Hidden aftershocks of the 2011 Mw 9.0 Tohoku, Japan earthquake imaged with the backprojection method

Authors

  • Eric Kiser,

    Corresponding author
    1. Department of Earth and Planetary Sciences, Harvard University, Cambridge, Massachusetts, USA
    2. Department of Earth Science, Rice University, Houston, Texas, USA
    • Corresponding author: E. Kiser, Department of Earth Science, Rice University, MS 126, 6100 Main St., Houston, TX 77005, USA. (edk2@rice.edu)

    Search for more papers by this author
  • Miaki Ishii

    1. Department of Earth and Planetary Sciences, Harvard University, Cambridge, Massachusetts, USA
    Search for more papers by this author

  • This article was corrected on 25 AUG 2014. See the end of the full text for details.

Abstract

[1] The first 25 h of the aftershock sequence following the 11 March 2011 Mw 9.0 Tohoku, Japan earthquake is investigated using a backprojection method. In total, 600 aftershocks are imaged during this time period. These aftershocks are distributed over a 500 by 300 km area and include many events in the outer rise. The backprojection events are compared with the Japan Meteorological Agency (JMA) catalogue, which is composed of earthquakes recorded by local seismic networks in Japan. Surprisingly, half of the backprojection events are not found in the JMA catalogue. These events cluster near the Japan Trench and in the outer rise and fill in gaps in the spatial distribution of the early aftershock sequence where large main shock slip is thought to have occurred. These results show that the JMA magnitude of completeness is very high near the trench following the 2011 Tohoku main shock, and earthquakes as large as magnitude 6.8 went undetected by local seismic networks.

1 Introduction

[2] The spatiotemporal evolution of aftershock sequences has been used by several studies to constrain fundamental aspects of stress transfer [e.g., King et al., 1994], postseismic deformation [e.g., Hsu et al., 2006; Sladen et al., 2010], rupture extent [e.g., McCann et al., 1979], the constitutive relationships that govern slip [e.g., Dieterich, 1994], and forecasting of future seismicity [e.g., Cocco et al., 2010] to name a few. Aftershock detection is usually limited by a lack of local instrumentation, which necessitates the use of teleseismic data. In addition, the detection and characterization of early aftershocks are usually hampered by the arrival of various seismic waves immediately following giant earthquakes [e.g., Kagan, 2004; Kagan and Houston, 2005; Lolli and Gasperini, 2006; Enescu et al., 2007; Peng et al., 2007; Enescu et al., 2009; Lengline et al., 2012]. This effect can be seen in most earthquake catalogues where gaps in magnitude and time exist directly following large events. For example, in the National Earthquake Information Center (NEIC) catalogue, the first aftershock of the 2004 Mw 9.1 Sumatra-Andaman earthquake does not occur until nearly 20 min after the main shock (earthquake.usgs.gov). This lack of early aftershock activity is generally considered to be a detection problem but also has important consequences for the seismic hazards and constitutive relationships acting following the large slip associated with a giant earthquake [e.g., Dieterich, 1994].

[3] In a few cases, vigorous aftershock sequences have been recorded by dense, local seismic arrays. The best example of this comes from the seismicity that followed the 11 March 2011 Mw 9.0 Tohoku, Japan earthquake, which was recorded by multiple seismic networks in Japan. An analysis of the data by the Japan Meteorological Agency (JMA) has led to a high-quality catalogue of the aftershock sequence. The JMA catalogue has been a valuable tool in many studies for understanding the rupture processes of the main shock and evaluating future seismic hazards in the region. In addition, this catalogue can be used for assessing the performance of teleseismic methods that are used for aftershock detection.

[4] In the current study, a teleseismic array processing method known as backprojection is used to study the early 2011 Tohoku aftershock sequence using data from North American stations, many of which are part of the USArray project (www.iris.edu/dms/dmc; www.earthquakescanada.nrcan.gc.ca). Array processing methods can filter out signals with unwanted slowness and back azimuths, and therefore, this approach can be useful for early aftershock detection. A comparison between the results of the analysis and the JMA catalogue will highlight some of the advantages and disadvantages of using teleseismic array data for monitoring seismic activity.

