The available geodetic and teleseismic data sets for the 1923 Kanto earthquake (Ms=8.1) have been combined into a joint inversion for both temporal and spatial slip variations. We assumed an initial faulting model to be consistent with the geometry determined by Kanamori [1971] on the basis of first-motion data, aftershock area, and the amplitude of surface waves at teleseismic distances and also to enclose the slipped area estimated by Matsu'ura et al. [1980] from the geodetic data employed here. We then invert for a heterogeneous distribution of slip of the fault plane. The leveling routes and triangulation stations used (consisting of 225 bench marks and 31 triangulation points) are from Matsu'ura et al. [1980]. We chose to first determine the overall, static slip distribution by inverting the geodetic data alone. We then proceeded to gradually increase the importance of the teleseismic data, always requiring a good fit to the geodetic leveling and horizontal displacements. In this way, we could provide a constraint on the overall static slip characteristics from the geodetic data and provide stability for the teleseismic inversion, yet determine the degree of slip heterogeneity and time history most suitable for matching the waveform data and for simulating strong ground motions. Our analysis yields a seismic moment of 7-8 x 1027 dyne-cm (Mw=7.8-7.9) with a maximum slip of approximately 8 meters. The most concentrated slip is in the shallow, central and western portion of the fault. The location of the concentrated slip on the fault plane has important consequences for the amplitude, duration and frequency content of the resulting ground motions as documented by Takeo and Kanamori [1993].
The Kanto earthquake of 1923 (Ms=8.1) was one of the most devastating earthquakes in history, killing nearly 140,000 in the environs of Tokyo and Yokohama. Most of the deaths and 95% of the property loss were attributed to fires following the earthquake, though shaking and tsunami damage was extensive (Imamura [1924]). Earthquake resistant design and emergency preparedness in Tokyo are based in large part on a repeat of the 1923 earthquake. Consequently, improvements in our understanding of the 1923 earthquake rupture process and its effects not only make a contribution to fundamental knowledge but are of practical importance as well.
Early seismometric studies of the hypocenter of the 1923 earthquake are summarized by Kanamori and Miyamura (1970), whose estimates of the hypocentral coordinates of the earthquake have been adopted in this study. These parameters are origin time 2:58:32, latitude 35.4 degress N, longitude 139.2 degrees E, and depth 0 to 10 km, with a epicentral location uncertainty of about plus or minus 15 km. Seismological and geodetic models of the 1923 earthquake were developed by Kanamori (1971) and Ando (1971) respectively, and the seismotectonics were discussed by Ando (1974). Faulting during the 1923 Kanto earthquake was interpreted as reverse, right-lateral faulting on a low-angle plane along the the Sagami Trough, the plate boundary between the northeastern edge of the Philippine Sea plate and the southeastern edge of the Eurasian plate (Figure 1). It was the second of two magnitude 8 or greater earthquakes in the southern Kanto region over the last 1,000 years (Matsuda et al. [1978]). The 1703 and the 1923 earthquakes both occurred along the Sagami Trough, but were considerably different in their rupture extent, tsunami generation and coastal uplift (Matsuda et al. [1978]) and appear to have ruptured adjacent sections of the plate boundary. The geodetic data from the 1923 earthquake were inverted for a slip model by Matsu'ura et al. (1980), who also checked the slip model for consistency with seismograms recorded in Tokyo. Refinements to the geodetic model were developed by Matsu'ura and Iwasaki (1983).
In this study, we have combined the available geodetic and teleseismic data sets for this event into a joint inversion for both temporal and spatial slip variations. The motivation for the determination of the slip characteristics is twofold. First, recurrence estimates and the spatial occurrence of future events may be estimated based on the distribution and heterogeneity of previous events. The specific location and variation of slip needs to be mapped out for this purpose. Second, both hazards mitigation and engineering design can be improved if the separate contributions from both source complexity and the effects of propagation within the Kanto basin to the the ground motion from the the 1923 earthquake events can be understood. The objective of this paper is to provide information on the source complexity of the earthquake which can be used to understand the local and regional strong motion data and to estimate the ground motions experienced in the Tokyo region during the 1923 earthquake.
