Intra-plate seismicity in the subducting Philippine Sea Plate, southwest Japan: magnitude–depth correlations

Physics of the Earth and Planetary Interiors 145 (2004) 179–202

Intra-plate seismicity in the subducting Philippine Sea Plate, southwest Japan: magnitude–depth correlations A.J. Smith a,∗ , P.R. Cummins b,a , T. Baba a , S. Kodaira a , Y. Kaneda a , H. Yamaguchi a a

b

IFREE/JAMSTEC, 2-15 Natsushima-cho, Yokosuka 237-0061, Japan Minerals and Geohazards Division, Geoscience Australia, P.O. Box 378, Canberra ACT 2601, Australia Received 12 August 2003; accepted 19 March 2004

Abstract Recent studies of intra-plate seismicity have focused on detailed earthquake locations and potential hypotheses for intermediate depth events. Three main competing hypotheses for intra-plate events are: (1) transformational faulting; (2) shear melt instabilities; and (3) dehydration reactions. Recent work, using precise hypocenters and thermal-petrological models, indicates that the spatial distribution of intra-plate events coincides with dehydration reactions. These studies have mainly focused on the location of events and the proposed reactions with little attention given to the magnitude distribution of events. Presented here is a magnitude–depth correlation associated with intra-plate events for southwest Japan. Earthquake hypocenters were obtained from relocating a data set consisting of Japan Meteorological Agency (JMA) seismicity data and Japan University Network Earthquake Catalog (JUNEC). Data from 1985 to 1996 were combined and relocated using a standard location routine (HypoInverse) that allows regional variations in velocity models and station corrections. HypoInverse results show a well defined plane of seismicity associated with the subducting Philippine Sea Plate. The data were further relocated using a double-difference relocation algorithm (HypoDD). HypoDD results show a magnitude–depth correlation for Kii peninsula and northern Kyushu. Histogram plots indicate that events near the plate interface were generally magnitude 3 or less, whereas most large events were located within the lower crust or possibly the subducting oceanic mantle. This observation also appears to hold true for intra-plate events beneath Tohoku (Japan trench) and Hokkaido (Kuriles), but is not clearly observed beneath the Tokai region. A dehydration reaction model can well explain the magnitude–depth correlation. © 2004 Elsevier B.V. All rights reserved. Keywords: Seismology; Earthquake magnitudes; Intra-plate; Subduction dynamics; Ocean crust

1. Introduction 1.1. Wadati–Benioff seismicity ∗ Corresponding author. Tel.: +81-457-78-5397; fax: +81-457-78-5439. E-mail addresses: [email protected] (A.J. Smith), [email protected] (P.R. Cummins), [email protected] (T. Baba), [email protected] (S. Kodaira), [email protected] (Y. Kaneda), [email protected] (H. Yamaguchi). 1 Fax: +61-2-6249-9999.

The details of intra-plate seismicity within the Wadati–Benioff zone, place important constraints on the thermal structure and state of stress in the subducting slab (Wang, 2002b; Hacker et al., 2003b; Peacock and Wang, 1999). As a result, a substantial amount of research has focused on Wadati–Benioff seismicity, its spatial distribution, and possible causes. The

0031-9201/$ – see front matter © 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.pepi.2004.03.011

180

A.J. Smith et al. / Physics of the Earth and Planetary Interiors 145 (2004) 179–202

three leading hypotheses for intermediate depth earthquakes were summarized by Hacker et al. (2003b): (1) tranformational faulting; (2) thermal shear instabilities; and (3) dehydration embrittlement. A strong spatial relationship between intra-plate seismicity and dehydration reactions has been demonstrated and appears to hold for models of both warm and cold subduction zones (Hacker et al., 2003a,b; Yamazaki and Seno, 2003; Omori et al., in press; Preston et al., 2003; Wang et al., 2004). Detailed intra-plate seismicity studies within warm and cold subduction zones have shown a correlation between magnitude and depth (Cassidy and Waldhauser, 2003; Igarashi et al., 2001). In the Cascadia subduction zone (warm), a band of smaller events (M < 3) was located near the upper part of the subducting oceanic crust, with larger events (M > 3) located near the oceanic crust/mantle boundary (Cassidy and Waldhauser, 2003). A similar observation was made in Tohoku (cold), where the Pacific Plate is subducting beneath Japan (Igarashi et al., 2001). Igarashi et al. (2001) observed a ‘triple seismic zone’ based on focal mechanism solutions. A key observation was the lack of large events within roughly 5 km of the plate interface (defined by the top of the seismicity band). Instead, smaller events characterized by normal faulting were typical and have been attributed to dehydration reactions in the uppermost oceanic crust. The small magnitude of these events may be associated with the weak, fractured nature of the uppermost crust, which would limit rupture propagation (Igarashi et al., 2001). Band 2 in their Fig. 13b occurs in the uppermost mantle or lowermost part of the subducting crust, consistent with slab bending models Wang (2002a). Wang (2002b) proposed that seismicity in band 2 contains larger events than the upper band, as the relatively unfractured nature of the lower crust and mantle allows greater rupture propagation over a wider area. Southwest Japan, where the Philippine Sea Plate (PHS) is subducting beneath the Japanese islands of Shikoku, Kyushu and Honshu, provides an ideal location to test this hypothesis (Fig. 1). In addition to a large, quality seismicity database being available, the seismicity variations along the margin allow for a range of physical conditions within the subducting slab to be examined (Nakamura et al., 1997). The age of the plate varies along the length of the subduction

zone (∼15–45 Ma old; Okino et al., 1999; Kobayashi et al., 1995; Hibbard and Karig, 1990; Deschamps and Lallemand, 2002, and references therein). The plate is subducting at a rate of ∼50 mm per year (Seno et al., 1993), producing a complex pattern of crustal and intra-plate seismicity (Yamazaki et al., 1989; Ishida, 1992; Nakamura et al., 1997). Beneath the Kyushu region, Wadati–Benioff (intra-plate) seismicity occurs at depths down to 200 km, whereas, beneath Shikoku island, most seismicity is shallower than 50 km (Fig. 1) (Nakamura et al., 1997). There are a number of proposed folds or tears in the subducting plate and previous studies distinguished two or more plate fragments (Cummins et al., 2002; Nakamura et al., 1997; Ishida, 1992; Yamazaki et al., 1989). Much of the intra-slab seismicity can be ascribed to dehydration embrittlement of the subducting crust (Peacock and Wang, 1999; Kirby et al., 1996), however, mantle seismicity observed beneath Kii peninsula, Kyushu, and the eastern edge of Shikoku Island, indicates large regions of mantle hydration (Seno et al., 2001; Kodaira et al., 2002). Here, we examine the distribution of earthquakes relative to the plate interface of the southwest Japan subduction zone, and compare the results with Wang’s (2002b) hypothesis. We focus on three regions in southwest Japan; beneath northern Kyushu, beneath Kii peninsula and beneath the Tokai region (see Fig. 1). The main aims are to define any possible magnitude–depth correlations in intra-plate seismicity and determine if these correlations can be explained by a dehydration embrittlement model. To accomplish this, improvements over previous hypocenter determinations have been achieved by using a larger arrival time data set consisting of an amalgam of available catalog data. In addition, we obtained improved relative earthquake locations by using the double-difference technique (Waldhauser and Ellsworth, 2000). 2. Seismicity data and relocation procedure 2.1. JMA and JUNEC data set Previous studies of seismicity in southwest Japan (e.g. Nakamura et al., 1997; Ishida, 1992; Yamazaki et al., 1989) have generally made use of regional networks, in part, because the coverage of the Japan

A.J. Smith et al. / Physics of the Earth and Planetary Interiors 145 (2004) 179–202

181

Fig. 1. Southwest Japan showing the location of the Philippine Sea Plate (PHS) subducting beneath the Eurasia plate (solid black line is plate boundary) and the JMA and JUNEC network configuration for SW Japan. The black vector indicates PHS motion relative to Eurasia. Solid black profile lines indicate the regions of study; northern Kyushu, Kii peninsula, and Tokai. Slab seismicity below 25 km depth is also plotted as well as 10 km depth contours to the top of the slab. The plate is subducting beneath the three main islands; Shikoku, Kyushu, and Honshu, at a rate of roughly 50 mm per year. JMA stations are indicated by black stars, and inverted white triangles denote JUNEC stations (Seno et al., 1993).