2 Method

[5] The backprojection method time reverses seismic array data to image the source of energy release. The method was first applied to the 2004 Mw 9.1 Sumatra-Andaman earthquake [Ishii et al., 2005] and has almost exclusively been used to study the rupture properties of main shocks [e.g., Walker et al., 2005; Ishii et al., 2007; Xu et al., 2009; Walker and Shearer, 2009; Lay et al., 2010; Ishii, 2011; Kiser and Ishii, 2011; Koper et al., 2011; Meng et al., 2011; Wang and Mori, 2011; Kiser and Ishii, 2012a, 2012b; Koper et al., 2012; Meng et al., 2012; Satriano et al., 2012; Wang et al., 2012; Yao et al., 2012; Yue et al., 2012; Ishii et al., 2013], though there have been a few applications to aftershock behavior [e.g., Kiser and Ishii, 2012b]. This method has the advantage that it does not require a priori constraints on the events being studied. The main disadvantage of the method is that amplitude information is either relative or distorted in such a way to deemphasize its importance [e.g., Ishii et al., 2007]. Therefore, absolute quantities such as slip or the focal mechanism cannot be estimated. In addition, since teleseismic P wave data are typically used, there is no depth resolution for shallow events [e.g., Kiser et al., 2011]. In this study, two forms of the backprojection method are used, the linear stacking and coherency function approaches. The basic idea of the linear stacking backprojection method is to time reverse and stack normalized seismic array data to a grid of possible source locations [Ishii et al., 2007]. This process can be expressed mathematically as

display math(1)

where si is the stack at the ith grid point, αk is the normalization factor at station k, uk is the seismogram at station k, tik is the theoretical traveltime between the ith grid point and kth station, Δtk is the empirical traveltime correction made at each station, t is time with respect to a reference event, and n is the number of stations. The coherency function analysis takes the additional step of calculating the correlation between the stacks at each grid point and the time-reversed data [Ishii, 2011] and takes the form

display math(2)

[6] Here xi(t) is the coherency function, pk equals 1 or −1 and corrects changes in polarization at station k, and T is the time window used to calculate the correlation. This additional step suppresses amplitude information, but leads to superior spatiotemporal resolution (Figure 1). Most of the results in this study come from the coherency function approach, but linear stacking results are used to compare relative amplitudes to magnitudes from the JMA catalogue.

Figure 1.

(a) The black dots compose the grid of possible source locations for the backprojection analysis. The three white stars are point source locations, with the northern and southern point sources having amplitudes one tenth that of the center point source. (b) Backprojection result of the three point sources using linear stacking. (c) The same as Figure 1b except using the coherency function.

3 Data and Data Processing

[7] An array that spans much of North America and is composed of multiple networks is used for this study (www.iris.edu/dms/dmc; www.earthquakescanada.nrcan.gc.ca; Figure 2a). The data from these stations are band-pass filtered to a frequency range of 0.8–2 Hz. Empirical traveltime and amplitude corrections are applied to each of these stations using a cross-correlation analysis of the initial P waves of a reference event [Ishii et al., 2007]. The use of a single event to make these corrections is preferred because all of the imaged energy will be with respect to a single point instead of multiple points associated with each individual event that may have their own independent hypocentral errors. In addition, this approach gives us the freedom to select an event with impulsive first arriving P waves that can be aligned readily. For this study, a magnitude 7.1 earthquake that occurred on 7 April 2011 at 38.204°N latitude, 141.920°E longitude, and 65.9 km depth is used to make the station corrections.

Figure 2.

(a) North American stations (white triangles) used in the backprojection analysis. This group of stations is composed of the USArray Transportable Array (IRIS and EarthScope), Caltech Regional Seismic Network (Caltech/USGS), Global Seismograph Network (IRIS), International Miscellaneous Stations, University of Utah Regional Network (University of Utah), Berkeley Digital Seismograph Network (Berkeley Seismological Laboratory), University of Oregon Regional Network (University of Oregon), Canadian National Seismograph Network (Geological Survey of Canada), United States National Seismic Network (ANSS Data Collection Center), and ANZA Regional Network (IGPP, University of California, San Diego). (b) Study area of the 2011 Tohoku aftershock sequence. The white and black stars are the epicenters of the 11 March 2011 main shock and 7 April 2011 Mw 7.1 earthquake used to make traveltime and amplitude corrections at each station, respectively. The gray focal mechanism is from the main shock. The thick black lines are plate boundaries, and the black dots are the grid points for the backprojection analysis.