In general, the geodetic data are more sensitive to fault geometry and the gross features of the slip patterns than the waveform data, but, usually, are not as powerful in resolving details of the slip distribution, particularly the roughness of slip. In the case of the 1923 Kanto earthquake, however, the dense geodetic coverage directly above the rupture plane provides a strong constraint on the slip variations. Even so, the static displacements provide no information on the temporal variations in slip, both in the slip function on the fault and the propagation of rupture on the fault. These parameters are critical to determining the nature of the ground motions. We are particularly interested in determining if the excitation of dominant long-period energy recorded in Tokyo at periods near 13 sec (Takeo and Kanamori [1993]) is source related or, alternatively, whether it can be attributed to wave propagation within the Kanto plain. Our source model, derived from static and teleseismic data, provides an independent estimate of the rupture characteristics that is relatively free of near-source propagation complications.
The geodetic data we use have been compiled and corrected for systematic errors by Matsu'ura et al. [1980]. We take advantage of their careful analysis and make use of the high-quality geodetic data available for the early part of the century from the Military Land Survey [1930].
In contrast, processing and interpreting the seismograms of the 1923 earthquake presented challenges. The large magnitude of the event often resulted in overlapping of surface-wave traces on the body-wave arrivals, and data for many stations were either not obtainable or poorly preserved. However, we believe that the historical data are valuable in spite of their limitations, and that it is desirable to obtain as much information from them as possible considering the importance of the 1923 Kanto earthquake.
In the inverse modeling approach, we chose to first determine the overall static slip patterns by inverting the geodetic data alone; this was done by severely down-weighting the teleseismic data. We then proceeded to gradually increase the importance of the teleseismic data, but always requiring an acceptable fit to the geodetic leveling and horizontal displacements. In this way, we could provide a constraint on the gross static slip characteristics from the geodetic data to provide stability for the teleseismic inversion.
The 1923 Kanto earthquake was recorded at numerous stations worldwide. Kanamori and Miyamura [1970] used P times from nearly one hundred stations to determine the hypocentral parameters (origin time, 2h 58m 32s; latitude, 35.4 degrees N; longitude, 139.2 degrees E; depth about 10 km). We examined the historical teleseismic body waveforms (the majority of which were collected by H. Kanamori) for useful recordings and selected several vertical and horizontal P and S waves, written on Galitzin, Wiechert, Milne-Shaw and Bosch-Omori seismographs. An example of the quality of the original analogue data is shown in Figure 2 for the stations at Riverview, Australia and Uppsala, Sweden as recorded on Wiechert inverted pendulum instruments. Absolute time was preserved at most stations, and time corrections from G.M.T. were provided along with instrument damping, magnification and free-period constants. The waveforms were scanned and digitized, and care was taken to remove any instrument pen arc. An example of the digitized record at Uppsala, Sweden is shown in Figure 3. Useful data were obtained for 7 stations, the locations and instrument constants of which are given in Table 1. The locations of these stations and the azimuthal coverage with respect to the source region are shown in Figure 4. Although the coverage is not ideal, three different azimuths are represented. The locations of these stations on P and Shttp://www.scecdc.scec.orgrce mechanism of Kanamori [1971] are shown in Figure 5. There is a change in polarity between the Riverview P and SH waves and the other stations. There is only a single horizontal component from Helwan, Egypt which prevented rotation into radial and transverse polarizations, so it was not used in the waveform modeling. For an event of this size, the P waves are often more easily digitized in that they have smaller amplitudes on the paper, and have fewer overlaps with traces above and below (Figure 3).
The most useful records were written by Wiechert and Galitzin instruments. The horizontal Wiechert instrument has a free period of about 7-10 sec and the magnification is on the order of 150-200 times (Table 1); the vertical Wiechert components have free periods ranging from 3-5 sec and magnifications from 50-200 times. The horizontal Galitzin has a longer-period response than the Wiechert, with a free period at 25 sec. Figure 6, from Kanamori [1988], shows the amplitude response curves for a variety of instruments including those employed in this study. The pendulum period, To, and damping constant, h, for each component are given in Table 1. The damping constant, h, is related to the damping ratio, e, given in Table 1 by the expression (e.g. Richter [1958], p. 219)
e = exp [ (pi* h) / sqrt(1-h**2)]
For uniformity and direct comparison, a common instrument response was used for each of the data sets for each of the three components. Specifically, for the vertical P waves, the response of Riverview was used; hence, the instrument responses of the Berkeley and Strasbourg records were deconvolved and the Riverview response convolved in. The north component at Riverview was the common response for the radial P waves and the tangential SH waves. For the data from Debilt, the Galitzin response was retained since it has a longer period response than other stations. Synthetic seismograms were convolved with the common response for each component.