Meteorological Agency’s (JMA) nationwide network was too sparse to accurately determine small earthquake hypocenters. This situation has improved since the amalgamation of the regional (university) network data into the Japan University Network Earthquake Catalog (JUNEC) (Tsuboi et al., 1994). Even so, the coverage of this catalog remains patchy, with observations very dense where a regional network is deployed, but sparse elsewhere. We have improved on the JUNEC catalog by merging it with the JMA data.

The JUNEC and JMA network distribution is shown in Fig. 1. Approximately 250 JMA and 209 JUNEC stations are used for relocating events in southwest Japan. We focused on the three profiles shown in Fig. 1, for which JMA stations provide good azimuthal coverage for these profiles. Fig. 2a shows the distance to nearest station for the JMA and JUNEC networks and the combined network. There is a considerable improvement in the distance to the nearest station along the Tokai and Kii peninsula lines, with only limited improvement along the Kyushu line. Fig. 2b shows

182

A.J. Smith et al. / Physics of the Earth and Planetary Interiors 145 (2004) 179–202

Fig. 2. (A) Plot of distance to closest station for JMA, JUNEC, and the combined network. The JMA network is well planned and very evenly spaced. The addition of the JUNEC stations improves the distance to nearest station in most areas. (B) Plot of maximum azimuthal gap between stations for JMA, JUNEC and the combined network. Due to the well-developed JMA network, addition of JUNEC stations does not contribute significantly to decreasing the maximum azimuthal gap between stations. There is some improvement along the Kii peninsula and Tokai lines, but it is less pronounced for the Kyushu profile.

the maximum azimuthal gap between stations for the JMA, JUNEC and the combined network. The calculation of azimuthal gap only includes stations within a 250 km radius of the grid point. The plots indicate the JMA network is distributed efficiently, with only moderate improvements to the azimuthal gap gained by combining the networks.

We combined all available JMA and JUNEC seismicity data from 1985 to 1996 and relocated them using HypoInverse (Klein, 2002). We defined an identical event in the JMA and JUNEC catalogue to have an origin time that differs by <15 s, a horizontal difference of <0.75◦ and a vertical difference of <80 km. After rejecting events with <6 P-arrivals, on average,

A.J. Smith et al. / Physics of the Earth and Planetary Interiors 145 (2004) 179–202

48% of the remaining data was matched and 52% was unmatched. We also chose to exclude M < 2 events, which accounted for 23% of the amalgamated data. All events (matched or unmatched) with at least 6 P-arrivals and M ≥ 2.0 were relocated. The accuracy of arrival times for the JUNEC data is generally better than 0.1 s (Salah and Zhao, 2003), while JMA accuracy is generally better than 0.2–0.3 s, but can be as low as several seconds (Ueno, personal communication). We used HypoInverse to relocate the events as it is a widely used program that allows regional variations in velocity models with either 1D layered or 1D layered with linear velocity gradients to be used. S-arrivals were used when available, but due to the potential for larger picking errors they were down-weighed. The events were relocated using two sets of models: one set consisting of several regional velocity models, and the second a single travel-time model (Table 1). The regional velocity models were taken from Nakamura et al. (1997) and in some cases were modified with information obtained from more recent land seismic experiments (Kodaira et al., 2002; Iidaka et al., 2003). The travel-time model was a single model obtained from averaging the models of Nakamura et al. (1997). The results shown here from Kyushu and Kii peninsula were obtained using the travel-time model. Results from the Tokai region were mainly obtained using a model based on data from a recent land seismic experiment by JAMSTEC (Iidaka et al., 2003). The events were relocated using either the JMA or JUNEC initial locations, or using the JUNEC locations for matched events. Poisson’s values, calculated from Wadati plots, vary from 0.24 to 0.26. A VP /VS ratio of 1.76 was obtained for seismicity below 25 km in depth, and was used to relocate the slab events. After

183

initial relocations were completed, station corrections were then obtained from the mean of the difference in observed and predicted travel-times for all events for each station. The events were again relocated, with station corrections applied, resulting in an improvement in standard errors. As we know very little about the temporal stability of the network stations, we elected to remove stations that had large station corrections or variable travel-time residuals. These questionable stations accounted for <5% of the total number of stations. 2.2. Double-difference method The double-difference method (Waldhauser and Ellsworth, 2000), combined with receiver function structure has been previously used to analyze intra-plate seismicity in the Cascadia subduction zone (Cassidy and Waldhauser, 2003). These results indicated a band of small events near the plate interface, with a second band of seismicity containing larger events located in the uppermost oceanic mantle. Similar to Cassidy and Waldhauser (2003), we used HypoDD (Waldhauser, 2001), a relative relocation program, to assess the distribution of seismicity within the subducting Philippine Sea Plate. HypoDD is a double-difference earthquake location program that minimizes residuals for pairs of earthquakes at each station by weighted least squares. HypoDD was applied to all events that were relocated with HypoInverse for both the travel-time model and the regional model locations. Having accurate initial locations is important, as event pairing in HypoDD is controlled by specifying the maximum separation distance between event pairs (Waldhauser, 2001). Separation

Table 1 Regional velocity models Layer

1 2 3 4 5 6

Hiroshima

Kochi

Tokushima

Wakayama

VP (km/s)

Depth to top of layer (km)

VP (km/s)

Depth to top of layer (km)

VP (km/s)

Depth to top of layer (km)

VP (km/s)

Depth to top of layer (km)

5.6 6.1 6.7 7.8 8

0 3 16 30 40

4.8 5.5 6.1 6.35 6.6 7.8

0 1 3 15 20 30

5.5 6.1 6.7 7.8

0 5 20 35

5.5 6 6.8 7.9

0 3 15 30

184

A.J. Smith et al. / Physics of the Earth and Planetary Interiors 145 (2004) 179–202

Table 2 HypoInverse quality ratings of JMA, JUNEC and combined data sets Quality

Aa Bb Cc Dd

JUNEC

JMA

Traveltime–HypoInverse

Traveltime with station corrections

Events

Percent

Events

Percent

Events

Percent

Event

Percent

2878 16090 20201 17807

5.05 28.24 35.46 31.25

7100 29651 18569 7915

11.23 46.89 29.37 12.52

33011 29209 8035 5448

43.61 38.58 10.61 7.20

35696 27332 7408 5267

47.15 36.10 9.79 6.96

RMS < 0.15 s, ERH < 1.0 km ERZ < 2.0 km. RMS < 0.30 s, ERH < 2.5 km ERZ < 5.0 km. c RMS < 0.50 s, ERH < 5.0 km. d Worse than the above. a

b

distances of 6, 9 and 12 km were used. This distance should be comparable to the error in the event locations, but not too large to lose geophysical significance. Before relocating with HypoDD, we filtered out poorly located events (vertical error >5.0 km or RMS error >0.5 s). Event pair selection and filtering of bad events generally removed 5–10% of events along any given profile. The majority of clusters were relocated to within 2 km of their original cluster centroid position. However, comparison of initial and relocated hypocenters indicated that individual events may vary by distances of up to 10 km. 2.3. Hypocenter error Comparing the quality of locations obtained using either the JUNEC or JMA data with those obtained by combining the two networks together highlights the significant benefit in using the larger database. Table 2 shows the quality of events obtained by JUNEC and JMA, using the quality rating A–D system of HypoInverse. Standard JMA and JUNEC locations have very few ‘A’ quality locations; ∼11 and ∼5%, respectively. Combining the data and relocating using a layered velocity model results in over 40% ‘A’ quality locations, with >80% of the events having depth errors of <5 km. For the purposes of this analysis, it is the location of earthquakes relative to the defined plate interface that is important. Although absolute hypocenter accuracy is useful, it is the relative accuracy that is essential and we consider the use of HypoDD to be appropriate for this purpose. Average HypoDD errors output for a subset of Kii peninsula data are shown in Table 3,

together with average deviations from these errors and the maximum errors observed. The subset contains 300 events that have been relocated in single value deconvolution (SVD) mode. Average errors for both M ≥ 3 events and M < 3 events differed by <150 m in both the horizontal and vertical, as did deviations from these errors (Table 3). The main errors in relative location are controlled by the velocity model, picking errors, phase mis-association, and station errors. An error in picking time of 0.1 s results in an error of up to ∼0.8 km for an event located within the subducting plate. For our dataset, the picking error would be one of the largest errors associated with the double-difference solutions. Our structural model may have errors of ±10% in the depth to layers and the layer velocities. However, since the subducting plate is not modeled by our 1D regional model, within the low velocity crustal layer of the subducting plate there may be deviations of over 1.0 km/s in the model (∼15%). A 3D relocation method may provide better absolute loTable 3 HypoDD errors Coordinate