[8] The first 25 h of data following the Mw 9.0 main shock is continuously backprojected to a 6° by 6° grid area around the epicenter of the 7 April earthquake, with a 0.1° grid spacing (Figure 2b). Events are initially manually picked from amplitude versus time plots, which are constructed by plotting the maximum squared coherency function amplitude within the gridded region as a function of time (Figure 3a). Only those events that have maximum amplitudes at least twice that of the local noise levels are initially considered as earthquakes. The time window of the earthquake is selected and the time of the maximum amplitude within this time window is defined as the earthquake time (Figure 3a). This definition of earthquake time is probably closer to the centroid time than the epicentral time, though this distinction will only become important for very large aftershocks.

Figure 3.

(a) A sample of backprojection amplitude with respect to time from the Tohoku aftershock sequence. The time windows of the four aftershocks are shown in red. The noise level is in black. The blue dots are the peak amplitudes of each event. Time is with respect to 11 March 2011, 19:45:18 UTC. (b) Distribution of energy release for the first event selected in Figure 3a. (c–e) The same as Figure 3b, except for the second, third, and fourth events, respectively.

[9] The next step is to determine the aftershock location. The spatial distributions of events within the time windows selected in the first step are evaluated. The centers of the energy kernels of these plots are selected as the event locations (Figures 3b–3e). In some cases, there are additional events that are clearly spatially separated from the event associated with the peak energy release. If the maximum amplitudes of these events have signal-to-noise ratios greater than or equal to two, then the centers of their energy kernels are selected as earthquake locations. Evaluating the spatial distribution of the backprojection results allows for additional quality control procedures to be performed. For example, occasionally, energy kernels are at the edge of the backprojection grid. These events are not included because of the possibility that the events are outside of the grid area and the imaged energy is only the projection of the event into this area instead of the actual source location and time. In addition, the areas of the energy kernels are sometimes much larger than would be expected given the duration of the events [e.g., Wells and Coppersmith, 1994; Houston, 2001]. These large features are caused by the arrival of surface waves, and other seismic phases, from the main shock and are discarded.

[10] It can also be difficult to determine if multiple events in the backprojection results that are close in both space and time are actually separate earthquakes or the complex rupture distribution of a single earthquake. We address this problem by discarding those events that occur within 0.4° and 25 s of an event of equal or larger amplitude. These space and time constraints are chosen based upon the distance and time for the energy from a 1 Hz synthetic point source to decay to the background noise level when evaluated using the backprojection method.

4 Results

[11] The backprojection event catalogue is provided in Table S1 in the supporting information. In addition to presenting the spatiotemporal distribution of this catalogue, the results will include a comparison with the JMA catalogue. Furthermore, brief comparisons will be made with a catalogue compiled using a waveform-matching technique [Lengline et al., 2012] and the NEIC global catalogue (earthquake.usgs.gov). The results of these comparisons, as well as the general characteristics of the backprojection catalogue, are robust and not strongly influenced by some of the subjective decisions made for the backprojection analysis. For example, using different choices for the frequency range of the filtered data and reference event leads to fewer detected events, but the overall results are similar to those given in this section (Figures S1 and S2).

4.1 The Backprojection Aftershock Sequence

[12] In total, the backprojection method detects 600 aftershocks during the first 25 h following the main shock hypocentral time (Table S1 in the supporting information). The spatial distribution of the aftershocks is very broad with a length of about 500 km along the strike of the Japan Trench (Figure 4a). The largest cluster of aftershocks is south and in the updip direction from the epicenter of the main shock. In addition, there is a large cluster of events in the outer rise that has an along-strike length of about 300 km. There are also a few events that are located near the west coast of Honshu that are not associated with the plate interface, or the outer rise, and instead are either shallow events taking place within the overriding Okhotsk/North American plate or deep events that occur within the subducting slab.

Figure 4.

(a) Spatial distribution of backprojection aftershocks (black dots). The white star is the epicenter of the 2011 main shock, and the thick black lines are the plate boundaries. (b) Seismicity rate with respect to time. Events are grouped into 1 h bins. Time is with respect to the 2011 main shock. The first 6 h of the aftershock sequence are excluded when determining the p value. The gray region shows the approximate arrival times of surface waves from the mainshock at stations in North America. (c) Time of backprojection events with respect to their along-strike distances from the 2011 main shock epicenter. Here negative values indicate locations south/southwest of the epicenter. (d) Time of backprojection events with respect to their along-dip distance from the trench. Here negative values indicate locations west of the trench.

