Strong Motion Seismograms Research Articles

Site-Specific Ground-Motion Waveform Generation Using a Conditional Generative Adversarial Network and Generalized Inversion Technique

ABSTRACT Accurate ground-motion simulations are essential for seismic hazard assessments and engineering practices. Herein, we propose a novel method combining conditional generative adversarial networks (cGANs) and the generalized inversion technique (GIT) to generate site-specific and variability-controlled strong-motion seismograms. The cGANs calculate synthetic seismogram without amplitude scales. The GIT is to separate the source, path, and site characteristics from the Fourier amplitude spectrum (FAS) of the observed seismograms. This method is applied to plate boundary earthquakes off the Pacific coast of Tohoku, Japan. It successfully generates a set of strong-motion seismograms at a given magnitude, distance, and observation station. The output waveforms reproduce the P and S waves as well as coda waves. We validate the method through a quantitative comparison with observed seismograms in terms of both time-domain duration and frequency-domain amplitude characteristics, using metrics of peak ground acceleration (PGA), peak ground velocity, FASs, response spectra, and waveform duration time. The validation results show that the variation in the PGA of the observed seismograms and the synthetic seismograms has a standard deviation of 0.643, and the duration of the seismograms has a standard deviation of 0.346, comparable to the standard deviations seen in the previous studies. Our approach offers high accuracy in stochastic finite-source modeling for a period of 1 s or shorter. The two features of the method, site-specificity and variability control, can contribute to further improvements in seismic hazard assessment by incorporating empirical information based on observed seismograms.

Automatic Digitization of JMA Strong-Motion Seismograms Recorded on Smoked Paper Using Deep Learning

Automatic Digitization of JMA Strong-Motion Seismograms Recorded on Smoked Paper Using Deep Learning

Automatic Digitization of JMA Strong-Motion Seismograms Recorded on Smoked Paper: An Attempt Using Deep Learning

Abstract Using deep learning, an attempt was made to automatically digitize waveforms in scanned Japan Meteorological Agency (JMA) analog seismograms of strong-motion seismographs recorded on smoked paper. We manually traced the waveforms in the scanned images to be used as supervised data for a convolutional neural network (CNN) model using these images. In the analog recording system, ground shaking was recorded on paper with a needle attached to the end of an arm. Because the other end of the arm is fixed on a pivot, the needle moves in an arc around the pivot for large shaking compared to the arm length. We carefully considered the effects of the finite arm length in the analog system and trained the CNNs. To validate the learning results, the trained CNN model was applied to images that were not used for the training. The automatic digitization by our method works fairly well, except for seismograms rich in high-frequency components and those with long-period large amplitudes. Even these images can be well digitized by resizing the image height by, for example, 1/2. To further improve the accuracy of automatic digitization, it would be effective to prepare many additional input and supervised data, and retrain the model. Other training is also necessary for other types of seismographs that have features and characteristics different from strong-motion seismograms.

Coseismic Slip and Early Afterslip of the M6.0 24 August 2014 South Napa, California, Earthquake

AbstractWe employ strong motion seismograms and static offsets from the Global Positioning System, Interferometric Synthetic Aperture Radar, and other measurements in order to derive a coseismic slip and afterslip model of the M6.0 24 August 2014 South Napa earthquake. This earthquake ruptured an ∼13‐km‐long portion of the West Napa fault with predominantly right‐lateral strike slip. In the kinematic seismic slip inversions, we couple the coseismic slip and afterslip distributions by requiring both distributions to involve right‐lateral strike slip with positive amplitude, with the net static slip being the sum of the two. We consider several candidate fault geometries: a first involving two steeply east dipping fault planes that reach Earth's surface at the western surface trace (STW), where most surface rupture was observed, a second involving a steeply west dipping plane that also reaches Earth's surface at the STW, and a third involving a combination of two variably west dipping planes constrained to pass through the locus of postseismic seismicity located ∼1 km west of the STW. The data are best fit using the model of two east dipping fault planes, with coseismic slip up to ∼1.2 m on a dominant shallow asperity about 10 km north of the hypocenter and on deeper asperities on the southern part of the rupture. Afterslip up to 1 m is concentrated along the southern part of the rupture at depths 5 km, consistent with surface observations of afterslip. Seismic moments associated with coseismic slip and afterslip are 1.13×1018 N m (Mw 6.00) and 3.64×1017 N m, respectively.

