GroningenNet: Deep Learning for Low-Magnitude Earthquake Detection on a Multi-Level Sensor Network

Area of Science:

• Seismology
• Geophysics
• Machine Learning

Background:

• Induced seismicity is rising globally, necessitating robust detection of low-magnitude earthquakes.
• Accurate micro-earthquake detection is vital for monitoring operations like hydraulic fracturing and understanding seismic mechanisms.
• Existing detection algorithms struggle to reliably distinguish weak seismic events from local noise.

Purpose of the Study:

• To develop and evaluate a convolutional neural network (CNN) for enhanced seismic event detection.
• To leverage multi-level borehole sensor data for improved discrimination of subsurface events from surface noise.
• To assess the CNN's performance against traditional seismic detection methods.

Main Methods:

• A CNN model was trained using seismic records from multi-level geophones (50-200m depths) at Groningen borehole stations.
• The CNN utilized energy moveout patterns across sensors as a key feature for event discrimination.
• Performance was compared against Short-Term Average/Long-Term Average (STA/LTA) and template matching algorithms.

Main Results:

• The CNN model demonstrated significantly superior performance in detecting previously uncatalogued events.
• The proposed CNN approach substantially reduced false positive detections compared to STA/LTA and template matching.
• Utilizing moveout features allowed for effective model training with reduced data requirements.

Conclusions:

• The developed CNN offers a robust and efficient solution for detecting low-magnitude seismic events.
• The multi-level sensor approach and moveout feature analysis enhance discrimination capabilities.
• This methodology is adaptable for microseismic monitoring in networks with similar multi-level sensor configurations.