[13] The temporal evolution of the backprojection events also exhibits interesting behavior. Seismicity rates are nearly constant during the first 6 h of the aftershock sequence and then decay according to Omori's law with a p value of 0.9 [e.g., Utsu et al., 1995; Figure 4b]. The initial constant seismicity rates are likely caused by the time-dependent catalogue completeness where smaller events can only be detected after the high noise levels associated with the main shock and early aftershocks have dissipated. Methods for correcting this time-dependent catalogue completeness can be applied when magnitude information is available [e.g., Lengline et al., 2012]. Also, note the dip in the seismicity rate during the second hour of aftershocks shown in Figure 4b. This dip is related to the arrival of surface waves, and possibly other seismic phases, in North America from the main shock. These waves obscure events that occur during the first and second hours of the aftershock sequence, and therefore, the decrease in seismicity is likely an artifact. Finally, Figure 4c shows that the spatial expansion of aftershocks occurs very quickly along strike, reaching the outer limits of the eventual aftershock distribution dimensions within about 30 min after the main shock. Excluding a few outliers, the expansion of aftershocks in the dip directions is more gradual, with only minor changes in the general expansion behavior across the trench (Figure 4d).

4.2 Comparison With the JMA Catalogue

[14] In order to evaluate the performance of the backprojection method for aftershock detection, we compare events from the JMA catalogue with those in the backprojection catalogue. In particular, we are interested in determining which events in the backprojection catalogue are also in the JMA catalogue and which events are not. It should be noted that an updated JMA catalogue that was released about a year after the Tohoku main shock is used for this comparison (K. Moriwaki, personal communication, 2012).

[15] The JMA catalogue is generally considered to be the most complete and homogeneous catalogue of the 2011 Tohoku aftershock sequence with about 2500 earthquakes during the first 25 h. This catalogue includes events over a magnitude range of 0.2 to 9.0, though the magnitude of completeness is around 2 during times of normal seismicity [e.g., Nanjo et al., 2010]. The spatial distribution of the JMA aftershocks covers an along-strike distance of about 500 km (Figure 5a), and there is very little decay in seismicity rates during the first 25 h (Figure 5b), though if only events with magnitudes greater than or equal to 5.0 are considered, seismicity rates decay with a p value of 0.9 (Figure 5c).

Figure 5.

(a) Spatial distribution of earthquakes (black dots) in the JMA catalogue during the first 25 h following the Mw 9.0 Tohoku earthquake. (b) Seismicity rate of the JMA catalogue with respect to time. Events are grouped into 1 h bins. Time is with respect to the 2011 main shock. (c) The same as Figure 5b, except only earthquakes with magnitudes greater than or equal to 5.0 are used. The first hour of aftershocks is excluded when determining the p value.

[16] To develop matching criteria between the backprojection and JMA catalogues, we first look at the distances of backprojection events from JMA events that occur within 100 s of each other (Figure 6). This comparison results in a large group of event pairs with distance separations less than 0.5°. In addition, the time differences between backprojection and JMA events are significantly smaller for these closest event pairs, and therefore, a 0.5° distance cutoff is used for earthquake matching (Figure 6). Once this distance constraint is applied, most of the event pairs occur within 20 s of each other, though a slightly larger time constraint of 30 s is used as the matching criterion.

Figure 6.

Distance separation between backprojection and JMA events that are within 100 s of each other. Events are grouped into 0.1° distance separation bins. The colors of the bars indicate the average absolute value of the time separations between the events.

[17] Surprisingly, using the matching criteria, exactly half of the events in the backprojection catalogue are not in the JMA catalogue (Table S1 in the supporting information). The exact number is strongly dependent upon the matching criteria used, especially the spatial constraint, but even when very large spatial (2.0°) and temporal (100.0 s) constraints are applied, 89 of the backprojection events are not in the JMA catalogue. In this case of extreme spatial allowance, note that the distance constraint spans the entire width of the seismogenic zone, so almost any JMA aftershock will be spatially matched to any event in the backprojection catalogue. Thus, the time constraint becomes the only discriminating parameter.

