Geophysical Journal International

SummaryWe revisit the challenge of estimating velocities from global navigation satellite system (GNSS) coordinate time series by incorporating stochastic processes to address the quasi-periodic oscillations (QPOs) in weighted least-squares estimation. We examine two advanced stochastic seasonal models (i.e., Fractional Sinusoidal Waveform (FSW) and Varying Periodic Band-Pass (VPBP)) and evaluate their efficacy on mitigating velocity biases by using both synthetic data and 14 Antarctic GNSS stations. The result shows that pure FSW cannot fully capture the spectral shape of the QPOs since the FSW's spectral shape is controlled by the fractional parameter d, which is unfortunately dependent on the global spectra slope of the GNSS data. In contrast, VPBP can fit the QPOs more freely by using the parameter phi. Nevertheless, both FSW and VPBP models, when used without augmentation, tend to underestimate velocity uncertainties due to spectral flattening at low frequencies. This poses a risk for applications requiring high-confidence geodetic velocity estimates. To solve this issue, adopting a hybrid noise model that is compensated with an additional appropriate noise background (e.g., power law) is recommended. This knowledge can guide future research on secular velocity estimation of GNSS stations.

SummarySpectral induced polarization (SIP) aims to characterize geological materials by measuring the dispersion of their complex conductivity in the frequency domain. Despite the complex-valued nature of SIP data, most machine learning models used for its analysis rely on real-valued representations that discard phase information and may limit performance. This study investigates the benefits of complex-valued neural networks (CVNN) for SIP applications by comparing their performance against real-valued neural networks (RVNN) across three tasks: mineral classification, Cole-Cole parameter estimation, and mechanistic modelling of ionic and electric potential perturbations around polarizable minerals. To ensure fair comparisons and emphasize the effect of complex-valued representations, we design CVNN and RVNN models with matched capacity, aspect ratio, and training duration. Our numerical experiments show that CVNNs consistently outperform RVNNs in the classification task, achieving lower validation loss and up to 5 percent higher classification metrics (p-value = 2.9 × 10−7). We test the Cole-Cole inversion networks on laboratory SIP measurements and validate the parameter estimation accuracy using synthetic data. Test results indicate that CVNNs produce curve fits that are ≈4 % more accurate for the imaginary part of resistivity (p-value = 3.1 × 10−4), and validation results show accuracy improvements of up to 2 percent for chargeability, relaxation time, and the Cole-Cole exponent (p-value = 1.7 × 10−7). CVNNs also yield more accurate approximations of mechanistic model variables, with error reductions of up to 1 percent for ionic concentrations (p-value = 1.6 × 10−4). Our experiments suggest that CVNNs provide modest but statistically significant benefits in SIP applications involving laboratory or synthetic data. While RVNNs may eventually reach comparable predictive accuracy if trained longer, we observe that CVNNs converge more rapidly under matched training conditions. This study provides a reproducible framework for benchmarking neural network architectures in SIP and supports the integration of CVNNs into geophysical workflows where phase responses encode physically meaningful information.

SummaryDistributed Acoustic Sensing (DAS) is a recent technology that turns optical fibers into multi-sensor arrays. In the marine environment, it offers new possibilities for measuring seismic and environmental signals. While DAS can be applied to existing fiber optic cables used for communications, a major limitation of such efforts is that the position of the cable is not always known with sufficient accuracy. In particular, for submarine telecommunication cables, the positioning accuracy decreases with increasing depth. This problem affects the accuracy of earthquake locations and source parameters based on DAS signals. This limitation calls for methods to retrieve the cable’s position and orientation. Here, we propose a method for relocating a linear section of cable “or multiple connected segments” using incidental acoustic sources, particularly boats moving in the vicinity of the cable. The method is based on Target Motion Analysis (TMA) for sources in uniform rectilinear motion. We consider Bearing-Only TMA (BO-TMA) and the Bearing and Frequency TMA (BF-TMA), which respectively use changes in back azimuth (called bearing in navigation) and changes in both back azimuth and Doppler frequency shift as the source moves. We adapt these methods to the 3D case to account for the difference in depth between the fiber and the sources. Both cases lead to a non-linear inverse problem, which we solve by the Levenberg-Marquardt method. On synthetic data, we test both TMA techniques on single and multiple source trajectories and evaluate their accuracy as a function of source trajectory and velocity. We then test the BO-TMA on real DAS recordings of acoustic signals produced by passing ships near a 42 km-long fiber optic cable off the coast of Toulon, southeastern France. In this study case, the position and characteristics of the acoustic source are known. While the Doppler frequency shift at low frequency (30 Hz) is difficult to measure with sufficient accuracy (<0.1○), we demonstrate that effective cable location can be achieved by BO-TMA using multiple ship passages with a variety of trajectories. Once the linear sections of the cable have been relocated, the stage is set to reconstruct the entire cable configuration. More generally, the three-dimensional TMA on linear antennas developed here can be used to locate either the sources or the antenna situated at different depths.

