This session invites contributions presenting new approaches to the analysis of large seismic datasets, whether in offline playback or (near) real-time applications, across a wide range of tectonic settings and spatial scales. We particularly welcome methods that integrate rigorous error and uncertainty analysis.
Submissions may focus on classical aspects of seismicity analysis, such as event detection, location, magnitude estimation, and source characterization, as well as novel instrumental or theoretical developments. We encourage contributions spanning diverse applications, including automated observatory workflows, enhanced geothermal systems (EGS), carbon capture and storage (CCS) monitoring, and studies from laboratory to regional scales.
Orals: Wed, 6 May, 08:30–10:15 | Room 0.96/97
The accumulation of decades of continuous seismic observations, combined with the emergence of new sensing technologies (e.g., distributed acoustic sensing-DAS- and nodes) and novel computing infrastructure, presents both outstanding challenges and opportunities for the observational geophysical community to tackle data processing at petabytes scale. Methodologies, open-source software practices, and cyberinfrastructure have advanced to a point where mining petabyte-scale archives can be done within a single day of cloud computation (Ni et al, 2025a,b). This contribution reviews research workflows centered around seismic event monitoring with large-scale seismometers and regional DAS networks. We evaluate strategies for both cloud infrastructure (Ni et al, 2025) and edge DAS units (Shi et al, 2025a,b) for a seismic event monitoring pipeline that leverages deep learning for rapid feature extraction, such as classification of seismic source type and picking of P- and S-wave arrivals. Leveraging and advocating for open-source software, we discuss computational considerations and strategies to improve the performance of pre-trained deep learning models through transfer-learning and model architecture adaptation. We illustrate these findings with the United States NSF-National Geophysics Facility archive operated by the EarthScope Consortium, as well as diverse experiments from the University of Washington FiberLab.
How to cite: Denolle, M., Ni, Y., Shi, Q., Rose, A., and Lipovsky, B.: PetaScale Data-Driven Seismology: Geohazard Discovery via Large Array Data Mining at the Edge, on Premise, and on the Cloud, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-15608, https://doi.org/10.5194/egusphere-egu26-15608, 2026.
Distributed Acoustic Sensing (DAS) enables continuous assessment of transport infrastructure conditions through passive surface wave analysis. However, operational deployment may be problematic as the dispersion curves from thousands of frequency-velocity dispersion images are generated during routine monitoring. Passive DAS recordings from train-induced vibrations exhibit severe fragmentation, strong higher-mode interference, and temporal variability driven by seasonal moisture changes that consistently challenge manual or semi-automated picking methods.
We present a training-free hybrid algorithm that combines marker-controlled watershed segmentation with physics-informed trajectory optimization to extract fundamental mode of dispersion curves from complex operational DAS data. Applied to a 15-month monitoring campaign on a 350 m railway embankment in the UK, the methodology operates directly on dispersion images exported from standard MASW processing software. The algorithm proceeds through five stages: (1) binary masking with morphological noise suppression establishes candidate energy regions; (2) watershed transformation with internal markers separates touching fragments that should constitute independent segments; (3) bidirectional amplitude-maximum propagation with adaptive vertical search radii extracts local trajectory estimates within each isolated fragment; (4) velocity band filtering combined with forward monotonic chaining reassembles disconnected segments by enforcing kinematic consistency and rejecting physically-implausible connections; and (5) global sigmoid fitting with constrained horizontal extension produces smooth, inversion-ready dispersion curves validated against aliasing boundaries.
Validation against manual picking demonstrates that the algorithm bridges spectral gaps exceeding 5 Hz, correctly isolates fundamental from higher modes even when energy amplitudes are comparable, and maintains trajectory continuity through severe fragmentation where conventional peak-following methods fail. Beyond immediate operational utility, automated extraction from real-world DAS railway data enables generating computationally labeled training datasets that preserve physical consistency and interpretability. We demonstrate how this computer vision approach produces high-quality dispersion curve labels across diverse geological settings and complexity levels.
How to cite: Shrivastava, S., Trafford, A., Saqlain, M., and Donohue, S.: Automated Extraction of Rayleigh Wave Dispersion Curves from DAS Railway Monitoring, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-13989, https://doi.org/10.5194/egusphere-egu26-13989, 2026.
Earthquake localization using Distributed Acoustic Sensing (DAS) is challenging due to the single-component directional sensitivity of DAS systems. We propose a novel approach for localization that is based on dense DAS recordings and constraints from known structural heterogeneity of the subsurface. Our method employs full-waveform simulations to generate synthetic DAS wavefield images for a range of potential earthquake source locations. A deep convolutional neural network (CNN), based on a U-Net architecture, is trained on these images to map DAS-recorded wavefield patterns to earthquake source coordinates, without the need for identification and picking of P- and S-wave arrivals. We evaluate this wavefield-based localization technique using a challenging synthetic case study involving DAS recordings in a single vertical borehole, representative of monitoring configurations commonly deployed at geothermal platforms. We consider different velocity models of varying geological complexity. The results show that the CNN effectively learns location-specific wavefield signatures influenced by subsurface heterogeneity. Uncertainties can be reduced significantly by adding recordings from a second borehole. While the results are based on idealized 2D synthetic modeling, the method offers a promising approach for improving microseismic monitoring when detailed information on the heterogeneous velocity structure is available (such as that derived from seismic surveys).
How to cite: Komeazi, A., rümpker, G., and Limberger, F.: Exploring wavefield-based location imaging in heterogeneous media: a borehole DAS example, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-5526, https://doi.org/10.5194/egusphere-egu26-5526, 2026.