Average error (m)

Average deviation (m)

Maximum error (m)

X-direction Y-direction Z-direction M ≥ 3 X-direction M ≥ 3 Y-direction M ≥ 3 Z-direction M < 3 X-direction M < 3 Y-direction M < 3 Z-direction

958 976 1819 863 933 1721 971 959 1821

347 348 460 304 382 478 348 310 442

3820 4473 4849 3327 4473 4849 3820 4210 4732

A.J. Smith et al. / Physics of the Earth and Planetary Interiors 145 (2004) 179–202

3. Tohoku intra-plate seismicity and histogram test

Table 4 Event clustering Dataset

JUNEC JMA HypoInv HypoDD

185

Average event separation between pairs (km)

Minimum separation (km)

Maximum separation (km)

10.31 8.68 6.59 6.54

2.86 3.00 1.96 2.05

20.18 15.02 10.35 10.98

cations, assuming an accurate velocity model is available, however, comparison of hypocenter relocations obtained using HypoDD with a 1D and a 3D model indicate that differences in event location are generally <2 km (Waldhauser, personal communication). Errors in the plate velocity model only scale the separation distance between two events by the factor of the error. Therefore, two events located 2 km apart in an 8 km/s model, would actually be 1.68 km apart in a layer whose velocity is actually 6.7 km/s. By keeping the event separation distance small (6–12 km), we limit the extent that velocity model errors can effect the event locations. We analyzed the variations in hypocenter location at each relocation step by examining the locations of a cluster of five events beneath Kii peninsula with similar stress axes. Table 4 shows the average separation distance between event pairs, as well as the maximum and minimum separation distance. The HypoInverse locations show a 2–4 km change in average event separation for the cluster. There is however a very nominal change in the separation for the HypoDD locations. The minimum and maximum separation distance between the events is also marginally smaller for the HypoInverse locations. Although the distance between event pairs stays nearly constant (maximum change of 0.6 km), the relative positions of the events change by 0.5–2.5 km. Therefore, the relative event locations changed by up to 2.5 km from the HypoInverse location, but still maintained a reasonably tight cluster. These five events are a subset of a much larger cluster (over 900 events) and the fact that their relocated positions remained in a tight cluster within the main cluster is reassuring. Since we are trying to discern magnitude–depth patterns on the scale of 2–5 km, relative changes in hypocenters on the order of 2.5 km are very important.

3.1. Three bands of intra-plate seismicity beneath Tohoku The three planes of seismicity observed by Igarashi et al. (2001) beneath Tohoku consist of distinct upper and lower planes of seismicity, with the upper plane being further segmented into two bands based on focal mechanisms. The upper band (band 1 in Igarashi et al., 2001) is within the subducting crust and contains low angle thrust (LT) and normal faulting (NF) events; the middle band is located 5–20 km beneath the slab surface (band 2) and contains mainly down-dip compression (DC) events; whereas the third band consists of down-dip extension (DE) events, starting at about 100 km depth and 25–50 km beneath the slab surface. If we exclude low angle thrust events near the seismogenic zone, band 1 consists of mainly small magnitude (M < 3) normal fault events within the upper 5 km of the subducting plate, whereas band 2 can contain larger events. If there are fewer small events from 0 to 5 km normal to the plate interface (S–N depth), then histogram plots of number of events versus depth for small events (M < 3) and larger events (M ≥ 3) should indicate a magnitude–depth correlation, with a shift in the larger event histogram and/or a reduced number of M > 3 events between 0 and 5 km S–N. We tested the magnitude–depth variation by relocating a 40 km wide band of seismicity across Tohoku (Fig. 3). All events below 35 km in depth were relocated, with obvious continental crust events excluded. We also excluded poor quality events (quality rating D). Our plate boundary for Tohoku is defined as a smooth curved surface along the upper edge of the seismicity band. Thus, we assumed that the plate interface coincides with the upper plane of seismicity, which may not be the case. The number of events versus slab normal (S–N) depth for M < 3 events and M ≥ 3 events is shown in the 2D histograms (Fig. 4). The 3D histograms show the number of events versus magnitude and S–N depth (Fig. 4). The 2D histograms indicate that larger earthquakes are generally deeper, although there are some large earthquakes near the plate interface (Fig. 4). The median for the smaller events occurs at ∼5.4 km S–N

186

A.J. Smith et al. / Physics of the Earth and Planetary Interiors 145 (2004) 179–202

Fig. 3. Map and profile view of Tohoku seismicity relocated using HypoDD. The location of the profile coincides with the region where Igarashi et al. (2001) observed three planes of seismicity based on focal mechanism solutions. Solid black line indicates the plate interface defined by seismicity. In the profile view, black circles indicate M < 3.5 events and squares indicate M ≥ 3.5 events.

Fig. 4. HypoDD histogram plots of Tohoku seismicity. It is clear from the histograms that larger events occurred at greater depths relative to the plate interface. The 3D histogram plots also indicate that, although limited in number, there were M > 4 events located near our defined plate interface (0–5 km slab normal distance). For all the 3D histogram plots, a magnitude of 2 indicates 2 ≤ M < 3, magnitude 3, 3 ≤ M < 4, magnitude 4, 4 ≤ M < 5, with magnitude 5 referring to all events with M ≥ 5, if they are present. For regular histograms, the black bars correspond to smaller events, while the white outline bars refer to larger events. The histogram medians for the M < 3 events vs. M ≥ 3 are 5.4 and 9.4 km, respectively.

A.J. Smith et al. / Physics of the Earth and Planetary Interiors 145 (2004) 179–202

distance, while the median for larger events occurs at ∼9.4 km S–N distance. The main band of seismicity occurs between 0 and 13 km S–N distance. The 4 km difference in histogram medians is significant. If earthquakes were distributed throughout the crust with no magnitude–depth correlations we would expect histograms for large and small events to have a similar median. Assuming a 6–7 km thick oceanic crust (Miura et al., 2003; Takahashi et al., 2000), it is quite possible that most events occurred within the subducting crust, taking hypocenter errors into account. However, recent studies indicate that larger intra-plate earthquakes generally occur within the lowermost oceanic crust and uppermost subducting mantle (Cassidy and Waldhauser, 2003; Ancorp Working Group, 1999; Wang et al., 2004). The larger events occurring deeper may support the presence of upper mantle slab seismicity. However, without better constraints on the subducting plate interface and higher precision hypocenters, it is not possible to confirm this association. 3.2. Potential histogram biases Three significant problems occur with our histogram method and may bias the results. The first is the assumption that the plate interface coincides with the upper boundary of seismicity. The second is that larger events may be biased because they sample a larger station distribution. Lastly, catalogue completeness may introduce a bias into the results (see Appendix A). Other factors affecting the method are uncertainties in earthquake magnitude and errors in earthquake locations. Smaller earthquakes may have large magnitude uncertainties. An analysis of JMA and JUNEC magnitudes for a combined event, indicates, on average, a 13% difference in the magnitude calculation, with the average magnitude of a combined event being 2.63 with a variation of 0.34. Errors in hypocenter location can affect the histograms. For a large enough data set, earthquake errors and magnitude uncertainties should only have a minor effect on the histograms, if they occur randomly. The plate boundary for the southwest Japan study is controlled by offshore seismic experiments in the shallow regions (<30 km depth) and intra-plate seismicity for deeper regions (see Baba et al., 2002). For the study region, there are very few detailed deeper structural

