Application of Neural Network Automatic Event Detection for Reservoir-Triggered Seismicity Monitoring Networks.

Area of Science:

• Geophysics
• Seismology
• Earthquake Science

Background:

• Reservoir-triggered seismicity (RTS) networks often have limited station coverage, making earthquake detection challenging.
• Inadequate P-wave data necessitates reliable S-wave identification for accurate event location.
• RTS datasets are typically small, requiring algorithms trained on limited data or external global datasets.

Purpose of the Study:

• To compare the effectiveness of different seismic signal detection methods for RTS.
• To evaluate deep learning models, transfer learning, specialized neural networks, and manual detection.
• To determine the optimal approach for enhancing earthquake catalogs in RTS regions.

Main Methods:

• Comparison of a global deep learning detection model, transfer learning applied to RTS data, a specialized RTS neural network, and manual detection.
• Focus on phase detection sensitivity over phase picking accuracy and specificity.
• Evaluation based on parameters related to seismic event location and phase association.

Main Results:

• Transfer learning efficiency is dependent on the specific database used.
• Neither automatic nor manual detection methods alone are sufficient for comprehensive RTS event detection.
• Combining automatic and manual methods significantly increases seismic event detectability.

Conclusions:

• A combined approach of automatic and manual seismic signal detection substantially improves RTS catalog completeness.
• The enhanced catalogs, covering up to 30% more events, aid in studying triggering mechanisms.
• Neural network detectors are crucial for increasing the number of detected seismic events and furthering RTS research.