Geophysical Journal International

SummaryWe introduce MTUQ, an open-source Python package for seismic source estimation and uncertainty quantification, emphasizing flexibility and operational scalability. MTUQ provides MPI-parallelized grid search and global optimization capabilities, compatibility with 1D and 3D Green’s function database formats, customizable data processing, C-accelerated waveform and first-motion polarity misfit functions, and utilities for plotting seismic waveforms and visualizing misfit and likelihood surfaces. Applicability to a range of full- and constrained-moment tensor, point force, and centroid inversion problems is possible via a documented application programming interface (API), accompanied by example scripts and integration tests. We demonstrate the software using three different types of seismic events: 1) a 2009 intra-slab earthquake near Anchorage, Alaska; 2) an episode of the 2021 Barry Arm landslide in Alaska; and 3) the 2017 Democratic People’s Republic of Korea (DPRK) underground nuclear test. With these events, we illustrate the well-known complementary character of body waves, surface waves, and polarities for constraining source parameters. We also convey the distinct misfit patterns that arise from each individual data type, the importance of uncertainty quantification for detecting multi-modal or otherwise poorly constrained solutions, and the software’s flexible, modular design.

SummaryThis short paper presents a new equation of state for condensed phases. The equation of state is built on the premise that K′, the first derivative of the bulk modulus, monotonically increases with volume according to a power law. The input parameters are the zero-pressure volume V0, bulk modulus K0, and first and second derivatives of the bulk modulus, $K^{\prime }_0$ and $K^{\prime \prime }_0$, and also $K^{\prime }_{\infty }$, the value of K′ at infinite compression. Expressions are provided for the internal energy, pressure, and bulk modulus. The equation of state is robust for all compressions as long as $K^{\prime \prime }_0 < 0$ and $K^{\prime }_{\infty } < K^{\prime }_0$. Heuristic values are suggested for situations in which available data is not sufficient to independently constrain $K^{\prime \prime }_0$ and $K^{\prime }_{\infty }$. The equation of state compares favourably with other equations of state using recently published experimental data on Au and Pt.

SummaryFaults exhibit dynamic weakening during large displacements (>1 m) at seismic slip velocities (>0.1 m/s), but the role of this weakening in small-displacement induced earthquakes (M 3–4), such as those in the Groningen Gas Field (the Netherlands), remains unclear. We conducted seismic slip-pulse experiments on Slochteren sandstone gouges (SSG) using a rotary-shear apparatus to investigate their dynamic behavior. Pre-sheared gouge layers, confined between ∼1.5 mm thick sandstone host blocks, were subjected to slip pulses at initial effective normal stresses of 4.9–16.6 MPa and pore fluid pressures of 0.1 and 1 MPa under undrained conditions. Slip pulses reached peak velocities of 1.8 m/s, accelerations up to 42 m/s², and displacements of 7.5–15 cm, using either dry Argon or water as pore fluid at ambient temperatures. Water-saturated gouges showed rapid weakening from a peak friction of ∼0.7 to ∼0.3, with early dilatancy followed by slower ongoing dilation. In contrast, Argon-filled samples exhibited only subtle weakening. Our findings confirm that water-saturated SSG weakens substantially during slip, with minimal dependence on normal stress, slip acceleration, or displacement, while dry samples do not. Microstructural analysis indicates no systematic relationship between PSZ width and frictional work or power input densities, suggesting that wear or heat production alone does not govern PSZ growth. Instead, thermal pore fluid pressurization, potentially involving water phase transitions at asperity scales, may drive weakening in short-displacement, induced seismic events.