A high-resolution earthquake catalog and detailed quantification of earthquake source parameters are essential for constraining fault structure and earthquake interactions. As a candidate site for the next-generation gravitational wave detector (i.e., Einstein Telescope), the Lower-Rhine Embayment region requires a comprehensive assessment of the fault distribution inferred through seismicity and earthquake source properties. In this study, we build an enhanced earthquake catalog using both permanent seismic stations and new data from temporary deployments together with AI-based techniques, including signal enhancement with a decoder-autoencoder denoiser, seismic phase detection using PhaseNet, and event association with PyOcto. We further refine earthquake locations with the NLL-SSST-coherence algorithm and then apply an automatic quality-control filter using event association to remove false detections resulting from the misinterpretation of teleseismic signals as local ones due to event denoising. We detect 3900 events for the period 2019 to 2025, with 2101 of them being classified as earthquakes. The enhanced catalog shows increased hypocentral depths along the NW-trending Sandgewand fault, with a maximum depth of 20 km at the southern end of the fault system.
We also present the first results of a source-parameter catalog for earthquakes that occurred between 2000 and 2025 based on the dataset of Hinzen et al. (2021). Focal mechanisms for selected earthquakes in the region are determined with SKHASH by combining S/P ratios and first-motion polarities obtained from PhaseNet+. We fit individual earthquake spectral parameters, including corner frequency and seismic moment, and calculate stress-drop values for ML≥1.5 events based on a circular crack model. Preliminary results indicate a median stress-drop value of 2.5 MPa across the region, with slightly higher stress-drop values observed on the Sandgewand fault relative to the Rurrand fault. In addition, we use the Distributed Acoustic Sensing (DAS) recordings to compute focal mechanism and corner frequency estimates and compare the results with broadband seismic stations for two earthquakes captured by DAS observations in the Netherlands in 2025. The enhanced spatial sampling density of DAS data provides additional constraints on earthquake source parameters and enables fault movement estimation that is difficult with seismic station data alone.
How to cite: Chen, X., Carrasco, S., Roth, M. P., Bocchini, G. M., and Harrington, R. M.: An enhanced catalog and earthquake source parameter estimations in the Lower-Rhine Embayment region, western Central Europe, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-11625, https://doi.org/10.5194/egusphere-egu26-11625, 2026.
In late 2023, the Reykjanes Peninsula in Iceland experienced an intense seismic and volcanic episode. A major earthquake swarm began on 24 October 2023, driven by magmatic intrusion beneath the region, and its frequency and intensity escalated dramatically on 10 November. This activity culminated in the Sundhnúksgígar crater chain eruption on 18 December. During the November swarm, the high density of tremors caused significant challenges for the automatic earthquake location system, reducing its reliability. To address this issue, we applied the extended Integrated Particle Filter (IPFx) method to continuous seismic data recorded during the eruption period.
The IPFx method, originally developed for Japan’s Earthquake Early Warning (EEW) system, integrates single-station P-wave detection with a network-based particle filter approach to estimate event locations and magnitudes in real time. It processes continuous waveform data from multiple stations, enabling rapid and accurate earthquake detection even during intense seismic sequences. We analyzed three days of continuous data (9–11 November 2023) and compared IPFx-derived locations with the manually reviewed catalog of the Icelandic Meteorological Office (IMO).
Initial application of the IPFx method using its default configuration—Japanese velocity structure and no historical seismicity—resulted in large offshore location uncertainties due to limited azimuthal coverage near the eruption site. To improve accuracy, we incorporated the South Iceland Lowland (SIL) velocity model used in Iceland and regional historical seismicity into the particle filter’s sampling and likelihood functions. These modifications reduced average location errors by approximately 50%. Furthermore, the IPFx method successfully distinguished multiple closely spaced events during periods of high seismicity, demonstrating its potential for generating reliable automatic earthquake catalogs under challenging conditions.
Our findings highlight the adaptability of the IPFx method for real-time seismic monitoring in volcanic regions with sparse station coverage. By improving earthquake location accuracy during swarm activity, this approach can enhance early warning capabilities and contribute to hazard mitigation efforts in Iceland and similar tectonic settings.
How to cite: Yamada, M., Jónsdóttir, K., and Erlendsson, P.: Enhancing Earthquake Detection During the 2023 Reykjanes Swarm Using the Extended IPF Method, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-466, https://doi.org/10.5194/egusphere-egu26-466, 2026.
The transition from manual to deep-learning automated seismic phase picking has revolutionized seismology applications such as seismic catalog building and fault structures analysis. However, the reliability of these AI-driven catalogs is often hindered by a "black-box" approach to model selection and decision thresholds. While deep learning models like PhaseNet (Zhu and Beroza, 2019) offer unprecedented efficiency, their performance is sensitive to the data they were trained on and to the probability thresholds used to define a "phase pick".
In this work, we present a comparative study focused on the Amatrice–Visso–Norcia 2016-2017 seismic sequence in Central Italy. We investigate the influence of the training models and the threshold variation on the phase picking detections.
In particular, we compare the performance of the default PhaseNet model (STEAD) against i) a model trained on the AQ2009 dataset (specific for the Central Apennines, from Bagagli et al., 2023) and ii) a model obtained through Transfer Learning on STEAD fine-tuned with the INSTANCE model (Michelini et al., 2021) via the SeisBench platform (Woollam et al., 2022). We also analyze how varying the confidence threshold (from 0.1 to 0.9) affects the final catalog's completeness and precision.
Preliminary results show that regional training significantly outperforms default models in specific noise conditions and that the optimal threshold is influenced by station geometry and signal-to-noise ratios. By providing a statistical framework for automated threshold calibration, this study offers a roadmap for more objective and reproducible signal detection, applicable not only to seismology but to any domain dealing with continuous time-series classification.
