High-rate phase association with travel time neural fields.

Area of Science:

• Earthquake science
• Seismology
• Geophysics

Background:

• Phase association is crucial for grouping seismic arrivals to identify earthquake sources.
• Deep learning has increased the detection rate of small seismic events, leading to high-rate arrival sequences.
• Associating these dense seismic data is challenging, particularly with heterogeneous or unknown wave speeds.

Purpose of the Study:

• To introduce HARPA, a novel phase-association framework.
• To improve the association of seismic arrivals in dense and complex scenarios.
• To jointly estimate earthquake parameters and wave-speed fields.

Main Methods:

• HARPA utilizes optimal transport metrics to compare probability distributions of observed and predicted arrival sequences.
• It employs travel-time neural fields (neural networks) to model wave speeds and estimate earthquake locations and origin times.
• The framework handles high-rate arrival sequences and heterogeneous wave-speed fields.

Main Results:

• HARPA performs comparably to state-of-the-art methods on standard, low-rate datasets with simple wave speeds.
• It outperforms existing methods significantly at high seismic event rates and with laterally heterogeneous or unknown wave speeds.
• The study highlights the importance of adaptive travel-time modeling for dense seismicity.

Conclusions:

• HARPA offers a robust solution for phase association, especially in challenging environments with dense seismicity.
• The framework's ability to jointly model wave speeds advances seismic data analysis.
• This work suggests a pathway for integrating phase association with passive-source tomography.