Journal of Seismology最新文献

{"title":"Author Correction: Stochastic ground-motion simulation using the EXtended SIMulation approach and synthetic aperture radar based damage assessment of the 2025 Myanmar earthquake","authors":"Monika, Ashwani Raju, Vikas Singh, Parveen Kumar, S. K. Pal, Saraah Imran,  Sandeep","doi":"10.1007/s10950-026-10422-3","DOIUrl":"10.1007/s10950-026-10422-3","url":null,"abstract":"","PeriodicalId":16994,"journal":{"name":"Journal of Seismology","volume":"30 5","pages":""},"PeriodicalIF":1.3,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148926333","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Explainable and statistically validated machine learning for earthquake alert-level classification using seismic intensity features","authors":"Yıldırım Özüpak","doi":"10.1007/s10950-026-10444-x","DOIUrl":"10.1007/s10950-026-10444-x","url":null,"abstract":"<div><p>Earthquake alert-level classification can support rapid decision-making by transforming available seismic descriptors into actionable warning categories. However, many machine learning studies in earthquake-related applications report high predictive performance on limited datasets without sufficient statistical validation, which may lead to over-optimistic conclusions. This study presents an explainable and statistically validated machine learning framework for multi-class earthquake alert-level classification using a seismic dataset consisting of 1,300 records and five input variables: magnitude, focal depth, Community Decimal Intensity (CDI), Modified Mercalli Intensity (MMI), and significance (SIG). An Optuna-tuned XGBoost model was developed and compared with conventional machine learning classifiers, including Random Forest, standard XGBoost, K-Nearest Neighbors, Logistic Regression, and Support Vector Machine. To improve the reliability of the evaluation, the experimental protocol was strengthened using stratified repeated cross-validation, macro-averaged performance metrics, bootstrap-based confidence intervals, ablation analysis, and feature-set sensitivity experiments. The optimized XGBoost model achieved strong classification performance; however, the results also indicate that intensity-related variables, particularly MMI, CDI, and SIG, contribute substantially to the predictive outcome. Therefore, the proposed framework should be interpreted as an alert-level classification and decision-support model based on available seismic descriptors rather than as a prospective earthquake prediction system. SHAP-based explainability analysis confirmed that the model decisions were mainly driven by physically meaningful intensity-related features. Overall, the study provides a transparent benchmark framework for earthquake alert classification while explicitly addressing the limitations associated with small datasets, synthetic oversampling, and possible target-related feature dependence.</p></div>","PeriodicalId":16994,"journal":{"name":"Journal of Seismology","volume":"30 5","pages":""},"PeriodicalIF":1.3,"publicationDate":"2026-08-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148928056","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Transient aseismic slip and two-stage evolution of the 2023-2025 Lesotho-South Africa earthquake swarm","authors":"Lindokuhle Hlatshwayo,&nbsp;Tamara Makhateng,&nbsp;Taufeeq Dhansay,&nbsp;Ian Saunders,&nbsp;Vunganai Midzi","doi":"10.1007/s10950-026-10441-0","DOIUrl":"10.1007/s10950-026-10441-0","url":null,"abstract":"<div><p>The earthquake swarm along the Lesotho-South Africa border began in late December 2023 and persisted into 2025, with activity declining during the latter part of the sequence. Earthquake swarms are of growing interest as they do not exhibit the typical mainshock-aftershock decay described by the Omori-Utsu law; instead, they reflect complex and often externally driven processes. In intraplate regions such as southern Africa, where tectonic earthquakes are relatively infrequent, such sequences provide a rare opportunity to investigate subsurface processes within a stable continental setting. This study aims to constrain the dominant driving mechanisms of the swarm using migration analysis, velocity-duration scaling, and non-stationary Epidemic-Type Aftershock Sequence (ETAS) modelling. The swarm evolves from a radial cluster into a linear NE-SW trend, propagating at 1.43 ± 0.13 km/day with diffusivity values &gt; 30 m²/s, suggesting that