應用深度學習即時微震監測系統解析2022年規模6.9池上、2024年規模7.2花蓮、及2025年規模6.4大埔地震序列

Abstract

建立即時且高解析度的地震目錄，對於瞭解地震序列發展過程及進行災害風險評估具有關鍵意義。本研究開發一套整合深度學習技術的即時微震監測系統，經測試及檢驗，確認透過該系統全自動化流程能快速提供可信的地震活動資訊。該系統主要資料處理流程包括：（1）利用SeedLink接收來自中央研究院地球科學研究所及國家地震工程研究中心寬頻地震觀測網的連續波形資料，儲存並建成連續波形資料庫；（2）透過以臺灣地震到時資料訓練的深度學習模型，判識及選取P波與S波到時並存成到時資料庫；（3）根據監測區域選取適當地震測站組合，提取對應P波及S波到時進行地震事件關聯並定位，生成初步AI地震目錄；（4）製作每日地震報告，透過電子郵件或LINE發送給相關人員。相較於現行的地震觀測網，本系統展現出在微震偵測解析能力與處理效率方面的優勢，特別適用於特定區域或需密集監測的場域，目前已建立三個即時微震監測系統：一、池上即時微震監測系統，觀測池上斷層滑移段及周圍地區的背景微震活動；二、花蓮地震即時微震監測系統，持續觀測2024年M7.2花蓮地震餘震序列變化；三、嘉南即時微震監測系統，自2024年起嘉南地區陸續發生數個規模大於五的地震，因此於2025年初建立此系統，觀測該地區中大型地震的主餘震序列。即時微震監測系統可迅速提供地震活動的變化，並建立長期地震目錄，經進一步資料處理後的地震目錄，例如絕對或相對重新定位，有助於後續地震地體構造解釋及其他地震參數研究，例如震源機制、地震規模、三維速度模型逆推等。綜合而言，本研究所提出之系統可作為現行地震觀測網的有效補強，顯著提升地震觀測的即時性與解析度。

Establishing a real-time and high-resolution earthquake catalog is crucial for understanding the development process of earthquake sequences and conducting disaster risk assessment. This study developed a real-time microseismic monitoring system that integrates deep learning technology. After testing and verification, it was confirmed that the system can quickly provide reliable earthquake activity information through a fully automated process. The main data processing process of the system includes: (1) using SeedLink to receive continuous waveform data from three broadband seismic networks maintained by the Institute of Earth Sciences of Academia Sinica and the National Center for Research on Earthquake Engineering, and store and build a continuous waveform database; (2) using a deep learning model trained with Taiwan earthquake arrival data to identify and selecte P- and S-wave arrival times and store them in an arrival database; (3) selecting appropriate seismic station combinations according to the monitoring area, extracting corresponding P- and S-wave arrival times to associate and locate earthquake events, and generating a preliminary deep learning earthquake catalog; (4) preparing daily earthquake reports and sending them to relevant personnel via email or LINE. Compared with the existing seismic observation network, this system has shown advantages in microseismic detection and analysis capabilities and processing efficiency. It is particularly suitable for specific areas or fields that require intensive monitoring. Currently, three real-time microseismic monitoring systems have been established: 1. Chihshang real-time microearthquake monitoring system, which observes the background microseismic activity of the creeping segmemt of the Chihshang fault and the surrounding areas; 2. Hualien earthquake real-time microseismic monitoring system, which continuously observes the changes in the aftershock sequence of the 2024 M7.2 Hualien earthquake; 3. the Chia-Nan real-time microseismic monitoring system, since 2024, several earthquakes with a magnitude greater than five have occurred in the Chia-Nan area, so this system was established in early 2025 to observe the main aftershock sequence of medium and large earthquakes in the area. The real-time microseismic monitoring system can quickly provide changes in seismic activity and establish a long-term earthquake catalog. After further data processing (such as absolute or relative relocationing), the earthquake catalog will help the subsequent interpretation of earthquake tectonic structures and other earthquake parameter studies, such as focal mechanism, earthquake magnitude, and three-dimensional velocity model inversion. In summary, the system proposed in this study can serve as an effective reinforcement for the current earthquake observation network, significantly improving the timeliness and resolution of earthquake observation.