Bulletin of the Seismological Society of America, Vol. 95, No. 6, pp. 2297.2317, December 2005, doi: 10.1785/0120040065 Assessment of Site Effects on Seismic Motion in Ashigara Valley, Japan by Tomiichi Uetake and Kazuyoshi Kudo Abstract We compared site ampli.cations at rock sites and sediment sites of Ashigara Valley, Japan, using ground-motion data from .ve remote (700 km) large (M7) events. The use of remote large events is advantageous to estimating site factors because the source and path effects are considered to be common with a suf.cient accuracy and the ground motions will cover a wide-frequency band. Ground motions at both sediment and rock sites were coherent in frequencies lower than 0.1 Hz. This means that the wavelength in these frequencies is longer than the size of the valley (12 km long and 5 km wide). Site ampli.cation factors were determined by taking spectral ratios with reference to one rock outcrop site. The ampli.cation factors of sediment sites deviated 2.10 times with respect to the rock site in the frequency range higher than 0.1 Hz, in which signi.cant peaks at about 1.2 Hz were found at most sites. These dominant ampli.cations in sedimentary basin are most essential for assessing earthquake hazard in the region. For sediment sites, the peak frequencies of spectral ratios to the rock sites were stable for different events and coincided with those of horizontal to vertical spectral ratios for the S-wave portion and those of relative site factors estimated separately by the generalized inversion method using local small-events data in the frequency range higher than 2 Hz. Although spectral ratios for frequencies lower than about 1 Hz should be affected by 3D basin structure, 1D S-wave responses represent the ampli.cation of ground motion in the sediment sites for frequencies higher than 2 Hz. Introduction The importance of site effects on seismic motion has and site factors. This method is excellent in understanding been realized since the early stages of seismology (e.g., Im-absolute ampli.cation/deampli.cation of body waves (S-amura, 1929; Ishimoto, 1932). Since then, various studies wave) at a site; however, uncertainties still remain because have been conducted. However, it is only recently that quan-of assumptions on source radiation pattern, geometrical titative studies have been conducted using strong-motion ar-spreading factor, and so on. A common issue for the spectral ray data. Several methods have been proposed for evaluating ratio method and the generalized inversion method is the site effects by using ground-motion records, such as soil-to-selection of a reference site. The third method using a hor-rock spectral ratios (e.g., Borchert, 1970), a generalized in-izontal-to-vertical spectral ratio is convenient in cases where version (e.g., Andrews, 1982; Iwata and Irikura, 1988; Boat-no reference site is available. However, there is still some wright etal., 1991), and horizontal-to-vertical spectral ratios controversy on its reliability. (e.g., Nakamura, 1988; Lermo and Chavez-Garcia, 1993; We used data from remote large events to discriminate Field and Jacob, 1995; Yamazaki and Ansary, 1997). Be-empirical site responses in the sedimentary basin by the cause of the assumptions in these methods and a limitation spectral ratio method, because both source and path effects in data, these methods are also limited in accuracy or appli-are assumed to be common with suf.cient accuracy and the cability. Therefore, a comparative study of these methods on ground motions contain a wide frequency band. From a simi-one area will be useful to validate the applicability of each lar viewpoint, Sasatani etal.