IDEAS Trainee Vivian Tang Applies ML Methods for EarthScope Discoveries

IDEAS NRT Fellow Vivian Tang systematically searched EarthScope data for seismic tremor and small earthquakes that occur within North America rather than on the North American plate boundary during transient stress fields.

She specifically explored searched USArray seismogram data for tremor and earthquakes dynamically triggered by the 2012 Mw8.6 Sumatra earthquake.

In addition, Tang newly observed dynamically triggered tremor near the Yellowstone hotspot, Wyoming, and triggered earthquakes in the Raton Basin and in central Utah. She identified additional triggered events for each of these three locations by investigating seismograms recorded between 2007 to 2017.

To advance our understanding and identification of the conditions that lead to dynamic triggering of seismic events, she applied a decision-tree machine-learning algorithm to these data sets.
Tang obtained her fundamental knowledge of Machine Learning from the DATA_SCI-423 course. She was also inspired by the elective course she took for the Integrated Data Science Certificate, EECS-349, to apply the decision tree algorithm to seismic data in order to explore the reasons for triggering earthquakes or tremor.

The algorithm found that dynamic stress estimates from teleseismic surface waves indeed appear to be a deciding factor in triggering tremor, though may be secondary to the back azimuth from which the surface waves arrive. Tang’s findings confirm that transient stresses generated by surface waves from strong earthquakes arriving from favorable directions can lead to triggered tectonic tremor in seismically active regions, such as central California and Yellowstone. These stresses do not appear to be deciding factors for the potentially dynamically triggered earthquakes in the Raton Basin and central Utah, while back azimuth does appear to be a deciding factor.