SummarySeismic anisotropy can inform us about convective flow in the mantle. Shear waves traveling through azimuthally anisotropic regions split into fast and slow pulses, and measuring the resulting shear-wave splitting provides some of the most direct insights into Earth’s interior dynamics. Shear-wave splitting is a constraint for path-averaged azimuthal anisotropy and is often studied regionally. Global compilations of these measurements also exist. Such compilations include measurements obtained using different data processing methodologies (e.g., filtering), which do not necessarily yield identical results, and reproducing a number of studies can be challenging given that not all provide the required information, e.g., about the source location. Here, we automatically determine SKS, SKKS and PKS shear-wave splitting parameters from a global dataset. This dataset includes all earthquakes with magnitudes ≥5.9 from 2000 to the present, collected from 24 data centers, totaling over 4,700 events and 16 million three-component seismograms. We obtain approximately 90,000 robust measurements for “fast azimuth”, φ, and delay time, δt, and 210,000 robust null measurements. Results generally agree with previous work but our measurements allow us to identify hundreds of “null stations” below which the mantle appears effectively isotropic with respect to azimuthal anisotropy, which are important for some splitting techniques. We make all measurements publicly available as a data product, along with detailed metadata. This serves two purposes: ensuring full reproducibility of results and providing all necessary information for future systematic use of our measurements, in tomography applications or comparisons with geodynamic flow predictions.

SummaryLove (LQ) and Rayleigh-wave (LR) dispersion data provide essential constraints on radially-anisotropic shear-wave velocity structure. Vertically-polarized S-wave velocity ($\rm S_V$) structure of India–Tibet are available from modeling of LR dispersion data, but reliable LQ dispersion measurements are limited. This is due to poor signal-to-noise (SNR) ratio on horizontal component waveforms, off great-circle-arc propagation and overtone interference. We overcome these limitations by performing systematic SNR tests, polarization analysis and minimization of overtone interference, to compute fundamental-mode LQ group-velocity dispersion and tomography across India, Himalaya and Tibet, for period range of 10–120 s. Fundamental-mode group-velocity dispersion, in this period range, is sensitive to the crust and upper-mantle structure. Lateral resolution of the group-velocity maps vary from 4$\rm ^{\circ }$ (10–40 s) to 9$\rm ^{\circ }$(90–120 s). Group-velocities (absolute and anomaly) and its lateral variations match the regional-scale geologic and tectonic features. Up to 20 s period low-velocity sedimentary layers are observed in the Bengal and Indus delta-fan complexes, Himalaya and Suleiman Mountain foreland basins, Qaidam, Tarim and Eastern Tadjik Basins. The Indian Cratons have higher group-velocities compared to the thickened Tibetan Plateau crust across the entire range of periods. At lower periods the western Tibetan Plateau is underlain by high-velocity anomaly and the central-eastern Plateau has a broad zone of low-velocity anomaly. To test the null-hypothesis of isotropy, we use isotropic $\rm S_V$ models to predict the observed LQ dispersion data. Our test negates the null hypothesis and suggest radially-anisotropic structure. Finally, we invert the LQ dispersion data to obtain horizontally-polarized shear-wave ($\rm S_H$) velocity structure beneath India–Tibet. $\rm S_H$ velocity models have an uncertainty of ∼5%. These models reveal high average $\rm S_H$ velocity (3.7–3.9 $\rm km \,\, s^{-1}$) in the Indian Cratonic crust, which extends beyond its geological outcrops, beneath the Deccan and Marwar Plateau, Southern Indian Granulite Terrane, and the Himalayan foreland basin (HFB). This is underlain by a high $\rm S_H$ velocity mantle-lid with variable thickness beneath cratons. The thickest mantle lids are beneath the Dharwar (∼150 km) and Bastar (∼180 km) Cratons. The high velocity Indian Cratonic crust and upper-mantle underthrust the Tibetan Plateau entirely in the west, up-to the Altyn-Tagh Fault (ATF); and half-way in the centre-east, up-to the Bangong-Nujiang Suture (BNS). This has lead to crustal-doubling beneath Tibet and a large-scale mid-crustal low velocity layer, bound by active faults along its edges. We use the $\rm S_H$ velocity of ≤3.0 $\rm km \,\, s^{-1}$ to delineate the regional-scale sedimentary basins. These include the Bengal and Indus delta-fan complexes with sedimentary layer thickness >15 km; the intra-cratonic HFB, Eastern Tadjik and Qaidam Basins with up-to 10 km sedimentary layer thickness. The Tarim basin has strong E-W variation in sedimentary thickness with the deepest depocentre beneath the eastern basin (>15 km), shallowing significantly west of the Mazhar-Tagh Fault to <10 km. The HFB is also marked by lateral variations in sedimentary thickness, with the thickest layers beneath the Eastern and Western Gangetic Basins (∼6–8 km) and the thinnest beneath the Indus Basin and Brahmaputra Valley (<2 km). These sub-basins are segmented by basement ridges, and the sedimentary thickness variation is controlled by flexural bending of the underthrusting Indian plate beneath the Himalaya.