On an historical note, the Riverview recording shown in Figure 3 was that viewed by F. Omori on a visit to a geophysical meeting in Australia in 1923. As recounted by Aki [1980], the observation at Riverview was a shock to Omori not only because of his concerns about damage and loss of life, but because it also marked a failure in his career. Nearly two decades prior to the Kanto earthquake, A. Imamura, Omori's younger colleague, published warnings of the possibility of a destructive earthquake in the near future. Imamura advised that the earthquake would be followed by a "general conflagration" should the fire protection system remain unimproved, and the possible loss of lives could reach 100,000 or more. Omori and other scientists had denounced Imamura's warnings and criticized him for causing social unrest. Unfortunately, Imamura's concerns were well founded, for the Kanto earthquake occurred in 1923, and nearly 140,000 lives were lost from the fires that resulted. Omori's health declined rapidly during his journey back to Japan, and he died shortly after his return. For a more detailed historical account, see Imamura [1924] and Aki [1980].
The coseismic elastic deformations produced by the Kanto earthquake are described by geodetic triangulation and leveling measurements by the Military Land Survey [1930]. For our investigation here, we rely heavily on the prior investigation of Matsu'ura et al. [1980] in which they used the geodetic data to estimate the optimal fault plane parameters. The parameters they determined include the rupture geometry and slip vector which was assumed to be uniform on the rupture surface. In a critical aspect of their study, both random errors in the data as well as systemic errors associated with the movement of 5 assumed reference stations were taken into account, and the movements of the reference points were determined explicitly in their inversion.
The leveling routes and triangulation stations from Matsu'ura et al. [1980] are shown in Figure 7. In all, there are 225 bench marks and 31 triangulation points. Also depicted in Figure 7 is the surface projection of our assumed fault plane (dashed rectangle), the epicenter (star) and the largest located aftershocks (filled circles). The faulting geometry chosen here is similar to that determined by Matsu'ura et al. [1980] and Kanamori [1971]. This faulting geometry is thus consistent with the first-motion data, aftershock area, and the amplitude of surface waves at teleseismic distances (Kanamori, 1971), but encloses the larger slipped area estimated by Matsu'ura et al [1980] from geodetic data. As shown by both Matsu'ura et al [1980] and Takeo and Kanamori [1993], this geometry is also consistent with the gross features of long-period strong motion data recorded at Hongo, Tokyo.
Since we assume a geometry similar to that of Matsu'ura et al. [1980] (with the exception of the dimensions and variable slip and rake in our model), we ignore corrections for the displacement of reference stations relative to those they determined since the overall calculated static displacements from both models are quite similar. The systematic error caused by differential movements of the reference points from our slip model to that of Matsu'ura et al. [1980] is very small.
The fault parameterization and modeling procedure we employ are described by Hartzell and Heaton [1983] in their study of the 1979 Imperial Valley earthquake. Faulting is represented as slip on a planar surface that is discretized into a number of subfaults. The ground motion or static displacement at a given station can be represented as a linear sum of subfault contributions, each appropriately delayed in time to simulate fault rupture.
A constrained, damped, linear least-squares inversion is then used to determine the slip distribution on each subfault that minimizes the difference between the observed and synthetic waveforms and, in this study, between the observed and predicted static uplift and horizontal displacements. We can also allow slip to occur in multiple time windows to allow for complexity of the slip function on each subfault and to allow for variations in rupture times (effectively allowing the rupture velocity to vary). The inversion is stabilized by requiring that the slip is everywhere positive and that the difference in dislocation between adjacent subfaults (during each time window) as well as the total moment is minimized. These constraints have been previously discussed by Hartzell and Heaton [1983].
The advantage of this approach is that we can use a consistent parameterization for both the static and teleseismic data sets. By representing slip on each subfault with the summation of many point sources over the subfault area, we can generate static displacements or teleseismic motions without changing the fault model, only the Green's functions. Further, we have some experience in producing ground motions with this parameterization (e.g., Wald et al. [1991]; Wald and Heaton [1994], facilitating the forward prediction of ground motions at local and regional sites in the vicinity of Tokyo.
To compute the static displacements, the analytic expressions of Mansinha and Smylie [1971] for computing surface displacements due to a uniform displacement at depth are used. We approximate the regional velocity structure with a half-space approximation (with rigidity of 3.0 x 1011 dyne/cm2).
