A total of 1314 very low frequency earthquakes (VLFEs) were identified along the Ryukyu trench from seismograms recorded at broadband networks in Japan (F-net) and Taiwan (BATS) in 2007. The spectra of typical VLFEs have peak frequencies between 0.02 to 0.1 Hz. Among those, waveforms from 120 VLFEs were inverted to obtain their centoroid moment tensor (CMT) solutions and locations using an examination grid to minimize a residual between the observed and synthetic waveforms within an area of 11° × 14° in latitude and longitude and at depths of 0 to 60 km. Most of the VLFEs occur on shallow thrust faults that are distributed along the Ryukyu trench, which are similar to those earthquakes found in Honshu and Hokkaido, Japan. The locations and mechanisms of VLFEs may be indicative of coupled regions within the accretionary prism or at the plate interface; this study highlights the need for further investigation of the Ryukyu trench to identify coupled regions within it.
 Unlike other parts of Japan where thrust earthquakes with a moment magnitude (Mw) greater than 8.0 occur along major subduction boundaries, the Ryukyu trench, at the northwestern boundary of the Philippine Sea plate, has had no known thrust earthquakes of similar magnitude in the last 300 years [Ando et al., 2009]. The Japan islands are surrounded by three major subduction plate boundaries between the Pacific, Philippine Sea, and Eurasian or Amur plates [Heki et al., 1999], each with many historic earthquakes with an Mw greater than 8.0. Therefore, because the Ryukyu trench is not associated with large thrust earthquakes, there is an assumption that the trench is aseismic [Wei and Seno, 1998; Kao, 1998]. If that is the case, the Ryukyu trench can be grouped with other Mariana-type subduction zones, as defined byUyeda and Kanamori , because of the presence of back-arc spreading [Kimura, 1985; Sibuet et al., 1987; Nakamura, 2004]. Nevertheless, in 1771 a large tsunami struck Ishigaki island with the run-up height up to 30 m. The 1771 source is still controversial and is suggested to be a tsunami earthquake (Mw = 8.0) near the trench axis as the result of a thrust earthquake along the Ryukyu subduction zone [Nakamura, 2009] or as a local earthquake (Mw = 7.2) accompanied with a submarine landslide [Goto et al., 2010].
 The seismicity recorded from the Ryukyu trench indicates that the earthquakes are distributed along the Wadachi-Benioff zone extending to a depth of approximately 300 km (Figure 1) [Nakamura et al., 2003]. The zone has a dip of approximately 30° to 40°, except above a depth of 50 km. The analysis of focal mechanisms demonstrates that the dominant stress provinces are an arc-parallel extensional stress in the forearc, and an extensional stress field in the Okinawa Trough along the Ryukyu trench [Kubo and Fukuyama, 2003]. Normal faults are dominantly oceanward of the trench axis and within the subducting slab. These features are similar to other subduction zones that exhibit back-arc rifting.
 Non-volcanic tremors were first found along both the Nankai trough [Obara, 2002] and the Cascadia margin [Rogers and Dragert, 2003] using seismic networks equipped with high-gain qualified seismometers, while slow-slips events (SSE) were found along both the Nankai trough [Ozawa et al., 2001] and the Cascadia margin [Dragert et al., 2001] using GPS networks. Subsequently, VLFEs were identified in a group of deep, low frequency earthquakes [Ide et al., 2007; Ito et al., 2007], while shallow VLFEs were recorded south of both Honshu and Hokkaido [Obara and Ito, 2005; Ito and Obara, 2006; Asano et al., 2008]. These VLFEs were located at shallow depths near either the slab interface or within the accretionary prism. Low-frequency earthquakes documented in the studies above have a frequency band predominantly between 1 and 4 Hz, whereas VLFEs characteristically have frequency bands between 0.02 and 0.1 Hz. Despite the variety of names used for these earthquakes, these events have a close relationship and, perhaps, a common origin [Schwartz and Rokosky, 2007; Ide et al., 2007; Beroza and Ide, 2011]. A common characteristic of the VLFEs is longer duration thrust fault motions near the plate interface, when compared to ordinary earthquakes with comparable seismic moments.