fluid diffusion alone is unlikely to explain the observed migration. The velocity-duration relationship places the sequence near the global Slow Slip-Driven (SSD) trend, indicating a transient aseismic slip phase early in the sequence. Non-stationary ETAS modelling was applied to separate background seismicity from earthquake triggering. Results show elevated background rates early in the sequence, followed by increased aftershock productivity, suggesting a transition from externally driven to internally triggered seismicity. Overall, the swarm was likely initiated by fluid circulation at depth that triggered a transient aseismic slip phase and was later sustained by earthquake interactions.</p></div>","PeriodicalId":16994,"journal":{"name":"Journal of Seismology","volume":"30 5","pages":""},"PeriodicalIF":1.3,"publicationDate":"2026-08-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148838324","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Far-field ground motion characteristics of the 2025 Mw 7.7 Myanmar earthquake and implications for tall buildings","authors":"Liu Yang,&nbsp;Qiang Ma,&nbsp;Quancai Xie,&nbsp;Dongwang Tao,&nbsp;Jiang Wang,&nbsp;Liang Qian","doi":"10.1007/s10950-026-10440-1","DOIUrl":"10.1007/s10950-026-10440-1","url":null,"abstract":"<div><p>The impact of far-field long-period ground motions (LPGMs) on tall buildings has become increasingly significant. The 28 March 2025 Mw 7.7 Myanmar earthquake was recorded at 675 CEEWN stations in Yunnan, China, at epicentral distances of about 271–800 km. One representative record is selected from five available CESMD records for comparison. To further investigate the characteristics of far-field LPGMs and the effects on tall buildings, differences in amplitude ratios, durations, and spectral characteristics are examined. The Empirical Mode Decomposition (EMD) method is employed to extract and analyze the various Intrinsic Mode Functions (IMFs) contained in the seismic records. Furthermore, an analysis framework combining EMD-based ground-motion decomposition and three-dimensional structural response analysis is adopted to evaluate the influence of individual IMF components on tall buildings. The study also evaluates whether the current Chinese Seismic Intensity Scale adequately represents tall buildings shaking. The results indicate that far-field ground motions recorded across Yunnan at epicentral distances of approximately 271–800 km are characterized by relatively low peak amplitudes, prolonged durations, and pronounced low-frequency content. These characteristics may pose potential threats to long-period buildings. Removing the dominant component whose period closely matches the natural vibration period of the structure can significantly reduce the response of tall buildings by up to 28.76%. A representative far-field record at an epicentral distance of 271 km exceeds the seismic design spectrum for intensity 8, indicating increased seismic risk on tall buildings under far-field large earthquakes. Tall buildings in western Yunnan, particularly in Ruili City, experienced strong shaking, reaching LPGM Level III and Violent shaking. However, the current Chinese Seismic Intensity Scale underestimates the actual shaking intensity experienced by tall buildings.</p></div>","PeriodicalId":16994,"journal":{"name":"Journal of Seismology","volume":"30 5","pages":""},"PeriodicalIF":1.3,"publicationDate":"2026-08-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148838060","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Early earthquake magnitude estimation using reference based spectral distance and MFCC-derived P-wave features: an interpretable machine learning framework","authors":"Rizwan Rehman,&nbsp;Dhiraj Baruah,&nbsp;Pranjal Kumar Bora","doi":"10.1007/s10950-026-10443-y","DOIUrl":"10.1007/s10950-026-10443-y","url":null,"abstract":"<div><p>Rapid and reliable prediction of earthquake magnitude is essential for seismic hazard analysis and timely seismic event characterization. Although machine learning has demonstrated considerable potential for modeling seismic waveforms, the development of informative and interpretable feature representations for early magnitude estimation remains an important challenge. In this study, we propose an interpretable machine learning framework for earthquake magnitude prediction using reference distance based spectral waveform features extracted from seismic signals. Mel-frequency