(1992) used the events imme-method to other sites (e.g., Field and Jacob, 1995; Riepl etdiately beneath the site at intermediate depth to evaluate the al., 1998). The .rst method practically gives reliable but basin effects. relative ampli.cation factors at a site, as a function of fre-Obtaining horizontal-to-vertical spectral ratios using the quency. This method, however, requires that the distance same data sets, we examined the reliability of this method from a reference site to a target site be close enough com-for empirical site effects. Site ampli.cation factors are also pared with the source distance, because it assumes the same estimated by the generalized inversion method (e.g., Iwata source and path effects. The generalized inversion method and Irikura, 1988) using local events near Ashigara valley. has an advantage in that it simultaneously gives source, path, Through a comparative study of these results, we will con- 2297 .rm that the spectral ratio method using remote large-event data is advantageous. Data Ashigara Valley is a sediment-.lled valley with middle-sized dimensions (12 km long and 5 km wide) located in the western Kanagawa Prefecture, Japan. By historical docu-ments since 1600, we note that this region has been severely damaged by several large earthquakes. The most recent dev-astating event was the 1923 Kanto earthquake. The valley is surrounded by the Hakone Volcano in the west, Tanzawa Mountains in the north, ans Oiso Hills in the east, and it opens to Sagami Bay in the south. The Kozu-Matsuda and Kannawa active fault systems (e.g., Yamazaki, 1992) are indicated at the east and north margins of the basin. For the most part, we used data acquired at the Ashigara valley strong-motion accelerograph array (Kudo and Shima, 1988; Kudo etal., 1988) installed and maintained by Earthquake Research Institute (ERI), University of Tokyo. The ERI de-ployed a strong-motion accelerograph array in the sedimen-tary basins and in rock outcrops at surrounding mountains to study the effects of surface geology (ESG) on seismic motions. The strong-motion observation network used in this study is shown in Figure 1. The network consists of 6 rock outcrop sites surrounding the valley and 15 sediment sites, including temporal stations, in the sedimentary basin. The Kuno area, which is located in the west margin of the valley, was selected as the international test site of the ESG study. Strong-motion records from the 1990 Odawara earthquake (MJMA 5.3) were used in the blind prediction studies (Kudo, 1992; Kudo and Sawada, 1998). The rock site KNO and the sediment sites KNP and KNS were used in the blind predic-tion study with the different station codes, namely, KR1, KS1, and KS2, respectively. The geology of the rock sites is complex; HYK, KNO, and SJJ are andesite of Quaternary and HSR, AKD, and KHZ are tuff breccia, basalt, and mud-stone of Tertiary, respectively. The sediment sites are also on various different geological conditions. KNP, KDW, and NGW are on the Kanto loam (volcanic sandy sediments), KRD and KNS are on humus soils, and other sediment sta-tions are on sandy soils of Quaternary deposits. We analyzed data sets obtained during the 1993 Ku-shiro-Oki earthquake (MJMA 7.5), the 1994 Near Vladivostok earthquake (MJMA 7.6), the 1994 Hokkaido Toho-Oki (Shi-kotan) earthquake (MJMA 8.2), the 1994 Sanriku Haruka-Oki earthquake (MJMA 7.6), and the 2000 Trishima-Kinkai earth-quake (MJMA 7.2). Locations of the epicenters are also shown in Figure 1. Earthquake parameters determined by the Japan Meteorological Agency (JMA) are shown in Table 1. The epicenters of these earthquakes are very distant from Ashi-gara Valley, so that we can expect to have plane incident wave and the common source and path effects within the network. The observed peak ground accelerations (PGAs) are listed in Table 2. Because the PGAs were mostly small, we can assume linear site responses. We should note that a T. Uetake and K. Kudo few sites (JNI, KDW, and NGW) could retrieve the ground motions during one or two events. The PGA ratios for sed-iment to rock sites are between 3 and 5. Ampli.cation fac-tors of sediment sites KNS, NRD, and TKD are quite large compared with the others, and the ampli.cation factor of HYK is the largest among the rock sites. For the generalized inversion analysis using small events, we