factor which is subsequently used to remove IE from the original dataset. The method is conservative in the sense that features not well matched by the one-dimensional inversion are preserved and no information gets lost. We demonstrate the effectiveness of the method with synthetic data, as well as field datasets from an unfrozen site in Germany, and from permafrost peatlands in Scandinavia. We provide evidence from reciprocal measurements that cable coupling effects, not included in the correction procedure, have been effectively minimized by the acquisition system. We further show that high-frequency phase shifts are strongly influenced by IE, and that the correction methodology successfully restores the spectral response. By applying the approach to a measured dataset, we demonstrate that two-dimensional inversion of the corrected data with a well-established code is both feasible and robust. The resulting model deviates markedly from that obtained with uncorrected data, highlighting the critical role of the correction procedure for reliable interpretation.

SummarySubduction zones are tectonic plate boundaries where one tectonic plate is being forced beneath another and known for producing some of the most powerful earthquakes in history. Understanding the regional-scale structural heterogeneity in the subduction zone is crucial for deciphering the genesis of megathrust earthquakes and the factors that control rupture dynamics. The rupture dynamics of the earthquakes in Andaman-Sumatra region are primarily hindered due to the lack of velocity and anisotropy information at the lithospheric level. Compression wave (Pn) waves propagate in the uppermost mantle and can provide velocity and anisotropy constraints for this portion of the mantle along their ray-path. We generated the first comprehensive high-resolution Pn-wave tomography, restricting the turning of diving ray paths less than 50 km deep in the collision zones, to map lithospheric velocity and anisotropy by inverting 65 297 Pn arrivals extracted from 6958 regional events recorded at 384 stations. The results show a strong variation in Pn-wave velocity and fast polarization directions (FPDs) in the entire study region that may be an essential factor for earthquake nucleation. We observed an abrupt change in the Pn-FPDs from Andaman to Sumatra and Sumatra to Java regions. We suggest that well defined plate boundaries between Indian and Capricorn plates, that was earlier reported to be diffused one. Such observation is also supported by the lower spreading rate and higher crustal age of the old Indian oceanic plate in the Andaman region, compared to the faster spreading rate and lower crustal age of the Capricorn oceanic plate in the north Sumatra region. We suggest that the Sunda plate strongly coupled with subducting oceanic slab causes the mega events of 2004 and 2005 supported by trench-normal Pn-FPDs along with rigid-slab nature, and both the rupture propagation terminate near Simeulue Island region due to abrupt material changes beneath it. Our study also reveals the presence of low Pn velocity anomalies in the west of the Andaman Sea, suggesting the existence of a magma reservoir pouring out lava through the steeply torn Indian oceanic slab.

SummaryThe converted wave technique, namely Receiver function (RF), has been routinely employed for estimating one-dimensional velocity models of the Earth’s crust and mantle structures. Physics-driven methods such as inversions are employed to receiver function data to estimate velocity models. Despite their wide utility, the presence of dipping and anisotropic geological structures often complicates this process. To address these complexities, here we introduce a Physics-Guided unsupervised deep learning approach for the inversion of receiver function data. An unsupervised deep learning approach together with implicit neural representations is developed here, for prediction of subsurface model parameters such as the layer thickness, S-wave velocity, anisotropy, trend, plunge, strike and dip without requiring any labelled data. In addition, we incorporate a dynamic layer setting criterion to automatically pick the optimal number of layers required to fit the data, which is found to be particularly effective for identifying pronounced discontinuities like the crust-mantle boundary. For determining the optimal model parameters, neural network output parameters (the model parameters) are used in forward modelling of the receiver functions. Inversion results from both synthetic and real field data from the Indian shield and Hi-CLIMB network suggest that the physics-guided unsupervised deep learning approach is effective in inversion tasks, particularly when dealing with dipping and anisotropic media.

SummaryWe explore the potential of utilizing Distributed Acoustic Sensing (DAS) for Back-projection (BP) to image earthquake rupture processes. Synthetic tests indicate that sensor geometry, azimuthal coverage, and velocity model are key factors controlling the quality of DAS-based BP images. We show that mitigation strategies and data processing modifications effectively stabilize the BP image in less optimal scenarios, such as asymmetric geometry, narrow azimuthal coverage, and poorly constrained velocity structures. We apply our method to the Mw7.6 2022 Michoacán earthquake recorded by a DAS array in Mexico City. We also conduct a BP analysis with teleseismic data for a reference. We identify three subevents from the DAS-based BP image, which exhibit a consistent rupture direction with the teleseismic results despite minor differences caused by uncertainties of BP with DAS data. We analyze the sources of the associated uncertainties and propose a transferrable analysis scheme to understand the feasibility of BP with known source-receiver geometries preliminarily. Our findings demonstrate that integrating DAS recordings into BP can help with earthquake rupture process imaging for a broad magnitude range at regional distances. It can enhance seismic hazard assessment, especially in regions with limited conventional seismic coverage.