Automatic imaging of earthquake rupture processes by iterative deconvolution and stacking of high‐rate GPS and strong motion seismograms

AbstractBy combining the complementary advantages of conventional network inversion and backprojection methods, we have developed an iterative deconvolution and stacking (IDS) approach for imaging earthquake rupture processes with near‐field complete waveform data. This new approach does not need any manual adjustment of the physical (empirical) constraints, such as restricting the rupture time and duration, and smoothing the spatiotemporal slip distribution. Therefore, it has the ability to image complex multiple ruptures automatically. The advantages of the IDS method over traditional linear or nonlinear optimization algorithms are demonstrated by the case studies of the 2008 Wenchuan and 2011 Tohoku earthquakes. For such large earthquakes, the IDS method is considerably more stable and efficient than previous inversion methods. Additionally, the robustness of this method is demonstrated by comprehensive synthetic tests, indicating its potential contribution to tsunami and earthquake early warning and rapid response systems. It is also shown that the IDS method can be used for teleseismic waveform inversions. For the two major earthquakes discussed here, the IDS method can provide, without tuning any physical or empirical constraints, teleseismic rupture models consistent with those derived from the near‐field GPS and strong motion data.

Robust features of the source process for the 2004 Parkfield, California, earthquake from strong-motion seismograms

SUMMARY We explore a recently developed procedure for kinematic inversion based on an elliptical subfault approximation. In this method, the slip is modelled by a small set of elliptical patches, each ellipse having a Gaussian distribution of slip. We invert near-field strong ground motion forthe2004September28Mw 6.0Parkfield,California,earthquake.Thedatasetconsistsof10 digital three-component 18-s long displacement seismograms. The best model gives a moment of 1.21 × 10 18 N m, with slip on two distinct ellipses, one with a high-slip amplitude of 0.91 m located 20 km northwest of the hypocentre. The average rupture speed of the rupture process is ∼2.7 kms −1 . We find no slip in the top 5 km. At this depth, a lineation of small aftershocks marks the transition from creeping above to locked below, in the interseismic period. The highslip patch coincides spatially with the hypocentre of the 1966 Mw6.0 Parkfield, California, earthquake. The larger earthquakes prior to the 2004 Parkfield earthquake and the aftershocks of the 2004 earthquake (Mw > 3) also lie around this high-slip patch, where our model images a sharp slip gradient. This observation suggests the presence of a permanent asperity that breaks during large earthquakes, and has important implications for the slip deficit observed on the Parkfield segment, which is necessary for reliable seismic hazard assessment.

Moment tensor inversions using strong motion waveforms of Taiwan TSMIP data, 1993–2009

Earthquake source parameters are important for earthquake studies and seismic hazard assessment. Moment tensors are among the most important earthquake source parameters, and are now routinely derived using modern broadband seismic networks around the world. Similar waveform inversion techniques can also apply to other available data, including strong-motion seismograms. Strong-motion waveforms are also broadband, and recorded in many regions since the 1980s. Thus, strong-motion data can be used to augment moment tensor catalogs with a much larger dataset than that available from the high-gain, broadband seismic networks. However, a systematic comparison between the moment tensors derived from strong motion waveforms and high-gain broadband waveforms has not been available. In this study, we inverted the source mechanisms of Taiwan earthquakes between 1993 and 2009 by using the regional moment tensor inversion method using digital data from several hundred stations in the Taiwan Strong Motion Instrumentation Program (TSMIP). By testing different velocity models and filter passbands, we were able to successfully derive moment tensor solutions for 107 earthquakes of M w >= 4.8. The solutions for large events agree well with other available moment tensor catalogs derived from local and global broadband networks. However, for M w = 5.0 or smaller events, we consistently over estimated the moment magnitudes by 0.5 to 1.0. We have tested accelerograms, and velocity waveforms integrated from accelerograms for the inversions, and found the results are similar. In addition, we used part of the catalogs to study important seismogenic structures in the area near Meishan Taiwan which was the site of a very damaging earthquake a century ago, and found that the structures were dominated by events with complex right-lateral strike-slip faulting during the recent decade. The procedures developed from this study may be applied to other strong-motion datasets to compliment or fill gaps in catalogs from regional broadband networks and teleseismic networks.