used 19 ERI stations, 4 K-NET (Kinoshita, 1998) stations deployed by the National Research Institute for Earth Science and Disaster Prevention (NIED), 6 stations de-ployed by the Central Research Institute of Electric Power Industry (CRIEPI), and 2 stations maintained by The Tokyo Electric Power Co. Ltd. (TEPCO). Shallow events with mag-nitudes from 4.0 to 5.5 were selected as shown in Table 3, and the list of station-event matrices is shown in Table 4. We classi.ed the sites simply as rock and sediment, indi-cated by roman and italic letters, respectively, in Tables 2 and 4, based on their surface geology and irrespective of their S-wave velocities or thickness of surface layers. The locations of observation stations and epicenters are shown in Figure 2. Epicentral distances are from 12 to 44 km, and are within 2.5 times the source depths. The instruments of the ERI stations use force-balanced accelerometers. They have .at responses from direct 0 to 30 Hz (100-Hz sampling), with clipping levels at 2000 cm/ sec/sec. Most digitizers of these instruments are 16 bits, al-though 14 bits are used for the real part whereas the rest, 2 bits, was used for gain-ranging (.1, .4, .16); that is, the least signi.cant acceleration is 0.015 cm/sec/sec for accel-eration less than 125 cm/sec/sec. Clocks of the recording systems were adjusted every 12 hr in a day by time signals received from broadcasting services of FM radio. In 1998, most accelerographs in Ashigara Valley were replaced to the K-NET95 type seismometers (24-bit digitizer; Kinoshita, 1998). The K-NET95 type seismometer has the same instru-mental response as the older one and uses the time code produced by Global Positioning System (GPS) timer. There-fore, the quality of data from event L5 is much higher than the others. Different types of accelerometers have been used at the stations of CRIEPI and TEPCO, but the instrumental responses are almost equal in the frequency range from 0.1 to 20 Hz. Waveform Features of Large Events We analyzed the data from the .ve large events in Ta-ble 1. In general, acceleration seismograms are suitable to study the characteristics of high-frequency components; however, we used velocity seismograms to study broadband waveform characteristics. To con.rm the reliability of low-frequency contents in the following analysis, we compared our records with that of STS-1 at near distance. Figure 3 shows the site locations, the comparison of velocity wave-forms of strong motions integrated without .lter (YGS) and the STS-1 (JIZ) record, and the spectra comparisons. Even in the worst case (east.west), we may use higher-frequency Figure 1. Epicenters of remote large events used in this study and the observation network in Ashigara Valley deployed by the Earthquake Research Institute, University of Tokyo. Station names in roman and italic letters indicate rock and sediment sites, respectively. Broken lines indicate active fault lines. The circles with numbers indicate the epicenters, and the numbers express the locations of earthquakes listed in Table 1. Table 1 List of Remote Large Earthquakes Used in This Study Origin Time (JST) Earthquake Latitude Longitude Depth Epicentral Distance Back Azimuth No. yyyy/mm/dd hr:min:sec (N) (E) (km) MJMA Mw (km) () L1 1993/01/15 20:06:07.2 42.917 144.357 101 7.5 7.6 962 30 L2 1994/07/22 03:36:31.5 42.277 133.550 552 7.6 7.3 916 326 L3 1994/10/04 22:22:56.9 43.372 147.678 28 8.2 8.3 1162 42 L4 1994/12/28 21:19:20.9 40.427 143.748 0 7.6 7.8 702 37 L5 2000/08/06 16:27:13.3 28.817 140.089 444 7.2 7.4 721 173 Source parameters were determined by the Japan Meteorological Agency. Epicentral distances and back azimuths were calculated from the KNO station. Mw values were determined by the U.S. Geological Survey. motion than 0.04 Hz. Numerical integration by Fast Fourier 0.04 Hz for event L5. Velocity waveforms are shown in transform (FFT) was applied to the acceleration records con-Figure 4. Every velocity waveform shown in Figure 4a.e sidering the signal-to-noise