187

studies, and we, therefore, defined the deeper subducting plate interface as a curved surface along the upper edge of the seismicity band. The assumption that the plate interface coincides with the upper plane of seismicity may not be entirely accurate. Crustal structure studies in areas such as Shikoku have observed relatively sparse subducting crust seismicity but abundant subducting mantle seismicity (Kodaira et al., 2002). This has been confirmed with OBS seismicity studies off Shikoku, which showed mainly two groups of seismicity; one in the mantle and one occurring in the accretionary prism near the top of the subducting plate (Obana et al., 2001, 2003). Structure beneath Kii peninsula has been resolved using receiver function analysis, and indicates that the deepest section of seismicity may reside within the oceanic mantle rather than the crust (Yamauchi et al., 2003). Unfortunately, for the Kyushu region, there is no detailed structural information for the subducting plate. Although recent structural data for the Tokai region has been obtained (Iidaka et al., 2003), a detailed analysis of the subducting plate at depth has yet to be made. Existing tomography data for southern and central Japan has insufficient resolution for defining an accurate plate boundary (Nakamura et al., 2002; Salah and Zhao, 2003). If there is a magnitude depth correlation within the band of seismicity, it should be observable with small errors in the plate interface. Generally, a visual observation of the seismicity profile coupled with the histogram plot is sufficient to overcome most of the biases associated with the defined plate boundary. Finally, even with the absence of crustal seismicity, as in Shikoku, there may still be a magnitude–depth correlation within the mantle events. A second bias is that larger events may relocate deeper because they sample a larger station distribution. To determine how this affects our histogram results, we relocated events using HypoDD and systematically decreased the radius of the sample stations for each of the southwest Japan study regions. We started with a 500 km radius and decreased this by 100 km intervals until we reached a radius of 100 km. For Kii peninsula and Tokai, there appears to be no significant change in the histograms for station sample radii of 100–500 km. The Kyushu results do change for station sample radii of <200 km. This is mainly due to the limited number of available stations (<25) and

188

A.J. Smith et al. / Physics of the Earth and Planetary Interiors 145 (2004) 179–202

unstable results using HypoDD. Using only HypoInverse with stations <100 km distance, most M > 4 events were located between 7 and 15 km S–N distance. Thus, the sampling of a larger station distribution by larger events does not appear to be a source of a magnitude–depth correlation. A third bias may be catalogue completeness. After combining earthquake catalogues, only a handful of events were removed due to filtering out of poor events, and automatic removal of airquakes by HypoDD. An example of the loss and removal of events for the Kii peninsula region can be found in Appendix A.

4. Southwest Japan intra-plate seismicity In southwest Japan, Kyushu and Kii peninsula contain high concentrations of intra-slab seismicity. Beneath Tokai, there also appear to be high concentrations of intra-slab seismicity relative to adjacent areas. We selected three profiles across these high seismicity concentrations to assess the potential magnitude–depth correlation of intra-plate seismicity (Fig. 1). Seno et al. (2001) proposed that Kyushu and Kii peninsula contain subducting mantle seismicity, on the basis that deeper slab events did not contain later crustal phases (Hori et al., 1985) and that back-arc volcanism would have caused mantle serpentinization and hydrated mantle. If mantle events are potentially larger, then we might expect to see a histogram shift for larger events with depth for these two regions. Seno et al. (2001) further argued that the mantle beneath Tokai is not hydrated and does not contain mantle events since the seismic zone is a single plane and most slab events have later crustal phases. In this case, we would not expect to see an offset in the histogram plot for larger events with depth. 4.1. Kii peninsula profile—double seismic zone Seismicity for initial JUNEC and JMA locations, relocated HypoInverse hypocenters and HypoDD hypocenters are shown in Fig. 5. For the HypoDD profile, we have only included events that are used in the histogram plots. Beneath Kii peninsula a single sparse layer of seismicity is observed down to about 45 km depth, where it broadens (Fig. 5). A second

seismic layer appears to occur at 55 km depth. Larger events (M > 3.5) appear to mainly occur at depths below 45 km. The histogram plots, obtained from HypoDD relocated events, are shown in Fig. 6. There is a clear offset in larger events with distance from the plate interface. Near the plate interface there is a 3 km band that contains few larger events. Histogram medians occur at 2.63 and 4.43 km S–N depth for small and large events, respectively. Most seismicity is contained within an 8 km thick band. This implies that the seismicity may be contained wholly within the subducting crust, which indicates that if mantle seismicity is present, it is relatively minor. This interpretation is dependent on the assumption that the plate boundary coincides with the top of the seismicity band. The 3D histogram indicates that M > 4 events occur between 2 and 8 km distance from the plate interface. 4.2. Northern Kyushu profile Initial JUNEC and JMA locations and relocated HypoDD hypocenters for the Kyushu region are shown in Fig. 7. For the HypoDD profile, we have only included events that are used in the histogram plots. A widening of the band of seismicity can be observed at a depth of 40 km (Fig. 7). From 0 to 55 km along the profile we observe a narrow band of seismicity (5 km thickness). At 60 km, the band of seismicity is >10 km thick. This widening also broadly coincides with the location of the main onset of larger intra-plate events (M > 3.5 at 80 km profile distance, Fig. 7). Note that this onset of larger events is not associated with an increase in seismicity. From 50 to 80 km depth, there is a high concentration of seismicity with very few large events associated with it. Although large events occur within 5 km of the defined plate interface, most M > 4 events occur on the lowermost part of the band of seismicity. Histograms for Kyushu are shown in Fig. 8. Most events are within a 12 km wide band. The histogram medians are 2.2 and 4.0 km S–N depth for small (M < 3.5) and large (M ≥ 3.5) events respectively. Note on the 3D histograms very few events with M ≥ 4 in the upper 3 km in the HypoDD results, which is also observed in the seismicity profile (Fig. 7). There is also a slight offset in the histogram for M ≥ 3.5 events, compared to the M < 3.5 events. The 3D

A.J. Smith et al. / Physics of the Earth and Planetary Interiors 145 (2004) 179–202

189

Fig. 5. Kii peninsula profile seismicity. There is substantial improvement in the merged data set locations when compared to the original JMA/JUNEC hypocenters. A thin band of seismicity is observed until it broadens at 45 km profile distance. The onset of larger events (M > 3.5) also coincides with the broadening of the band of seismicity. At 60 km depth, the band of seismicity starts to thin again. For Figs. 5, 7, 9, 11 and 12, red stars denote M ≥ 4 events, green diamonds indicate 3.5 ≤ M < 4.0 events and black circles signify M < 3.5 events. For the HypoDD locations, we only show events that were used in calculating the histograms.

190

A.J. Smith et al. / Physics of the Earth and Planetary Interiors 145 (2004) 179–202

Fig. 6. Histogram plots for Kii peninsula. There is a clear offset in the histogram for larger events. There is, however, a number of M > 3 events that occurred near the plate interface. The 3D histograms show that the M > 4 events were located between 2 and 8 km depth normal to the plate interface. Histogram medians for small events (M < 3) and larger events (M ≥ 3) are 2.63 and 4.43 km, respectively.

histograms show that the majority of M ≥ 4 events occurred between 0 and 10 km slab normal distance, and the M > 5 events were deeper than 5 km slab normal distance. 4.3. Tokai region The HypoDD results for the Tokai region (Fig. 9) differ from those of Kii peninsula and northern Kyushu. Seismicity across the profile appears to contain several drops or jumps. This may be due to a tear in the subducting plate (see Yamazaki et al., 1989), or thickened oceanic crust associated with cyclic ridge subduction in the region, as observed by Kodaira et al. (2003). The onset of large intra-plate earthquakes occurs at 40 km depth (60 km along the profile). There is scarce seismicity before 60 km profile distance, with the exception of a large cluster near the down-dip end of the thermally controlled seismogenic zone (Hyndman et al., 1995). The histogram median occurs at 1.75 and 2.14 km S–N distance for small (M < 3) and large (M ≥ 3) events respectively (Fig. 10). Most seismicity occurs between 0 and 8 km S–N distance, supporting Seno et al.’s (2001) observations that all the seismicity is contained within the oceanic crust. The far end of the profile has limited seismicity, but the distribution of seismicity indicates a band >15 km thick. The HypoDD histograms show a slight shift in large events, however, the 3D his-

Table 5 Event magnitudes Location

Northern Kyushu

Kii peninsula

Tokai

Tohoku

M<3 4 > M >3 M≥4 Percent M > 3 events Percent M > 4 events

697 168 20 19 2.3

628 181 12 22 1.5

979 204 7 17 0.6

626 235 11 26.9 1.3

tograms and the seismicity profile indicate that the largest events (M > 4) occurred in the uppermost part of the seismicity band. Compared to the Kii peninsula and Kyushu, the Tokai area exhibits a lower frequency of M > 3 event (Table 5).