4.3 Events in Both the JMA and Backprojection Catalogues

[18] The backprojection events that have corresponding entries in the JMA catalogue span almost the entire main shock region, though there is a noticeable lack of events near the trench and in the outer rise (Figure 7a). Most of these events have corresponding JMA magnitudes that are greater than 4, with a peak at about magnitude 5 (Figure 7b). It is likely that this peak represents the magnitude of completeness of the backprojection catalogue, with the steady decrease at larger magnitudes being a real feature of the aftershock sequence, i.e., the Gutenberg-Richter relationship [Gutenberg and Richter, 1944].

Figure 7.

(a) Spatial distribution of backprojection events that are in the JMA catalogue (black dots). (b) Number of detected backprojection events with respect to the corresponding JMA magnitude. These events have been grouped into 0.2 magnitude bins. (c) Logarithm of the linear stacking backprojection amplitude with respect to JMA magnitude. The best fitting line (black line) through these data is log(amplitude) = 1.4(JMA magnitude) − 11.0. In this plot, 1 is the logarithm of the peak amplitude during the Mw 9.0 main shock, though this event is not used to estimate the amplitude/magnitude relationship.

[19] The magnitudes in the JMA catalogue can also be evaluated with respect to the peak amplitudes of the backprojection events when the linear stacking approach is applied. For this comparison, event time windows that are picked from the coherency function analysis are used to find the peak amplitudes in the linear stacking results (Figure 3a). The linear stacking approach has inferior spatiotemporal resolution compared to the coherency function approach used thus far; therefore, in some cases the events identified from the coherency function analysis are not as clearly defined when linear stacking is used. Nevertheless, the maximum amplitude of the linear stacks within the time windows of the events does correlate well with JMA magnitude (Figure 7c) with a best fitting line through these data of log(amplitude) = 1.4(JMA magnitude) − 11.0. This relationship will be useful for characterizing events for which a JMA magnitude does not exist.

4.4 Events Only in the Backprojection Catalogue

[20] In contrast to the events that are detected by both backprojection and the JMA, most of the backprojection events not in the JMA catalogue are located near the trench and in the outer rise (Figure 8a). Though this observation is unexpected, the general result holds even when different matching criteria are used (Figure 8b), and different processing choices are made in the backprojection analysis (Figures S1 and S2). Furthermore, the percentage of events only in the backprojection catalogue does not decrease significantly with time from the main shock (Figure 8c). This indicates that high noise levels following the main shock, which decrease with time as seismicity rates decay, are not the primary cause of the unmatched backprojection events, as may be expected. Instead of this time dependence, the results point toward a spatial control on the events that are in the backprojection catalogue, but not in the JMA catalogue.

Figure 8.

(a) Spatial distribution of backprojection events that are not in the JMA catalogue (black dots). (b) The same as Figure 8a, except spatial and temporal matching criteria of 2.0° and 50 s, respectively, are applied. (c) Percentage of the backprojection events that are not in the JMA catalogue with respect to time. Events are grouped into 1 h bins. (d) Number of backprojection events that are not in the JMA catalogue with respect to magnitude. Events are grouped into 0.2 magnitude bins. Magnitudes are estimated using the scaling relationship between the logarithm of backprojection amplitude, using linear stacking, and JMA magnitude from the backprojection events that are in the JMA catalogue (Figure 7c).

[21] Using the best fit scaling relationship between amplitude and JMA magnitude determined in the previous section (Figure 7c), estimates of the magnitudes of these events are made. There is a peak in the magnitude distribution of the events only in the backprojection catalogue between 4.4 and 4.6 (Figure 8d). The number of events in this magnitude range may be an indicator of the background noise level in the linear stacking results, instead of a distinct peak in amplitude associated with an event. In other words, smaller events that are detected with the coherency function analysis may not be prominent features in the linear stacking results, making magnitude estimation difficult. The largest event that is only in the backprojection catalogue has a magnitude of 6.8. Figure 9 shows that this is a very distinct event in both space and time. In addition, there is a Mw 6.6 event with a similar location and time in the NEIC catalogue, which lends support to this result. Despite its large magnitude, local Hi-net data show only a very noisy arrival from this event that is difficult to identify if the data are high-pass filtered (Figure S3).

Figure 9.

