← Back to Sessions

Classifying Central and Eastern U.S. Seismic Events in the Earthscope Database Using Machine Learning and Lg-Wave Spectral Ratios.

Description: 

Historical earthquake catalogs allow for the characterization of natural seismicity prior to subsurface fluid injection, which can induce earthquakes. To facilitate large-scale seismicity characterization in the central and eastern United States (CEUS), our team applied a machine-learning-based waveform classifier (WC) to events in the EarthScope database to augment existing earthquake catalogs. We selected EarthScope’s Transportable Array because it transected the contiguous United States from west to east with uniform station coverage in the otherwise nonuniformly—and, in some regions, sparsely—monitored CEUS. EarthScope’s Array Network Facility (ANF) detected seismic events that occurred within the array but did not classify them as earthquakes or mining-related events. The WC was employed to calculate the probabilities that waveforms were produced by earthquakes or mining activities such as blasts and mining-induced collapses (MIC). Using a subset of 502 manually classified events, we determined the WC reliably classified earthquakes at probabilities ≥ 0.7 and blasts at probabilities ≥ 0.5. Events with lower probabilities, with waveforms from fewer than four stations, or with standard deviations of the individual-station probabilities ≥ 0.40 were manually classified. Of the 6,634 events, 14% required manual review. Because some events falsely classified as earthquakes occurred in mining areas, and some events in those areas classified as blasts occurred during the nighttime, when blasting is typically restricted, we also calculated Lg spectral ratios to assist with discriminating between MICs, blasts, and earthquakes. Preliminary results using 1-2 to 2-4 Hz Lg-wave spectral ratios on a subset of known events shows that MICs have higher Lg spectral ratios relative to blasts and to earthquakes. In total, we found that 61 earthquakes were uniquely detected by the ANF.

Session: Opportunities and Challenges for Machine Learning Applications in Seismology [Poster]

Type: Poster

Date: 4/19/2023

Presentation Time: 08:00 AM (local time)

Presenting Author: Jonathan P. Schmidt

Student Presenter: No

Invited Presentation: 

Authors