PyOcto: A high-throughput seismic phase associator

TL;DR

PyOcto is a new seismic phase associator that significantly improves speed and maintains high detection sensitivity, effectively handling dense seismic sequences using 4D space-time partitioning and velocity models.

Contribution

PyOcto introduces a novel, high-throughput phase associator that outperforms existing methods in speed while maintaining or improving detection sensitivity.

Findings

Achieves at least 10x faster runtimes than state-of-the-art methods.

Maintains detection sensitivity on par or above existing algorithms.

Successfully applied to dense seismic sequences like the 2014 Iquique earthquake.

Abstract

Seismic phase association is an essential task for characterising seismicity: given a collection of phase picks, identify all seismic events in the data. In recent years, machine learning pickers have lead to a rapid growth in the number of seismic phase picks. Even though new associators have been suggested, these suffer from long runtimes and sensitivity issues when faced with dense seismic sequences. Here we introduce PyOcto, a novel phase associator tackling these issues. PyOcto uses 4D space-time partitioning and can employ homogeneous and 1D velocity models. We benchmark PyOcto against popular state of the art associators on two synthetic scenarios and a real, dense aftershock sequence. PyOcto consistently achieves detection sensitivities on par or above current algorithms. Furthermore, its runtime is consistently at least 10 times lower, with many scenarios reaching speedup factors above 50. On the challenging 2014 Iquique earthquake sequence, PyOcto achieves excellent detection capability while maintaining a speedup factor of at least 70 against the other models. PyOcto is available as an open source tool for Python on Github and through PyPI.