How to cite: Fonzetti, R., Bailo, D., Valoroso, L., De Gori, P., and Chiarabba, C.: The Impact of Probability Thresholds and Model Training on Phase Picking Performance, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-11512, https://doi.org/10.5194/egusphere-egu26-11512, 2026.
Local magnitude (ML) is widely used for reporting the size of small earthquakes, but to achieve consistent physical scaling and cross-regional comparability, moment magnitude (MW) and related source parameters such as corner frequency (fc) are required. Routine MW estimation for small events is often hampered by low signal-to-noise ratios and unstable spectral fitting in conventional frequency-domain approaches. Recent time-domain approaches have therefore estimated MW from peak S-wave displacement amplitudes measured in multiple narrow-band filters, but commonly rely on frequency and distance-dependent empirical attenuation curves that are inherently region specific. We present a generalized time-domain method that estimates moment magnitudes from peak S-wave displacement amplitudes measured on vertical-component seismograms filtered into ten fixed narrow bands with center frequencies spanning 0.1-30 Hz.
The method applies one-way Butterworth bandpass filters with constant bandwidth (0.2 Hz) and three poles, selected to provide stable spectral equivalence across center frequencies. For each band, the peak S-wave displacement is converted to an equivalent displacement spectral amplitude using a constant factor calibrated against Fourier displacement spectra. Source and path effects are corrected using geometrical spreading and intrinsic anelastic attenuation. Multi-band amplitudes are interpreted with a Brune source model (Brune, 1970) using a nested grid search to retrieve the long-period level (Ω0) and corner frequency (fc). MW is then calculated from the resulting seismic moment.
We validated the approach using 12,025 records from 490 earthquakes in and around the southern Korean Peninsula (2017–2022). The time-domain MW agrees closely with reference MW from displacement-spectral fitting with uncertainty assessment (R² = 0.97). To validate the generalization of our method, we applied it to a two-week subset of the 2019 Ridgecrest sequence (4–18 July 2019), comprising 115,309 seismograms from 5,073 earthquakes. The resulting magnitudes also showed strong agreement with Trugman (2020) (R² = 0.91) without a region-specific correction curve.
The proposed method is directly compatible with real-time workflows, and we are integrating it into an Earthworm-based pipeline to output source parameter estimates shortly after S-wave arrival. To support this implementation, we developed modules for real-time IIR filtering and moment-magnitude estimation by adapting and extending Earthworm modules. This provides an efficient and practical route to real-time MW estimation in operational settings.
How to cite: Hong, Y., Doo, M., and Sheen, D.-H.: A generalized time-domain approach for routine moment magnitude estimation from S-wave peak displacement amplitudes, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-8875, https://doi.org/10.5194/egusphere-egu26-8875, 2026.
The seismic moment tensor (MT) delivers valuable information about the physical source process of a seismic event. It allows to distinguish between earthquakes, explosions and volcanic events, constraints the orientation of an earthquake rupture, and represents the most accurate estimate of the released seismic energy.
The computation of absolute MTs by waveform inversion is a data intensive task that is oftentimes feasible only for the largest events in a data set. Relative MTs rely on less subsurface information and may be computed for a large number of closely spaced weaker seismic events that are connected to an absolute MT through relative amplitude measurements. The relative MT method assumes that the Green’s functions between events is similar and that relative amplitudes are measured below the corner frequency of the largest event. Under these assumptions, the relative amplitude between seismograms can be attributed to the difference in moment tensor between events. Compared to absolute methods, path and site effects cancel out and do not need to be considered.
We here present relMT, an accessible, research-grade, open-source software package that facilitates computation of relative moment tensors for a large variety of data sets. The software takes as inputs seismic waveform, event locations, ray take-off angles, and a reference MT, as well as waveform headers and a configuration file. In synopsis, the similar waveforms are aligned to sub-sample accuracy under consideration of possible polarity reversals for P-waves and planar polarization of S-waves. Amplitude ratios between the aligned seismograms are measured on single seismic stations in a principal component framework. The relative amplitudes are combined mathematically with ray take-off angles, relative event distances and one absolute reference moment tensor in a linear system of equations. The solution of the equation system with algebraic methods yields all relative moment tensors at once. The uncertainty of the solutions is quantified using the bootstrap method. The software is under active development on GitHub (https://github.com/wasjabloch/relmt).
We illustrate the application of relMT using data sets of induced seismicity and tectonic aftershock seismicity. For the induced seismicity of the enhanced geothermal system in Helsinki, Finland, we are able to lower the magnitude threshold for which MTs can be computed from 0.5 to -0.5. For aftershocks in the Pamir highlands of Central Asia from 4.0 to 2.0. For the data sets, this represents a 3- to 30-fold increase in the number of recovered MTs.
How to cite: Bloch, W., Oye, V., Drolet, D., Plourde, A., and Bostock, M.: relMT – Software to Determine Relative Seismic Moment Tensors Illustrated with Tectonic and Induced Seismicity, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-7124, https://doi.org/10.5194/egusphere-egu26-7124, 2026.
Differential travel-time observations from waveform cross-correlation are among the most precise measurements in observational seismology and underpin high-resolution relative relocation. In large modern catalogs, however, HypoDD-style pair files (dt.cc) often contain outliers, cycle skipping, and internally inconsistent links that can bias downstream workflows, particularly cluster-based relocation pipelines. We present DDSync, a lightweight preprocessing method that treats the differential-time measurements for each station–phase as a weighted event graph and solves a graph synchronization problem to recover a self-consistent set of relative arrival-time proxies (one scalar per event and station–phase). The baseline estimator is a sparse weighted least-squares solution of a graph Laplacian system, which implicitly averages over redundant constraints in dense graphs and yields stable long-baseline differentials without enumerating paths.