弱震時における表層地盤のグリーン関数に基づく強震時の非線形応答に関する研究―２００８年岩手・宮城内陸地震におけるＩＷＴＨ２５観測点の記録―

Two noteworthy points shown in strong motion seismograms recently recorded in near source regions, are the transient response of tilt motion and the nonlinear response of near-surface layer. In the present study, we investigated the nonlinear response of near-surface layer by means of borehole array recordings. First, we estimated the Green’s function of near-surface layer for SH waves assuming that response of near-surface layer is linear for weak motion. Two Green’s functions are estimated for two different data sets which are constructed from the earthquakes occurred before the 2008 Iwate-Miyagi Inland Earthquake and its aftershocks, respectively. Furthermore, we estimated another Green’s function of near-surface layer using the coda part of the main shock of the 2008 Iwate-Miyagi Inland Earthquake. This Green’s function estimated using the coda part of main shock is signicantly different from the other two Green’s functions, although these two Green’s functions are practically identical. This fact reveals that the rigidity of surface layer reduced during the main shock and recovered after that. The coda part of main shock showed the reduced state of near-surface layer’s rigidity. Next, for explaining the strong motion part of main shock, we constructed a time-series model using the Green’s functions estimated using weak motion data. The nonlinear response of near-surface layer in the strong motion part is represented by the time-dependent pole conguration involving in the time-series model, which is constructed by an exponential auto-regressive model of second order. Hence, our constructed model is applicable to the whole process of main shock. Finally, we validated the practicability of model to the data involving the nonlinear response of near-surface layer by using borehole array recordings obtained at the IWTH25 station.

Parkfield earthquakes: Characteristic or complementary?

We model the two most recent Mw∼6 Parkfield, California, earthquakes, which occurred in 1966 and 2004, from a nonlinear global inversion of near‐fault strong motion seismograms. Our rupture models are characterized by spatially variable slip amplitude and rake, rupture velocity, and risetime. The rupture models indicate that the two earthquakes generated slip in regions of the fault that are not identical, as earlier suggested. Given the sparse seismic data set available for the 1966 earthquake, we conduct a series of tests to verify our results: (1) we perform synthetic tests in order to study the resolution of the 1966 seismic data set; (2) we perform an inversion of the 2004 earthquake using a data set equivalent to the 1966 earthquake; and (3) we model the 1966 data set under the a priori assumption that it was similar to the 2004 earthquake. All of the tests, as well as independent observations, indicate that slip during the 1966 and 2004 Parkfield earthquakes occurred in different regions of the fault. This result implies that regions of a fault that are frictionally locked may remain locked even during a main shock (moderate‐size earthquake). In this scenario, large earthquakes occur when all the locked regions of a fault are “synchronized” and ready to slip at the same time.

Seismic Microzoning for Urban Planning

Since the Nobi Earthquake of 1891, seismology, engineering seismology, earthquake engineering, structural engineering and sciences which are related to the phenomena of earthquakes were developed in Japan. Especially pure sciences and engineering made great advances individually. However, actual earthquake damage has not yet ceased to exist, and this damage leads the loss of a great number of human lives. Undeveloped areas and not so important buildings have always suffered these disasters. From worldwise point of view, there is a more serious problem, sometimes more than 10,000 people are killed by an earthquake. The importance of these sciences and engineering is centered in such problems, but at their present stage of development they can not solve these serious problems. Primarily because there are many problems regarding the economical conditions. Seismic microzoning has its purpose in the solution of the serious problems mentioned above. The results of seismic microzoning should cover whole areas and structures, for saving money in construction and to minimize the danger posed to human life. Also seismic microzoning works are related to the disaster prevention programs of various areas. Investigations into seismic microzoning are mainly related to the field of engineering seismology and its applications are in earthquake engineering, but fundamental investigations and applications can not be separated, they are closely related. Thus difficult points still exist, because only a few investigators have interests in the applications of seismic microzoning work. In this paper the author tries to explain the theoretical background and the technique of application of seismic microzoning which developed mainly in Japan. But these theoretical and technical results are not yet completed, and the facts interpreted from the strong motion seismograms and other results of field observations will correct these conclusions.