SummaryWe present a full-wave inversion algorithm (FWI) to accurately delineate the subsalt body using seismic borehole data. This ill-posed inverse problem is constrained by introducing geological a priori information through the parameterization of the salt boundary using a level set function. The implicit level set function is spanned by a set of B-spline basis functions for their ability to represent a wide range of shapes. Furthermore, the proposed FWI algorithm combines a meshed discretization with the implicit representation of shapes throughout the inversion process. A weak deformation of the mesh is applied at each iteration of the inversion to maintain the explicit discretization of the shapes when the level set boundary is updated. This method is very accurate when it comes to modelling the scattered wavefields and computing the Fréchet derivatives at interfaces. Three numerical examples using synthetic borehole seismic data illustrate the ability of the method to accurately retrieve the size, location and shape of the salt body when the density and seismic velocities are known.

SummaryFractures in reservoirs are potential conduits for fluid flow. Therefore, it is crucial to know to what extent fluid flowing through a fracture could be lost by seepage to its porous background. For this reason, the hydraulic contact between the porous background and the fracture should be characterized, ideally based on seismic reflections. The representation of a fracture as a thin porous layer can provide insight into this seepage from a dynamic poroelasticity perspective. This is possible because the seismic waves reflected from a fracture are partially converted into the slow P-wave, which is the fluid motion relative to the solid-frame, and are sensitive to hydraulic contact being sealed or leak. It is well known that the P-wave reflectivity of fractures exhibits a marked difference between sealed and leaky cases for small angles of incidence (below twenty degrees) because of the variation in conversion scattering to slow P-wave. Drawing from a recent finding that a vertically polarized shear wave (SV-wave) can also generate a robust slow P-wave, we analyze the SV-wave reflectivity at fractures that can be hydraulically connected or disconnected from the surrounding porous medium, with the aim of advancing fracture characterization. We find that the reflectivity of the SV-wave is sensitive to fluid seepage, particularly at larger incident angles (above thirty degrees) where the amplitude is diminished substantially. Therefore, SV-wave reflectivity can also be used to identify leaky fractures, complementing the information provided by P-wave reflectivity.

SummaryEarthquake early warning systems are designed to provide critical seconds of warning before strong ground shaking, facilitating emergency mitigation efforts. Existing methods, such as neural networks and ground motion prediction equation-based approaches, rely on manually defined parameters and physics-based computations, which introduce human bias and hinder the efficiency of real-time applications. Furthermore, current studies primarily focus on scalar metrics such as peak ground acceleration and peak ground velocity to evaluate earthquake impacts. These metrics are limited to measuring ground shaking intensity and fail to capture the spectral characteristics of ground motion. Therefore, a ground-motion and structural-oriented deep learning-based model is proposed to predict uniform hazard spectral acceleration values across 111 periods ranging from 0.01 to 20 seconds. The framework is initially trained and evaluated on 17,500 ground-motion records from the crustal Next Generation Attenuation West 2 project. Spectral acceleration values are predicted by two subsets: deep learning-based uniform hazard spectral acceleration models 1 and 2. These models effectively utilize feature information from the initial seconds of seismic waveforms, eliminating the need for empirically defined parameters. Two deep learning-based models are developed for two datasets representing two distinct broad geographical regions. Both models utilize a similar deep-learning architecture but vary in input settings and hyperparameters to account for regional seismic characteristics. To assess the model's goodness-of-fit between observed and predicted values, as well as its generalization ability, we rigorously compare the two models with the latest data from the U.S. Geological Survey Earthquake Hazard Toolbox and the Japanese Strong-Motion Earthquake Network, respectively. An explainable artificial intelligence technique has been applied to better understand the framework and analyze how individual input features influence the outputs of the trained models. Integrating cutting-edge deep learning technologies into ground motion and engineering seismology reveals the significant potential of the model in enhancing real-time early warning systems. This integration also provides valuable support to various end-users involved in seismic monitoring, facilitating well-informed decisions in both real-time and near-real-time scenarios.