(a) Amplitude versus time plot for the backprojection event with an estimated magnitude of 6.8 that is not in the JMA catalogue. The three red vertical lines are the times of the three nearest JMA events within the study region with magnitudes of 4.7, 4.5, and 5.7. The blue vertical line is the time of an Mw 6.6 event in the NEIC catalogue. Time is with respect to 11 March, 05:45:18 UTC. (b) Spatial distribution of energy release from the M6.8 backprojection event where dark blue represents low energy release and white represents high energy release. The three circles with red outlines are the epicenters of the three JMA events. The circle with a blue outline is the epicenter of the Mw 6.6 NEIC event. The yellow line is the Japan Trench and the white star is the epicenter of the Mw 9.0 main shock.

4.5 Events Only in the JMA Catalogue

[22] An alternative approach to comparing catalogues is to investigate the properties of earthquakes in the JMA catalogue that do not have corresponding events in the backprojection catalogue. Most of these events occur near the east coast of Honshu (Figure 10a) and have magnitudes between 1 and 4.5 (Figure 10b). Almost all of the larger aftershocks with magnitudes greater than 6.0 are detected by the backprojection analysis, though there are three large JMA earthquakes that are not in the backprojection catalogue. The largest of these events (M 6.8) is in the outer rise. Its imaged energy is clearly defined in space and time, but cut off at the edge of the backprojection grid. Therefore, using the procedure outlined above, this backprojection event is discarded. The remaining two large events that are only in the JMA catalogue have predicted P wave arrivals in North America that overlap with the arrival of surface waves from the main shock. This result underscores the large degree to which detection is affected during this time.

Figure 10.

(a) JMA events that are not in the backprojection catalogue (black dots). (b) Number of JMA earthquakes that are not in the backprojection catalogue with respect to JMA magnitude. Events have been grouped into magnitude bins of 0.2.

4.6 Combining Catalogues

[23] Combining the backprojection events not in the JMA catalogue with the JMA catalogue leads to some important changes in the spatiotemporal distribution of the aftershock sequence. For example, the large spatial gap observed near the trench in the JMA catalogue (Figure 5a) is almost entirely filled with the combined catalogue (Figure 11a). In addition, seismicity rates decay faster in the combined catalogue with a p value of 1.1 (Figure 11b). Finally, the spatial evolution of the aftershock sequence is similar to that of the JMA catalogue with very rapid expansion of the aftershocks along the strike of the subduction zone (Figure 11c). However, the new events near the trench from the backprojection analysis exhibit a much more gradual along-strike expansion that originates near the along-strike position of the global centroid moment tensor (CMT) solution from the main shock (Figure 11d) [Dziewonski et al., 1981; Ekström et al., 2012].

Figure 11.

(a) Combined spatial distribution of the aftershock sequence when the JMA catalogue (black dots) is combined with the backprojection events that are not in the JMA catalogue (red dots). (b) Number of earthquakes with respect to time for the combined catalogue. Only events with magnitudes greater than or equal to 5 are included, and events are grouped into 1 h bins. The first hour of aftershocks is excluded when determining the p value of these seismicity rates. (c) Event time with respect to the along-strike distance from the main shock epicenter. Here negative values indicate locations south/southwest of the epicenter. Only events with magnitudes greater than or equal to 5 are included. Time is with respect to the main shock hypocentral time. (d) The same as Figure 11c except only events that are west of and within 0.5° of the Japan Trench are plotted. The gray line is the along-strike location of the CMT solution of the main shock [Dziewonski et al., 1981; Ekström et al., 2012].

4.7 Comparison With the Waveform Matching Catalogue

[24] The comparison between the backprojection and JMA aftershocks indicates that there is spatial incompleteness near the trench and in the outer rise in the JMA catalogue. The implications of this observation will be discussed later, but it should first be noted that even when more advanced local array processing methods are used, this incompleteness persists. For example, Lengline et al. [2012] applied a waveform-matching technique, which detects signals that are similar to those produced by catalogued earthquakes, to the first 12 h of the Tohoku aftershock sequence along the plate interface using Hi-net data. This method detected 881 new events during this time period. When this waveform-matching catalogue is compared to the backprojection catalogue during the first 12 h, 156 of the 335 backprojection events that are west of the Japan Trench are not in the waveform-matching catalogue and cluster near the trench (Figure 12). Since the waveform-matching technique searches for signals produced by earthquakes identified in the Hi-net catalogue, it is not capable of detecting events that are significantly different than typical seismicity in the region (e.g., events near the trench).

