Suppressing Coda Events with a Bayesian Model of Global Scale Seismology

NET-VISA stands for NETwork processing - Vertically Integrated Seismic Analysis. The package comprises a physics-based, probabilistic model and a heuristic inference algorithm to find the most probable set of seismic events to explain a series of arrivals detected by a global seismic network. It has been extended to find events in any of three mediums – rock, air, and water- and supports seismic, hydro-acoustic, and infrasound sensors.

Large seismic events often trigger a wave train of slow decaying energy known as the coda that can mislead signal detectors into forming coda detections that look like regular phase detections. These coda detections can confuse event formation algorithms into building false events known as coda events. Naive solutions to this problem by dropping any detection that looks like coda detection can have the negative consequence of missing real events. 

We propose to address this issue by extending an existing Bayesian Approach, designed to build event bulletins using a generative model of global-scale seismology. Our extensions significantly boost to the existing work by reducing the total number of false events and virtually eliminating coda events at the cost of a very small drop in the number of real events.