Skip to main content
Open Access Publications from the University of California


UCLA Electronic Theses and Dissertations bannerUCLA

Navigating Microseismicity Characterization with Backprojection, Matched-filter, and Deep Learning


Benefiting from the recent deployments of dense seismic arrays, seismologists have the opportunity to detect earthquakes of small magnitudes (M < 2), referring to the occurrence of these events as microseismicity. The characterization of microseismicity is essential in solving many geophysical problems, such as probing fault structures, investigating the process of hydraulic fracturing, and imaging the preseismic, coseismic, and postseismic slip of megathrust earthquakes. Traditionally, earthquakes are identified by picking seismic phases on continuous waveforms using the short-time average to long-time average ratio, kurtosis and skewness, waveform polarization, Akaike information criterion, and discrete wavelet transforms. These single-station methods, however, may fail to detect microseismicity with weak phase arrivals hidden in the noise.

This thesis introduces three new approaches to detecting microseismicity with multistation data: backprojection, matched-filter, and deep learning. The backprojection approach tracks and back-projects the coherent seismic pulses to the target region to determine the timing and location of any microseismic earthquakes. The matched-filter approach searches for similar patterns of existing template earthquakes in the continuous recordings as suggestive of a new event. We characterize the aftershock sequence of the 2011 M 9 Tohoku earthquake with the backprojection and matched-filter methods. Based on the spatial consistency between aftershock-depleted zones and large coseismic slip, we identify a possible large coseismic slip zone in the near-trench region offshore Fukushima. The deep learning approach is data-driven and it learns to characterize microseismicity with a neural network based on a large number of microseismicity recordings. We trained a Deep Learning Phase-picking model named EdgePhase with a Southern California dataset and applied it to detect the early aftershocks following the 2020 M 7 Samos, Greece earthquake. Compared to a local earthquake catalog, EdgePhase showed 190% more detections with an event distribution that is more conformative to a planar fault interface, suggesting higher fidelity in event locations. In addition to characterizing microseismicity, backprojection can also improve the prediction of earthquake ground motions. We propose a high-frequency distance metric based on backprojection, which outperforms traditional distance metrics in predicting the ground shaking intensity of megathrust earthquakes in Japan and Chile.

Main Content
For improved accessibility of PDF content, download the file to your device.
Current View