5. Discussion 5.1. Magnitude–depth correlations Two magnitude depth correlations were observed for SW Japan. The first showed that there appears to be a down-dip depth at which large events start to occur. This was observed in all three regions (Kii peninsula, Tokai and Kyushu). The second magnitude depth correlation indicates that larger events (M ≥ 3.5) are generally deeper within the band of seismicity (farther from the top of the seismicity band). Note, large events

A.J. Smith et al. / Physics of the Earth and Planetary Interiors 145 (2004) 179–202

191

Fig. 7. Kyushu profile seismicity. Original locations of JMA and JUNEC data shown followed by the HypoDD relocations. HypoDD locations show only the events used to calculate the histograms. There was a substantial improvement over the original locations by using the combined data set. HypoDD plots show a thin band of seismicity to about 50 km profile distance, where the band thickens. Seismicity was mainly contained within a 10–12 km band. Unlike Kii peninsula, which contained few if any M > 3.5 events near the top of the seismicity band, Kyushu had several large events near the plate interface. Between 60 and 100 km depth, a large number of the M ≥ 4 events occurred along the lower portion of the seismicity band.

192

A.J. Smith et al. / Physics of the Earth and Planetary Interiors 145 (2004) 179–202

Fig. 8. Histogram plots for the Kyushu profile. Although a histogram offset is observed for the larger events, it is not as prominent as that observed for the Tohoku or Kii peninsula histograms. Most large events (M > 4) occurred between 0 and 10 km slab normal distance, as indicated in the 3D histograms. The two largest events occurred between 5 and 10 km slab normal distance. Another important feature is that M > 4 events mainly occurred almost evenly distributed to 10 km slab normal distance, and were not clustered around 2 km where one would expect if they were distributed where the highest concentration of seismicity occurs. Histogram medians for M < 3.5 events and M ≥ 3.5 events are 2.2 and 4.0 km, respectively.

can and do occur near the upper portion of the band of seismicity as seen in northern Kyushu, however, our results show that they are more abundant deeper in the band of seismicity. An interesting feature of the visible double seismic zones is that larger events occur in the oceanic mantle (Cassidy and Waldhauser, 2003). Figs. 11 and 12 show profiles through Japan trench and the Kurile trench, both well-defined double seismic zones. Japan trench seismicity are HypoDD relocated results from the combined JMA-JUNEC catalogue. The Hokkaido data (Kuriles trench) is from the JMA catalogue, 1997–2001. A comparison of structure and seismicity indicates that the lower band is clearly within the subducting oceanic mantle (Miura et al., 2003; Nakanishi et al., in press; Takahashi et al., 2000, in press). If we neglect events near the seismogenic zone (upper plane to 70 km depth), visual observation indicates that very few large events (M > 3.5) occur near the top of the upper band of seismicity, and that a greater proportion of large events occur within the mantle band.

spatial distribution of hypocenters and predicted hydrous phases (Hacker et al., 2003a,b; Yamazaki and Seno, 2003; Omori et al., in press). Igarashi et al. (2001) proposed a link between dehydration reactions and earthquake magnitudes by associating the small normal faulting earthquakes observed near the top of the plate with dehydration of basalt to eclogite. A model to explain the variations in magnitude within the subducting plate has yet to be made. Such a model would be required to have the following characteristics: it must explain a low velocity wave-guide near the plate interface, as observed in late arrival studies of seismic waves (Abers et al., 2003; Abers and Sarker, 1996; Abers, 2000; Ohkura, 2000; Hori et al., 1985), it should be consistent with the spatial distribution of earthquakes, and it should explain the magnitude–depth correlations observed within the slab. Next, we consider two possible models: cold and warm subduction.

5.2. Dehydration reactions and intra-plate seismicity

Igarashi et al.’s (2001) proposed a cold subduction zone model for the Japan trench subduction zone where small events occur in a zone of mixed basalt and eclogite. Because, the thermal contours within a cold

Studies of dehydration reactions and their links to intra-slab seismicity have generally dealt with the

5.3. Cold subduction model

A.J. Smith et al. / Physics of the Earth and Planetary Interiors 145 (2004) 179–202

193

Fig. 9. Tokai profile seismicity. Once again there were substantial improvements in the relocated events when compared to the original JMA/JUNEC data. The relocations show a very complex pattern of seismicity, suggesting either subduction of a very jagged plate, tears within the subducting plate, or a highly deformed subducting plate. The band of seismicity was generally <10 km thickness with some scatter. There is a large cluster of events at 25 km profile distance which accounts for a large portion of the seismicity. Large events (M > 3.5) occurred down-dip, mainly from 55 to 100 km profile distance.

plate are such that the interior (middle) of the plate is the coldest (Peacock and Wang, 1999), the eclogite/basalt layer would migrate down through the ocean plate resulting in a 100% eclogite layer overlying a 100% basaltic layer below. Hacker et al.’s (2003b) model for northern Japan has dehydration reactions occurring near the plate interface and, with increasing distance from the trench, slowly migrating down

through the subducting basalt into the gabbro layer and finally into the oceanic mantle. The 4–8% slower low velocity layer (LVL) for northern Japan is 3.7 ± 0.9 km thick (Abers, 2000), although this value may be underestimated by up to 50%. This is consistent with the thickness of the NF zone observed by Igarashi et al. (2001). However, the 5 km layer of NF events is likely thinner, due

194

A.J. Smith et al. / Physics of the Earth and Planetary Interiors 145 (2004) 179–202

Fig. 10. Histogram plots for the Tokai profile. The Tokai histograms differ from those for Kii peninsula and Kyushu, in that they do not show a pronounced offset for larger events. There is some indication of larger events being deeper from the regular histogram plots, but the 3D histograms show that the largest events occurred near our defined plate interface. There was however, a decreased number of large events in the Tokai region when compared to areas such as Kii peninsula and Kyushu. The histogram median for the smaller events (M < 3) occurs at 1.75 km, whereas the median for larger events (M ≥ 3.5) occurs at 2.14 km distance normal to the slab.

Fig. 11. Tohoku profile highlighting the distribution of larger events. Profile location is shown in Fig. 3. If we look at down-dip events away from the seismogenic zone (below 70 km depth), we observe that the lower plane of seismicity within the subducting oceanic mantle contained a greater proportion of the large events when compared to the oceanic crustal plane of seismicity. The lower plane also contained the largest events from 90 km depth to 150 km depth. In the upper oceanic crustal plane of seismicity, most of the larger events do not appear to have been located near the surface of the seismicity plane.

A.J. Smith et al. / Physics of the Earth and Planetary Interiors 145 (2004) 179–202

195

Fig. 12. Profile showing Kurile seismicity beneath Hokkaido, northern Japan. The profile location is shown in the inset map. Similar to Tohoku, the lower plane of seismicity within the subducting mantle contains the largest events when compared to the upper crustal plane.

to relocation errors. Abers (2000) noted that most events are likely associated with or very near the LVL. Assuming seismicity is distributed throughout the subducting crust suggests the LVL is associated with the entire subducting crust (basalt and gabbro). This layer would disappear at ∼150 km as all the basalt/gabbro transforms to eclogite. We modify the model of Igarashi et al. (2001) to better fit the observed magnitude–depth correlations observed in Tohoku (Fig. 11) and the Kuriles (Fig. 12), and to be consistent with the LVL within the subducting plate (Abers, 2000). We first divided the oceanic crust into a layer of basaltic crust on top of a layer of gabbroic crust. The oceanic crust in Tohoku and the western Kuriles is comprised of a ∼2 km thick band of basalt overlying a ∼5 km thick gabbro layer (Miura et al., 2003; Nakanishi et al., in press; Takahashi et al., 2000, in press). The region of smaller events near the plate interface is associated with the basalt layer. Beneath this, events would occur within the less fractured gabbro layer. Dehydration within the basalt layer

would occur rapidly, producing many small events. Dehydration within the stronger gabbro layer would be sluggish and allow larger rupture propagation (Hacker et al., 2003b). The third band of seismicity, within the mantle, often contains a greater frequency of large earthquakes and the largest intra-plate events, when compared with the crustal seismicity bands (see Figs. 11 and 12). Mantle material is stronger than the crustal material and therefore requires more energy to rupture. A larger stress buildup would therefore promote larger ruptures. The LVL is associated with the entire oceanic crust. Dehydration to eclogite would occur simultaneously within the basalt and gabbro layers. At temperatures and pressures below 450 ◦ C and 3 GPa (roughly ∼120 km depth), MORB and Gabbro would have ∼8–15% slower seismic velocities than the surrounding mantle material. At this depth, MORB should transform to lawsonite amphibole eclogite (Hacker et al., 2003a), which has seismic velocities comparable to mantle peridotite. Dehydration of the gabbro