DDSync adds two robustness layers. First, it computes an edgewise inconsistency diagnostic from the global fit (a loop-closure-style residual) and prunes grossly inconsistent links using a MAD-based threshold, with automatic re-identification of connected components. Second, it refines the solution using iteratively reweighted least squares, with a Huber loss to downweight remaining heavy tails while preserving connectivity. Beyond producing cleaned and synchronized dt.cc files, DDSync estimates uncertainty in the synchronized results by approximating per-event variance of the inferred potentials using a stochastic diagonal estimator of the inverse reduced Laplacian, and propagating these to conservative pairwise σ estimates and weights for downstream inversions.
We evaluate DDSync on the Ridgecrest synthetic benchmark of Yu et al. (2025), where differential times are perturbed with Laplacian-distributed errors and outliers added, and show near complete removal of gross outliers and strong tightening of residual distributions relative to ground truth, reducing error by about a factor of 5. The inferred uncertainty on the denoised observations capture station–phase and event-specific constraint quality and provide a practical, uncertainty-aware weighting scheme for relocation and related inverse problems. We also highlight a subglacial volcanic example with distinct event family types (icequake and VT events) where pruning preferentially removes inconsistent links that bridge otherwise separated event families, improving interpretability and robustness for analysis of dense catalogs.
How to cite: Heimisson, E. R., Winder, T., and Yu, Y.: DDSync: Denoising and outlier removal in differential travel-time observations using graph synchronization, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-8169, https://doi.org/10.5194/egusphere-egu26-8169, 2026.
Posters on site: Thu, 7 May, 10:45–12:30 | Hall X1
Accurate knowledge of existing fiber-optic cable geometry is essential for applications of distributed acoustic sensing (DAS), however the true positions of buried or installed fibers are often uncertain due to slack, bends, or deviations from documented routes. We present two passive, seismology-based approaches for cable localization that exploit information contained in DAS recordings. Case A employs ambient noise cross-correlations with reference points to estimate relative travel times, whereas Case B uses the differential arrivals of plane waves from distant earthquakes with linearly independent slowness vectors. Both approaches can be formulated in a least-squares framework that allows for the joint estimation of propagation velocity and geometry, thereby reducing biases from noise and model assumptions. Synthetic experiments show that cable positions can be recovered with an accuracy better than 100 m, even when apparent velocities are uncertain or the medium exhibits heterogeneity. The two methods provide independent geometric constraints that complement other sources of information on cable routing.
How to cite: Rümpker, G., Limberger, F., and Komeazi, A.: Passive seismological approaches for localizing near-surface fiber-optic cables with DAS, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-5599, https://doi.org/10.5194/egusphere-egu26-5599, 2026.
Moment tensor (MT) determination is a key component of real-time seismology, with applications in moment magnitude estimation, tsunami early warning, volcano monitoring and shake map generation. Despite its importance, the reliable inversion of MT components, especially the non-double-couple ones, presents significant challenges. Indeed, the non–DC components are highly sensitive and often exhibit large fluctuations, making reliable estimation of the full moment tensor difficult. These limitations highlight the need for robust uncertainty quantification of MTs. To efficiently address this issue, we propose a Bayesian bootstrapping approach. The approach assumes that Signal to Noise Ratio (SNR) is fair and the velocity model is not systematically biased. The method relies on a series of weighted inversions, in which station contributions are stochastically varied using Bayesian weights. This procedure produces an ensemble of plausible MT solutions enabling statistical characterization of the inversion results (e.g., median MT, confidence intervals of ISO and CLVD components, etc.). This approach, free of the assumption of Gaussianity of data error, provides meaningful uncertainty estimates and improves the interpretability of non–double-couple components. The proposed methodology has been integrated into GISOLA, an open-source, highly efficient near–real-time MT inversion software, currently in routine operation in several seismic networks. The resulting automated operational framework handles multiple data streams in heterogeneous formats, interfaces with diverse processing modules, applies a systematic preprocessing workflow to identify the most reliable stations and corresponding signals, performs parallelized inversions, and provides robust uncertainty quantification. This enhances the reliability of source characterization in operational environments and supports more informed use of MT results in time-critical seismic monitoring.
This work is supported by TRANSFORM² which is funded by the European Union under project number 101188365 within the HORIZON-INFRA-2024-DEV-01-01 call
How to cite: Velentzas, V., Zahradník, J., Psarakis, E., Evangelidis, C., and Sokos, E.: Bayesian bootstrapping extension of GISOLA automatic moment tensor inversion software, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-21600, https://doi.org/10.5194/egusphere-egu26-21600, 2026.
Recent advances in seismology, driven by the deployment of dense seismic networks and the development of machine-learning-based earthquake detection, have enabled the generation of high-quality seismic catalogs with unprecedented spatial and temporal resolution. These dense microseismic datasets provide a robust foundation for detailed waveform-based analyses, that allow individual earthquakes to be reliably linked to fault segments and enable constraints on fault geometry and slip style at fine spatial scales.
We present an integrated seismological workflow that starts from automated earthquake detection using machine-learning techniques (PhaseNet) applied to continuous seismic recordings in the Val d’Agri region (Southern Italy). The resulting high-resolution microseismic catalog is then analyzed through waveform similarity-based clustering to identify events associated with the same seismogenic structures, followed by high-precision relative relocation to delineate fault segments, and Bayesian moment tensor inversion to robustly characterize faulting style.
This waveform-based workflow enables the association of earthquakes with individual seismogenic structures, allowing to resolve fault geometries and slip styles at fine spatial scales. Results indicate that seismicity predominantly clusters on steeply southwest-dipping normal faults, with focal mechanisms consistent with the regional extensional stress regime.
