Machine learning reveals cyclic changes in seismic source spectra in Geysers geothermal field.

Area of Science:

• Geophysics
• Machine Learning
• Seismology

Background:

• Earthquake rupture involves complex interactions of stress, fracture, and friction.
• Seismic signal analysis is crucial for understanding faulting processes.
• Geothermal fields experience seismicity influenced by fluid injection.

Purpose of the Study:

• To apply machine learning to identify temporal patterns in seismic spectral properties.
• To investigate the relationship between seismic event characteristics and geothermal reservoir operations.
• To explore the potential of these methods for monitoring faulting processes.

Main Methods:

• Clustering of 46,000 seismic events (0.3 < ML < 1.5) from The Geysers geothermal field.
• Analysis of time-dependent spectral properties of seismic signals.
• Correlation of seismic event patterns with water injection rates.

Main Results:

• Clustering revealed clear temporal patterns, not spatial ones, in seismic events.
• Events with similar spectral properties exhibited annual cycles.
• These cycles tracked changes in water injection rates into the geothermal reservoir.
• Observed changes in acoustic and faulting properties correlated with thermomechanical state changes.

Conclusions:

• Machine learning effectively identifies subtle changes in seismic source properties.
• Seismic spectral patterns provide insights into faulting processes influenced by geothermal operations.
• The methods offer new applications for characterizing tectonic and geothermal seismicity.