SummaryMarine observation data are plentiful for constructing seafloor topography, and the integration of multi-sources data to construct a more accurate topography model remains a significant subject that continues to be explored and studied. In this study, we use geoid height (GH), Gravity (VG) and Vertical Gravity Gradient (VGG) derived from a single rectangular prism to establish the foundational observation equations for predicting topography. The effectiveness of the foundational observation equations is verified through study cases without the use of the ship measurement depth data. Additionally, the single- and multi-beam soundings data are employed as control points to integrate into the foundational observation equations for predicting topography. The prediction results demonstrate that the prediction accuracy of combined VG anomalies with ship soundings is better than GH and VGG anomalies, which is primarily because VG anomalies are effective than GH amplify high frequency signals of topography and stronger than VGG anomalies in suppressing high frequency errors. Additionally, considering the limited accuracy of marine gravity in sea region with islands and reefs, this study incorporates satellite imagery data to identify the location and size of the islands. Then, the topography of the islands is introduced and the control equations is established to jointly predict topography. The prediction results reveal the RMS errors between prediction results and single- and multi-beam sounding data are 67.4 m, which is 37.4%, 57.8% and 62.8% higher than that of SRTM 15+, DTU and ETOPO-1 models respectively. Notably, compared with the STRM 15+ model, the algorithm improves the topography accuracy of the sea area near the islands by nearly 60.8%.

SummaryGeological storage of captured CO₂ is essential for reducing anthropogenic greenhouse gas emissions and ensuring the sustainable use of fossil fuels. Understanding the influence of saturation, pressure, and temperature on the elastic properties of brine-saturated sandstone flooded with supercritical CO₂ is critical for interpreting seismic and sonic logging data, which aids in monitoring and quantifying subsurface changes associated with CO₂ injection. The experimental results indicate that as scCO₂ saturation increases (0∼60 per cent), the P-wave velocity decreases significantly with an average of 14 per cent drop, while the S-wave velocity increases slightly. Temperature variations (80–110°C) have a minimal effect on both velocities (1∼2 per cent), whereas elastic features show noticeable sensitivity to the variation of confining pressure (20-30 MPa) and pore pressure (10-20 MPa). Ignoring the effects of pore pressure might lead to the bias of interpreting seismic data for monitoring scCO₂ saturation change. The constructed rock physics models well capture the coupled effects of porosity and scCO₂ saturation on the P-impedance and P- and S-wave velocity ratio, which show good agreement with the experimental results. These findings are crucial for improving monitoring methods and enhancing the accuracy of predictive models for CO₂ geological storage.

SummaryThe causes of intraplate volcanoes in northeastern (NE) China and, in particular, how asthenospheric upwelling interacts with the lithosphere remain poorly constrained. In this study, we use teleseismic data to measure the phase and group velocities of Rayleigh and Love waves, and invert for the shear wave velocity and radial anisotropy within the crust and uppermost mantle. Our results show that there are significant low-velocity anomalies and negative radial anisotropy to the northeast of Changbaishan, suggesting an asthenospheric melt reservoir. This is underlain by mantle upwelling, with regional lithospheric structure focusing melt beneath the volcano. In addition, our results show a high-velocity body in the mantle beneath the southwestern Songliao Basin. This exhibits negative radial anisotropy at its margins, suggesting vertical flow. We suggest that lithospheric delamination here may drive intraplate volcanism beneath the Great Xing'an range.

