Comparison of source areas of M4.8±0.1 repeating earthquakes off Kamaishi, NE Japan

- Are asperities persistent features ?

Tomomi Okada, Toru Matsuzawa, Akira Hasegawa

ResearchCenter for Prediction of Earthquakes and Volcanic Eruptions, Graduate School of Science, TohokuUniversity, Sendai 980-8578, Japan

Fax: +81-22-264-3292

Phone: +81-22-225-1950

E-mail:

Abstract

M4.8 +- 0.1 earthquakes have occurred regularly since 1957 with a recurrence interval of 5.52 +- 0.68 yrs at the same location of the plate boundary off Kamaishi, Iwate Prefecture. The last event (M4.7) occurred on 11/13/200116:44 (JST) and the previous one (M4.8) on 3/11/199510:30 (JST). Waveforms of these events are similar to each other. To examine the hypothesis that this characteristic earthquake sequence is caused by repeating ruptures of an asperity surrounded by creeping areas, we compared the rupture area of the 2001 event with that of the 1995 event by a waveform inversion method. Estimated spatial extents of the rupture area of the 2001 and 1995 events are almost the same and are estimated to be about 1.5 x 1.5 km^2. The rupture area of the 2001 event are mostly overlapped with that of the 1995 event, although the ruptures of the two events were initiated from different points respectively. The present observations clearly show that the 1995 and 2001 events are caused by the repeating ruptures of the same asperity patch, and support the hypothesis of persistent asperities.

Keyword: earthquake recurrence, repeating earthquakes, subduction zone, interplate earthquake

1. Introduction

Earthquake recurrence is a basal problem for seismology and earth sciences as well as long-term earthquake prediction. Numerical simulations based on the recent friction laws predict that an isolated coupled area surrounded by a creep region will slip with a constant recurrence interval when the creep region slips steadily (e.g. Kato and Hirasawa [1]). For real earthquakes, however, aperiodicities of earthquake recurrence have been observed. For example, Shimazaki and Nakata [2] concluded that the recurrence of large thrust earthquakes around Japan followed a time-predictable model rather than a slip-predictable model or a constant recurrence model.

Characteristic or repeating earthquake sequences along the San Andreas Fault have been investigated in detail. Bakun and McEvilly [3] compared source parameters of the three Parkfield earthquakes (1922, 1934 and 1966) and attempted to explain the aperiodicity of these earthquakes. Small-size repeating earthquake sequences have been found in StoneCanyon[4], Calaveras fault [5], Parkfield [6], and northern Hayward fault [7]. These small-size earthquake sequences were interpreted as repeating ruptures of small isolated patches and, recently, some studies have attempted to construct physical models of repeating microearthquake (e.g. [8], [9], and [10]). Vidale et al. [5] analyzed spectral differences between 18 small repeating earthquakes in Calaveras fault. They showed that events belonging to a cluster with longer recurrence interval had smaller source duration. Nadeau and McEvilly [11] compared the recurrence intervals of repeating earthquakes with surface laser ranging data and concluded that episodic creeps control the recurrence intervals of the small repeating earthquakes.

Many large earthquakes have occurred along the plate boundary east off northeastern (NE) Japan, but no large earthquakes have occurred in the area of N39-40, E142-143 (Fig. 1 (a)). Smaller-size earthquakes, however, occur very actively in this area as in the other areas (Fig. 1 (b)). One possible interpretation for this is that only smaller-size coupled areasthat slip seismically during earthquakes sparsely distributed on the plate boundary there. We [12] have detected in this area, M4.8 +- 0.1 earthquakes which have occurred regularly since 1957 with a recurrence interval of 5.52 +- 0.68 yrs at the same location off Kamaishi, IwatePrefecture (Fig. 2). We interpreted that this characteristic earthquake sequence was caused by repeating ruptures of a coupled area with a dimension of ~1km. Its very regular occurrence is perhaps due to the repeating slips of the isolated coupled area surrounded by the stable sliding area slipping at nearly a constant rate. Based on this interpretation we [14] reported in 1999 that the next event was expected to occur by the end of November 2001 with 99 % probability. An earthquake with M4.8 actually occurred on November 13, 2001 as expected [12]. The coupled area is reflected as a large slip area i.e., asperity at the time of each rupture. Therefore, theasperities by the repeating earthquakes should be consistent with each other, if our interpretation described above is correct. In the present paper, we tested the hypothesis of persistent asperities by comparing slip distributions of the recent two events based on wave form inversions.

