Towards an Improved Earthquake Catalog for Northern California Using Deep-Learning-Based Arrival Time Picking and Graph-Based Phase Association

Session: Leveraging Advanced Detection, Association and Source Characterization in Network Seismology
Type: Oral
Date: 4/30/2020
Time: 04:15 PM
Room: 120 + 130
Description: 

Developing accurate, comprehensive earthquake catalogs to low magnitude of completeness requires robustly detecting low signal-to-noise ratio (SNR) arrivals and properly associating those arrivals across a seismic network while identifying false arrivals and untangling moveouts from multiple earthquakes that occur close together in time. Recent advances in phase picking have shown that appropriately trained deep-learning-based phase pickers, such as PhaseNet (Zhu and Beroza, 2018), can accurately pick low-SNR arrivals. For association, a recent method has been proposed which uses an efficient implementation of back-projection with graph-theory-based clustering and integer linear optimization to resolve the association problem when arrivals are either false or missing (McBrearty et al., 2019) and when there are high rates of seismicity that produce overlapping moveouts across a seismic network. In this work we apply both techniques to develop a catalog of northern California seismicity using continuous seismic data recorded by the Northern California Seismic Network. We explore the sensitivity of these methods to hyperparameter choices, as well as observational complexities such as event rates and uncertainties in the velocity model. We compare our findings to existing catalogs to assess our results.

Presenting Author: Ian W. McBrearty

Authors