Geophysical Research Letters

A persistent localized microseismic source near the Kyushu Island, Japan



[1] Very strong signals with apparent velocity higher than Rayleigh wave velocity are observed on noise correlation functions (NCFs) between seismic stations in East Asia. These signals are present on one-month NCFs in ten years period with stable arrival times, indicating their persistent and localized nature. The signals are strong in the frequency band of 0.07–0.12Hz, and their amplitudes show inter-annual but not seasonal variation. Location obtained from two algorithms with GSN and FNET data indicates that the source is situated in Kyushu Island, Japan. After an earthquake is used to account for heterogeneity effects, the location is closer to the Ariake bay but still in the island. The non-seasonal amplitude variation and the peak frequency of 0.1Hz suggest the signals are probably not generated by oceanic sources. This persistent localized microseismic source needs to be taken into consideration in ambient noise tomography studies in East Asia.

1. Introduction

[2] Even when earthquakes are absent, seismometers still register constant ground motion referred to as ambient seismic noise. Usually, the noise energy in the frequency band of (0.05–0.1 Hz) and (0.1–0.3 Hz) are called as primary and secondary microseisms respectively. Secondary microseism is much stronger and is generated by nonlinear interaction between two ocean waves travelling in opposite directions [Longuet-Higgins, 1950; Kedar and Webb, 2005]. In contrast, primary microseism is weaker and is caused by interaction between ocean waves and ocean bottom in coastal regions [Hasselmann, 1963]. But there are still controversies regarding whether microseisms are generated in open ocean or only in coastal regions [Stehly et al., 2006; Yang and Ritzwoller, 2008; Gerstoft et al., 2008; Landès et al., 2010].

[3] In recent years, microseisms are found to be valuable for imaging interior of the Earth via the estimated Green's function (EGF) obtained from noise correlation function (NCF) between two seismograph stations [Lobkis and Weaver, 2001; Derode et al., 2003a, 2003b; Snieder, 2004]. Surface wave tomography based on EGF has been applied at scales ranging from regional to global, and provides much higher resolution as compared to earthquake surface wave tomography [Shapiro et al., 2005; Yao et al., 2006; Cho et al., 2007; Lin et al., 2008; Nishida et al., 2008, 2009; Zheng et al., 2008]. Recently there are also reports of body waves extracted from NCFs [Roux et al., 2005; Zhan et al., 2010].

[4] Various studies have demonstrated that NCF can be used as good approximate of true Green's function only when the noise sources are randomly distributed or the noise wave field is diffusive [Snieder, 2004; Yang and Ritzwoller, 2008; Yang et al., 2007]. However, from modeling of precursors to inter-station Rayleigh waves, researchers find some areas particularly effective in generating microseisms, for example, southern Italy [Gu et al., 2007] and northern Italy [Marzorati and Bindi, 2008]. In order to suppress effects due to inhomogeneity of noise sources, Yang and Ritzwoller [2008] proposed that sufficiently long time series of ambient seismic noise need to be used for computing EGF. They found that typically one year is long enough to average out bias in EGF due to seasonal variation of noise sources. They also noticed that EGF would fail to approximate true Green's function if there are persistent localized microseismic sources. Indeed, there are very few reported persistent localized sources. The 26 sec microseismic source in the Gulf of Guinea (and probably its antipode counterpart in the North Fiji basin) is probably the only such source well established [Shapiro et al., 2006].

[5] Here we report another case of persistent localized microseismic source in southwestern Japan. This source shows up with high signal noise ratio on NCFs between stations in China, Korea and Japan, and it features inter annual but not seasonal variation. Since East Asia is one of the most active tectonic region in the world, ambient noise surface wave tomography has been applied in this region and many interesting results have been published [Cho et al., 2007; Nishida et al., 2008; Zheng et al., 2008]. This persistent localized microseismic source may bias surface wave dispersion measurement for certain paths.

