Geosci

本日選文（共 4 篇）。 [1843] 以GIT為基礎之場址響應去卷積建立歐地中海區域參考硬岩條件下的地震動模型 Ground‐Motion Model for Reference Hard‐Rock Conditions in the Euro‐Mediterranean Context by GIT‐Based Site‐Response Deconvolution 期刊：Geophysics (GeoScienceWorld) — Advance Access Published：Fri, 15 May 2026 00:00:00 GMT DOI：10.1785/0120250024/730834 原文連結：連結 Summary (EN) The paper addresses the lack of hard‑rock reference motions in strong‑motion databases and develops a method to retrieve and model reference ground motion for shallow crustal earthquakes in the Euro‑Mediterranean region. The authors merge multiple strong‑motion datasets, estimate station site responses using parametric and nonparametric generalized inversion techniques (GITs), select reference hard‑rock stations, and deconvolve site terms to virtually project recordings to very high VS (bedrock) conditions. Resulting ground‑motion models for reference hard‑rock show limited site‑to‑site variability (0.1–0.2 in natural log), indicating effective removal of site effects, and an average VS30‑based amplification model is proposed for cases without site‑specific soil estimates. ...

本日選文（共 5 篇）。 [1794] 針對希臘北部 Prinos CO2 儲存場址潛在誘發地震之混合地震動模型（混合 GMM） Hybrid Ground‐Motion Model for Potential Induced Seismicity at the Prinos CO 2 Storage Site (Northern Greece) 期刊：Geophysics (GeoScienceWorld) — Advance Access Published：Wed, 13 May 2026 00:00:00 GMT DOI：10.1785/0120250287/730782 原文連結：連結 Summary (EN) The paper presents the first region-specific hybrid ground-motion model (GMM) developed for potential injection-induced seismicity at the Prinos CO2 storage site in northern Greece. The framework combines stochastic simulations calibrated against regional recordings and constrained by an a priori seismological model, with iterative parameter optimization and quantification of epistemic uncertainty via suites of acceptable parameter combinations. A synthetic dataset (2≤Mw≤6, hypocentral distances up to 80 km) and an expanded empirical dataset including machine-learning-derived events are integrated to support regional hazard assessment and to inform methodologies for induced-earthquake ground-motion characterization in Europe. ...

本日選文（共 2 篇）。 [1755] 將基於深度神經網路與地震水平對垂直頻譜比之地震動模型應用於日本2024年能登半島地震 Application of Ground‐Motion Model Based on Deep Neural Network and Earthquake Horizontal‐to‐Vertical Spectral Ratio to the 2024 Noto Peninsula Earthquake in Japan 期刊：Geophysics (GeoScienceWorld) — Advance Access Published：Mon, 11 May 2026 00:00:00 GMT DOI：10.1785/0120250265/730571 原文連結：連結 Summary (EN) The paper tests a deep‑neural‑network ground‑motion model that incorporates earthquake horizontal‑to‑vertical spectral ratios (EHVRs) against recordings from the 2024 Noto Peninsula (Mw 7.5) earthquake. The DNN was retrained on an updated strong‑motion database with site‑specific EHVRs and used data augmentation (adding random noise to response spectra and EHVRs) to better represent large‑amplitude motions. Evaluation against observed spectra and independent 2024 events shows the augmented model outperformed the original, especially in near‑field areas, and generalized well while capturing complex site‑specific spectral features including peak shifts under strong shaking. ...

本日選文（共 1 篇）。 [1317] CapsNet 強化之地震事件分類：在 DiTing 2.0 資料集上的基準測試與系統部署 CapsNet-enhanced seismic event classification: benchmarking and system deployment on the DiTing 2.0 data set 期刊：GJI (OUP) — RSS Published：Fri, 03 Apr 2026 00:00:00 GMT DOI：10.1093/gji/ggag129 原文連結：連結 Summary (EN) This study provides the first systematic benchmark of mainstream deep learning models (CNNs, Transformer-based architectures, and Capsule Networks) on the large-scale DiTing 2.0 seismic dataset, which contains 19,384 three-component waveforms labeled as natural earthquakes, quarry blasts, or mine collapses. Optimized CapsNet variants, particularly a CapsNet+Res model using MFCC features and data augmentation, achieved 91.08% accuracy (weighted F1 = 91.10%) on the held-out test set; multi-station voting raised event-level accuracy to 97.52%, and a companion noise-event classifier reached 98.47% accuracy. Demonstration tests on continuous records show reliable end-to-end operation in a transferable, user-friendly system, and the authors highlight large-scale continuous-data evaluation, cross-regional transfer, and adaptive learning as important future directions. ...

本日選文（共 8 篇）。 [1168] 熱活化靜摩擦可解釋地震相互作用 Thermally Activated Static Friction Can Explain Earthquake Interactions 期刊：Earth and Space Science (AGU/Wiley) — eTOC Published：Thu, 02 Apr 2026 03:54:53 -0700 DOI：10.1029/2025JB032266 原文連結：連結 Summary (EN) Earthquake time asymmetry and the ubiquity of aftershocks remain poorly understood. Using a spring–slider model that combines elastic stress transfer with reaction-rate theory for thermally activated static friction, the authors introduce a microscopic timescale and temperature and reproduce many features of natural seismicity, including a power-law distribution of seismic moments, Omori and productivity laws for aftershocks, clustering of aftershocks near rupture edges, and a strong time asymmetry in the seismic cycle. ...

