Implementation of AI/ML Detection of Seismicity as a Real-time SeisComP Module
Description:
Rapid and precise detection of seismic phases is a priority for seismic monitoring networks. While various methodologies exist, recent advancements in machine learning (ML) techniques have demonstrated exceptional performance in automatically identifying seismic events. However, the real-time application of these methods remains challenging due to their high computational demands and limited integration with real-time processing workflows. Since 2023, the Texas Seismological Network (TexNet) has used the EarthQuake Compact Convolutional Transformer (EQCCT), a deep-learning ML model, to detect seismic phases automatically. This approach, matched with scanloc (gempa.de), has increased the detection of low-magnitude earthquakes by a factor of 50 compared to the traditional STA/LTA method. However, this implementation operates in a post-event mode, introducing delays of 10–45 minutes relative to the phase onset time.
To address this limitation, we developed a new SeisComP module, sceqcct, integrating EQCCT into TexNet’s real-time data workflow. The development focused on three objectives: (1) optimizing the computational efficiency of EQCCT, (2) integrating EQCCT within SeisComP’s framework, and (3) tuning critical parameters (e.g., bandpass filter) to enhance real-time performance. The sceqcct module processes real-time seismic data, achieving pick detections with delays less than one minute after phase onset time. It also supports playback mode for analyzing archived waveforms. We evaluated the sceqcct’s performance by comparing its real-time results with TexNet’s post-event EQCCT implementation. The new module achieved a high correspondence in pick and, eventually, event detection, with a near-perfect agreement for events of magnitude ≥1.2. After tuning parameters with region-specific settings, additional arrivals were detected, further enhancing performance. Overall, these advancements represent a significant step forward in real-time seismic monitoring, enabling rapid and reliable automatic event detection with minimal delays (P and S arrival detection within 1–2 minutes) using an AI/ML-based methodology.
Session: Network Seismology: Recent Developments, Challenges and Lessons Learned - II
Type: Oral
Date: 4/15/2025
Presentation Time: 11:30 AM (local time)
Presenting Author: Camilo
Student Presenter: No
Invited Presentation:
Poster Number:
Authors
Camilo Muñoz Lopez Presenting Author Corresponding Author camilo.munoz@beg.utexas.edu University of Texas at Austin |
Victor Salles victor.salles@beg.utexas.edu University of Texas at Austin |
Constantinos Skevofilax constantinos.skevofilax@austin.utexas.edu University of Texas at Austin |
Yangkang Chen yangkang.chen@beg.utexas.edu University of Texas at Austin |
Alexandros Savvaidis alexandros.savvaidis@beg.utexas.edu University of Texas at Austin |
|
|
|
|
Implementation of AI/ML Detection of Seismicity as a Real-time SeisComP Module
Category
Network Seismology: Recent Developments, Challenges and Lessons Learned