2. Observations of Localized Microseismic Source

[6] We collected continuous vertical component waveform data from the New China Digital Seismograph Network(IC) and IRIS/USGS network between 1999 and 2009 (open triangles in Figure 1a, referred to as GSN stations hereafter), and waveform data from FNET of Japan for the months of June and December 2009. The continuous waveforms are divided into one-day-long segments, then instrument responses, linear trend and mean are removed. Earthquake signals are suppressed with the running-absolute-average method proposed by Bensen et al. [2007]. The spectral whitening procedure was applied between 0.02 and 0.3 Hz, which covers the frequency band if both primary and secondary microseisms. After single station data preprocessing, daily cross-correlations between stations were computed and then stacked to enhance signal noise ratio of NCFs.

Figure 1.

(a) GSN stations (open triangles) and FNET stations (solid triangles) used in this study. (b) One-year NCFs of SSE-INCN from 2006 to 2009. The Rayleigh waves propagating along inter-station great circle path are highlighted with gray shadows. The PL signals are marked by the box. (c) Composite record section of NCFs between FNET stations and two GSN stations (SSE, INCN). The arrivals of inter-station Rayleigh waves are denoted with dashed gray lines. The upper and lower solid lines show the PL signal on the NCFs of SSE-FNET, INCN-FNET respectively.

[7] We computed one-year NCFs among GSN stations (referred to as GSN NCF dataset hereafter), and one-month NCFs between FNET stations and two GSN stations (SSE, INCN) (referred to as FNET NCF dataset hereafter). All the NCFs show high signal noise ratio, and 22 one-year NCFs as well as 146 one-month NCFs were obtained. Examples of one-year NCF between SSE and INCN are shown in Figure 1b. Obviously, a significant energy (around lag time of 120 seconds) arrives well before the Rayleigh waves (indicated with gray bars). This signal is constant in arrival time from 2006 to 2009, demonstrating its persistent nature. Also because of its short duration, the source region for this signal should be fairly localized, similar to the case of the 26 second source [Shapiro et al., 2006]. We also observed such signals in the NCFs between other GSN stations. Similar signals also show up in one-month NCF between FNET stations and two GSN stations (SSE, INCN) during June, 2009. For convenience, this signal is referred to as the PL (persistent localized) signal later on.

3. Location of the Microseismic Source

[8] Various methods have been used in microseismic source locations, including the methods of travel-time misfit minimization [Shapiro et al., 2006], back projection [Stehly et al., 2006, Yang and Ritzwoller, 2008], migration [Gu et al., 2007]. Here, we employed the grid-search algorithm based on energy stacking to obtain the source location. This method is essentially the migration method. As a first step, we assumed that surface wave speed is homogeneous in the study region, so that travel time between a seismograph station and a candidate source location can be easily computed. Homogeneous surface wave speed assumption is also adopted by Shapiro et al. [2006] to locate the 26 second source. Then the L2 norm energy is stacked in a time window of NCFs, and center of the time window is set according to the difference between arrival times of the two stations for the candidate source location (equation (1)). The length of time windows is set as 50 second which is close to the duration of the PL signal. The candidate source location is obtained with spatial grid spacing of 0.1 degree in both latitude and longitude. Since the period of the PL signal shows dominant period about 10 seconds, we chose the average velocity ranging from 2.25 to 3.5 km/s with a 0.25 km/s step. The candidate location with highest stacked energy (E in equation (1)) is assumed to be the location of this microseismic source. We applied the location algorithm to the GSN NCF dataset and FNET NCF dataset separately, since these two datasets are processed for different time span (1 year for GSN data set, and 1 month for FNET dataset). The source location was searched in a rectangular region defined by (25–40°N, 120–140°E)

equation image

where CCij is the NCF between i-th and j-th stations, tij = di/vdj/v, diis the distance between i-th station and the candidate source location, and v is the velocity.