2. Hypocenter relocation

First, we precisely relocated hypocenters of the 1995 and 2000 events, the initial points of the ruptures．We adopted the homogeneous station method [15] to avoid the effect of inhomogeneous structure of the crust and upper mantle caused by the use of different station set. We used twelve stations for picking P-wave arrival times and three stations for S-wave arrival times; the station sets were the same for both of the two events. We carefully compared P- and S- wave waveforms of the 2001 event with those of the 1995 event and tried to pick the same phases of the two events. The seismic velocity model routinely used in the TohokuUniversity seismic network [16] was adopted in the calculation of travel times.

Hypocenter of the 2001 event is located about 200m to the west of the 1995 event. Relative location errors are about 50m in north-south direction, 70m in east-west direction, and 100m in depth, respectively.

3. Moment tensors

We determined moment tensors of the two events using regional broad-band seismograms. The method is based on Dreger and Helmberger [17] and we used the program TDMT-INVC developed by Dreger in the inversion (see also [18]). Green’s functions are calculated by using the program FKRPROG by Saikia [19]. The seismic velocity structure used in the calculation of Green’s function is shown in Table 1. This is the same as the structure used in Fukuyama et al. [20].

Waveform data are from the TohokuUniversity’s broad-band seismic network. STS-1 or STS-2 type seismograph is deployed at each station. Waveform data from each station are digitized at a sampling frequency of 100Hz with a wide dynamic range of 20 or 22 bits, and they are transmitted to the observation center by telephone or satellite telemetry system.In the present analysis, waveform data thus recorded were low-pass filtered with a pass-band between 0.02 and 0.05Hz and resampled at a sampling rate of 1Hz. Waveform data were from common three stations (SWU; dist. 68km, baz. 279degrees, HSK; dist. 130km, baz. 338 degrees, HMK; dist. 96 km, baz. 309 degrees) of the Tohoku university's broad-band seismograph network.

Obtained moment tensors of the two events are almost the same (Fig. 3). They are thrust-type solutions with low-angle westward dipping planes, which are consistent with the local geometry of the plate boundary between the subducting Pacific plate and the overlying plate. Non-double-couple component is quite small (3 to 12 %) compared to double-couple component. Scalar moments of the 1995 and 2001 events are 1.05 and 1.12 x 10^23 dyne.cm, respectively.

4. Source processes

We determined moment release distributionson the fault planes of the two events by a waveform inversion method [21] based on the concept of the empirical Green’s function method (e.g., [22], [23], and [24]). Observed waveforms are inverted to determine the source time function of each grid distributed on the fault plane. In the inversion, we used the multiple time window method [25]. We corrected the difference in rise time between the target earthquake and a smaller event used as empirical Green's function and the perturbation of rupture velocity is represented by the multiple time window method. Slip direction of the target earthquake was assumed to be the same as that of the empirical Green's function event over the whole fault plane. Relative source time function for each grid is assumed to be expressed as a sum of isoscales triangles with a base of trise. The interval between the triangles having a half of duration trise. Thus, the unknown parameters are these heights of the triangles. To stabilize the solution, we add smoothing constraints in the inversion. The equation to be solved by the inversion is

,

where A is the matrix of the Green’s function, b is the observed seismograms (data), x is the amplitude of relative moment release (i.e. the heights of the triangles) for each time window of each grid, D is a first-degree differential operator for the time and space domains, and s is a constant to weight the smoothing.

Rupture front is assumed to propagate circularly with a velocity of 3.8 km/s (80% of S-wave velocity at the source depth). We attempted to resolve the rupture velocity using the variance [23]. The adopted range of rupture velocity was 2.6 to 4.4km/s and the rupture velocity of 3.8 km/s gives a comparatively small variance.Amount of relative moment release with duration of 0.1 sec is determined at each grid point and at four time intervals of 0.05s by waveform inversion. Their fault planes were assumed as a low-angle westward dipping plane of the moment tensor solutions. We used a grid net of 3km x 3km with 81 points spaced 300m on the fault plane. An event (8/17/200112:14 M3.1) located near the 1995 and 2001 events was selected as the empirical Green's function. We used the seismic wave velocity model routinely used in the TohokuUniversity seismic network [16] for calculating the ray paths.