 In this study, we focus on the identification of VLFEs using broadband data from the F-net network along the Ryukyu trench supplemented with data from the BATS network in Taiwan [Kao et al., 1998]. We anticipate that the identification of VLFEs could help elucidate the nature of coupling along the Ryukyu subduction zone.
2. Data and Methods
 Our analysis uses data from 2007 when the data quality from the BATS network first became suitable; all of the seismological data were recorded at broadband stations. Our analysis found that VLFEs were not recognized on the unfiltered, raw broadband seismograms because the short-period (0.2–1.0 Hz) noise levels are too high compared with VLFE signals (0.02–0.1 Hz; Figure S1 in theauxiliary material). First we applied two filters, 0.02–0.05 Hz bandpass and 1 Hz highpass, to all raw broadband seismograms. After identifying low-frequency events from the 0.02–0.05 Hz bandpass filtered seismograms, local and teleseismic earthquakes were removed using the PDE and hypocenter catalogs from the Japan Meteorological Agency and the Central Weather Bureau (CWB) of Taiwan. Then, from these low-frequency events, we removed local small events that are not listed in any hypocenter catalogues using the 1 Hz highpass filtered seismograms. In this process we removed 10 normal small events as a local event, which is determined to be smaller than M2.5. The first arrivals of approximately 1,314 VLFEs were selected from a station above the Ryukyus trench (Figure 1). Figure 2depicts an example of a series of VLFEs that were recorded by the F-net and BATS networks along a distance of 3,000 km. The amplitude and arrival times recorded at stations YNG and IGK for this earthquake imply that the VLFEs occurred proximal. Along the Ryukyu trench, the majority of the VLFEs form a swarm that can continue for 1 to 3 hours, as depicted inFigure 2; however, isolated events that are bigger than the swarm-type events, occur occasionally (Figure S1).
3. Hypocenters and Mechanisms
 The inversion program of Nakano et al. was used to systematically locate hypocenters and determine mechanisms of the VLFEs. This program performs the inversion assuming a pure double-couple source mechanism. According to the numerical experiments byNakano et al. , the assumption of double-couple is crucial for stable estimations of the source location and focal mechanism for the inversion of data with small number of stations. In this method, a target earthquake is assumed to occur at a spatially distinct grid point. Assuming a double-couple model, the strike, dip, and rake angles are measured at each node using theP-wave andS-wave velocity structural model (AK135) ofKennett et al. .
 First, the Green's functions of full waves were calculated at an interval of 0.5°in an area between 22° and 30°N and 122° and 132°E at depths of 5, 10, 20, 30, 40, 50 and 60 km. To conduct a stable inversion we selected 120 events that occurred under low ambient noise levels and that had maximum velocity amplitudes greater than 15 nm/s. After roughly locating the VLFEs within the map (Figure 1), the horizontal grid spacings were reduced to 0.2° within Areas 1 to 3. We chose a pair of stations for each area: YNG and IGK for Area 1, ZMM and KGM for Area 2 and AMM and KYK for Area 3. These stations are located at epicentral distances less than 300 km, which are appropriate for our analysis. The location and centroid moment tensor (CMT) solution for an example VLFE are depicted in Figure 3. The best-fit model requires motion along a reverse fault with anMw of 4.0 located southeast of Ishigaki Island near the trench axis. Considering the location and mechanism, this event occurred either in the accretionary prism or at the plate interface along the Ryukyu trench and could be similar to events recorded off Central Honshu [Ito and Obara, 2006].
 We applied this methodology to obtain the locations and focal mechanisms for 120 VLFEs. We evaluated and ranked the reliability of our results as ranks A (good), B (fair) and C (poor) based on three criteria, including, residual, spatial resolution and waveform (see Table S1 and Figures S2a–S2c). The number of VLFEs assigned to ranks A, B and C are 22, 66 and 32, respectively. This study only used events with ranks A and B for discussion. VLFEs of ranks A and B are mainly distributed near the trench axis (Figure 4) whereas VLFEs of rank C were scattered and located away from the trench axis. The latter have complicated waveforms and occasionally have normal fault solutions, possibly due to the misidentification of the phase. The focal mechanisms of the 22 VLFEs of rank A are depicted in Figure 4.
