Neural Network Interpretation as a Denoising Tool for Automated Tremor Location

Session: From Aseismic Deformation to Seismic Transient Detection, Location and Characterization
Type: Oral
Date: 4/29/2020
Time: 04:30 PM
Room: 230 + 235
Description: 

We present a new methodology for automated tremor detection and location, relying on neural network interpretation to reduce the noise level of seismic waveforms. By identifying and extracting core tremor characteristics such as patterns in the frequency domain, this denoising process allows us to uncover structure in the data below the noise level. We then use the resulting, cleaned waveforms to locate detected tremor events applying standard array-based techniques.

We first apply our methodology to the Cascadia subduction zone. We analyze a slow slip event that occurred in 2018 below the southern end of the Vancouver Island, Canada and validate our approach through comparison with existing catalogs. We find our tremor locations to be consistent with results from the literature and detect more events previously lying below the noise level.

Having validated our new methodology in a well-studied area, we further apply it to various tectonic contexts and discuss the implications of tremor occurrences in the scope of exploring the interplay between seismic and aseismic slip.

Presenting Author: Claudia Hulbert

Authors