Tremor Hypocenters Form a Narrow Zone at the Plate Interface in Two Areas of SW Japan
Wednesday, 16 December 2015
Poster Hall (Moscone South)
The tremor detectors developed for accurately locating tectonic tremor in Cascadia [Armbruster et al., JGR 2014] have been applied to data from the HINET seismic network in Japan. In the overview by Obara [Science 2002] there are three strong sources of tectonic tremor in southwest Japan: Shikoku, Kii Pen. and Tokai. The daily epicentral distributions of tremor on the HINET web site allow the identification of days when tremor in each source is active. The worst results were obtained in Shikoku, in spite of the high level of tremor activity observed there by others. This method requires a clear direct arrival of the S and P waves at the stations for coherence to be seen, so scattering and shear wave splitting are possible reasons for poor results there. Relatively wide station spacing, 19-30 km, is another possible reason. The best results were obtained in Tokai with stations STR, HRY and TYE spacing 18-19 km, and Kii Pen. with stations KRT, HYS and KAW spacing 15-22 km. In both of those areas the three station detectors see strong episodes of tremor. If detections with three stations are located by constraining them to the plate interface, a pattern of persistent sources is seen, with some intense sources. This is similar to what was seen in Cascadia. Detections with four stations give S and P arrival times of high accuracy. In Tokai the hypocenters form a narrow, 2-3 km thick, zone dipping to the north, consistent with the plate interface there. In Kii Pen. the hypocenters dip to the northwest in a thin, 2-3 km thick, zone but approximately 5 km shallower than a plate interface model for this area [Yoshioka and Murakami, GJI 2007]. The overlap of tremor sources in the 12 years analyzed here suggests relative hypocentral location errors as small as 2-3 km. We conclude that the methods developed in Cascadia will work in Japan but the typical spacing of HINET stations, ~20 km, is greater than the optimum distance found in analysis of data from Cascadia, 8 to 15 km.