Figure 12.

The updated Hi-net catalogue (black dots) compiled by Lengline et al. [2012] using a waveform-matching technique and backprojection events (red dots) not in this catalogue. Only subduction interface events are detected in Lengline et al. [2012], and therefore, only backprojection events west of the Japan Trench are used in the comparison.

4.8 Comparison With the NEIC Catalogue

[25] Figure 8d shows that some of the backprojection events that are not in the JMA catalogue are large enough to be detected in global earthquake catalogues. Therefore, a comparison between the backprojection catalogue and the NEIC global catalogue (Figure 13a) may provide further insight into the nature of these unmatched events. Using the same matching criteria as for the JMA catalogue comparison, 191 of the backprojection events are not found in the NEIC catalogue. In contrast to the JMA catalogue comparison, backprojection events that are in the NEIC catalogue and those not in the NEIC catalogue show similar spatial distributions that include events near the trench and in the outer rise (Figures 13b and 13c). In addition, the percentage of events only in the backprojection catalogue decreases with time (Figure 13d), which is expected if background noise levels are the primary cause of earthquakes not being detected in the NEIC catalogue.

Figure 13.

(a) Earthquakes in the NEIC catalogue (black dots) during the first 25 h of the Tohoku aftershock sequence. (b) Backprojection events (black dots) that are also in the NEIC catalogue. (c) Backprojection events (black dots) that are not found in the NEIC catalogue. (d) Percentage of the backprojection events not in the NEIC catalogue with respect to time. Events have been grouped into 1 h bins.

5 Discussion

[26] The differences between the JMA and NEIC comparisons indicate that the spatial incompleteness observed in the JMA catalogue is likely caused by some combination of the local velocity structure [e.g., Bogiatzis et al., 2012], differences in the rupture properties between near-trench earthquakes and those farther downdip that lead to low-quality high-frequency data at local stations [e.g., Bilek and Lay, 1998; Lay et al., 2012, Figure S3], and limitations (e.g., distances and azimuths of stations) in the local arrays used for the JMA catalogue. These effects can be seen in spatial variations in the magnitude of completeness during normal seismicity. In the Tohoku region, the magnitude of completeness for the JMA catalogue is around 1.0 on the mainland but gradually increases to as large as 3 near the Japan Trench [e.g., Nanjo et al., 2010]. Similar spatial patterns in the magnitude of completeness seem to persist during the 2011 aftershock sequence, though the absolute magnitudes jump considerably to values greater than 6.5 near the trench. This jump is likely caused by increased background noise levels associated with the aftershock sequence. Though these values seem high for a local catalogue, similar estimates have been made for the Hi-net catalogue during the early aftershock sequence [Lengline et al., 2012].

5.1 Near-Trench Seismicity

[27] Regardless of the causes of the detection problems, the backprojection events near the trench and in the outer rise have important consequences for seismic and tsunami hazard assessment following giant earthquakes. For example, as Figure 8d shows, the near-trench events can be quite large, and may pose their own set of seismic hazards. For the 2011 Tohoku aftershock sequence, these hazards paled in comparison to those from the main shock, but events around magnitude 7 near the trench can produce local seismic and tsunami hazards (e.g., at local islands) that may be more significant for future events in Japan and other subduction zones [e.g., Newman et al., 2011].

[28] In addition, multiple studies have used the 2011 aftershock sequence reported in various catalogues as either direct evidence of or as support for estimates of the main shock rupture area, postseismic deformation, and stress changes [e.g., Ozawa et al., 2011; Toda et al., 2011; Kato and Igarashi, 2012; Lengline et al., 2012; Ozawa et al., 2012; Tajima and Kennett, 2012]. For example, there are many references to the apparent anticorrelation between early aftershocks and areas of large slip as an indicator of the negative Coulomb stress changes within the Tohoku main shock rupture area and positive Coulomb stress changes at the periphery of this area [e.g., Ide et al., 2011; Simons et al., 2011; Kato and Igarashi, 2012; Lengline et al., 2012; Yao et al., 2013]. The identification of events near the trench eliminates any spatial gaps in the aftershock distribution (Figure 11a), which indicates that unstable sliding resumed very quickly within the lateral dimensions of the main shock rupture area. Given that many models put highest slip near the trench, the seismicity may suggest dynamic overshoot during the main shock, which would reverse the orientation of stress on the plate interface and cause a positive Coulomb stress change [e.g., Ide et al., 2011]. Alternatively, these events may represent increased seismicity within the subducting and overriding plates due to large slip on the adjacent plate interface. Many of the aftershocks that took place near the trench in the months following the main shock have been interpreted as intraplate earthquakes [e.g., Asano et al., 2011], and it is possible that the backprojection analysis is imaging the initial stages of this intraplate seismicity.