These analyses illustrate how machine-learning-driven seismic monitoring combined with waveform-based analysis can bridge the gap between large microseismic datasets and fault-scale imaging. Beyond increasing detection rates, this workflow provides new insights into the geometry and kinematics of active faults in regions affected by diffuse seismicity.
How to cite: Caredda, E., Cesca, S., and Morelli, A.: From Machine-Learning detection to fault imaging: high-resolution seismology in the Val d'Agri (Southern Italy), EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-7959, https://doi.org/10.5194/egusphere-egu26-7959, 2026.
Deep learning methods have gained significant attention in seismic denoising due to their superior ability to extract weak signals from prestack data, which are often smoothed out by traditional techniques. While conventional convolutional neural networks can be trained in an end-to-end manner, they often fail to capture the underlying data distribution. Generative models are capable of reconstructing more realistic seismic signals by learning the distribution. The primary generative models include Variational Auto-Encoders (VAEs), Generative Adversarial Networks (GANs), and the more recently proposed Denoising Diffusion Probabilistic Models (DDPMs). VAEs offer stable training but tend to yield results of limited quality. GANs can produce high-quality outputs via an adversarial discriminator but suffer from unstable training. DDPM could provide a favorable balance between output quality and training stability. However, the supervised training paradigm relies on high-quality labeled data, which is often scarce in geophysical applications. This limitation frequently leads to constrained generalization ability. Consequently, there is significant signal leakage for strong reflection events when applied to unseen data from different work areas.
To address this, we propose a novel discriminator-constrained diffusion model. Our key innovation is the integration of a discriminator into the DDPM framework. This adversarial component provides a powerful constraint during the training process. The hybrid training objective combines the standard diffusion loss with the adversarial loss, guiding the model to preserve critical reflections while removing noise.
We validate our method through comprehensive experiments. On synthetic data containing various noise intensities. Our method has an improvement of 0.7 dB for noisy data with 1% Gaussian noise compared to standard DDPM. More importantly, in cross-field tests, the proposed method has an improvement of 2 dB. Visualizations of denoised sections and difference profiles confirm that our approach better preserves reflections.
In conclusion, the incorporation of adversarial training into the diffusion process offers a robust solution to the generalization challenge in deep learning-based seismic denoising. Our work demonstrates a promising pathway for applying advanced generative models to practical geophysical data with limited labels.
How to cite: Zhang, Y.: Discriminator-Augmented Denoising Diffusion Probabilistic Models for Seismic Data, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-10528, https://doi.org/10.5194/egusphere-egu26-10528, 2026.
We present new local magnitude models for Switzerland, derived using a top-quality input dataset of ground-motion intensity measures, obtained through an automated and consistent waveform processing workflow (https://doi.org/10.1785/0120250032). The input dataset includes Wood-Anderson displacement response amplitudes calculated from 150,000 waveforms generated by about 15,000 earthquakes in the region of interest. This dataset is substantially larger than those used in previous magnitude studies in Switzerland and contains a large number of recordings at short distances for smaller events (about 5,000 earthquakes with magnitudes lower than 1, and about 1500 records at hypocentral distances shorter than 5 km), which were sparse in earlier studies. The parametrization of the local magnitude model is based on a detailed investigation of attenuation characteristics of the Wood-Anderson response amplitudes, comprising: (i) linear and logarithmic distance terms to represent different physical mechanisms of seismic wave propagation; (ii) hinge distances to allow modelling the effects of Moho reflections and the associated changes in attenuation rate; (iii) regional adjustments (Swiss Alps vs Swiss northern Foreland) based on the length and location of the surface projection of the source-to-site ray paths. Model coefficients are determined through mixed-effects regressions, thus allowing the derivation of station-magnitude correction terms consistent with the reference rock-like ground type used for mapping seismic hazard in Switzerland. We assess and validate the model’s performance via uncertainty analyses, validation against recent data not used in model calibration, validation against recordings of major Swiss events, and comparisons with results obtained from alternative, fully independent modelling approaches (including non-parametric and 2D cell-based methods). The new model yields systematically lower magnitudes, with an average difference of 0.1-0.2 magnitude units for smaller events with magnitudes below 2.5 and for earthquakes located in the Swiss northern Foreland, compared to the currently authoritative catalogue magnitudes. Based on the new candidate magnitude model, we present and discuss an updated empirical scaling relationship between local and moment magnitudes.
How to cite: Altindal, A., Cauzzi, C., Bindi, D., Diehl, T., Deichmann, N., Clinton, J., and Wiemer, S.: Towards New Magnitude Models for Switzerland, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-10958, https://doi.org/10.5194/egusphere-egu26-10958, 2026.
We present an updated and extended seismotectonic earthquake catalog of Switzerland and surrounding regions. This SECOS25 catalog serves as input for the seismic-source component of Switzerland’s next generation seismic hazard model to be developed by the Swiss Seismological Service (SED) over the coming years. The SECOS25 catalog includes 51 years of instrumental seismicity detected and located by the SED between 1975 and 2025. For the digital era of the SED catalog (phase picks and seismograms available in digital form) starting in 1984, hypocenters were consistently relocated in absolute terms using a recent 3D P and S-wave crustal velocity model. Starting from these improved hypocenters, double-difference relative relocations were performed at different scales (local clusters as well as regional scales), combining differential times from manual picks and waveform cross correlations. A merging procedure was developed that selects the preferred location method (bulletin location, absolute relocation, relative relocation) based on location-quality criteria for each hypocenter. The proposed procedure provides various hypocenter uncertainty measures and ensures the maximum possible hypocenter-location accuracy and precision for each event. Local magnitudes were revised using a new set of consistent amplitude measurements (doi.org/10.1785/0120250032) and a new magnitude model. For each earthquake, we provide moment magnitudes from native methods if available (either from moment tensors or spectral fitting) or revised scaling relationships. Finally, we link the hypocenters with catalogs of moment tensors (containing about 80 solutions) as well as first-motion focal mechanisms (containing about 600 solutions).