The point source responses, or Green's functions, for teleseismic P or SH body wave synthetic seismograms are computed using the generalized ray method [ Langston and Helmberger, 1975]. We include the responses of all rays up to two internal reflections in a layered velocity model, including free surface and internal phase conversions. An attenuation operator [ Futterman, 1962] is applied with the attenuation time constant t* equal to 1.0 and 4.0 sec for P and SH waves, respectively. The velocity model used to compute the teleseismic Green's functions (given in Table 2) is from Takeo and Kanamori [1993]. The subfault synthetic seismograms are convolved with a dislocation time history which we represent by the integral of an isosceles triangle with a duration of 4.0 sec.
The parameters of the rupture model are summarized in Table 3. Following Matsu'ura et al. [1980], we represent the 1923 Kanto rupture as a 130-km-long plane striking N 70 degress W and dipping 25 degrees toward the northeast (Figure 7). The fault extends from a depth of 2.0 km to 31.6 km, giving a downdip width of 70 km. Although the planar fault approximation may be a simplification of the true rupture surface, it is consistent with the geometry of the Pacific and Philippines Sea plates beneath the Kanto District as determined by Ishida (1992) on the basis of the regional seismicity, focal mechanisms, and seismic velocity structure.
We chose the overall dimensions of the fault to encompass the models of Takeo and Kanamori [1993] and Matsu'ura et al. [1980]. We fixed the epicenter based on the location determined by Kanamori and Miyamura [1970] using local and teleseismic P-wave arrival times. The fault plane geometry chosen here, requires a hypocentral depth of 14.6 km and nearly unilateral rupture towards the east (Figure 7).
We discretized the fault area into 10 subfault elements along strike and 7 elements downdip, giving each subfault a length of 13 km and a downdip width of 10 km. The subfault elements are shown as a gridded overlay on later figures, which display the modeled slip distribution. This subfault area is a compromise chosen to give sufficient freedom to allow slip variations necessary to model the surface displacements and teleseismic waveforms without over-parameterizing the inverse problem. The rupture velocity is assumed to be 3.0 km/sec, or about 85% of the shear wave velocity in the source region (Table 2). Some flexibility in the rupture velocity and slip time history is obtained by introducing time windows Hartzell and Heaton, 1983]. Each subfault is allowed to slip in any of three identical 4.0-sec time windows following the passage of the rupture front, thereby allowing for the possibility of a longer slip duration or a locally slower rupture velocity. The time windows are separated by 2.0 sec, allowing an overlap of 2.0 sec of each 4.0-s duration subfault source-time function windows.
Since the static displacements are computed for a half-space approximation of the layered velocity model used for computing the teleseismic Green's functions, there will be a slight discrepancy between seismic moment estimates from the half-space and the layer velocity model. This is an inherent problem with the use of "seismic moment" and can be mitigated by using "seismic potency", the integral of slip over the rupture area (Heaton and Heaton, [1989]). In this study, we will report the seismic moment based on both the half-space structure and the more realistic earth structure, that we used for the teleseismic Green's functions (Table 2).
Geodetic Data Alone
In the inverse modeling, we weight the horizontal vectors at all station equally; the vertical displacement vectors at the leveling stations are also equally weighted. However, the horizontal weights are 3 times the weight vertical weights to compensate for the larger number of leveling data points. The results of using only the geodetic data in the inversion for slip are shown in Figure 8. Note that the left side of the figure is northwest and the right side is southeast in order to correspond to the map view (Figure 7), hence, the vertical scale goes from deep depths (31.6 km) to shallow depths (2 km). Our geodetic model requires a seismic moment of about 8 x 1027 dyne-cm (Mw=7.9) and a maximum slip near 8 meters. Figure 9 shows the observed horizontal displacements (solid arrows) and those predicted from the geodetic data (dashed arrows); the observed (solid squares) and predicted (open triangles) vertical displacements are shown in Figure 10. There are two regions of concentrated slip (Figure 8), one near the hypocenter and another shallower and southeast of the first. The shallower concentration is required to fit both the large horizontal displacements and the vertical uplift on the Miura and Boso Peninsulas (Figure 9} and Figure 10, R-3 and R-9).