SummaryLong-period underside SS wave reflections have been widely used to furnish global constraints on the presence and depth of mantle discontinuities and to document evidence for their origins, e.g., mineral phase-transformations in the transition zone, compositional changes in the mid-mantle, and dehydration-induced melting above and below the transition zone. For higher-resolution imaging, it is necessary to separate the signature of the source wavelet (SS arrival) from that of the distortion caused by the mantle reflectivity (SS precursors). Classical solutions to the general deconvolution problem include frequency-domain or time-domain deconvolution. However, these algorithms do not easily generalize when (1) the reflectivity series is of a much shorter period compared to the source wavelet, (2) the bounce point sampling is sparse, or (3) the source wavelet is noisy or hard to estimate. To address these problems, we propose a new technique called SHARP-SS: Sparse High-Resolution Algorithm for Reflection Profiling with SS waves. SHARP-SS is a Bayesian deconvolution algorithm that makes minimal a-priori assumptions on the noise model, source signature, and reflectivity structure. We test SHARP-SS using real data examples beneath the NoMelt Pacific Ocean region. We recover a low-velocity discontinuity at a depth of ∼69 ± 4 km which marks the base of the oceanic lithosphere, consistent with previous work derived from surface waves, body wave conversions, and ScS reverberations. We anticipate high-resolution fine mantle stratification imaging using SHARP-SS at locations where seismic stations are sparsely distributed.

SummaryReflection imaging at volcanoes presents significant challenges due to the highly heterogeneous subsurface, which generates complex wavefields characterized by substantial wave scattering. These scattered waves obscure coherent energy, such as reflections from geological structures in the subsurface. In this study, we develop processing strategies to address the limitations of high-frequency (5-20 Hz) passive reflection imaging at Krafla, a volcanic caldera in NE Iceland. Krafla is among the few locations worldwide where magma has been encountered at 2.1 km depth when drilling the IDDP1 borehole. We analyze over 300 local microearthquakes and industrial noise recorded during five weeks in the summer of 2022. We show that wavefields lack coherency even between stations spaced at 30-meter intervals due to the dominance of site effects beneath the stations. However, data coherency improves in the common-station domain, where different earthquakes recorded by a fixed station are analyzed, thereby stabilizing the site effect. Spectral analyses in this domain reveal that site effects are partly due to resonances at the stations, likely caused by lava flows and cavities in the heterogeneous near-surface. By constructing a resonance removal filter, we successfully deconvolve resonance effects from the data, revealing previously masked coherent energy. We further reduce site effects by applying linear stacking of clustered earthquake traces and non-linear amplitude weighting. Our approach significantly enhances coherency between stations and enables the identification of reflections in microearthquakes likely originating from the known magma-rock interface beneath the IDDP1 borehole.

SummaryIn seismology, wavefield injection refers to the propagation of seismic waves generated by remote sources into local domains bounded by enclosed surfaces. The simulations of wavefield injection, primarily focused on the interaction between incoming seismic waves and local structures, are key to earthquake hazard modeling and full-waveform seismic tomography using tele-seismic waves. In this paper, we show that simulating wavefield injection is equivalent to solving the wave equation subject to interface discontinuity conditions. To provide a general framework to study wavefield injection, we formally define the interface discontinuity problem, and discuss its representation theorem and uniqueness. We also develop an efficient interface-discontinuity-based numerical algorithm to solve the wavefield injection problem through implementations of spectral-element methods, and show with numerical examples that wavefield injection can be accurately simulated at different scales with this algorithm. Under this framework, we draw connections with previously proposed wavefield injection algorithms/hybrid methods, and clarify several theoretical questions on wavefield injection from previous research. We demonstrate the efficiency and accuracy of our approach through wavefield injection examples at local and continental scales. Furthermore, we illustrate the applicability of the interface discontinuity approach to performing kinematic fault simulations through an numerical example.