SummaryShort-duration instrumental earthquake catalogues, sparsity in seismic stations, and paucity of near-source earthquake data impose strong limitations on data-driven seismic hazard assessment. In contrast, regional-scale multi-cycle earthquake simulations that generate realistic rupture scenarios provide physics-based earthquake rupture forecasts for seismic hazard assessment. In this study, we use the physics-based multi-cycle earthquake simulation engine MCQsim to compute long-term synthetic seismic catalogues that include multi-segment ruptures on a complex-geometry fault system. We expand on previous research by conducting systematic statistical analyses of synthetic earthquake catalogues for the Gulf of Aqaba (GoA, Saudi Arabia) that forms the southern extension of the Dead Sea Fault (DSF). The GoA is characterized by predominantly left-lateral strike-slip faulting and its seismic activity is revealed by past seismic swarms and large-magnitude events, including the 1995 M 7.2 Nuweiba earthquake. To address epistemic uncertainties in the seismic source characterization (SSC) of this region, we explore several modelling realisations by altering the fault system’s geometric configuration and frictional properties. Our simulations reveal that multi-segment ruptures reaching M 7.6 are possible in the GoA. The simulated catalogues reveal source-scaling properties of rupture area and stress drop consistent with global observations and empirical scaling laws. Additionally, the statistical properties of the synthetic catalogues, including the earthquake magnitude-frequency distribution, align with the short-term recorded seismicity in the study region. In summary, our simulation-based study provides insights into the GoA’s seismic behaviour through comprehensive parameter-space exploration and sensitivity analyses that document the possibility for geometrically complex, large multi-segment magnitude earthquakes.

SummaryFull waveform inversion (FWI) is a high-precision subsurface imaging technique that inverts subsurface parameter models by minimizing the discrepancy between observed and synthetic seismic data. However, complicated wave propagation mechanisms, non-convexity of the loss function, and limited seismic acquisition system necessitate the incorporation of sufficient prior and physical constraints to alleviate the ill-posedness and cycle-skipping. Although the implicit FWI (IFWI) can encode implicit spectral bias (i.e., inverting model parameters from low frequencies to high frequencies) to reduce the dependency on an accurate initial model, its limited high-frequency inversion capability results in thousands of iterations for the final results. In this paper, we indicate that the frequency hyperparameters of the sine activation function in IFWI modulate the spectral bias, making a trade-off between inversion accuracy and stability, i.e., lower frequencies yield robust FWI but lower accuracy, while higher frequencies achieve higher accuracy on the premise of an accurate initial model. To improve both the stability and accuracy of IFWI, we propose a novel implicit FWI method with an adaptive Fourier reparameterization strategy (termed FR-IFWI), which explicitly encodes multi-frequency information by reparameterizing the network weights using a learnable coefficient matrix and fixed Fourier frequency bases. The role of learnable matrices in neural networks can evolve from determining frequencies in IFWI to actively selecting frequencies from fixed frequency bases through FR-IFWI, which alleviates the dependence on activation function frequency and obtains more robust and accurate inversion results. Extensive numerical experiments on the modified Marmousi, 2D SEG/EAGE Salt and Overthrust models confirm that FR-IFWI successfully achieves superior inversion efficiency and accuracy compared with conventional FWI and IFWI methods.

SummaryA growing body of literature has contributed to linking the presence of bacteria with SIP signals. Yet, there are still unresolved questions concerning the contribution of cell density and microbial metabolic activity in porous media (soils and sediments) to SIP signals. Moreover, there is continued debate on whether cells themselves polarize or whether a cell-sediment interaction is a prerequisite for the measured responses. This study investigates the SIP response of Shewanella oneidensis MR-1 in isolation, that is, in the absence of a mineral porous medium using two setups (i) cells in aqueous suspension and (ii) alginate bead-packed reactors. Results from experiments conducted with static cell suspensions shed light on the strong control of cell settling that drives erratic, poorly reproducible and difficult to interpret SIP signals. However, incubating cells in bead packed reactors yielded reproducible trends in σ″, with strong (3 – 10 mrad) signals that followed the expected cell growth behaviour. Relating σ″ to measured cell density and metabolic activity (using ATP) highlighted the strongly linked contribution of both activity and cell density and SIP. Here, we report a lower frequency polarization peak between 0.01 and 0.1 Hz in the bead reactors, which we attribute to the polarization of cell colonies in the densely packed reactors. In summary, our findings shed light on the direct contribution of cells and their activity to polarization, in the absence of cell-sediment interactions and provide a novel approach for studying cell polarization in static incubation in a porous environment.