The SECOS25 catalog serves as a base for further down-stream seismotectonic components of the seismic-source model. This includes heat maps of seismic activity and moment release across Switzerland as well as maps of deformation regimes and stress orientations derived from the analysis and inversion of focal mechanisms. Finally, we apply an enhanced version of the HyFi method (doi.org/10.1029/2023JB026352) to the SECOS25 catalog to systematically identify previously unknown seismically active fault segments and their orientations. The derived information will contribute to a refined definition of seismic zonation as well as faulting styles and preferred rupture orientations required for hazard computations. The SECOS25 catalog and derived products therefore also contribute to an improved understanding of present-day seismotectonic processes in the Central Alpine region.
How to cite: Diehl, T., Heilig, J., Altindal, A., Truttmann, S., Cauzzi, C., Deichmann, N., Clinton, J., Herwegh, M., and Wiemer, S.: Towards an Improved Seismic Source Catalog for Switzerland, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-14799, https://doi.org/10.5194/egusphere-egu26-14799, 2026.
Probabilistic seismic hazard assessment (PSHA) is challenging in cratonic regions such as Sweden, where the characteristics of strong motion (detrimental for structures) are uncertain due to a lack of data. The current approach is therefore to use available data sets of small-to-moderate earthquakes to identify seismogenic areas and to adapt models from more seismically active regions. One issue encountered in this process is estimating the moment magnitude (Mw) of these earthquakes. In fact, evaluation of local magnitude (Ml) is preferred for magnitude <4 earthquakes due to the difficulty of deriving Mw using standard methods.
In this work, we propose to estimate the Mw of earthquakes in Sweden using a generalised inversion technique (GIT), and to investigate the uncertainty of this measurement in this seismic context. To do so, a ground-motion data set of Ml=0-4.1 earthquakes recorded by the Swedish National Seismological Network (SNSN) is compiled, and the non-parametric generalised inversion of Oth et al. (2011) is applied. This approach identifies the source spectrum of earthquakes, the apparent attenuation of the region and the site amplification of stations by performing a spectral decomposition of the Fourier amplitude spectrum (FAS). The parameters of the Brune (1970) source model and the frequency-dependent elastic and anelastic attenuation models are derived from these terms in post-inversion.
We present here the resulting source parameters (Mw and corner frequency fc) that are analysed and compared with the moment based Ml computed by the SNSN. The consistency of these estimates is then evaluated using higher-sample-rate recordings from a temporary network that was deployed for four years in Sweden’s most seismic active region. The 100 Hz sampling rate of the permanent stations limits the estimation of Mw and fc for earthquakes of magnitude <2, as their fc is expected to exceed the Nyquist frequency (~45 Hz). The 200 Hz sample rate set for the temporary network enables the source spectrum to be derived up to 90 Hz and therefore allows the veracity of the Brune (1970) model (which has been derived from a shorter frequency band) to be analysed.
Applying this inversion algorithm also provides new insights into anelastic attenuation in cratonic regions and into site amplification observed at hard rock/bedrock sites (which constitute the majority of SNSN installations). Knowing the apparent attenuation and site conditions is useful for future studies, especially for correcting the FAS at stations to measure Mw using other approaches or models, and for achieving a quasi real time estimate.
How to cite: Buscetti, M. and Lund, B.: Estimation of the moment magnitude and its uncertainty for small-to-moderate earthquakes in Sweden using a generalised inversion approach., EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-17337, https://doi.org/10.5194/egusphere-egu26-17337, 2026.
Estimating the stress drop (Δσ) occurring during an earthquake allows us to characterize the mechanical state of the area surrounding its source. The magnitude (Mw) and the source radius (r) of an earthquake are usually assumed to scale such that Δσ is constant. We typically refer to this assumption as self-similarity of earthquakes.
However, estimates of Δσ can vary by at least four orders of magnitude (e.g., Cocco et al., 2016, DOI: 10.1007/s10950-016-9594-4). It is an open question to what extent this variability is caused by methodological and data uncertainties, or it is a manifestation of different physical processes (Abercrombie et al., 2025, DOI: 10.1785/0120240158). This is especially true for smaller events, due to a strong correlation between source and propagation terms in the waveform modeling.
In this study, we obtain precise relative source parameters (seismic moment, corner frequency and stress drop) estimates analyzing the spectral ratios of co-located events with similar source mechanisms. This allows us to get rid of the propagation term in the waveform modeling, and to focus on the effects of the assumed source model on the stress drop.
We analyze pairs of events recorded during 2017 in North Ibaraki region (Japan) by a high-resolution temporary seismic network operated by AIST and by Hi-net stations, and during Pawnee and Prague earthquakes in Oklahoma (2011 and 2016, respectively). Overall, we explore a range of magnitudes from M = 0.7 to M = 5.8.
We find that source model and self-similarity are not always compatible and that in general a strong correlation exists between stress drop estimates and the source parameter γ that describes the decay of source spectrum at high frequencies (γ = 2 in the model). This emphasizes the importance of considering γ as a free parameter when modeling earthquake source spectra.