SummaryThe time-varying geomagnetic field is a superposition of contributions from multiple internal and external current systems. A major source of geomagnetic variations at periods less than a few years are current systems external to the solid Earth, namely the ionospheric and magnetospheric currents, as well as associated induced currents. The separation of these three sources is mathematically underdetermined using either ground or satellite measurements alone, but becomes tractable when the two datasets are combined. Based on this concept, we developed a new geomagnetic field modelling approach that allows us to simultaneously characterise the mid-latitude ionospheric, magnetospheric and the internal induced magnetic fields using ground and satellite observations for all local times and magnetic conditions, and without prescribing any harmonic behaviour on these current systems in time, as is typical in other models. By applying this new method to a 10-year dataset of ground observatory and multi-satellite measurements from 2014 to 2023, we obtained the time series of the spherical harmonic coefficients of the ionospheric, magnetospheric and induced fields. These new time series allow the study of complex non-periodic dynamics of the external magnetic fields during global geomagnetic storms, as well as periodicities in the magnetospheric coefficients linked to solar activities and periodic ionospheric magnetic fields linked to lunar daily variations, contributing to a more complete picture of the dynamics of the external currents and magnetosphere-ionosphere interactions, and facilitating more accurate space weather nowcast and forecast. Finally, the new approach allows for a better characterisation of internal induced field sources, leading to higher quality electromagnetic transfer functions.

SummaryThis paper examines the linear stability of sliding on faults embedded in a 2D elastic medium that obey rate and state friction and have a finite length and/or are near a traction-free surface. Results are obtained using a numerical technique that allows for analysis of systems with geometrical complexity and heterogeneous material properties; however only systems with homogeneous frictional and material properties are examined. Some analytical results are also obtained for the special case of a fault that is parallel to a traction-free surface. For velocity-weakening faults with finite length, there is a critical fault length L* for unstable sliding that is analogous to the critical wavelength h* that is usually derived from infinite fault systems. Faults longer than L* are linearly unstable to perturbations of any length. On vertical strike-slip faults or faults in a full-space L* ≈ h*/e, where e is Euler’s number. For dip-slip faults near a traction-free surface L* ≤ h*/e and is a function of dip angle β, burial depth d of the fault’s up-dip edge, and friction coefficient. In particular, L* is at least an order of magnitude smaller than h* on shallowly dipping (β < 10○) faults that intersect the traction-free surface. Additionally, L* ≈ h*/e on dip-slip faults with burial depths d ≥ h*. For sliding systems that can be treated as a thin layer, such as landslides, glaciers, or ice streams, L* = h*/2. Finally, conditions are established for unstable sliding on infinitely-long, velocity-strengthening faults that are parallel to a traction-free surface.

SummaryIn this paper, we present a catalogue of relocated seismic events in the North Sea spanning 1961 to 2022. Data from all relevant agencies were combined, incorporating all available seismic phase readings, thereby enhancing station coverage. As a result, our updated locations reveal a more clustered and aligned seismicity pattern compared with the original catalogue. Even with our combined dataset, only 157 of the 7,089 relocated events have azimuthal gaps of less than 90 degrees. Additionally, the distances between onshore stations and offshore events are considerable. Both of these factors lead to relatively poorly constrained hypocentres for most events. We therefore evaluate the performance of 1D velocity models routinely used by different North Sea adjacent monitoring agencies for earthquake location estimations in the North Sea. The variations in assessments due to the seismic velocity model used are significantly larger than the uncertainty ellipses calculated in the relocation, demonstrating that arithmetic uncertainties systematically underestimate location uncertainties in this setting. Obtaining a realistic estimate of location uncertainty is however crucial, particularly for distinguishing between natural and induced seismicity. This is fundamental to safe monitoring of the North Sea offshore industries, including geological CO2 storage. To overcome these discrepancies between the uncertainty ellipses and our multiple relocations, we introduce an alternative method that accounts for variability in the 1D velocity models. This approach enhances the reliability of the earthquake catalogue, and provides a more robust assessment of seismic activity in the North Sea.

