Probing the Nature of Low-Frequency Earthquakes Through the Deconvolution of Tectonic Tremor

We resolve the low-frequency earthquake (LFE) activity within tremor from a compact area beneath southern Vancouver Island. Using LFE templates made from stacking thousands of nearly co-located LFEs as empirical Green's functions, we detect LFEs using time-domain iterative deconvolution of the “optimal” horizontal component (the dominant particle motion direction) of tremor velocity seismograms. The deconvolution is guided by the correlation between seismic stations. During identified tremor bursts, our three-station catalog yields more than one detection per second. Sixty percent of these are validated at a fourth station, but judging from the consistency of their migration patterns, many of the rejected detections are also legitimate. Waveforms predicted from these detections simultaneously reduce the residual on the unused vertical and “orthogonal” horizontal components of the seismograms. The unresolved portions of the seismograms are likely times of significant tremor activity outside the target area. Deconvolution reveals clear LFE migrations with speeds near 5–6 m/s during 20% of the bursts, consistent with contemporaneous 4-s tremor detections in the same region. Given the number of consecutive seismogram peaks with detections, it is doubtful that each detection represents just one LFE. More likely, tremor is usually saturated in time with LFEs that are on average separated by less than the characteristic LFE duration. Such time saturation is consistent with stochastic models of tremor generation designed to explain tremor spectra, and also disguises the real number of LFEs within tremor, making it difficult to constrain the seismic moment and duration of individual LFEs.