Kinematic Inversion of the 2004 M 6.0 Parkfield Earthquake Including an Approximation to Site Effects

The 2004 M 6.0 Parkfield earthquake yielded one of the largest amounts of near-source strong ground motion seismic data ever. We invert strong-motion seismograms to obtain a model for the space-time distribution of coseismic slip on the fault. To reduce noise in the inversion, we take into account local amplifications that affect each station by using records of the 1983 M 6.5 Coalinga earthquake. Site amplification correlates well with large peak ground velocities registered during the 2004 Parkfield mainshock. The inversion for a kinematic rupture model yields a nonunique solution; we therefore analyze various rupture models that explain the data equally well. Our preferred rupture model identifies a primary zone of high slip surrounding the hypocenter, where the maximum slip is 57 cm. A secondary slip area, over which contours are not well resolved, is located northwest of the hypo- center. The rupture speed is highly heterogeneous. We infer an average rupture ve- locity of 2.8 km/sec close to the hypocenter, and of 3.3 km/sec in the secondary region of large slip to the northwest of the hypocenter. By correlation of our rupture model with both microseismicity and velocity structure, we identify six patches on the fault plane that behave in seismically distinct ways. Online material: Kinematic rupture model parameters.

A High-Frequency View of the 1999 Chi-Chi, Taiwan, Source Rupture and Fault Mechanics

High-frequency bandpass filtering of broadband strong-motion seismograms recorded immediately adjacent to the fault plane of the 1999 Chi-Chi, Taiwan, earthquake reveals a sequence of distinct bursts, many of which contain quasi-periodic subbursts with repeat times on the order of a few tenths of a second. The subevents that produce these bursts do not appear in conventional slip-maps, presumably because of the low-pass filtering used in the waveform inversions. The origin times, locations, and magnitudes of of about 500 of these subevents were determined. Those closest to the hypocenter appear to have been triggered by the P wave, while the earliest subevents at greater distances are consistent with a rupture front propagating at about 2.0 km/sec. Later subevents at a given distance follow Omori9s law and may be interpreted as aftershocks that begin before the Chi-Chi rupture has terminated. The frequency-magnitude distribution of these subevents has a b -value equal to 1.1. The subevents are clustered in space, with most clusters located at shallow depth along the Chelungpu surface rupture. The larger subevents tend to occur at greater depth, while the small subevents are only located at shallower depths. Cluster locations generally coincide with the large-amplitude slip-patches found by source inversions at lower frequencies. Relocation of the deeper subevents suggests a nonplaner rupture surface where dip increases with depth at the southern end.

Source process and near-source ground motions of the 2005 West Off Fukuoka Prefecture earthquake

AbstractA large shallow crustal earthquake occurred at the western off-shore of Fukuoka Prefecture, northern Kyushu, Japan, at 10:53 on March 20, 2005 (JST). Source rupture processes of the mainshock and the largest aftershock on April 20, 2005, are estimated by the kinematic waveform inversion of strong motion seismograms. The rupture of the mainshock started with relatively small slip, and the largest slip was observed at the southeast of the hypocenter. The inverted source models showed that both of the ruptures of the mainshock and the largest aftershock mainly propagated to southeast from the hypocenters, and the rupture area of those events did not overlap each other. Three-dimensional ground motion simulation by the finite difference method considering three-dimensional bedrock structure was also conducted to see the spatial variation of the near-source ground motion of the mainshock. The result of the simulation shows that expected groud motions are relatively large in and arround Genkai Island, Shikanoshima Island, and the center of Fukuoka City compared to the other area because of the rupture heterogeneity and the deep basin structure in Fukuoka City.

Rupture process of the 1948 Fukui earthquake (M 7.1) from the joint inversion of seismic waveform and geodetic data

We used teleseismic body waves, regional low‐gain strong motion seismograms, and horizontal geodetic displacements from triangulation surveys to invert for the slip distribution of the 1948 Fukui, Japan, earthquake using the multiple time window method. The earthquake ruptured unilaterally from a depth of 10 km to the north with very little slip (0–0.5 m) evenly distributed across a fault of 24 by 18 km over a period of about 10 s. At 9 s after the initiation of rupture, an adjacent fault along the same strike ruptured to the south from the hypocenter to about 20–30 km toward Fukui city mostly along a shallow asperity between 0 and 12 km depth. The large amount of shallow slip (1–1.5 m) below and within the sedimentary basin and strong rupture directivity likely contributed to the high amount of damage in Fukui city. We obtain a total seismic moment of 1.6–2.2 × 1026 dyn cm (Mw 6.74–6.84) which is similar to results by Kanamori (1973), Sagiya (1999), and Kikuchi et al. (1999) using portions of the same regional strong motion and geodetic data sets. We needed to use very low rupture velocities (1.5 km/s) and a hypocenter 10–20 km north of the original location to fit the combined geodetic and seismic data sets with a single fault model. We developed an alternative fault model with two segments allowing for different rupture dynamics for each segment. This resulted in a slightly better fit to the data and allows for a more reasonable rupture velocity up to 2.4 km/s.

