Identification of low-frequency earthquakes in non-volcanic tremor using the subspace detector method

Shelly et al. identified 188 low frequency earthquakes (LFEs) in a one-hour episode of tremor recorded at Japanese Hi-Net borehole stations. Later they proposed that non-volcanic tremor may consist entirely of a sequence of LFEs. We search for additional LFEs using the subspace detector technique. This method uses a matrix of template waveforms and, via singular value decomposition, builds a set of basis vectors that, in some linear combination, can reproduce the templates or similar events that fall within the subspace. We explore the utility of the method to search for additional LFEs within the subspace spanned by our basis vectors. The results compare well with previous LFE detections. We also compare our results with an independent measure of signal polarization and find that the polarization takes on a distinct character during times of known LFEs. This suggests that signal polarization may also have potential for tremor detection, characterization, and monitoring.