 Because we used only two stations for each inversion, the azimuthal coverage of the stations is poor, and consequently, the depth between 0 and 30 km cannot be resolved for events near the trench axis. Most solutions show thrust faulting but a few strike-slip faulting are also found (Figure 4). These strike-slip solutions could suggest that strike-slip VLFEs occur along the Ryukyu trench. Considering the number of stations used in the inversion and the poor azimuthal coverage, errors cannot be discounted. Nevertheless, it is likely that these VLFEs are distributed at shallow depths, possibly in the accretional prism or around the deep seismic plane.
 Even though the location and mechanism of a VLFE cannot be determined with high accuracy, its location is defined roughly by region, based on arrival times. Assuming that VLFEs are located near the trench axis, the approximate locations were determined from arrival times at the stations. The VLFEs were located as follows, 638 in Area 1, 260 in Area 2 and 416 in Area 3. The VLFEs identified in Area 1 have a simple waveform; the VLFEs in other areas, however, have complicated waveforms with several peaks and small amplitudes ranging from approximately 1/3 to 1/5 of typical VLFEs in Area 1.
 There is another active VLFE area outside of this study area, located east to southeast of Kyushu, first identified by Obara and Ito . More than 1,200 VLFEs were found in this area in 2007. These events were often triggered by large teleseismic or local events that continued for several days. We did not, however, focus on this area because the VLFEs were numerous, smaller and the recurrence intervals of their activity are quite short. A detailed study is needed to clarify the activity of this area.
 We attempted to estimate the magnitude-frequency relationship for the three areas; however, this was completed for Area 1 only due to the limited number of available data with CMT solutions. First, we identified an empirical relationship between magnitude and amplitude of VLFEs recorded at YNG and IGK (Figure S3). Appling this relationship to events in Area 1, we determined a b-value of 1.7 (Figure S4). This b-value is significantly higher than that typical of ordinary earthquakes and even higher than values obtained from volcanic swarms [Farrell et al., 2009]. The high b-value could be due to the high pore pressure, low applied stress or the presence of heterogeneous materials as measured in laboratory experiments [McNutt, 2002]. Another explanation for the high b-value may be the narrow range of magnitudes (3.3 to 4.5) that were available in this study.
Heki and Kataoka  concluded that the episodic slow slip events (SSE) found in the Ryukyu subduction zone occur every six months at depths of 40 to 50 km. This depth range is comparable to the transition zone between the stable sliding and the coupled or the seismic areas in other seismogenic subduction zones [Dragert et al., 2001; Ozawa et al., 2001]. The accumulated SSEs are similar to the relative plate motion between the Eurasia and Philippine Sea plates at depth. Furthermore, Nakamura  found a similar event west of the SSE area identified by Heki and Kataoka . Considering the association of other major plate boundaries with SSEs and historical large events, it is very possible that the Ryukyu subduction zone has a coupled portion. Further studies are necessary, particularly those that will locate VLFEs in the Ryukyu trench more accurately and to ascertain the extent of the coupled portion of the Ryukyu trench.
 VLFEs along the Ryukyu trench were analyzed systematically using seismograms from broadband networks F-net and BATS. Seismograms were band-path filtered (0.02–0.05 Hz); next, local and teleseismic events were removed from identified events using the hypocenter catalogs. Spectra of typical VLFEs had peak frequencies between 0.02 and 0.1 Hz. The CMT solutions and locations were obtained for these events using inversion techniques. Most of these events were distributed near the trench axis. The location and depth of the VLFEs suggest that these earthquakes occurred around the plate interface along the Ryukyu trench. Furthermore, considering the recent identification of slow slip events in the western Ryukyu trench from GPS observations, it is inferred that the subduction in the Ryukyu trench is similar to the subduction occurring beneath Central Honshu and Hokkaido, Japan.
 This work used GMT and SAC software. Broadband seismograms from the F-net and BATS networks were used in this study. We thank M. Nakano for useful discussions and two anonymous reviewers for constructive comments. This study was supported by NSC grant 99-2116 M-001-004.
 The Editor thanks the two anonymous reviewers for their assistance in evaluating this paper.