[9] Figure 2 shows the location results. For the GSN NCF data set (Figure 2a), the maximum energy location is near the Kyushu Island for different velocities. Much sharper stacked energy is achieved with the FNET NCF data set (Figure 2b). Both datasets show the largest stacked energy for candidate source in northeastern part of Kyushu Island, Japan (v = 2.75 km/s). This velocity is consistent with Rayleigh wave velocity for period of 10s from surface tomography studies (Zheng et al., 2008). For the FNET NCF dataset, the stacked energy reaches maximum at (33°N, 131°E) with velocity of 2.75km/s, which is probably the location for the PL source. When NCFs from the FNET dataset are corrected for the difference between the travel times of the two stations assuming the location of (33°N, 131°E), the PL signals line up at zero, suggesting that the source location is probably correct (Figure S1).

Figure 2.

Results of the grid-search location of the PL signal source. (a) Result from one-year of GSN dataset between 2006 and 2009. (b) Result from one-month (June, 2009) FNET dataset. The average velocities used in grid-search are also shown.

[10] However, the energy stack algorithm has some drawbacks. When the PL source and two stations are situated on a great circle, the inter-station Rayleigh wave from ambient noise and the PL signal coincide, therefore making the energy stack algorithm fail. To improve the reliability, we verified the result by the method of travel-time misfit minimization [Shapiro et al., 2006] with the GSN NCF dataset. The arrival was picked at the maximum of PL signal envelope. Only when the PL signal is separate from the inter-station Rayleigh waves, its arrival is included for location. The spatial grid search range is same as the one used in the energy stacking method. The location result agrees with that from energy stacking method (Figure S2). The little difference of velocity (2.75 km/s for the energy stacking method, and 3.0 km/s for the travel time minimization method) could due to the different objective function and errors in picking arrivals.

4. Discussion

[11] Microseism source location provides important information of its generation mechanism. Various studies demonstrated that microseism sources are either in open ocean or coastal region [Stehly et al., 2006; Yang and Ritzwoller, 2008]. But the PL source seems to be situated on Kyushu Island, instead of in the water body. If the PL source is indeed on land, then the source is not due to ocean waves. However, we only assumed very simple surface wave velocity model in locating the PL signal source, the location error is probably large. Therefore the PL source could be in water body. In order to locate the source better, we used an earthquake as calibration. We measured Rayleigh wave group velocity (Figure S4) for an earthquake near Kyushu Island (Figure S5). Indeed surface wave velocity shows substantial lateral variation (Figure S4), and the path from Kyushu Island to station SSE shows lower group velocity (2.5 km/s around period of 12 second) as compared to velocities greater than 2.6 kms/ for other paths. With the propagation path effects taken into account, we relocated the PL source and it shifts about 50 km westward, but it is still situated on the island. However, location accuracy with energy stack mechanism is not very high because the PL signal has finite duration (about 50 seconds). Similar problem arises in the study of the 26 second microseism near Gulf of Guinea [Shapiro et al., 2006]. And relocation with earthquakes also has errors from effects due to source terms on surface wave propagation [Levshin et al., 1999]. Probably location accuracy for the PL signal is worse than 50 km, therefore the PL source could still be in water bodies near Kyushu (such as Philippine sea to the east, the Ariake bay to the west). We also located the PL signal source with FNET NCF data set for December 2009 (Figure S3), and the location is identical to that located with data from FNET NCF data set for June 2009, also arguing that the PL signal source location does not change over time.