196

A.J. Smith et al. / Physics of the Earth and Planetary Interiors 145 (2004) 179–202

Fig. 13. Warm subduction zone model. Upper band (∼2 km thick) of seismicity is contained within the basaltic layer of the oceanic crust. Events within the basaltic crust are generally small due to the fractured nature of the crust. Below this, events occur within the gabbroic crust. Events within the gabbroic crust can be larger due to the less fractured nature of the crust. The third band of seismicity is within the oceanic mantle and has a higher frequency of larger events compared to the gabbro band. This mantle band roughly follows a geotherm for a warm subduction zone and would, therefore, not necessarily be parallel to the subducting plate.

layer may continue to occur to greater depths, due to the colder temperature of the layer and sluggish reaction rates. This would provide a mechanism for the observed LVL zone extending to 150 km depth. 5.4. Warm subduction model For a warm subduction zone (SW Japan, Cascadia), we need only to modify the cold model. An important feature is that the thermal regime has a greater influence on the dehydration reactions than pressure does. In contrast to the cold slab subduction regime considered above, in a warm subduction zone, there is likely to be very little variation in the temperature from the top of the oceanic crust to the crust/mantle boundary (Wang et al., 1995; Peacock and Wang, 1999), and we would, therefore, expect dehydration reactions within the MORB and gabbro layers to occur near simultaneously, with little or no downward migration of the zone of dehydration.

In the warm subduction model (Fig. 13), the first band of seismicity is contained within weak oceanic basalt that would tend to produce only small magnitude events. Events beneath this occur within the gabbro layer (band 2). The third band of seismicity occurs in the oceanic mantle. This band is difficult to differentiate from the crustal events due to the spatial distribution of the dehydration of serpentinite. Yamazaki and Seno (2003) indicated that this layer of dehydrated serpentinite could start as shallow as 25 km depth and extend to nearly 80 km depth. For the Kii peninsula region, we observed mantle events starting at ∼50 km depth and extending to 75 km depth (Fig. 14), however, mantle events have also been observed offshore at 20–30 km depth using OBS data (Obana et al., 2003). A important feature of the mantle events is that they do not parallel the oceanic crust events, but appear to parallel temperature contours (Peacock and Wang, 1999). This feature has also been observed in Cascadia by Preston et al. (2003). This observation also

A.J. Smith et al. / Physics of the Earth and Planetary Interiors 145 (2004) 179–202

197

Fig. 14. Warm subduction model bands applied to Kii peninsula. Light green shaded region, we have associated with seismicity within the basalt layer (band 1). Light blue shaded region is seismicity associated with the gabbroic layer (band 2). Gray shaded region we define as seismicity associated with the subducting oceanic mantle (band 3).

has hazard assessment implications, since the largest intraslab events for warm subduction zones may be localized where dehydration reactions in the subducting oceanic crust merge with the area where mantle dehydration reactions are occurring. Yamazaki and Seno’s (2003) model predicts dehydration in MORB to a depth of 50 km, slightly shallower than the observed crustal seismicity. It is difficult to determine where the oceanic crust seismicity terminates, because it merges with the mantle seismicity. If our band 1 (basalt) seismicity terminates near the occurrence of M > 3.5 events, the basalt seismicity band would only reach depths of ∼55 km (Fig. 14). The gabbro band, however, would extend to a depth of ∼60 km, possibly due to sluggish reactions or a marginally colder plate environment. For southwest Japan, there appears to be a specific depth at which larger events commence. For Kii peninsula and Tokai, larger events usually occur at depths below 40 km. For Kyushu, the depth appears to be around 60 km. This difference may be a result of only looking at a 12-year data set. However, dehydration reactions can also provide an explanation for this observation. As dehydration occurs, water is released and cracks form in the dehydrated material. If enough of

these cracks align and the stress regime is appropriate, rupture can occur. The number of cracks and amount of released water would increase with depth. Unlike the basalt layer, the water in the coarser grained gabbro layer would find it difficult to escape, thus elevating fluid pressures with increasing down-dip depth. There would also be an increasing number of fractures with increasing down-dip depth. Alignment of these fractures would cause rupture to propagate further, producing larger magnitude events. Abers (2000) suggested that the LVL layer near the plate interface, although unlikely, may be contained within the mantle wedge. Therefore, the band of small events near the plate interface may also be associated with the mantle wedge. However, our observations coupled with slow tremors observed by Obara (2002) indicate that the LVL and small events do not occur within the mantle wedge. Slow tremors occur above the Kii peninsula seismicity band (Obara, 2003) and are located at mantle wedge depths. They also correlate spatially with dehydration in the basaltic band of the oceanic crust being slightly up-dip of the end of band 1 (basalt) of our model. This is also consistent with water escaping the basalt layer (band 1), but not the gabbro layer (band 2).

198

A.J. Smith et al. / Physics of the Earth and Planetary Interiors 145 (2004) 179–202

5.5. Model fit The warm subduction model fits the Kii peninsula seismicity quite well (Fig. 14). Applying the model to northern Kyushu and the Tokai region has mixed results. The northern Kyushu profile would contain the upper band of seismicity in the basalt layer. This would likely terminate around 65 km depth. The second band of seismicity within the gabbro layer would continue on to nearly 80 km depth. Mantle events would be associated with the larger events occurring at 60 km depth and continuing to over 120 km depth. This is supported by the events observed below 60 km depth that contain no later phases (Ohkura, 2000). The Tokai region has larger events near the plate interface, with one large cluster of bigger events at 90 km profile distance. Kodaira et al. (2003) observed a cyclic ridge subduction offshore Tokai in a wide angle OBS study. The study showed the ongoing subduction of the paleo-Zenisu ridge and an older subducted ridge beneath the accreted prism. A region of thickened crust (13–20 km) is associated with each ridge. This repeated structure has a wavelength of 35–50 km, very similar to the wavelength of the thickened bands of seismicity in our observations (∼20–30 km in Fig. 9). We interpret that the higher frequency of seismicity near the upper 6 km of the band of seismicity is associated with the hydrated portion of the thickened oceanic crust. Below this main band of seismicity is a region of hydrated thickened oceanic crust where fewer events occur. Oceanic mantle seismicity is unlikely, as events recorded beneath Tokai contain later phases, and are therefore within or near the LVL (Hori et al., 1985; Abers et al., 2000). The apparent lack of a magnitude–depth correlation may be associated with the ridge subduction. In the Tokai region only 0.5% of events are M > 4 and 17% are M > 3. There are noticeably fewer large magnitude events than in the Kyushu, Kii peninsula and Tohoku regions (Table 5). 6. Conclusions Dehydration reactions can explain the spatial distribution of intermediate depth earthquakes in subduction zones (Yamazaki and Seno, 2003; Hacker et al., 2003b; Omori et al., in press). The focus of these studies was on the spatial distribution of earthquakes and

did not stress any magnitude–depth correlations. Both Hacker et al. (2003b) and Yamazaki and Seno (2003) did detailed thermo-petrologic studies of Tohoku (northern Japan) and Nankai (southwest Japan) and found a reasonably good spatial correlations between intra-plate events and dehydration reactions. In these regions, we have also observed a magnitude–depth correlation for intra-plate earthquakes. The relocated seismicity shows a significant improvement in quality ratings over the JUNEC and JMA standard locations. For the Kyushu region, the newly relocated intra-plate seismicity clearly defines the subducting slab. The seismicity imaged beneath Kii peninsula and Tokai also show considerable improvement in defining the location of the subducting plate, compared with previous studies (Nakamura et al., 1997). The HypoInverse locations obtained by using a travel-time model show very little clustering around the Moho or Conrad discontinuities. HypoDD further improves this by producing a more defined slab dip near Moho depths and dense clustering of seismicity along the upper plane. A clear pattern of larger events (M > 3.5) occurring at greater distances normal to the plate interface is observed beneath northern Kyushu and Kii peninsula. These results are consistent with the magnitude–depth correlation observed in northern Japan (Igarashi et al., 2001) and Cascadia (Cassidy and Waldhauser, 2003). In the visible double seismic zone beneath northern Japan (Figs. 11 and 12), although large events do occur within the crustal layer, a greater proportion of larger events were observed within the subducting oceanic mantle. At Kii peninsula, the seismicity can be separated into three bands on the basis of the spatial distribution and magnitude (Fig. 14). Near the plate interface there is a thin band (2–3 km) of seismicity containing mainly smaller events (M < 3.5). We interpret this band to be in the subducting oceanic basalt layer. Beneath this is a thicker band of seismicity (∼5 km) contained within the oceanic gabbro layer. Larger events can occur in this layer because the higher strength and larger crystal size requires more energy (stress) to rupture. The higher energy required for rupture to occur, coupled with a less fractured, stronger medium, allows for larger rupture propagation. The third band of seismicity beneath Kii peninsula occurs within the subducting mantle. This band contains both the largest events, and a higher frequency of large events when compared to bands 1 and 2.