Waveform data were from the TohokuUniversity's seismic network. Figure 4 shows the locations of stations used in the present analysis. Data used are of short-period seismographs (1Hz) or broad-band (STS-1 or STS-2) seismographs. For the 2001 event, we also used waveform data from JOF station of JMA and those from OB3 station, which is one of cabled ocean bottom seismometers (accelerometers) off Kamaishi, of ERI, Univ. of Tokyo. Waveform data recorded at these stations were low-pass filtered with a cut off frequency of 5Hz and resampled at a sampling rate of 20Hz. We used 5 second, three-component (east-west, north-south and vertical) seismograms starting 2 second before S wave arrival.

Figure 5 shows the obtained moment release distribution of the 1995 and 2001 events. For the 2001 event, spatial extent of the rupture area is about 1.5 x 1.5 km^2. Moment release distribution has a simple shape with a single peak. The peak of moment release is located near the hypocenter, the initial point of the rupture. For the 1995 event, spatial extent of rupture area is about 1.5 x 1.5 km^2, which is the same as that of the 2001 event. Moment release distribution has a slightly complicated shape compared with the 2001 event. The peak of moment release is located about 300m to the west of the hypocenter. Scalar moments of the 1995 and 2001 events are 0.82 and 1.12 x 10^23 dyne.cm, respectively. Ratio of seismic moment of the 2001 event to the 1995 event is 1.35. Observed and synthesized seismograms are shown in Fig. 6 for the 1995 event and in Fig. 7 for the 2001 event. They are very consistent with each other.

Figure 5 (c) shows a comparison of the rupture area of the 2001 event with that of the 1995 event. They are almost overlapped with each other, although their hypocenters are different from each other. Especially,spatial extents of the asperities (the areas with larger moment release) are also same and locations of the peaks of the two moment release distributions are closely located to each other. The present observations clearly show that the 1995 and 2001 events are caused by the repeating ruptures of the same asperity patch, and that the asperities are persistent features.

5. Discussion

Source inversions using an EGF are strongly affected by the selection of the EGF event. To test the stability of the present source process inversion, we used a different small event as empirical Green’s function and did the same procedure. We selected a small event with a magnitude of M3.1 that occurred on Oct. 13, 2001. Figure 8 (b) shows the result obtained by using this small event as EGF event. In this case, rupture areas of the 2001 event and of the 1995 event extend in areas with about 1 x 1.5 km^2 and they are almost overlapped with each other as in the case shown in Figure 5 and Fig. 8 (a). Selection of waveform data also affects the results of inversions. We also inverted different sets of data to check the stability of the present source inversion. We showed the moment release distributions obtained by using only vertical componentseismograms, and those obtained by using waveforms from common 5 stations (see Fig. 4) in Figs.8 (c) and (d), respectively. Again, these cases, rupture areas of the 2001 event and of the 1995 event are almost overlapped with each other as in the case shown in Figure 5 and Fig. 8 (a).

The present 'characteristic' earthquake sequence off Kamaishi has periodically occurred in an average period of 5.6 yrs with a standard deviation of about 10% of the mean [12]. No earthquakes which size is equal to or larger than that of the M4.8 event off Kamaishi have occurred nearby the M4.8 event and the asperity of the M4.8 event is isolated from other asperities. The interaction between the asperities sparsely distributed would be neglible and the recurence interval will be constant. For example, the 1990 event have occurred at 5.3 yrs after the previous event (the 1985 event). But the recurrence intervals of the two recent events fluctuated from the average interval. The 1995 event occurred at 4.7 yrs after the 1990 event, but, in contrast, the 2001 event occurred at 6.7 yrs after the 1995 event. We re-examined seismograms of the 1990, 1995, 2001 events to determine their seismic moments, and compared the seismic moments with their recurrence intervals.

Seismic moments of the 1990, 1995 events relative to that of the 2001 event. Matsuzawa et al. [12] described that the waveform similarities of these events are quite good up to about 3 Hz. As a simple procedure made in Bakun and McEvilly [3], we measured maximum peak-to-peak amplitudes of these low-pass (to 2Hz) filtered seismograms and calculated the ratios of these amplitudes.