[12] The temporal variation of microseism strength is another important clue for understanding the generation mechanism, for example the seasonal variation of the 26 second microseism strength is used to argue its origin in ocean [Shapiro et al., 2006]. We presented evidences of temporal variation of the PL signal in Figure 3. The PL signal typically shows up on ten-day NCFs for station pair of SSE-INCN (Figure 3a). The PL signals were highlighted with gray bar, and obviously the PL signal shows temporal variation. It seems that the signal is very strong from 60–150th days of year 2006, in contrast to strongest microseisms in winter [McNamara and Buland, 2004]. Figure 3b displays strength variation of the PL signal on ten-day NCF from year 2006 to 2009. The PL strength is taken as the energy for lag time window of 100–150 second. Seasonal variation is not clear on four years of NCF (Figure 3b). For year 2007 and 2008, the PL signal is strong from 120–240th day. For year 2009, the signal is strong around 120th day. Though the strength variation is clear on ten-day NCFs, the arrival time and shape of the PL signal are very stable over the years (Figure 3c). Similar pattern is also observed on NCF for station pair SSE-BJT (Figure S8). Inter annual variation in the PL signal strength is displayed in Figure S6b, which shows stronger PL signal (in the box) from 2005–2009 and weaker signal from 1999–2002), with the PL signal particularly weak in 2001. Indeed, we used the relative strength of PL signal which could be affected by seasonal and inter annual variation of inter-station Rayleigh waves. But seasonable amplitude variation of inter-station Rayleigh wave can only produce false seasonal pattern of PL signals, and will not produce non-seasonal patterns on PL signals. Moreover, for frequency lower than 0.05 Hz, the inter-station Rayleigh wave in one-year NCFs seems to be stable (Figure S7a.). Unlike the 26 second microseism signal which show clear seasonal pattern, this PL source seems not strongly affected by season factors such as winter storm or tropical storm. Therefore, this PL source is probably not related to oceanographic phenomena.

Figure 3.

(a) Ten-day NCFs of SSE-INCN during 2006. Gray bar indicates the PL signal. (b) The temporal variation of the PL signal strength from 2006 to 2009. (c) The PL signal in one-year NCFs from 2006 to 2009. The shape and arrival of the PL signal are almost identical over the years. (d) The amplitude spectrums computed from one-year NCFs of SSE-INCN. The dominant frequency ranges from 0.07 to 0.12 Hz. Amplitude spectrums are also stable over years.

[13] The dominant frequency of microseisms may provide clues to their generation mechanisms. For example, secondary microseism has peak frequency around 0.2Hz, and can be explained as result of interacting of standing ocean waves which feature dominant period 10 seconds [Young, 1999]. The primary microseism has peak frequency about 0.06Hz [Peterson, 1993], and is proposed to be generated in coastal regions due to the pressure on the ocean floor caused by the ocean waves. The PL signal on NCFs shows dominant frequency of 0.07–0.12 Hz, which falls between the frequency bands of primary (< 0.1 Hz) and secondary (>0.1 Hz) microseism (Figure 3c). Therefore, if the PL signal is generated by oceanic sources, the excitation could be due to the mechanism of primary microseism (shoaling) or secondary microseism (standing ocean waves). Either way, we might observe strong 0.1Hz signals for stations in Kyushu Island. We computed daily spectrogram of the microseism recorded by station TKD and TMC for year 2009 (Figure S6). However, it seems there is no good correspondence between PL signal strength (black lines) and microseismic energy in the frequency band of 0.07–0.12Hz at TKD and TMC. Instead there seem to be anti-correlation around the 100th day, when PL signal is strong, but microseimic spectral amplitudes at TKD and TMC are weak. The anti-correlation implies that the PL signals are probably not generated from oceanic processes. And the non-oceanic origin of the PL signals may also explain their peak frequency of 0.1Hz, in contrast to microseisms due to ocean waves which show minimal energy around 0.1Hz [Peterson, 1993].

5. Conclusion

[14] We probed a persistent localized microseism source on or near the Kyushu Island, Japan with continuous waveform data from seismic stations in East Asia. This source shows inter annual but not seasonal variability in strength, and the source shows stability in location over a span of ten years. The signal features dominant frequency in 0.07–0.12Hz, between the primary and secondary microseism frequency peaks, suggesting that the signals are probably not caused by ocean waves. If it is generated from oceanic sources, probably it is due to the particular bathymetry near Kyushu Island in resonance to ocean wave of the same dominant period [Shapiro et al., 2006]. However we are not sure about its generation mechanism, and more data (buoy data, barograph data) need to be cross-examined to understand its origin better. Since the signal is very strong on NCFs in 0.07–0.12Hz, special attention should be paid to Rayleigh wave dispersion measurement for station pairs aligned close to the great circle path connecting the stations and the persistent localized source on or near Kyushu Island.


[15] Supported by NSFC fund 40821160549, 41074032, CAS fund KZCX2-YW-116-1 and CEA fund (200808078). Data provided by IRIS and F-NET.