Identification of Overlapping Earthquakes using a moveout-based matching pursuit method

Clara Daniels, Zhigang Peng, James McClellan, & Lijun Zhu

Published August 15, 2019, SCEC Contribution #9760, 2019 SCEC Annual Meeting Poster #069

The time right after a large earthquake, as well as other periods of increased earthquake activity, can reveal spatio-temporal evolution of seismicity and give insight into the physical mechanism of earthquake triggering and aftershock forecasting. Large earthquakes are typically followed by numerous aftershocks, some of which are readily identifiable as earthquakes using automatic methods such as the matched filter technique. However, the standard methods often fail to detect all of them because many aftershocks overlap in time. This study uses a Moveout-Based Matching Pursuit (MBMP) method to detect earthquakes whose waveforms overlap in time at individual recording stations. The inputs include a dictionary consisting of known earthquake waveforms (templates) in a region, and an observed signal at each station. The MBMP algorithm is a greedy algorithm which iteratively chooses the next best dictionary template found in the overlapping signal and its time shift, by minimizing the l2 norm of the residual signal. The residual signal of a template at a certain time shift is the part of the query signal orthogonal to the template at that time shift, subtracted from the query signal. The algorithm searches for the best template match, as well as scaling factor from the projection, and time shift with the constraint that the arrival times of a template at each station are within an acceptable range of moveouts from one another. Waveforms following the 2004 Mw 6.0 Parkfield earthquake recorded at several stations in the HRSN network with a high pass filter of 2 Hz are used as the primary dataset for this study. Initial results on both synthetic and real-world overlapping signals indicate that the algorithm can successfully detect overlapping earthquakes with different magnitudes. Repeating earthquakes are detected especially well, as their waveforms are nearly identical. Earthquakes with significantly lower cross-correlation matches can be successfully detected as well. The effectiveness depends on factors such as the spatial density of earthquakes in the dictionary, station coverage, and the signal-to-noise ratio of the observed signals.

Citation
Daniels, C., Peng, Z., McClellan, J., & Zhu, L. (2019, 08). Identification of Overlapping Earthquakes using a moveout-based matching pursuit method. Poster Presentation at 2019 SCEC Annual Meeting.

Related Projects & Working Groups
Seismology