cepstral coefficients (MFCCs) are computed from the vertical, north and east components of seismic waveforms within a short time window around the primary wave arrival. These spectral descriptors are subsequently combined with station related attributes to form a comprehensive feature representation. Regression and classification models including XGBoost, Random Forest, Support Vector Machines, k-Nearest Neighbors, Decision Trees and Artificial Neural Networks are evaluated for earthquake magnitude estimation and event classification. Experiments on the STanford EArthquake Dataset (STEAD) using an event level train-test partitioning strategy demonstrate strong predictive performance across both regression and classification tasks. For earthquake magnitude classification, Random Forest achieved the highest accuracy of 88.26% and F1-score of 83.84%, whereas XGBoost produced the highest ROC-AUC value of 94.36%, indicating superior class separability. For regression, XGBoost demonstrated strong predictive performance together with high computational efficiency suitable for near real-time seismic applications. Furthermore, SHapley Additive exPlanations (SHAP) based explainability analysis and systematic ablation experiments reveals the relative contributions of spectral and station-based features to the model predictions, providing insights into the underlying physical characteristics captured by the learning algorithms. The results demonstrate that the proposed combined feature representation provides an effective and transparent approach for earthquake magnitude estimation and classification within the STEAD dataset. The ablation analysis further indicates that the reference-based spectral distance descriptors provide complementary information to the MFCC representation.</p></div>","PeriodicalId":16994,"journal":{"name":"Journal of Seismology","volume":"30 5","pages":""},"PeriodicalIF":1.3,"publicationDate":"2026-08-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148838064","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Noise-robust and transferable seismic phase picking for regional earthquake monitoring using a multi-scale temporal fusion network","authors":"Runping Ma,&nbsp;Maofa Wang,&nbsp;Lianqing Zhou,&nbsp;Jiajun He,&nbsp;Zhiang Zeng","doi":"10.1007/s10950-026-10442-z","DOIUrl":"10.1007/s10950-026-10442-z","url":null,"abstract":"<div><p>Automatic P- and S-wave arrival-time picking remains challenging under low signal-to-noise ratio (SNR) conditions and across seismic networks with different waveform characteristics. We propose MST-FuseNet, a multi-scale temporal fusion network that integrates dilated convolution, bidirectional temporal–channel attention, and LSTM-based sequence modeling for three-component seismic waveforms. MST-FuseNet is evaluated on event-disjoint PNW-CSD and INSTANCE datasets, together with low-SNR, background-noise, regional, and zero-shot cross-domain experiments. At a strict 0.1 s tolerance, it achieves P- and S-wave F1 scores of 90.00% and 85.94% on PNW-CSD. On 10,000 independent PNW noise records, it yields the lowest total false-pick rate of 8.203 picks per station-hour, compared with 9.973 for PhaseNet and 13.752 for EQTransformer. On INSTANCE, MST-FuseNet achieves the highest F1 scores for both phase types. In zero-shot transfer from PNW-CSD to INSTANCE, it obtains P- and S-wave F1 scores of 82.97% and 78.72%, respectively, together with the lowest mean absolute timing errors. These results demonstrate a favorable balance among phase-picking accuracy, noise robustness, false-trigger control, and cross-domain transferability.</p></div>","PeriodicalId":16994,"journal":{"name":"Journal of Seismology","volume":"30 5","pages":""},"PeriodicalIF":1.3,"publicationDate":"2026-08-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148838062","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Operational evaluation of pretrained deep learning phase pickers in the Marmara Region","authors":"Timur Tezel,&nbsp;Caner Erden,&nbsp;Muhammed Yusuf Küçükkara,&nbsp;Hazal Arıkan,&nbsp;Kenan Yanik,&nbsp;Murat Utkucu","doi":"10.1007/s10950-026-10445-w","DOIUrl":"10.1007/s10950-026-10445-w","url":null,"abstract":"<div><p>The Marmara region, located within the western segment of the North Anatolian Fault Zone (NAFZ) in north-west Turkey, is a high-risk seismic zone with a history of catastrophic earthquakes. Precise and rapid detection of seismic phase arrival times (P- and