Cross-Correlation Beamforming for Simultaneous Event Detection and Location in Conjunction With Logistic Regression for Event Discrimination

Date: 4/24/2019

Time: 05:15 PM

Room: Puget Sound

Seismic event detection, location, and characterization (DLC) problems are fundamental problems in seismology, given that to study subsequent seismological questions, one or more of these elements needs to have been addressed first. Thus, improving DLC algorithms has been a major research focus in seismology over the past several decades. Recently, earthquake DLC algorithms involving the use of machine learning have been shown in many cases to be particularly effective tools at DLC related tasks, often surpassing methods from the previous generation.

In this study we combine cross-correlation beamforming, which allows us to simultaneously detect and locate seismic events, with logistic regression, a supervised machine learning algorithm that allows us to discriminate between different types of events such as earthquakes and tremors. A significant advantage to combining these methods is that the results of cross-correlation beamforming can be directly used to generate meaningful features for logistic regression. We apply this combined methodology to data recorded by networks of seismometers located along the Cascadia subduction zone. Using several high quality earthquake and tremor catalogs available for this region, we assess the robustness of our detections and locations, as well as our ability to automatically distinguish different event types from one another. In addition to the features generated by the cross-correlation beamforming results, we further explore those features that allow our algorithm to discriminate between different seismic events.

Presenting Author: Stephen G. Mosher

Authors