The Use of Engineering Seismological Models to Interpret Archaeoseismological Findings in Tolbiacum, Germany: A Case Study

Archaeological excavations carried out from 2001 to 2003 at the vicus Tolbiacum site, located in the present day city of Zulpich, Germany, unearthed heavily damaged and destroyed late Roman fortification works. Damage includes 0.17-m-wide tensile cracks in a 3.1-m-wide wall, tilted walls and fortification towers, horizontal displacement of wall sections of 0.95 m, and rotation of wall fragments. Engineering seismological models are used to test the hypothesis that the observed building damage is of seismogenic nature. The site is located at the western edge of the Lower Rhine Embayment (lre), in the northern part of the Rhine rift valley. More than 20 damaging earthquakes occurred in the area over the past 300 years. Several paleoseismic events with magnitudes of at least 6.4 occurred during the Holocene. Site-specific strong motion seismograms are modeled for assumed large earthquakes at active faults in the lre using stochastic methods and calibrated Q models and duration models. Nonlinear site amplification is calculated based on an S -velocity model from refraction seismic experiments and downhole measurements. Finite element simulation of the dynamic behavior of a fortification tower shows a natural period of the soil-building system of about 3.0 sec. Horizontal and vertical displacements at the top of the 8-m-tall tower reached 0.12 m and 0.06 m, respectively, for the simulated event. No indications of man-made activity exist that could explain the damage. Running water can also be excluded as a threat to the buildings due to the orographic situation, and slow-acting gravitational causes can be excluded due to the damage structure. The destruction and damage is regarded as most probably seismogenic, and site intensity is assessed at IX on the European Macroseismic Scale (ems98) (Grunthal, 1998). This result marks the first time an intensity above VIII is described for a tectonic earthquake in Germany. Preliminary dating results from three carbon-14 (14C) accelerator mass spectometry (ams) samples lead to the second half of the fourth century a.d.

Space‐time model for regional seismicity and detection of crustal stress changes

The regional occurrence rate of earthquakes is modeled as a function of previous activity, the specific form of which is based on empirical laws in time and space such as the modified Omori formula and the Utsu‐Seki scaling law of aftershock area against magnitude, respectively. Its parameters, including the p‐value of the aftershock decay rate, vary from place to place in such a way that, for example, the regional p‐value at the locations around the main shock is similar to the one estimated from the corresponding individual aftershock sequence. The estimated space‐time model is used to visualize features of the regional seismic activity. The utility is demonstrated by applying the model to the events of M ≥ 5 in and around Japan for the period 1926–1995. Among the parameters of the model, this paper is particularly concerned with the normalized K‐value of the modified Omori function indicating the aftershock productivity. This takes high values around the boundaries of asperities that are estimated in the wide area off the East Coast of Tohoku District, Japan, by the inversion of historical strong motion seismograms. Furthermore, this space‐time model enables us to identify the anomalous period and regions where the actual occurrence rates deviate systematically from the modeled one. Such an anomaly (activation and quiescence) relative to the model could sensitively reveal the change of stresses in the regions. We attempt to relate this relative activation and lowering of the seismicity to the pattern of Coulomb stress changes due to a rupture or silent slip elsewhere. For example, such anomalies seen in the seismic activity in the central Japan (M ≥ 2.5) during 1995–1999 and during 2001 are respectively likely the consequence of the Coulomb stress changes due to the 1995 Kobe rupture and the interplate aseismic slip during 2001 beneath the western Tokai region.

