Abstract:

At the SCSN, we've introduced the deep-learning phase picker PhaseNet into our event post-processing to improve realtime event solutions and reduce analyst workload. This change has increased the number of automatic picks, especially S phases, associated with an event, with similar or slightly better pick accuracy. The new processing also results in more accurate epicentral locations, allowing more events to be finalized from the automatic solutions with minimal analyst review. We've also been working on using PhaseNet with the deep-learning associator GaMMA to automatically process subnet triggers, which are collections of potentially related picks that were not associated with an event by the realtime system. Our initial tests demonstrated that this processing can accurately recover events from triggers, as long as enough stations in the subnet recorded the event, with a low false event rate.

All are welcome.