本日選文（共 3 篇）。 [1024] SPIDER：用於雙差地震重定位的可擴展機率推論 SPIDER: Scalable Probabilistic Inference for Differential Earthquake Relocation 期刊：Earth and Space Science (AGU/Wiley) — eTOC Published：Wed, 25 Mar 2026 05:58:33 -0700 DOI：10.1029/2025JB032769 原文連結：連結 Summary (EN) Seismicity catalogs have grown hugely, but Bayesian double‐difference earthquake relocation struggles to scale to the millions of parameters involved. The authors introduce SPIDER, a scalable Bayesian framework for double‐difference hypocenter relocation that pairs a physics‑informed neural network Eikonal solver with Stochastic Gradient Langevin Dynamics to generate posterior samples jointly for entire catalogs. SPIDER explicitly whitens residual correlations that traditional formulations neglect, is designed for multi‑GPU parallelization, and is demonstrated on a synthetic catalog and three real catalogs from California and Japan, with proposed tools to analyze high‑dimensional posteriors. ...

本日選文（共 4 篇）。 [1004] 利用KiK-net二十年資料估算日本岩手縣地下速度與共振頻率變化 Estimating subsurface velocity and resonance frequency changes using two decades of KiK-net data in the Iwate Prefecture, Japan 期刊：GJI (OUP) — RSS Published：Fri, 13 Mar 2026 00:00:00 GMT DOI：10.1093/gji/ggag088 原文連結：連結 Summary (EN) Using over two decades of KiK-net borehole–surface recordings at 28 stations in Iwate Prefecture, the authors show consistent frequency-domain (stacked Stockwell PSD) and time-domain (multitaper deconvolution and phase cross-correlation) evidence that resonance frequencies and seismic velocities systematically decrease with increasing peak ground acceleration (PGA). Frequency shifts mainly reflect the very shallow layers, while borehole-referenced velocity changes represent a path-averaged perturbation that depends on borehole depth. Across stations, relative velocity reductions average ~12% for the 200–400 cm/s^2 PGA range (implying a shear modulus reduction of ~23%), with stronger nonlinear effects at sites with thicker sedimentary deposits and no clear correlation with VS30. ...

本日選文（共 4 篇）。 [907] 利用新建立的 N-net 海底地震儀網於南海溝日向灘檢測淺部構造性顫動 Detection of Shallow Tectonic Tremors in Hyuga‐nada, Nankai Trough, Using the Newly Established N‐Net OBS Network 期刊：JGR: Solid Earth (AGU/Wiley) — eTOC Published：Tue, 17 Mar 2026 02:06:18 -0700 DOI：10.1029/2025GL121198 原文連結：連結 Summary (EN) Using machine-learning on newly deployed offshore N‑net OBS data, the authors detected shallow tectonic tremors in Hyuga‑nada between July 2024 and September 2025. Two major tremor episodes occurred: one immediately after the August 2024 M7.0 earthquake and another three days before the January 2025 M6.7 event, with several tremors migrating alongside seismicity—suggesting slow‑slip transients promoted stress transfer along the plate boundary. Tremors clustered in fluid‑rich plate‑boundary zones and were absent in drained areas, indicating structural heterogeneity controls shallow slow‑earthquake dynamics and providing new insights into slow‑earthquake mechanics. ...

本日選文（共 2 篇）。 [907] 利用新建立的 N-net 海底地震儀網於南海溝日向灘檢測淺部構造性顫動 Detection of Shallow Tectonic Tremors in Hyuga‐nada, Nankai Trough, Using the Newly Established N‐Net OBS Network 期刊：JGR: Solid Earth (AGU/Wiley) — eTOC Published：Tue, 17 Mar 2026 02:06:18 -0700 DOI：10.1029/2025GL121198 原文連結：連結 Summary (EN) Using machine-learning on newly deployed offshore N‑net OBS data, the authors detected shallow tectonic tremors in Hyuga‑nada between July 2024 and September 2025. Two major tremor episodes occurred: one immediately after the August 2024 M7.0 earthquake and another three days before the January 2025 M6.7 event, with several tremors migrating alongside seismicity—suggesting slow‑slip transients promoted stress transfer along the plate boundary. Tremors clustered in fluid‑rich plate‑boundary zones and were absent in drained areas, indicating structural heterogeneity controls shallow slow‑earthquake dynamics and providing new insights into slow‑earthquake mechanics. ...

本日選文（共 6 篇）。 [809] DeepSubDAS：基於海底分布式聲學感測（DAS）資料的地震相位挑選器 DeepSubDAS: an earthquake phase picker from submarine distributed acoustic sensing data 期刊：GJI (OUP) — RSS Published：Mon, 09 Feb 2026 00:00:00 GMT DOI：10.1093/gji/ggag061 原文連結：連結 Summary (EN) Submarine Distributed Acoustic Sensing (DAS) can address the scarcity of seismometers in oceanic regions, but land-trained models perform poorly on submarine data. This study develops a DeepLab v3–based machine learning model specifically for submarine DAS to detect seismic events and pick P- and S-waves, trained and validated on nearly 57 million manually and semi-automatically labelled records from multiple global submarine sites. The model is designed to handle different cable lengths, configurations and channel spacings, providing an adaptable tool for automated earthquake analysis and with potential to support real-time monitoring and tsunami early warning. ...