The Effect of Bandwidth Limitations on the Inference of Earthquake Slip-Weakening Distance from Seismograms

Numerous researchers have obtained estimates of slip-weakening distance, Dc , and fracture energy for recent earthquakes. Dc is often observed to be a significant fraction of the total slip and tends to correlate with total slip. Although these observations may well be true of real earthquakes, we show that low-pass filtering of strong-motion seismograms can also produce some of these effects in inverted rupture models. We test the accuracy of Dc estimates by calculating them in low-pass-filtered versions of models A and B of Guatteri and Spudich (2000). Models A and B are two different rupture models for a hypothetical M 6.5 earthquake, and they have nearly identical rupture time, slip, and stress-drop distributions, and nearly identical predicted seismograms, but Dc for model B is about twice that for model A. By low-pass filtering slip models A and B at 1.0 Hz, we simulate the blurring effects of band-limited waveform inversions on these slip models. At each point on a fault, ![Graphic][1] is defined to be the slip at the time of the peak slip speed at that point. Low-pass filtering the slip models causes an upward bias in Dc inferred from stress-slip curves, and it causes an artificial correlation between ![Graphic][2] and the total slip. Low-pass filtering might also bias fracture energy high and radiated energy low. These biases should be considered when interpreting Dc derived from band-limited slip models of real earthquakes. [1]: /embed/inline-graphic-1.gif [2]: /embed/inline-graphic-2.gif

A hybrid method for calculating near-source, broadband seismograms: application to strong motion prediction

We present a hybrid method for computing broadband strong motion seismograms in the near-field of large earthquakes. We combine complete seismograms at low-frequency with ray-theory seismograms at high-frequency to form a composite broadband seismogram that spans the entire frequency range of interest. In our approach, the amplitude spectra of the two sets of synthetic seismograms are reconciled at intermediate frequencies where their domain of validity overlaps. We demonstrate the method with scenario earthquakes based on the spatial random-field model for complex earthquake slip [J. Geophys. Res. 107 (B11) (2002) 2308]. The hybrid near-source, broadband seismograms are useful both for detailed source modeling and for incorporating source effects into probabilistic seismic hazard analysis.

Simulation of earthquakes in the Upper Rhinegraben using empirical Green functions

SUMMARY In low seismicity areas such as southwestern Germany, quantitative statements on strong ground, motion suffer from the lack of instrumental strong motion data although the historic catalogue indicates that events of moment magnitudes up to MW = 6 are conceivable. This is the situation in the Upper Rhinegraben area were moderate earthquakes with low probability but high impact caused by the significant aggregation of population, infrastructure and industrial facilities are expected to happen. We have developed scenarios of crustal earthquakes with moment magnitudes MW = 6.0 for a source region southeast of Basel (Switzerland), itself being struck by a strong earthquake in 1356 with an intensity of up to IX (EMS98 intensity). In order to synthesize strong ground motion seismograms, we are using an empirical Green’s function (EGF) approach. We utilize records in the frequency range of 0.5‐25 Hz of permanent stations in the Rhinegraben and the adjacent Black Forest in southwestern Germany as EGFs. To be consistent with the local seismicity, we model earthquakes with a left-lateral strike slip focal mechanism. The applied EGF approach allows one to define independently a variety of rupture scenarios on the fault plane by varying the hypocentre location, rupture velocity, fault plane geometry, slip distribution and rise time. Therefore, we can create worst-case scenarios for the chosen sites. We compare peak horizontal acceleration (PHA) and response spectra in terms of spectral accelerations and PHAs with attenuation laws proposed for Europe. Our results encourage the application of the EGF approach as a supplementary tool for site-specific strong ground motion prediction in low-seismicity regions such as the Upper Rhinegraben area.

地震記象から見た１９４６年南海地震の震源過程

Subevents of the 1946 Nankai earthquake (Mjma8.0) were distinguished from seismograms recorded on smoked papers with strong motion seismograms (T0=6s, amplification=2) and seismological data of Japan Meteorological Agency. Location and origin time of each subevent were roughly estimated from a hypocenter determination with phase readings of P and S waves for the subevent. Rupture of the Nankai earthquake started at a point southward about 50km off Cape Shionomisaki (first subevent). The rupture propagated in the NNW direction with a rupture velocity of 1.9 km/s and a second subevent that grew to class M8 started after about 16 sec from the initiation of the first subevent. Then the rupture propagated westward, and third subevent of M8 occurred around Tosa Bay, located about 200 km from the second event, about 53 sec after the initiation of the second event. Seismic waves were excited near Tosa Bay during the 1946 Nankai earthquake.