Our model indicates that there is a significant change in the rake vector from about 160 degrees around the hypocentral region to about 110 degrees in the shallower asperity (Figure 8). The need for the change in rake from predominantly strike-slip to near dip-slip is controlled by both the ratio of vertical uplift to horizontal displacement, and the change in azimuth of displacement vectors on the western side of Boso Peninsula toward a more southerly direction than the rest of the displacements (Figure 9). A similar change in rake was found by various investigators for models of thttp://www.scecdc.scec.orgake (discussed by Wald et al. [1991]).
The results of our geodetic analysis (Table 3) compare favorably with those of Matsu'ura et al. [1980]. Both seismic moments are approximately 8 times 1027 dyne-cm and the average slip values are 4.8 m for the single fault model of Matsu'ura et al. [1980], whereas we find a average slip of 3.5 m, indicating a slightly larger slip area determined in our study. The average rake angle in our study is approximately 138 degrees, consistent with that of Matsu'ura et al. [1980] (140 degrees). The two-fault model of Matsu'ura et al. [1980] indicates a change in rake from 147 degrees to a slightly more thrusting slip angle of 138 degrees towards the southeast, in agreement with the change in rake angle found in our model (discussed above).
We have considered whether vertical and horizontal deformation of the Miura and Boso Peninsulas could be attributable to local surface faulting rather than slip on the plate interface. In Figure 7 and Figure 9 we show the locations of two faults that experienced surface displacement during the 1923 earthquake, as documented by Yamasaki (1925, 1926) and Kaine (1967). The southeastern end of the Shitaura fault at the southeast tip of the Miura Peninsula experienced predominantly vertical movement in the range of 30 to 150 cm with the southwest side down over a rupture length of 1 km. The Emmeiji fault near the southern tip of the Boso Peninsula experienced predominantly vertical movement of about 100 cm with the south side down over a rupture length of 3 km. Neither of these faults is close to any benchmarks or intersects the leveling lines, and there is little evidence of perturbation of the broad vertical uplift seen in the leveling lines; the affected parts of these lines would be near station 14 on Line R-3 and station 20 on Line R-9 in Figure 10. We conclude that the geodetic data predominantly reflect the slip on the plate interface.
We also tried a fault dip of 34 degrees based on the fault model of Kanamori [1971]. The resulting model was judged slightly inferior to that described above (dip of 25 degrees). The steeper dip model required a slight shallowing of the larger central asperity (Figure 8) such that slip was unrealistically concentrated toward the shallow edge of the fault, although a slight change in the dip does not violate any of the seismic first motion polarities (Kanamori [1971], Figures 1 and 2) nor does it worsen the teleseismic waveform fits. We note the horizontal displacement misfit at the Oshima station (Figure 14). This site was slightly better fit by the steeper, 34 degrees dip model, but is still poorly fit relative to the other stations. Without more data southwest of the fault plane, it is difficult to assess the significance of this data point.
The geodetic data reflect the deformation resulting from not only the mainshock but also the aftershocks that occurred prior to the geodetic resurveys following the earthquake. We have examined whether any of the aftershocks were sufficiently large and sufficiently close to the geodetic network to have contributed significantly to the measured geodetic deformation. The aftershocks shown in Figure 17 and Figure 18 include four events having JMA (Japan Meteorological Agency) magnitudes larger than 7 (Table 5). The MJMA 7.3 aftershock near the mainshock epicenter in the Sagami Bay, the MJMA 7.0 aftershock located below Tokyo Bay, and the MJMA 7.1 aftershock located furthest off the Boso Peninsula have peak Rayleigh wave amplitudes at Uppsala that are an order of magnitude smaller than those of the mainshock (Table 5), suggesting that their seismic moments are also much smaller. It seems unlikely that any of these events caused significant modifications to the static displacement field of the mainshock.
The MJMA 7.3 aftershock that occurred off the Boso Peninsula at the southeast corner of the inferred rupture plane was significantly larger than the other aftershocks, and has peak Rayleigh wave amplitudes at Uppsala that are about one-half those of the mainshock. This suggests that this event had a fairly large seismic moment, and that its effects might be seen in the slip model derived from the geodetic data. However, the slip model derived from the geodetic data alone (Figure 8) does not have significant slip near the hypocenter of this aftershock. The apparent lack of influence of this aftershock on the geodetic data may indicate that it ruptured to the east, away from land and towards the 1703 rupture zone, and had little slip in the hypocentral region just off the Boso Peninsula.