SummaryElastic full-waveform inversion has recently been utilized to estimate the physical properties of the upper tens of meters of the subsurface, leveraging its capability to exploit the complete information contained in recorded seismograms. However, due to the non-linear and ill-posed nature of the problem, standard approaches typically require an optimal starting model to avoid producing non-physical solutions. Additionally, conventional optimization methods lack a robust uncertainty quantification, which is essential for subsequent informed decision-making.Bayesian inference offers a framework for estimating the posterior probability density function through the application of Bayes’ theorem. Methods based on Markov Chain Monte Carlo processes use multiple sample chains to quantify and characterize the uncertainty of the solution.However, despite their ability to theoretically handle any form of distribution, these methods are computationally expensive, limiting their usage in large-scale problems with computationally expensive forward modelings, as in the case of full-waveform inversion. Variational Inference provides an alternative approach to estimating the posterior distribution through a parametric or non-parametric proposal distribution. Among this class of methods, Stein Variational Gradient Descent stands out for its ability to iteratively refine a set of samples, usually referred to as particles, to approximate the target distribution through an optimization process. However, mode and variance-collapse issues affect this approach when applied to high-dimensional inverse problems.To address these challenges, in this work we propose to utilize an annealed variant of the Stein Variational Gradient Descent algorithm and apply this method to solve the elastic full-waveform inversion of surface waves. We validate our proposed approach with a synthetic test, where the velocity model is characterized by significant lateral and vertical velocity variations. Then, we invert a field dataset from the InterPACIFIC project, proving that our method is robust against cycle-skipping issues and can provide reasonable uncertainty estimations with a limited computational cost.

SummaryProbabilistic forecasts of earthquakes caused by anthropogenic changes in subsurface stresses require seismicity models that link rupture nucleation to stress states in geological faults. The recently introduced time-dependent stress response (TDSR) model is based on an exponential dependence of the time-to-failure on stress and is a generalization of the well-known rate-and-state (RS) seismicity model. Unlike RS, TDSR can directly incorporate estimates of the initial stress distribution on affected faults in the seismogenic zone. For the case of the Groningen gas field in the Netherlands, we utilize detailed field and borehole studies to estimate the initial stress distribution and rock properties of the reservoir faults. Using these initial conditions, we show that TDSR outperforms the Coulomb failure model, which assumes instantaneous failure, as well as the RS model, which relies on simplified pre-stress assumptions. Furthermore, an instantaneous Coulomb failure model cannot explain the effect of seasonal gas production in Groningen on the timing of induced earthquakes, in contrast to the TDSR model, which shows a good agreement between prediction and observation. Pseudo-prospective tests show that the seismic response to the reduced production since 2014 could have been predicted as early as 2010 if the production scenario had been known.

SummaryIn the past decade, six Mw ≥5.5 earthquakes struck the mountainous Golden Triangle region (Laos, Thailand, Myanmar) of the southeast India-Eurasia collision zone. The largest of them, the 2019 Mw 6.2 Sainyabuli earthquake in western Laos, shook river communities, dams, and a UNESCO World Heritage Site, prompting a need to understand regional earthquake potential. We used Interferometric Synthetic Aperture Radar (InSAR) data and modelling to solve for the 2019 mainshock source parameters, revealing right-lateral strike-slip along a 24 km-long NNW-trending fault which has limited topographic expression and was previously unmapped. InSAR modelling of its largest (Mw 5.5) aftershock in 2021 revealed a 7 km-long splay fault, also previously unrecognized. The 2022 Mw 5.9 Keng Tung earthquake in the northern Golden Triangle also ruptured an unknown, NW-trending right-lateral fault conjugate to longer, NE-trending faults nearby. Collectively, this shows that the region contains faults which are little evident in global digital topography and/or obscured by vegetation but long enough to generate sizeable earthquakes that should be accounted for in seismic hazard assessments. We relocated well-recorded aftershocks and other background seismicity (1978–2023) from across the Golden Triangle using the mloc software. Calibrated hypocenters span focal depths of 5–24 km and are distributed away from the main InSAR-modelled fault traces, another indication of fault structural immaturity. For the three 2019–2022 InSAR-constrained events, we also obtained moment tensor solutions from regional seismic waveform inversion. InSAR-derived peak slip depths and seismological centroid depths are mostly shallow (3–5 km), while focal depths are generally located in areas of low coseismic slip near the bottom of InSAR model faults. More broadly, we estimate a regional seismogenic thickness of ∼17 km (the 90% seismicity cut-off depth), a crucial parameter for seismic hazard calculations and building codes. Our integration of remote-sensing and seismologic analyses could be a blueprint for assessing earthquake potential of other regions with sparse instrumentation and limited topographic fault expression.