In the first case, we measured maximum peak-to-peak amplitudes of vertical seismograms in a length of 30 sec from P-wave arrival. Each waveforms are low-pass filtered with a passband of 2 Hz. These waveforms of the 1995 and the 2001 events are very similar to each other as discussed in Matsuzawa et al. [12]. We can measure the maximum amplitudes of the 1995 and the 2001 events at 14 stations with epicentral distances ranging from about 80km to 320km. Figure 9 shows relative amplitudes of the 2001 event to the 1995 event as a function of epicentral distance. Relative amplitudes are greater than 1.0 at most of stations, although the data is some what scattered. The mean and standard deviation is 1.3 ±0.2.

For the events before the 1995 event, we cannot measure the maximum amplitudes due to saturations of the seismograms. This is because the dynamic range of our telemetry system was low (12 or 16 bits) before the 1995 event occurred. However, fortunately, P-wave first motions were not saturated with a good S/N at a few stations. Then, in the second case, we measured the maximum peak-to-peak amplitudes of vertical components of P-wave first motions for the 1985, 1990, 1995, and 2001 events. Figure 10 (a) and (b) shows examples of waveforms observed at AOB station. These waveforms were not saturated from P-wave arrivals (0 sec) to 5 sec. They are very similar to each other. Figure 10 (c) shows relative amplitudes measured from vertical components of P-wave arrivals of the 1985, 1990 and 2001 events to that of the 1995 event. We could measure peak-to-peak amplitudes of P-wave first motions at four stations. The 2001 event has larger amplitudes, which is same as in the first case. The mean and standard deviation is 1.1±0.1．In contrast, the amplitude of the 1990 event at each station was nearly identical with that of the 1995 event. The mean and standard deviation is 1.0±0.1．The amplitude of the 1985 event, which could be obtained at only one station, was also nearly identical with those of the 1995 event.

Seismic moments obtained in this study are summarized in Table 2. In all cases, the estimated seismic moment of the 2001 event is slightly larger than those of the 1995 and 1990 events. The ratio of the moment of the 2001 event to that of the 1990 event (about 1.2 times) is consistent with the ratio of the time interval from the previous event of the 2001 event to that of the 1990 event. This positive correlation between seismic moment and recurrence interval leads to the relative change in static stress drop of about 0.2MPa/year (2MPa/decade) by assuming the spatial extent of the rupture area of the 1990 event is the same as that of the 2001 event. This supports the typical values previously reported for large earthquakes (4MPa/decade; [26]) and for microearthquakes (2MPa/decade; [27]).

If the slip rate of the aseismic slip in the surrounding creeping area is constant and the spatial extent of the 1990 event is the same as that of the 2001 event, the slip-predictable model would be suitable for the interpretation of the aperiodicity of the recurrence of this ‘characteristic’ earthquake sequence off Kamaishi. In contrast, the ratio of the moment of the 1995 event to that of the 1990 event (almost identical) is larger than the ratio of the time interval from the previous event of the 1995 event to that of the 1990 event (about 0.8 times). In the interval between the 1995 and the 1990 events, the 1994 M7.5 Far off Sanriku earthquake occurred at Dec. 28, 1994, about 150km north from the focal area of this ‘characteristic’ earthquake sequence. Based on GPS data surrounding the focal area of this M7.5 event, Nishimura et al. [28] revealed the spatio-temporal distribution of the after-slip. The estimated after-slip area was extended to the focal area of this ‘characteristic’ earthquake sequence off Kamaishi. We infer that the after-slip of the 1994 M7.4 Far off Sanriku earthquake accelerated the occurrence of the 1995 event and the rate of the aseismic slip in the surrounding creeping area became larger before the 1995 event.

6. Conclusions

M4.8 +- 0.1 earthquakes have occurred regularly since 1957 with a recurrence interval of 5.52 +- 0.68 yrs at the same location off Kamaishi, Iwate Prefecture, NEJapan. We interpreted that this peculiar characteristic earthquake sequence is caused by repeating slips of an isolated small asperity surrounded by stable sliding area on the plate boundary. In order to test whether this interpretation is correct or not, we compared the rupture area of the latest (2001) event with that of the previous (1995) event by a waveform inversion method. The result shows that the asperities by them are almost overlapped with each other, although their hypocenters are separated by about 200m from each other.