Figure 11 shows a comparison of the observed teleseismic waveforms (top, labeled) and synthetic seismograms (bottom) predicted from the geodetic model. A single time window is allowed, with rupture propagating uniformly away from the hypocenter. The components are labelled on each observation, with "Z" indicating vertical P waves, "R" indicating radial P waves, and "T" indicating SH waveforms. Note that these data were not used in the inversion, yet the overall amplitude, durations and character are reasonably well fit. We expect to obtain a better fit when these data are included in the inversion.
Geodetic and Teleseismic Data Combined
When inverting for the best slip distribution for both the geodetic and teleseismic data sets combined, the final slip distribution for a joint inversion might not predict the overall amplitudes of each data set equally well, but will represent an average solution. In this study, we chose to independently scale up the slip to best fit each data set (on average) and report those values. This allowed us to determine if there is any discrepancy between the total slips required by the geodetic and teleseismic data. The final model slip and moment estimates are given in Table 5. Note that we are inverting for the slip distribution values, which are the common unknown model parameters for the two data sets.
Figure 12 shows the slip distribution results for equally weighting of the geodetic and teleseismic data sets. In general, the overall features are quite similar to the result using the geodetic data alone, with the slip partitioned into two main asperities. However, the addition of the waveform data has the effect of sharpening the asperity features. Also, there is more variability in both the slip and the rake vectors compared to the geodetic model, indicating that the teleseismic data require a rougher character than the static displacements. In general, geodetic data can be fit with a smoother slip function than waveform data; the geodetic data are sensitive to the total local slip, whereas the waveform data are more dependent on the gradient or variability of the slip.
With the teleseismic data included, we add time windows to allow for temporal slip variations not needed in the geodetic modelling. The geodetic slip is simply computed for the total slip on each subfault obtained by summing the slip in each of the three time windows. The slip contributions for each time window are shown in Figure 13. These are not snapshots of slip on the fault at given times, but instead show the slip that occurs in discrete time intervals following the onset of slip at each point on the fault, with the delay due to rupture removed.
Most of the slip in the hypocentral region occurs within the first time window, while slip in the largest asperity occurs over the three time windows indicating a longer duration on average. This suggests a longer slip duration in regions of larger slip, though the resolution of the teleseismic data and the use of coarse subfault elements precludes more definitive observations. This model has a larger total seismic moment than the geodetic model (see Table 4), and some areas near the fault edges now contribute to the solution. This indicates that some source contributions to the synthetic seismograms may be compensating for inadequate teleseismic Green's functions, most likely on the radial P waves which may have unmodeled receiver complications.
The observed horizontal displacements and those predicted by the fault model derived from the combined data are shown in Figure 13. The observed (solid squares) and predicted (open triangles) vertical displacements are shown in Figure 15. Figure 16 shows a comparison of the observed waveforms (top) and synthetic seismograms predicted from the combined model. There is a much better data fit to the waveforms compared to the geodetic model prediction, in part because we now allow for temporal variations in slip, and also because these data are used explicitly to determine the slip, and so this is the preferred model.
Using only teleseismic data in the inversion does not results in a robust solution for slip, indicating that the waveform data alone are insufficient for resolving the slip distribution. However, Wald et al. [1991] found similar slip distributions could be obtained from separate teleseismic and local strong motion data sets for the 1989 Loma Prieta earthquake. Clearly, the modern broadband teleseismic data coverage and quality played an important role in allowing a solution based on the teleseismic data alone. Wald et al. [1993] found that with suitable teleseismic recordings, the overall rupture characteristics of the 1906 San Francisco earthquake (Mw=7.8) could be determined using turn-of-the-century waveform data and an empirical Green's function summation process. They found that the slip concentrations determined from the teleseismic data were in agreement with many aspects of the geodetic and surface slip data. This was accomplishttp://www.scecdc.scec.orgake recordings from a modern event having a similar mechanism and location to the 1906 rupture (the 1984 Morgan Hill earthquake) as empirical Green's functions for modeling the 1906 mainshock rupture. Unfortunately, the apparent variability of the aftershock mechanisms with respect to the 1923 mainshock mechanism, the lack of suitable modern events in the region, and the need for a variety of depths of aftershocks for use as empirical Green's functions makes such an approach difficult for the 1923 Kanto earthquake.