S-waves) is crucial for real-time earthquake monitoring, early warning systems, and fault characterisation. Although deep learning (DL) models have demonstrated superior performance in automated phase picking compared with traditional methods, their capacity to generalise across different tectonic settings remains insufficiently explored. This study comprehensively evaluates seven leading pre-trained DL models using a dataset of three-component seismograms from 389 earthquakes recorded by the Marmara Region Seismic Network from 2015 to 2025, comprising approximately 1.7 million phase picks. The evaluated models include PhaseNet, PhaseNetLight, EQTransformer, BasicPhaseAE, LFEDetect, Skynet, and OBSTransformer, each tested with multiple pretrained checkpoints from different seismic datasets. Model performance is assessed using confidence distributions, station-level confidence statistics, and picking accuracy metrics such as mean absolute error (MAE), root-mean-square error (RMSE), and median absolute error (MedianAE). Results show that P-wave picks consistently have higher confidence values and lower timing errors than S-wave picks across all models. Transformer-based architectures and PhaseNet variants demonstrate the most stable performance, with several configurations achieving MAE values below 1 s. Station-level analysis indicates that the Marmara seismic network provides relatively uniform waveform quality, although S-phase detection is more affected by station conditions. A trade-off between detection coverage and confidence reliability is observed among different models. The findings offer practical guidance for choosing deep-learning phase pickers for regional seismic monitoring systems. Our findings highlight the importance of region-specific model validation and seismic phase picking in tectonically complex areas. This study fills a critical gap by providing the first large-scale, multi-model evaluation of pretrained deep learning pickers in the Marmara region.</p></div>","PeriodicalId":16994,"journal":{"name":"Journal of Seismology","volume":"30 5","pages":""},"PeriodicalIF":1.3,"publicationDate":"2026-08-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148838143","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Kernel density-based quantification of uncertainties in location and magnitude of past earthquakes, towards probabilistic seismic hazard analysis of North-East India","authors":"Niranjan Borah,&nbsp;Abhishek Kumar","doi":"10.1007/s10950-026-10430-3","DOIUrl":"10.1007/s10950-026-10430-3","url":null,"abstract":"<div><p>This study employs a Kernel density-based approach for the probabilistic seismic hazard analysis (PSHA) for the seismically active and complex region of North-East India (NEI). A novel strategy by generating multiple synthetic earthquake (EQ) catalogues to account for uncertainties in the locations and magnitudes of instrumental, pre-instrumental, and paleo EQs is adopted. Both isotropic and anisotropic Kernel densities are used to model the spatial variations of the seismic activity rates within the region, to perform PSHA in the region. The final results indicate that the Indo-Burma Ranges exhibit the highest seismic activity and hazard levels, followed by Arunachal Pradesh's western regions, East Assam, the Shillong Plateau, and southern Sikkim. Peak Ground Acceleration (PGA) values for the Maximum Credible Earthquake (MCE) range from 0.425 g to 0.7 g, while for the Design Basis Earthquake (DBE), PGA ranges from 0.22 g to 0.43 g. Besides, the anisotropic Kernels, which incorporate fault directionality, show higher PGA values near the faults compared to isotropic Kernels. The Uniform Hazard Spectra (UHS) are also generated in the present study for the eight cities in NEI. This novel framework represents a significant advancement in the PSHA methodology to integrate various data sources and uncertainties.