SummaryWhile the Earth’s magnetic field has existed for 4Gyr or more, most recent estimates for the age of the inner core nucleation date no further back than 1.5Gyr. As a consequence, the relevant geometry for the Earth’s dynamo has been a full sphere for much of its life, fundamentally different from the present day dynamo operating in a spherical shell. We therefore systematically study magnetic field generation in a rapidly-rotating full sphere where convection is driven by heat sources uniformly-distributed throughout the fluid. We observe a rich diversity of behaviour in our solutions, including dipolar and multipolar dominated fields, together with vacillating and chaotically-reversing magnetic fields. At Prandtl number of unity, we construct regime diagrams as a function of three control parameters, namely the Rayleigh, Ekman and magnetic Prandtl number, which show some similarities with the corresponding diagrams for spherical shell dynamos. This study comprehensively demonstrates the feasibility of early-Earth dynamos that operate based on secular cooling of the core.

SummaryThe rapid and accurate prediction of peak ground acceleration (PGA) few seconds after earthquake start is crucial for assessing the potential damage in target areas in impact-based earthquake early warning systems. However, it is difficult to substantially improve the performance of PGA prediction methods based on empirically defined ground motion prediction equations. In this study, we proposed a hybrid deep learning network (HDL-Net) model for PGA prediction based on Japanese and Chinese datasets. The HDL-Net model is capable of extracting useful spatial and temporal features from the input physical feature parameters and three-component waveforms. The test results showed that HDL-Net outperformed the traditional empirical approaches in terms of timeliness and accuracy. To further validate the robustness of the HDL-Net model for PGA prediction, we conducted a potential damage analysis for five earthquakes in Japan. The results showed that the successful alarm (SA) rate reached 95.22%, the successful no alarm (SNA) rate was 100%, and there was no false alarm (FA). The HDL-Net model provides a potential method for earthquake early warning (EEW) and seismological PGA prediction.

SummaryBeamforming (BF) has been demonstrated to extract multi-mode surface wave dispersion curves from ambient seismic noise. However, due to the limited sampling of the array and the complex distribution of the noise sources, the dispersion image generated by the array-based technique is usually contaminated by aliasing or artifacts. According to seismic interferometry (SI) theory, the Green's function (GF) in the time domain can be retrieved using the noise cross-correlation function (NCF). The Fourier transform of NCFs, i.e. the spatial coherence function, is related to the imaginary part of the frequency domain GF. For the vertical component of the surface wave, it corresponds to the zero-order Bessel function of the first kind, i.e. the standing wave containing propagating waves in two directions described by positive and negative vector wavenumber. In array techniques based on wavefield transforms, it is common to adopt the propagating wave instead of the standing wave to eliminate the aliasing associated with the negative wavenumber, i.e. to replace the Bessel function using the Hankel function or to construct a complete GF via the Hilbert transform. In this paper, we quantitatively analyze the characteristics of three types of aliasing, i.e. the aliasing associated with the period extension of the positive wavenumber, the aliasing associated with the negative wavenumber and those associated with the constant wavenumber. The theoretical representations of different imaging conditions are derived for the finite sampling of the wavefield. A new BF imaging condition is then proposed to remove the crossed artifacts, a type of aliasing associated with the negative wavenumber. The new imaging condition relies only on the computed NCFs and does not require reconstruction of the complete GF via the Hilbert transform. The advantage of random sampling in removing artifacts is illustrated. A random array design scheme is suggested by investigating the array performance of the random array and the array designed using tiles of the Hat family newly discovered in the field of monotile aperiodic tiling. We show the artifacts associated with the constant wavenumber, which are usually manifested as a straight line in the dispersion image of the frequency-velocity domain, also known as radial artifacts, can be eliminated by windowing the NCFs.