Non-volcanic seismic tremor was discovered in the Nankai trough subduction zone in southwest Japan and subsequently identified in the Cascadia subduction zone. In both locations, tremor is observed to coincide temporally with large, slow slip events on the plate interface downdip of the seismogenic zone. The relationship between tremor and aseismic slip remains uncertain, however, largely owing to difficulty in constraining the source depth of tremor. In southwest Japan, a high quality borehole seismic network allows identification of coherent S-wave (and sometimes P-wave) arrivals within the tremor, whose sources are classified as low-frequency earthquakes. As low-frequency earthquakes comprise at least a portion of tremor, understanding their mechanism is critical to understanding tremor as a whole. Here, we provide strong evidence that these earthquakes occur on the plate interface, coincident with the inferred zone of slow slip. The locations and characteristics of these events suggest that they are generated by shear slip during otherwise aseismic transients, rather than by fluid flow. High pore-fluid pressure in the immediate vicinity, as implied by our estimates of seismic P- and S-wave speeds, may act to promote this transient mode of failure. Low-frequency earthquakes could potentially contribute to seismic hazard forecasting by providing a new means to monitor slow slip at depth.

An automatic detection and a precise picking of the arrival times of seismic waves using digital seismograms are important for earthquake early detection systems. Here we suggest a new method for detecting and picking Pand S-wave signals automatically. Compared to methods currently in use, our method requires fewer assumption with properties of the data time series. We divide a record into intervals of equal lengths and check the "local and weak stationarity" of each interval using the theory of the KM 2 O-Langevin equations. The intervals are stationary when these include only background noise, but the stationarity breaks abruptly when a seismic signal arrives and the intervals include both the background noise and the P-wave. This break of stationarity makes us possible to detect P-wave arrival. We expand the method for picking of S-waves. We applied our method to earthquake data from Hi-net Japan, and 90% of P-wave auto-picks were found to be within 0.1 s of the corresponding manual picks, and 70% of S-wave picks were within 0.1 s of the manual picks. This means that our method is accurate enough to use as a part of the seismic early detection system.

SUMMARY Since the discovery of deep low‐frequency tremors in southwestern Japan, various types of low‐frequency oscillation phenomena have been reported and analysed. Among these phenomena, non‐volcanic, deep, low‐frequency tremors in southwestern Japan are outstanding low‐frequency oscillation phenomena occurring over a wide region along the Philippine Sea Plate. However, the weakness of these signals makes it difficult to apply usual analyses to derive source characteristics. In this paper, we apply a new algorithm to obtain a frequency structure from highly noisy data and reveal the characteristic frequency structure of deep low‐frequency tremors with peaks lined up from 1 to 5 Hz at intervals of 0.5 Hz. In addition, we present one piece of evidence that the deep low‐frequency tremors and low‐frequency earthquakes have a common source mechanism.