ratio (S/N) in the low-frequency has a duration longer than 90 sec, and most events were range. A cosine-tapered, low-cut .lter was applied before retrieved before the initial Sarrivals. As is clearly evident in integration with cutoff frequency of 0.05 Hz for events L1 the .gures, the amplitudes at rock sites (indicated by roman and L2, 0.02 Hz for event L3, 0.03 Hz for event L4, and letters) are very small compared with those at sediment sites 2300 T. Uetake and K. Kudo Table 2 Peak Ground Accelerations of Observed Records Peak Ground Acceleration (cm/sec/sec) Station L1 L2 L3 L4 L5 HSR SJJ AKD KNO KHZ HYK 1993/01/15 3.2 . 1.8 2.6 2.5 4.7 1994/07/22 2.4 1.8 . 1.8 . 1.8 1994/10/04 4.5 4.8 . 3.7 3.5 6.7 1994/12/28 1.7 2.2 1.8 2.3 2.6 3.7 2000/08/06 1.0 1.6 1.0 1.3 1.3 1.7 KNP KNS KRD KYM NRD TKD CTS SHS NSK HKC SKW SCH JNI KDW NGW . 10.2 . 7.7 8.8 9.4 7.7 . 5.4 . 7.3 . 9.4 . . . 9.0 . 3.7 6.8 6.8 5.8 4.5 3.8 3.0 3.6 . . . . 15.7 14.7 12.5 7.8 14.2 20.5 12.0 11.1 8.2 9.2 10.3 11.5 11.7 . . . . 7.6 4.2 8.4 . 7.3 7.4 5.4 4.7 5.6 5.0 . . . . 4.7 . 2.6 4.2 8.7 4.6 4.8 3.2 2.3 2.6 . . 5.5 4.5 Earthquake numbers are the same as those in Table 1. The bigger value between the north.south and east. west components is taken as the value in the table. Station codes in roman and italic letters denote rock and sediment sites, respectively. Table 3 List of Events Used for Generalized Inversions Origin Time (JST) Earthquake Latitude Longitude Depth No. yyyy/mm/dd hr:min:sec (N) (E) (km) MIMA 1 1990/08/05 16:13:02.1 35.207 139.095 13.6 5.3 2 1994/10/04 02:56:00.4 35.480 139.078 23.6 4.4 3 1996/03/06 23:12:27.6 35.471 138.945 19.3 4.6 4 1996/03/06 23:35:28.7 35.473 138.951 19.5 5.5 5 1996/10/25 12:25:17.6 35.452 139.005 22.6 4.7 6 1997/11/04 10:31:08.3 35.249 139.107 14.6 4.1 7 1999/05/22 09:48:15.5 35.456 139.180 20.9 4.3 8 2000/02/11 20:57:04.0 35.496 139.047 16.8 4.4 9 2000/05/02 23:32:59.4 35.300 139.068 15.6 4.0 Source parameters were determined by the Japan Meteorological Agency. (indicated by italic letters). One exception is event L3, show-ing similar peak ground velocities (PGVs) almost all sites irrespective of the site geology. The waveforms of event L3 show a clear dispersion of surface waves (Uetake and Kudo, 1998), predominantly in the frequency of 0.03.0.04 Hz. The PGV is mostly controlled by the low-frequency surface waves, but if we pay attention to ground motions for fre-quencies higher than 0.1 Hz (see Fig. 5a), the same ampli-.cation in the sediment site as seen in the other events is found. Bandpass-.ltering technique is applied to study the difference of site response for broad-frequency bands. The .ltered waveforms of events L3 and L5 are shown in Figure 5. Frequency bands were selected for logarithmically equal width of frequency. Bandpass-.ltered waveforms for both events in the frequency range from 0.0464 to 0.1 Hz are very coherent, although they possessed different features for dif-ferent events, i.e., later arrivals dominate for event L3, whereas pulselike Swaves are clear for event L5. This sug-gests that seismic wavelengths in these frequencies are suf-.ciently longer than the size of the valley and that the S/N of the records are preserved in these low frequencies. In the frequency range from 0.1 to 0.215 Hz, sediment site records are different from those of rock sites with respect to ampli-tude and phases. Waveforms at sediment sites tend to have larger and longer wavepackets than those at rock sites. How-ever, even though it is a rock site, the KHZ waveform is similar to those of sediment sites. On the other hand, the waveform at sediment site KYM is similar to those of rock sites. In the higher-frequency range of over 0.215 Hz, dif-ferences between waveforms from sediment and rock sites become large. The peak values at the low-frequency band appear at the same phase for all stations, but those at the high-frequency band are found at different phases. Reference Site of Ashigara Valley Rock types or physical parameters at the observation sites are not the same among the rock outcrop sites of the Assessment of Site Effects on Seismic Motion in Ashigara Valley, Japan 2301 Table 4 Data List for Inversion Analysis Latitude Longitude Earthquake No. Institutions Station (N) (E) 1 2 3 4 5 6 7 8 9 K-NET KNG01035.3322 139.3536 . . . . . . . . . 