In order to summarize our slip model we show a map view of the distribution of slip on the model fault plane projected to the surface as determined from the geodetic model in Figure 17 Also shown in Figure 17 are the locations of the largest aftershocks of the 1923 earthquake. Although the locations of the aftershock have some uncertainty, there is reasonable correlation between large slip and the absence of aftershocks, and a tendency for large aftershocks to cluster near the edge of regions of concentrated slip as has been found in other events (e.g., Mendoza and Hartzell, [1988]).
As expected from the location of the largest observed horizontal displacement near the tip of the Miura Peninsula of nearly 3.4 meters (Figure 12), the region of maximum slip is centered below this area, and reaches at least 7.6 meters of mostly thrusting motion (Table 4). Similarly, the region around and southeast of the epicenter has large horizontal surface displacements and vertical uplifts and shows correspondingly large slip at depth, nearly 6.8 meters. The change of slip vector as a function of fault position is shown in Figure 18. There may be some correspondence with the increased degree of thrusting determined along the shallow, central portion of the fault model and the change in the azimuth of the surface manifestation of the Sagami Trough.
The overall area of slip is comparable to that suggested by Matsuda et al. [1978, figure 12], though the extent of slip to the west is larger in our model. Thus the size and likelihood of a future earthquake of the "Oiso" type west of the Kanto rupture zone as suggested by Matsuda et al. [1978] may need to be reevaluated.
The sensitivity of the amplitude, duration and frequency content of the ground motions calculated at Hongo, Tokyo from a wide range of rupture models of the 1923 Kanto earthquake has been documented by Takeo and Kanamori (1993). They found that the best correspondence between their predicted ground velocities and those recorded on the Ewing seismoscope in Hongo, Tokyo was produced by a rupture model (model c5) having concentrated slip in the shallow eastern and western parts of the fault. However, the c5 rupture model is quite different from the one obtained here, which has slip concentrated more in the middle of the fault than at its ends, and at intermediate rather than shallow depth, except in the middle of the fault where it has a large dip-slip component. Further, the c5 rupture model does not satisfy the geodetic data. The predominantly strike-slip motion of the c5 model produces horizontal displacements on the hanging wall in the Miura and Boso Peninsulas that are rotated about 25 degree counterclockwise with respect to the observations, and vertical displacements along leveling lines R-3 in the Miura Peninsula, R-4 along the Shonan coast, and R-9 in the Boso Peninsula that are much smaller than those observed. Also, the shallow hypocenter location at the southwest corner of the fault in the c5 model is incompatible with the location of the epicenter determined by Kanamori and Miyamura (1970), which has an estimated uncertainty of 15 km and which was adopted in our model. This suggests that in order to explain the ground motions inferred from the Ewing seismoscope recording at Hongo, it may be necessary to consider not only source effects but also more complex wave propagation effects, such as the 3-D response of the Kanto basin. It is expected that by using the rupture model presented here, more accurate estimates of the character of the ground motions of the 1923 Kanto earthquake can be obtained.
This paper was improved by reviews from J. Cassidy, T. Heaton, K. Hudnut and an anonymous reviewer. H. Kanamori provided most of the waveform data in the form of original sized copies of seismograms as well as considerable insight into several aspects of this work. K. Shimazaki kindly located the information about surface faulting. Discussions with D. Helmberger, T. Heaton, R. Graves, T. Sato and S. Kataoka were helpful. We are grateful to M. Takeo for providing an early preprint of this manuscript; discussions with him were also useful. This study was supported by the Ohsaki Research Institute of Shimizu Corporation.
Aki, K. (1980). Presidental Address: Possibilities of seismology in the 1980's, Bull. Seism. Soc. Am., 70, 1969-1976.
Ando, M. (1971). A fault-origin model of the great Kanto earthquake of 1923 as deduced from geodetic data, Bull. Earthquake Res. Inst. Tokyo Univ., 49, 19-32.
Ando, M. (1974). Seismotectonics of the 1923 Kanto earthquake, J. Phys. Earth., 22, 263-277.
Futterman, W. I. (1962). Dispersive body waves, J. Geophys. Res., 67, 5279-5291.
Hartzell, S., and T. H. Heaton (1985). Teleseismic time functions for large, shallow subduction zone earthquakes, Bull. Seism. Soc. Am., 75, 965-1004.
Hartzell, S. H. and T. H. Heaton (1983). Inversion of strong ground motion and teleseismic waveform data for the fault rupture history of the 1979 Imperial Valley, California earthquake, Bull. Seism. Soc. Am., 73, 1553-1583.