A.J. Smith et al. / Physics of the Earth and Planetary Interiors 145 (2004) 179–202

Acknowledgements Many thanks to our colleagues at IFREE and special thanks to Kelin Wang, and Simon Johnson for their time, comments and useful discussions. Thanks for reviews by Dr. Felix Waldhauser and an anonymous reviewer. Special thanks to Monica Handler for extensive editing of the paper. Figures in this report have been produced using GMT (Wessels and Smith, 1995). We thank the Japan Meteorological Agency and Japan University Network Earthquake Catalog for the use of the earthquake datasets.

Appendix A The JMA earthquake catalogue from 1985 to 1996 contains over 269,000 events, the JUNEC catalogue over 156,000. During the amalgamation of the JUNEC and JMA data on average 6% of JUNEC events and

199

30% of JMA events were discarded because they contained <6 P-arrivals. This equates to a 21% rejection rate for all events (JUNEC + JMA). The remainder of the events were either matched up, or they became a JMA-only or JUNEC-only events. For the complete catalogue, 48% of the events were matched up. JMA-only events comprised 35% of the events and JUNEC-only events made up 17%. In some cases, events that were matched were probably not the same event. In these instances the matched event generally had a very large RMS errors after relocation by HypoInverse. Most were removed manually from the archive file after relocation, consisting of only a handful of events. Others were filtered out as a D type event because of the large errors. Our criterion for matching events may cause some sort of bias. We removed events with <6 P-arrivals. If we lower this constraint to <4 P-arrivals it has a negligible effect on increasing the number of matched events. It does, however, increase the number of JMA

Fig. 15. Comparison of HypoInverse relocated events before filtering out bad events (black circles), and after filtering out bad events (white circles). Total number of unfiltered events is 997, which becomes 986 after filtering. Grey boxes indicate the location of 10 events that were filtered out. Five of the events would be associated with continental crust seismicity and would be removed before calculating histograms.

200

A.J. Smith et al. / Physics of the Earth and Planetary Interiors 145 (2004) 179–202

Fig. 16. Comparison of HypoDD relocated events (white circles) and original locations (larger black circles). There are 997 original locations and 935 after relocating with HypoInverse, removing bad events and then relocating with HypoDD. Of the 62 filtered events, 53 have been marked with boxes. We can see that about one-third (23) are likely associated with continental crust.

and JUNEC only events, and these events would only have 4 or 5 P-arrivals. We also removed events of M < 2 after the amalgamation. This accounted for approximately 23% of the data. To check for a bias caused by removing these events, we relocated datasets containing all events with M > 0. Comparison of histogram medians is shown in Table 1.A after relocation with HypoDD. There was no real significant difference caused by removing events with M < 2. Table A.1 Region

Kii peninsula (km) Tohoku (km) Tokai (km)

Kyushu (km)

M > 0 dataset

M ≥ 2.0 dataset

M<3

M≥3

M<3

M≥3

2.65 5.67 1.49

4.33 9.43 1.91

2.63 5.4 1.75

4.43 9.4 2.14

M < 3.5

M ≥ 3.5

M < 3.5

M ≥ 3.5

2.25

4.01 km

2.2

4.00

After relocating events with HypoInverse we then filtered out “bad” or D-type events. An example is given using a Kii peninsula data set. Before filtering out D events the Kii Dataset contained 997 events. Removing the ‘bad’ events leaves 986 events. Fig. 15 shows the plotted events before and after filtering. This represents a loss of only 11 events or 1.1%. Locating events using HypoDD also removed some events. During the relocation, events that became airquakes or isolated events were excluded from the relocation. Our final location produced 935 events in the area of interest, for a total loss of 62 events or 6.2%. Fig. 16 compares the initial 997 events to the 935 events. We have marked 53 of the filtered events on the figure. More than one-third (23) of the filtered 62 events appear to be associated with continental crust seismicity rather than slab seismicity. The remainder of the filtered events appear to be randomly distributed. It is unlikely that removal of <4% of the events would cause a significant bias in the histogram means.

A.J. Smith et al. / Physics of the Earth and Planetary Interiors 145 (2004) 179–202

References Abers, G.A., 2000. Hydrated subducted crust at 100–250 km depth. Earth Planet Sci. Lett. 176, 323–330. Abers, G.A., Sarker, G., 1996. Dispersion of regional body waves at 100–150 km depth beneath Alaska: in situ constraints on metamorphism of subducted crust. Geophys. Res. Lett. 23, 1171–1174. Abers, G.A., Plank, T., Hacker, B.R., 2003. The wet Nicaraguan slab, Geophys. Res. Lett. 30 (2), 1098, 10.1029/2002GL015649. Ancorp Working Group, 1999. Seismic reflection image revealing offset of Andean subduction-zone earthquake locations into oceanic mantle. Nature 397, 341–344. Baba, T., Tanioka, Y., Cummins, P.R., Uhira, K., 2002. The slip distribution of the 1946 Nankai earthquake estimated from tsunami inversion using a new plate model. Phys. Earth Planet Int. 132, 59–73. Cassidy, J.F., Waldhauser, F., 2003. Evidence for both crustal and mantle earthquakes in the subducting Juan de Fuca plate. Geophys. Res. Lett. 30 (2), 109510.1029/2002GL015511. Cummins, P.R., Baba, T., Kodaira, S., Kaneda, Y., 2002. The 1946 Nankai earthquake and segmentation of the Nankai trough. Phys. Earth Planet Int. 132, 75–87. Deschamps, A., Lallemand, S., 2002. The west Philippine basin: an Eocene to early oligocene back-arc basin opened two opposed subduction zones. J. Geophys. Res. 107, 232210.1029/2001JB001706. Hacker, B.R., Abers, G.A., Peacock, S.M., 2003a. Subduction Factory. 1. Theoretical mineralogy, densities, seismic wave speeds, and H2 O contents. J. Geophys. Res. 108, 202910.19/2001JB001127. Hacker, B.R., Peacock, S.M., Abers, G.A., Holloway, S.D., 2003b. Subduction Factory. 2. Are intermediate-depth earthquakes in subducting slabs linked to metamorphic dehydration reactions? J. Geophys. Res. 108, 203010.1029/2001JB001129. Hibbard, J.P., Karig, D.E., 1990. Alternative plate model for the early Miocene evolution of the southwest Japan margin. Geology 18, 170–174. Hori, S., Inoue, H., Fukao, Y., Ukawa, M., 1985. Seismic detection of the untransformed basaltic oceanic crust subducting into the mantle. Geophys. J. R. Astron. Soc. 83, 169–197. Hyndman, R.D., Wang, K., Yamano, M., 1995. Thermal constraints on the seismogenic portion of the southwestern Japan subduction thrust. J. Geophys. Res. 100, 15,373–15,392. Igarashi, T., Matsuzawa, T., Umino, N., Hasegawa, A., 2001. Spatial distribution of focal mechanisms for interplate and intra-plate earthquakes associated with the subducting Pacific plate beneath the northeastern Japan arc: a triple-planed deep seismic zone. J. Geophys. Res. 106, 2177–2191. Iidaka, T., Iwasaki, T., Takeda, T., Moriya, T., Kumakawa, I., Kurashimo, E., Kawamura, T., Yamazaki, F., Koike, K., Aoki, G., 2003. Configuration of subduction Philippine Sea Plate and crustal structure in the central Japan region. Geophys. Res. Lett. 30 (5), 121910.1029/2002GL016517. Ishida, M., 1992. Geometry and relative motion of the Philippine Sea Plate and Pacific plate beneath the Kanto-Tokai district, Japan. J. Geophys. Res. 97, 489–513.

