Abstract
In principle, a goal of any regional seismic network (RSN) is simple -
detect and review seismicity in the network’s region of interest.
Vigorous aftershock sequences however present a unique challenge to an
RSN’s mission in that they can overwhelm automated real-time and
analyst-driven post-processing systems. This inundation with seismicity
can the lead to earthquake catalogs that are incomplete and erroneous in
the area around the aftershock sequence. In response to this challenge,
this abstract explores integrating machine learning as a tool to
expedite the University of Utah Seismogram Station’s (UUSS)’s response
to the September, 2017 Soda Springs event in southeast Idaho. In the
months following the event the UUSS was able to locate over 1000
aftershocks and was catalog complete to approximately 2.5 M_L. Owing to
the fact that a temporary deployment of 6 six-channel and 2
strong-motion instruments were deployed to the area shortly after the
event, it is anticipated that many more high-quality locations can be
obtained without resorting to template matching. Therefore, it is our
goal to attempt to recover the existing UUSS catalog with machine
learning enabled workflows that can produce analyst-grade picks for
location and subsequent magnitude estimation. Should this be successful
then the anticipated consequence is that analysts can quickly screen and
approve more events and by extension lower the network’s magnitude
completeness during the aftershock sequence.