4 KNG012 35.3761 139.2080 . . . . . . . . . 4 KNG013 35.2608 139.1552 . . . . . . . . . 4 KNG014 35.3575 139.0858 . . . . . . . . . 4 CRIEPI MNZ 35.1392 139.1553 . . . . . . . . . 4 JZD 35.3092 139.0286 . . . . . . . . . 8 KND 35.2700 139.1508 . . . . . . . . . 9 KKM 35.2086 139.1439 . . . . . . . . . 9 KZR 35.2464 139.1275 . . . . . . . . . 7 OYM 35.3592 139.0025 . . . . . . . . . 9 ERI AKD 35.3331 139.1894 . . . . . . . . . 6 CTS35.2744 139.1914 ... . . . . . . 5 HKC 35.2653 139.1883 . . . . . . . . . 5 HSR 35.3669 139.1039 . . . . . . . . . 9 HYK 35.2383 139.1464 . . . . . . . . . 9 JNI 35.2464 139.1600 . . . . . . . . . 5 KHZ 35.2861 139.2106 . . . . . . . . . 8 KNO 35.2706 139.1250 . . . . . . . . . 9 KNP 35.2696 139.1402 . . . . . . . . . 6 KNS 35.2666 139.1516 . . . . . . . . . 9 KRD 35.2616 139.1506 . . . . . . . . . 5 KYM 35.3103 139.1511 . . . . . . . . . 9 NRD 35.2839 139.1700 . . . . . . . . . 9 NSK 35.2650 139.1817 . . . . . . . . . 8 SCH 35.2616 139.1904 . . . . . . . . . 5 SHS 35.2715 139.1891 . . . . . . . . . 5 SJJ 35.2994 139.0786 . . . . . . . . . 9 SKW 35.2664 139.1947 . . . . . . . . . 9 TKD 35.2847 139.1947 . . . . . . . . . 8 TEPCO NSG35.2878 139.1214 .. . . . . . . . 4 SFJ 35.3692 138.9603 . . . . . . . . . 3 Total 18 21 24 24 30 29 22 21 18 207 Station codes in roman and italic letters denote rock and sediment sites, respectively. The left column lists the observation institutions: K-NET; CRIEPI, The Central Research Institute of Electric Power Industry; ERI, Earthquake Research Institute, University of Tokyo; and TEPCO, Tokyo Electric Power Company. Ashigara Valley network. By using datasets from events L4 and L5 (see Table 1), which were recorded at all rock sites, we obtained spectral ratios of individual rock sites with re-spect to the average of all rock sites, as shown in Figure 6. Data over 5 Hz from event L4 were not used because of its low S/N. The deviations of spectral ratios are within a factor of 2 in a wide-frequency band; in particular, it is within 20% in frequencies lower than 0.1 Hz. This implies that the wave-length in this frequency range is longer than the size of ir-regularities in Ashigara Valley and that the quality of ob-served records are good, even at very low frequencies. The spectral ratios of horizontal components at KNO and SJJ are close to the average in the broad-frequency range. Those at HSR and AKD are slightly smaller in the intermediate-frequency range (0.1.1.0 Hz); on the other hand, those at HYK and KHZ are larger in the same frequency range. In the high-frequency range over 2 Hz, HSR and HYK show large ampli.cation factors of 2 or 3. These spectral responses at rock sites may be attributed to the effects of weathering rock (Steidl etal., 1996), rock types and/or surface topog-raphy. For the reference site, it is desirable that the spectral response of the site is smooth with respect to frequency and minimum on average compared with others. For these rea-sons, AKD and SJJ are good choices for the following anal-ysis. Unfortunately, some events could not be retrieved (see Table 2), and so, KNO was .nally chosen as the next-best reference site, considering that KNO was also used as a ref-erence site in the ESG blind prediction study (Kudo, 1992; Kudo and Sawada, 1998). Relative Site Factors Evaluated UsingLarge-Events Data In general, Fourier spectra of observed record at station jis represented by, Figure 2. Epicenters and observation stations used for inversion analyses. Open circles show the epicen-ters. Event numbers are the same as those in Table 3. Filled squares show the observation stations. Oj( f) .S( f) .Pj( f) .Gj( f) .Ij( f), (1) where, fis frequency, S(f) shows source spectra. Oj(f), Pj(f), Gj(f), and