201

Kirby, S., Engdahl, E.R., Denlinger, R., 1996. Intermediate-depth intraslab earthquakes and arc volcanism as physical expressions of crustal and uppermost mantle metamorphism in subducting slabs. In: Bebout, G.E., et al. (Eds.), Subduction: Top to Bottom: American Geophysical Union Geophysical Monograph, vol. 96. pp. 195–214. Klein, F.W., 2002. User’s Guide to HYPOINVERSE-2000, a Fortran Program to Solve for Earthquake Locations and Magnitudes: USGS Open File Report 02-171, 123. Kobayashi, K., Kasuga, S., Okino, K., 1995. In: Taylor, B. (Ed.), Shikoku Basin and its Margins: Backarc Basins: Tectonics and Magnatism. Plenum Press, NY, pp. 381–405. Kodaira, S., Kurashimo, E., Park, J.-O., Takahashi, N., Nakanishi, A., Miura, S., Iwasaki, T., Hirata, N., Ito, K., Kaneda, Y., 2002. Structural factors controlling the rupture process of a megathrust earthquake at the Nankai trough seismogenic zone. Geophys. J. Int. 149, 815–835. Kodaira, S., Nakanishi, A., Park, J.-O., Ito, A., Tsuru, T., Kaneda, Y., 2003. Cyclic ridge subduction at an inter-plate locked zone off central Japan. Geophys. Res. Lett. 30 (6), 133910.1029/2002GL016595, 2003. Miura, S., Kodaira, S., Nakanishi, A., Tsuru, T., Takahashi, N., Hirata, N., Kaneda, Y., 2003. Structural characteristics controlling the seismicity of southern Japan trench fore-arc region revealed by ocean bottom seismographic data. Tectonophysics 363, 79–102. Nakamura, M., Watanabe, H., Konomi, T., Kimura, S., Miura, K., 1997. Characteristic activities of subcrustal earthquakes along the outer zone of southwestern Japan. Ann. Disaster. Prev. Res. Inst. Kyoto Univ. 40 (B-1), 1–20. Nakamura, M., Yoshida, Y., Zhao, D., Yoshikawa, K., Yakayama, H., Aoki, G., Kuroki, H., Yamazaki, T., Kasahara, J., Kanazawa, T., Sato, T., Shiobara, H., Shimamura, H., Nakanishi, A., 2002. Three-dimensional P and S wave velocity structure beneath central Japan. Pap. Meteorol. Geophys. 53 (1), 1–28. Nakanishi, A., Smith, A.J., Miura, S., Tsuru, T., Kodaira, S., Obana, K., Takahashi, N., Cummins, P.R., Kaneda, Y. Crustal and thermal structure across the coseismic rupture zone of the 1973 Nemuro-Oki earthquake, the southern Kuril seismogenic zone. J. Geophy. Res., in press. Obana, K., Kodaira, S., Mochizuki, K., Shinohara, M., 2001. Micro-seismicity around the seaward updip limit of the 1946 Nankaido earthquake dislocation data. Geophys. Res. Lett. 28, 2333–2336. Obana, K., Kodaira, S., Kaneda, Y., Mochizuki, K., Shinohara, M., Suyehiro, K., 2003. Micro-seismicity at the seaward up-dip limit of the western Nankai Trough seismogenic zone. J. Geophys. Res 108 (B10), 2459. Obara, K., 2002. Nonvolcanic deep tremor associated with subduction in southwest Japan. Science 296, 1679–1681. Obara, K., 2003. Non-Volcanic Deep Low-Frequency Tremor Associated with Subduction in Southwest Japan. IUGG, Sapporo, Japan. Okino, K., Ohara, Y., Kasuga, S., Kato, Y., 1999. The Philippine sea: new survey results reveal the structure and the history of the marginal basins. Geophys. Res. Lett. 26 (15), 2287–2290.

202

A.J. Smith et al. / Physics of the Earth and Planetary Interiors 145 (2004) 179–202

Ohkura, T., 2000. Structure of the upper part of the Philippine Sea Plate estimated by later phases of upper mantle earthquakes in and around Shikoku, Japan. Tectonophysics 321, 17–35. Omori, S., Komabayashi, T., Maruyama, S. Dehydration and earthquakes in the subducting slab: empirical link in intermediate and deep seismic zones. Phys. Earth Planet Int., in press. Peacock, S.M., Wang, K., 1999. Seismic consequences of warm versus cool subduction metamorphism: examples from southwest and northeast Japan. Science 286, 937–939. Preston, L.A., Creager, K.C., Crosson, R.S., Brocher, T.M., Trehu, A.M., 2003. Intraslab earthquakes: dehydration of the Cascadia slab. Science 302, 1197–1200. Salah, M.K., Zhao, D., 2003. 3D seismic structure of Kii peninsula in southwest Japan: evidence for slab dehydration in the forearc. Tectonophysics 364, 191–213. Seno, T., Stein, S., Gripp, A.E., 1993. A model for the motion of the Philippine Sea Plate consistent with NUVEL-1 and geological data. J. Geophys. Res. 98, 17941–17948. Seno, T., Zhao, D., Kobayashi, Y., Nakamura, M., 2001. Dehydration of serpentinized slab mantle: seismic evidence from southwest Japan. Earth Planets Space 53, 861–871. Tsuboi, S., Abe, K., Ishikawa, Y., 1994. Determination of fault plane solutions for small earthquakes in Japan. J. Phys. Earth 42, 45–67. Takahashi, N., Kodaira, S., Tsuru, T., Park, J.O., Kaneda, Y., Kinoshita, H., Abe, S., Nishino, M., Hino, R., 2000. Detailed plate boundary structure off northeast Japan coast. Geophys. Res. Lett. 27, 1977–1980. Takahashi, N., Kodaira, S., Tsuru, T., Park, J.O., Kaneda, Y., Suyehiro, K., Kinoshita, H., Abe, S., Nishino, M., Hino, R.

Seismic structure of the seismogenic zone of off Sanriku, Japan. Geophys. J. Int., in press. Waldhauser, F., Ellsworth, W.L., 2000. A double-difference earthquake location algorithm: method and application to the Northern Hayward Fault, California. Bull. Seismol. Soc. Am. 90 (6), 1353–1368. Waldhauser, F., 2001. A Program to Compute Double-Difference Hypocenter Locations. US Geological Survey, Open-File Report 01-113, Menlo Park, CA. Wang, K., Cassidy, J.F., Wada, I., Smith, A.J., 2004. Effects of metamorphic crustal densification on earthquake size in warm slabs. Geophys. Res. Lett. 31, L0160510.1029/2003GL018644, 2004. Wang, K., 2002a. Unbending combined with dehydration embrittlement as a cause for double and triple seismic zones. Geophys. Res. Lett. 29I 18, 188910.1029/2002GL015441. Wang, K., 2002b. Stresses and earthquakes in warm slabs: Nankai, Mexico, and Cascadia. Eos Trans. AGU, 83(47), Fall Meet. Suppl., Abstract S21C-09. Wessels, P., Smith, W.H.F., 1995. New version of generic mapping tools released: EOS transactions. Am. Geophys. Union 76, 329. Yamauchi, M., Hirahara, K., Shibutani, T., 2003. High resolution receiver function imaging of the seismic velocity discontinuities in the crust and the uppermost mantle beneath southwest Japan. Earth Planets Space 55, 59–64. Yamazaki, F., Ooida, T., Aoki, H., 1989. Subduction of the Philippine Sea Plate beneath the Tokai area, central Japan. J. Earth Sci. Nagoya Univ. 36, 15–26. Yamazaki, T., Seno, T., 2003. Double seismic zone and dehydration embrittlement of the subducting slab. J. Geophys. Res. 108, 221210.1029/2002JB001918.