New Unsupervised Machine Learning Model for Seismic event Discrimination and its application to discriminate Icequakes and Tectonic quakes in Southeastern Alaska

Akash Kharita, Peter Henrik Voss, Trine Dahl-Jensen, Michael West

Research output: Contribution to conferenceAbstract at conference

Abstract

New Unsupervised Machine Learning Model for Seismic event Discrimination and its application to discriminate Icequakes and Tectonic quakes in Southeastern Alaska
Akash Kharita1,*, Peter H. Voss2, Trine Dahl-Jensen2, Michael West3
1. Department of Earth Sciences, Indian Institute of Technology Roorkee
2. Geological Survey of Denmark and Greenland
3. Alaska Earthquake Center, University of Alaska Fairbanks

Seismic observatories around the world often require a classification system that can discriminate tectonic and non-tectonic seismicity, e.g. for seismic hazard assessment, especially in the areas where these events overlap in magnitude, space and time. Although many machine learning models have been proposed for event detection and location, very few models are being reported for providing a reliable automatic classification of events.
Automatic classification systems are necessary to improve the quality of earthquake catalogs, reduce the workload of analysts and also to produce a reliable classification especially in the regions where seismic coverage is not strong.
In this study, we present a new unsupervised machine learning model for seismic event discrimination which is based on applying Principal Component Analysis to absolute frequency spectrum of different event series and labelling the clusters using Gaussian Mixture Models. The model can work on data from a single station using just a single component and provide reliable classification in near real time.
We have done synthetic tests where we have shown the separations of synthetic sine curves having different frequency contents at different signal to noise ratios ranging from 0.1 to 5. We also applied our model to show the discrimination of 300 well reviewed icequakes and tectonic quakes in southeastern Alaska with accuracy of about 90%. Our model thus has a potential to come out as a general automatic event classifier and can also be used to discriminate other non-tectonic events such as quarry blasts, mining events and explosions from tectonic events.

Original languageEnglish
Publication statusPublished - 22 Sep 2021
Event37th General Assembly of the European Seismological Commission - ESC - Online / Greece
Duration: 19 Sep 202124 Sep 2021
https://www.erasmus.gr/microsites/1193/virtual-platform-guide

Conference

Conference37th General Assembly of the European Seismological Commission - ESC
Period19/09/2124/09/21
Internet address

Programme Area

  • Programme Area 3: Energy Resources

Cite this