Moreover, our approach allows us to investigate potential differences among catalog and moment magnitudes through the inferred relative seismic moment estimates. For the small events in Ibaraki, we find differences larger than 0.5 units between local magnitude (Ml) and moment magnitude, supporting previous evidence of non-linear relationships between Ml and Mw (e.g., Uchide and Imanishi, 2018, DOI:10.1002/2017JB014697). This highlights the importance of expanding Mw catalogs to smaller magnitudes, as statistical analysis of Ml catalogs may be affected by systematic biases.
This study was supported by TRANSFORM², funded by the European Commission under project number 101188365 within the HORIZON-INFRA-2024-DEV-01-01 call.
How to cite: Maille-Okada, N., Supino, M., Satriano, C., Uchide, T., and Marzocchi, W.: Relative stress drop and source parameters estimation from spectral ratio method, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-18401, https://doi.org/10.5194/egusphere-egu26-18401, 2026.
Seismological studies are traditionally based on the observation of ground motions recorded by translation sensors. However, to assess a comprehensive description of any wavefield produced by seismic sources, the three components of rotation are as important as the three translation components. Due to improvements in instrumentation through the last two decades, the ground rotational motion is now an observable. Several recent publications show that 6 degrees of freedom (-dof) seismic station, recording the 3 translation and 3 rotation components of the ground motion, provides valuable information to locate events or to infer their source mechanism.
Rotation motions can be recorded directly via dedicated sensors, or numerically derived via a dense array of seismometers. The formers primarily use technologies based on fiber optic gyroscopes, liquid-based systems, or mechanical principles. However, these broadband instruments do not have the high sensitivity required to detect "weak" ground movements. Conventional sensor arrays, on the other hand, use finite differences techniques to estimate reliable indirect rotation measurements, but these are limited at high frequencies (wavelength > 4 × array aperture).
Since the end of 2025, a temporary experimental campaign is set at the Low Noise Underground Laboratory (LSBB) in Rustrel, France. Two rotational rate sensors, namely a BlueSeis 3A and a Eentec R3, are deployed in the underground galleries, jointly with five seismometers which complete the permanent seismic array. The dense seismic array co-located around the rotation sensors is used to compute array derived rotations (ADR), and validate the direct observations of rotational ground motions. Furthermore, six seismometers are deployed at the surface to form an array with an aperture of around fifty kilometers. It has been designed for long-term observations of regional and global seismicity and array processing analysis, such as beamforming techniques. This campaign allows to compare the performance of classical seismic array processing with innovative gradiometric approaches based on a single 6-dof station, focusing on detection, location, and characterization of seismic events within a common frequency band. A sensitivity study on rotation signals, in terms of instrumental conditions (array geometry, station quality) and processing parameters (signal duration, filtering), is carried out with the help of numerical full waveform modelling in order to quantify the uncertainty of the estimated source parameters. We present the benefit of such multi-component seismic wavefield recording, illustrated on several events of interest such as the recent (2025/07/29) Kamchatka event (Mw 8.8).
How to cite: Morin, M., Sèbe, O., Beucler, É., Capdeville, Y., Rouille, G., Boyer, D., Decitre, J.-B., Bremaud, V., Lallemand, C., Lepoint, F., and Govindin, G.: From Seismic Arrays to Single-Station Wavefield Gradiometry: Results from a Multi-Scale Experimental Campaign, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-19637, https://doi.org/10.5194/egusphere-egu26-19637, 2026.
The open-source, seismological software SeisComP is widely used for earthquake monitoring world-wide. Its default automatic phase association and location module is scautoloc, which was originally developed to monitor large earthquakes with global or large-regional networks. It detects and locates earthquakes iteratively, beginning with P picks from only the nearest stations and improving the earthquake location as additional picks arrive. This process may take more than 15 minutes and the availability of early intermediate results is therefore essential for time-critical applications like tsunami early warning.
Requirements for local earthquake monitoring are quite different. The number of stations is usually smaller and the seismic wave travel times are typically below one minute. Small networks greatly benefit from the use of S picks and custom velocity models to optimize their locations, neither of which are currently supported by scautoloc.
In an effort to improve local and regional monitoring capabilities in SeisComP using only open-source software, we adapted the phase associator PyOcto [1, 2] to the SeisComP framework. PyOcto was specifically designed for fast processing of large amounts of regional network picks. This is particularly important because of the vast improvements in the amount of picks obtained using machine learning techniques [3, 4]. Its efficient implementation and low computational overhead also make PyOcto a perfect phase associator for real-time earthquake monitoring.
The new SeisComP module scoctoloc [5] leverages the use of PyOcto to improve processing of local and regional network data in SeisComP. The ability to process both P and S picks and the convenient support for either homogeneous (0D) or custom layered (1D) velocity models overcome the main limitations of scautoloc. Small networks may choose to run scoctoloc as a drop-in replacement of scautoloc, though both modules may also be run in parallel.
The factors limiting the real-time processing speed are the network dimension (and hence seismic travel times) and pick latency. The latter depends on the latency of the data telemetry and on the processing delay imposed by the picking technique used. In our real-time test setup, a virtual seismic network in Northern Chile, earthquake locations are usually produced within three minutes after an event using P and S picks produced by the scdlpicker module [4]. Where processing speed is more crucial than optimum pick accuracy, the standard SeisComP scautopick with S picking enabled is still the picker of choice.
The use case that PyOcto was developed for originally, the bulk processing of huge amounts of picks read from a database, is supported by scoctoloc as well. In addition it is possible to run "pick playbacks" from a SeisComP database in order to simulate real-time operation. This mode is useful in order to fine-tune the configuration parameters relevant for real-time monitoring based on past events.
[1] Münchmeyer, J. (2024). PyOcto: A high-throughput seismic phase associator. Seismica. doi:10.26443/seismica.v3i1.1130.
