Wavefield and source spectra of non-volcanic low-frequency tremors in a southwest Japan subduction zone

We investigate wave-field properties and source spectra of non-volcanic low-frequency tremors in a southwest Japan subduction zone, which have not been fully understood due to a low signal-to-noise ratio of tremor signals. Geological Survey of Japan, AIST, has recently started an integrated borehole observation of water levels, strains, tilts, water temperatures and seismic waves in southwest Japan for monitoring of the anticipated Tonankai and Nankai earthquakes. At each observation site, we have drilled three boreholes with different depths (about 30 m, 200 m and 600 m) and installed high-sensitivity seismometers at a bottom of every borehole. In this study, we use borehole seismic data recorded by this vertical seismic array network as well as NIED Hi-net. In order to investigate wave-field properties of non-volcanic low-frequency tremors, the vertical array data are used. During tremor activities, high semblance values continuously appear at a specific apparent velocity. Because the S-wave velocity estimated by borehole logging is consistent with the apparent velocity, we conclude that the non-volcanic low-frequency tremors are composed of near vertically incident S-waves. This is supported by polarization analysis, where the particle motion is linearly polarized and the incident angle is steep. The polarization analysis also reveals that the polarization angles are almost same during each tremor activity. Because the polarization angle will have various values if the tremor signals contain large amount of scattered waves, it is likely that they are composed mostly of seismic waves coming directly from the source. A comparison of source spectrum between regular earthquakes and non-volcanic low-frequency tremors is a key to understand the difference in their physical processes. We first estimate model source spectrum for co- located regular earthquakes by Multi-Window Spectral Ratio method [Imanishi and Ellsworth, 2006]. By averaging the ratios between observed and model spectrum over all events in a cluster, we obtain the average attenuation function, which is assumed to represent the path and site effects between the source and station. The source spectrum of each tremor is then calculated by the ratio between the observed spectrum and the attenuation function. The estimated velocity spectra of tremors are flat for higher frequency, which clearly differ from those for regular earthquakes that follow f-1 decay at high frequency. It is noted that this feature of the velocity spectra can be explained by the Brownian walk model for slow earthquakes that was recently proposed by Ide [2008].