5.2 Improving Aftershock Detection

[29] The results of this paper demonstrate that backprojection is a useful seismicity monitoring tool that can be used to complement local earthquake catalogues. The magnitude of completeness of the backprojection catalogue is similar to other global catalogues, and the method can be applied relatively quickly following large events, though an automated picking system needs to be established to apply the method in real time [e.g., Earle and Shearer, 1994]. In addition, as Figure S4 shows, some of the subjective decisions used for the manual picking process can cause imaged events to be missed.

[30] Changes to some of the processing steps could also improve the backprojection analysis. For example, in general, the JMA events occur before the matched backprojection events (Table S1 in the supporting information). There are a few different explanations for this behavior. First, estimates of the backprojection event times likely represent the centroid times of the events instead of the hypocentral times; therefore, the time difference reflects the time difference between the largest energy release and the beginning of the event. Another possibility is that interference from depth phases causes the time shift. Depth phases tend to shift the imaged energy of earthquakes toward the seismic array and later in time (Figure S5). The time shifts may also result from the relatively deep set of grid points that are used. When a depth discrepancy exists between the grid points and the earthquake hypocenter, the backprojection results are a projection of the event energy onto the grid plane along the P wave raypath, which results in a time delay when the grid plane is below the event hypocenter (Figures S6 and S7a). Regardless of the cause, more advanced picking algorithms, deconvolution procedures for reducing the effects of depth phases [e.g., Yagi et al., 2012], and the use of grid points that better match the geometry of the plate interface (Figure S7b) may all improve estimates of event times and locations.

[31] An additional source of error may be variability in earthquake focal mechanisms. The amplitude and polarity corrections made with the cross-correlation procedure [Ishii et al., 2007] are specific to the focal mechanism of the reference event. For this study, the 7 April 2011 Mw 7.1 aftershock used as the reference event has a thrust mechanism that is typical of many events in the Tohoku aftershock sequence. However, some events in the aftershock sequence have significantly different focal mechanisms than this reference event. Based upon the few events that have focal mechanisms in the Global CMT catalogue during the early aftershock sequence, we confirm that earthquakes with a variety of focal mechanisms are imaged with the backprojection analysis (Figure S8). Nevertheless, it may be useful to utilize multiple reference events with different focal mechanisms in future studies.

6 Conclusions

[32] The JMA catalogue provides an excellent resource for monitoring seismicity around Japan. Similar catalogues have been compiled in different regions of the world using high-quality local data (e.g., the Southern California Seismic Network). In general, these catalogues provide the best characterization of local seismicity, but the current work shows that they may also be susceptible to significant spatial incompleteness during aftershock sequences. In order to fully utilize the information within these aftershock sequences, teleseismic catalogues should be combined with local catalogues when appropriate. An important issue that is not addressed in this paper is the time evolution of the spatial incompleteness observed in the JMA catalogue. During times of normal seismicity the magnitude of completeness near the trench is thought to be as large as 3 [Nanjo et al., 2010]. This value jumps to between 6.5 and 7 directly following the Mw 9.0 main shock, but additional work is required to determine how this value changes during the days and months following the main shock.

Acknowledgments

[33] This paper benefitted greatly from the thoughtful comments of Bogdan Enescu, an anonymous reviewer, and the Associate Editor. The authors thank the IRIS Data Management Center and the Canadian Geological Survey for making North America data available and the National Research Institute for Earth Science and Disaster Prevention for making Hi-net data available. The authors also thank the Japan Meteorological Agency and the National Earthquake Information Center for making earthquake catalogues available. Figures are generated using the Generic Mapping Tools [Wessel and Smith, 1991].

Erratum

  1. In the originally published version of this article, supplementary table 1 is missing. The error has since been corrected and this version may be considered the authoritative version of record.

Ancillary