</p></div>","PeriodicalId":16994,"journal":{"name":"Journal of Seismology","volume":"30 5","pages":""},"PeriodicalIF":1.3,"publicationDate":"2026-08-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148837532","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Ground motion prediction based on source physics for earthquake early warning","authors":"Jiang Wang,&nbsp;Qiang Ma,&nbsp;Dongwang Tao,&nbsp;Quancai Xie,&nbsp;Huiyu Zhu","doi":"10.1007/s10950-026-10437-w","DOIUrl":"10.1007/s10950-026-10437-w","url":null,"abstract":"<div><p>The complex rupture process of destructive major earthquakes governs the intensity of near-field ground motions. It is well established that the low-frequency amplitude of the source spectrum is primarily controlled by the seismic moment, whereas the high-frequency energy is significantly modulated by the stress drop. However, current earthquake early warning systems predominantly rely on empirical ground-motion models that oversimplify the dynamic rupture process, thereby failing to capture the influence of fault-zone stress drop on high-frequency shaking. In this study, we develop a ground-motion prediction equation that integrates both seismic moment and stress drop, based on an elliptical source model and the physical relationship between source parameters and ground motions. Its key advantage lies in the ability to invert source parameters in real time directly from early P-wave recordings in the near field, enabling physically driven, real-time ground-motion prediction. Validation with Japan’s KiK-net data that shows that the method yields accurate and low-error predictions of peak ground motions (acceleration and velocity), with errors converging stably as the time window extends. The proposed method is insensitive to local site conditions and requires no complex site corrections. This work provides a new paradigm and theoretical foundation for overcoming the limitations of traditional empirical ground-motion models and for developing ground-motion prediction approaches with explicit source physics. </p></div>","PeriodicalId":16994,"journal":{"name":"Journal of Seismology","volume":"30 5","pages":""},"PeriodicalIF":1.3,"publicationDate":"2026-08-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148837723","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Physics-based 3D ground motion simulations of the 1703 L’Aquila earthquake: from historical reconstruction to seismic scenarios","authors":"F. Di Michele,&nbsp;D. Pera,&nbsp;E. Stagnini,&nbsp;I. Mazzieri,&nbsp;J. B. May,&nbsp;V. Kastelic,&nbsp;M. M. C. Carafa,&nbsp;B. Rubino,&nbsp;R. Aloisio,&nbsp;C. Smerzini,&nbsp;P. Marcati","doi":"10.1007/s10950-026-10434-z","DOIUrl":"10.1007/s10950-026-10434-z","url":null,"abstract":"<div><p>The city of L’Aquila (central Italy) and its surrounding villages have been struck several times throughout history by strong earthquakes, most recently by the <span>(M_w)</span> 6.1 event of April 6, 2009. Previously, on February 2, 1703, L’Aquila was partially destroyed by a <span>(M_w)</span> 6.7 earthquake, which has remained in the city’s collective memory as the <i>Grande Terremoto </i> (“Big Earthquake”). In this paper, we present a physics-based simulation of this event, together with a set of plausible scenarios associated with the same fault structure believed to be responsible for the 1703 earthquake. Assuming the geometry of the fault plane as assigned, we simulate 120 stochastic rupture scenarios corresponding to different magnitudes, epicentral locations, and slip distributions, selected within ranges compatible with the available literature. For each scenario, several estimates of macroseismic intensity are obtained from peak ground velocity and peak ground acceleration values using a set of empirical models specifically calibrated for the Italian territory. The performance of the scenarios is then evaluated through comparison with the macroseismic intensity field provided by historical catalogues. This study demonstrates that a satisfactory reconstruction of ground shaking from major historical earthquakes can be achieved using a stochastic source model constrained by a limited amount of ex-ante information. The proposed PBS-based approach helps integrate historical catalogues, filling data gaps. However, it is affected by limitations related to near-source numerical modeling, uncertainties in the empirical correlations used to derive macroseismic intensity, and assumptions on the source geometry. Nevertheless, the generated scenarios provide valuable insights into the potential impacts of future earthquakes with characteristics similar to the investigated event.</p></div>","PeriodicalId":16994,"journal":{"name":"Journal of Seismology","volume":"30 5","pages":""},"PeriodicalIF":1.3,"publicationDate":"2026-08-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10950-026-10434-z.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148837852","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}