[2] https://github.com/yetinam/pyocto
[3] Münchmeyer et al. (2022), https://doi.org/10.1029/2021JB023499
[4] https://github.com/SeisComP/scdlpicker
[5] https://github.com/jsaul/scoctoloc
How to cite: Saul, J., Münchmeyer, J., and Tilmann, F.: scoctoloc - A regional phase associator and locator module for SeisComP, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-14439, https://doi.org/10.5194/egusphere-egu26-14439, 2026.
Automatic monitoring of local seismicity produces events of varying quality. Some events will be poorly located, and some event solutions will not represent a real seismic event, arising only due to noise. Noise events are removed and location quality improved during manual revision, but that is not always feasible for large catalogs. In cases with >10000 events manual review is not tenable, so an automatic quality score calculation is beneficial in improving catalog quality.
Machine learning methods are a useful tool for this purpose, both for classification and calculating a quality score. We explore machine learning methods for solutions to this problem, with a special focus on feature extraction from travel-time information.
The models are evaluated on data from three seismic networks in Iceland. The dataset contains both automatic and manual solutions for a large number of earthquakes, so direct comparisons between manual and automatic solutions can be made. The manual locations can then be used as a ground truth solution that the automatic solutions attempt to approximate. The models are ranked on their classification score as well as their ability to estimate the spatial distance from the ground truth solution.
How to cite: Magnússon, R. L.: Event classification and quality assessment for local seismic events using machine learning, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-19111, https://doi.org/10.5194/egusphere-egu26-19111, 2026.
In regions with limited tectonic information, dense deployments of local seismic networks significantly improve the detection of low-magnitude events. The consequent enrichment of seismic catalogs is proved to have a crucial role for understanding seismotectonic processes and assessing seismic hazards. This study evaluates the performance of machine learning algorithms (MLA) for P- and S-wave picking and event association, using data recorded between April 2013 and June 2025 by the OTRIONS seismic network (FDSN code OT), operating in the Gargano Promontory (GP, Southern Italy) since 2013.
The MLA workflow consists of PhaseNet, a deep learning-based phase picker, in combination with GaMMA, an association algorithm, were employed and approximately 27,000 seismic events were detected. NonLinLoc was employed for event locations. The visual inspection confirmed that about 51% of the events were local earthquakes, while the remainder events were classified as quarry blasts, false events, or events located outside the network. The visual revision procedure was essential at this step.
Compared to the previous manual catalog (based on the STA/LTA detection algorithm) in the same area, the MLA workflow brougth to a new enriched automatic catalog. The quality assessment of the new catalog indicates that the automatic picking is reliable and confirms the OT network’s ability to detect a high rate of low-magnitude seismicity. The NonLinLoc-SSST-Coherence algorithm was also applied to better identify the structures on which seismicity is accomodated and the results suggest that NonLinLoc-SSST-Coherence has better permormances when applied to small seismic sequences than to the widespread seismicity of GP.
From a seismotectonic perspective, the already known seismogenic layer deepening northeasternward characterizing the GP seismicity here appears for the first time to be splitted in two structures located at different depth. This study highlights the crucial role of dense local networks and MLA tools in managing and analyzing large volumes of low-energy seismic data.
How to cite: Ferreri, A. P., Panebiano, S., Satriano, C., Filippucci, M., Cecere, G., Serlenga, V., Stabile, T. A., Selvaggi, G., and Tallarico, A.: A machine-learning workflow for event detection and relocation in the Gargano Promontory (Southern Italy), EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-21190, https://doi.org/10.5194/egusphere-egu26-21190, 2026.
Posters virtual: Tue, 5 May, 14:00–18:00 | vPoster spot 1b
EGU26-22669 | ECS | Posters virtual | VPS24
An integrated geodynamic analysis of seismic sources in the Eastern Rif: Insights from geological, seismological, gravimetric, and aeromagnetic dataTue, 05 May, 14:03–14:06 (CEST) vPoster spot 1b
The Rif’s belt is characterized by low to moderate seismic activity resulting from the continental collision between the African and Eurasian plates. This seismic activity, which involves devastation and human losses, requires an in-depth study of its origins and mechanisms. This study aims to identify the geological structures responsible for seismic activity in the eastern Rif by adopting an integrated methodological approach. The methodology relies on the use of a Geographic Information System (GIS) to process and analyze multiple geological, seismological, and geophysical datasets. Various filters were applied to magnetic and gravimetric data (vertical derivatives) to characterize the subsurface. The analysis of earthquake focal mechanisms helped identify active faults. The results show that the seismicity, with a NW-SE orientation, is localized within a fragile depression south of the city of Selouane. The final geological model highlights a system of faults and strike-slips oriented NE-SW and NW-SE. A significant spatial correlation is observed between epicenters and Messinian-aged NW-SE strike-slips, suggesting their reactivation. The analysis indicates that a system of dextral strike-slips is likely the source of this seismic activity. The proposed geodynamic model represents a major advancement in understanding local seismic activities and serves as an essential reference for future studies. These results significantly contribute to the assessment and management of seismic risks, thereby enhancing the safety and resilience of populations in this high-risk area.
KEYWORDS: Geodynamic model; Seismotectonic; Focal mechanism; Magnetic; Gravimetric; ·
Eastern Rif.
How to cite: Iken, H., Nouayti, A., Nouayti, N., and Khattach, D.: An integrated geodynamic analysis of seismic sources in the Eastern Rif: Insights from geological, seismological, gravimetric, and aeromagnetic data, EGU General Assembly 2026, Vienna, Austria, 3–8 May 2026, EGU26-22669, https://doi.org/10.5194/egusphere-egu26-22669, 2026.