Heaton, T. H. (1982). The 1971 San Fernando earthquake: A double event? Bull. Seism. Soc. Am., 72, 2037-2062.
Heaton, T. H., and R. E. Heaton (1989). Static deformations from point forces and force couples located in welded elastic Poissonian half-spaces: implications for seismic moment tensors, Bull. Seism. Soc. Am., 79, 813-841.
Imamura, A. (1924), Preliminary note on the great earthquake of S. E. Japan on Sept. 1, 1923, Seismological Notes of Imp. Earthq. Invest. Comm., No. 6.
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.
Kanamori, H. (1971). Faulting of the great Kanto earthquake of 1923 as revealed by seismological data, Bull. Earthquake Res. Inst. Tokyo Univ., 49, 13-18.
Kanamori, H. (1988). in Lee, W. H. K., Meyers, Importance of historical seismograms for geophysical research, H. and K. Shimazaki, eds., Historical seismograms and earthquakes of the world, San Diego, Calif., Academic Press, p.16-36.
Kanamori, H. and S. Miyamura (1970). Seismometrical Re-Evaluation of the Great Kanto, Earthquake of September 1, 1923, Bull. Earthquake Res. Inst. Tokyo Univ., 48, 115-125.
Kanie, Y. (1967). Activity of the Shitaura fault during the great Kanto earthquake, Miura archives (Keihin Express), No. 3, 115-120 (in Japanese).
Langston, C. A., and D. V. Helmberger (1975). A procedure for modeling shallow dislocation sources, Geophys. J., 42, 117-130.
Mansinha, L., and D. E. Smylie (1971). The displacement field of inclined faults, Bull. Seism. Soc. Am., 61, 1433-1440.
Matsuda, T., Y. Ota, M. Ando, and N. Yonekura (1978). Fault mechanism and recurrence time of major earthquakes in southern Kanto district, Japan, as deduced from coastal terrace data, Geol. Soc. Am. Bull., 89, 1610-1618.
Matsu'ura, M. and T. Iwasaki (1983). Study on the coseismic and postseismic crustal movements associated with the 1923 Kanto earthquake, Tectonophysics, 97, 201-215.
Matsu'ura, M., T. Iwasaki, Y. Suzuki, and R. Sato (1980). Statical and dynamical study on the faulting mechanism of the 1923 Kanto earthquake, J. Phys. Earth, 28, 119-143.
Mendoza C., and S. H. Hartzell (1988). Aftershock patterns and main shock faulting, Bull. Seism. Soc. Am., 78, 1438-1449.
Military Land Survey (1930). Re-survey of the Kwanto district after the great earthquake of 1923, Bull. Imp. Earthq. Invest. Comm., 11, 1-6.
Richter, C. F. (1958). Elementary Seismology, W. H. Freeman and Company, San Francisco and London, 768 pp.
Takeo, M, and H. Kanamori (1993). Simulation of long-period ground motions for the 1923 Kanto earthquake (M=8), Bull. Earthquake Res. Inst. Tokyo Univ., 67, 389-436.
Utsu, T. (1982). Japanese Earthquakes 1885-1925, Bull. Earthquake Res. Inst. Tokyo Univ., 57, 111-117.
Wald, D. J., T. H. Heaton and D. V. Helmberger (1991). Rupture model of the 1989 Loma Prieta earthquake from the inversion of strong motion and broadband teleseismic data, Bull. Seism. Soc. Am.,, 81 1540-1572.
Wald, D. J., H. Kanamori and D. V. Helmberger, and T. H. Heaton (1993). Source Study of the 1906 San Francisco Earthquake, Bull. Seism. Soc. Am., 83, 981-1019.
Wald, D. J., and T. H. Heaton (1994), Spatial and temporal distribution of slip for the 1992 Landers, California, earthquake, Bull. Seism. Soc. Am, 84, 668-691, 1994.
Yamasaki, N. (1925). Physiographic studies of the great Kanto earthquake, Report to the Research Group for the Preveition of Earthquake Disasters, No. 100-B, 11-54 (in Japanese).
Yamasaki, N. (1926). Physiographic studies of the great earthquake of the Kwanto district, 1923, J. Fac. Sci. Imp. Univ. Tokyo, Ser. 2, Vol. 2, 77-119.