Ij(f) represent observed record, pass effect, site response, and instrumental response of station j, respec-tively. Assuming that the source and path effects from re-mote events are common for all stations, and that Ij(f) is also the same for all stations, the spectral ratio of the observed record at a site jto the reference site record 0 gives the relative site response, Oj( f)/ O0( f) .Gj( f)/ G0( f). (2) Expression (2) gives an empirical (relative) ampli.ca-tion factor if an input motion from the base rock is approx-imately equal for all the stations. Note that, as discussed in the preceding section, KNO was chosen as the next-best ref-erence site. Spectral ratios of individual events and their overall av-erage are plotted in Figure 7a.c. These spectra were com-puted using a time window of 81.92 sec, taking the initial part of the S-wave arrival. The spectral ratios for stations that recorded more than three events are shown. Data over 5 Hz from event L4 are not plotted because of low S/N and were not included in obtaining the average. The spectral ratios of every component are stable in a broad-frequency band, irrespective of the event. The spectral ratios of horizontal motions, even for the rock sites, have strong variations against frequency. Signi.cant troughs at about 0.5 Hz appear at almost all north.south (NS) com-ponents of rock sites, although it is not clear in the east. west (EW) component. This is due to the spectral peak at about 0.5 Hz of the reference site KNO. In addition, up. T. Uetake and K. Kudo down (UD) components show different features, with the rock sites at AKD and KHZ having strong peaks at about 0.2 Hz. Thus, every rock site, including KNO, has local site response. However, their averages are 0.9.1.3 in the fre-quency range lower than 0.1 Hz and 0.5.2.0 for the rest, except for frequencies higher than 4 Hz at a few sites (HSR, HYK, and SJJ). At sediment sites, ampli.cation factors in frequencies lower than 0.1 Hz are similar to those at rock sites; however, they increase from about 0.1 Hz, reaching up to 5.10 times at about 1.2 Hz and at higher frequencies. Troughs at about 0.5 Hz in the spectral ratios of almost all sites, both sediment and rock sites, are again probably brought about by the am-pli.cation at the reference (KNO) site. Large peaks at about 1 Hz are commonly found for horizontal spectral ratios of sediment sites, and relatively large ratios at low frequencies of 0.15 and 0.35 Hz are roughly three times at the southeast valley (CTS, SHS, NSK, HKC, and SKW). The trough at about 2 Hz and peak at about 3 Hz are also common features of the southeast valley. These peaks cannot be attributed to the spectral response of KNO, because strong peaks or troughs are not found at rock sites. In the up.down com-ponent, the spectral ratios are smaller than those of horizon-tal components but the peak at about 0.2 Hz is very distinct, especially in the southeast valley (CTS, SHS, NSK, HKC, and SKW), including rock sites AKD and KHZ. A predom-inance of 1-Hz ground motion in Ashigara Valley was sug-gested by Kawase and Sato (1992), caused by the ampli.-cation due to both strati.cation of quaternary deposits, whose predominant frequency is about 1 Hz, and basin-induced Love waves from the 1987 Chiba-ken Toho-oki earthquake. Spectral ratios of north.south and east.west components are not the same at some sites, especially at KNS. The north.south component KNS/KNO ratio has two signi.cant peaks at about 1 and 2 Hz, whereas the peak at about 1 Hz can not be found in the east.west component. Sato etal.(1998) interpreted the difference as a 2D effect of the small-basin structure, and Kudo and Sawada (1998) interpreted it as an effect of anisotropy of volcanic sediments at shallow intermediate layers. Both effects may contribute to this difference. Spatial distributions of the spectral ratios with respect to KNO are plotted in Figure 8. These .gures were made by spatial interpolation of spectral ratios using the surface-gridding algorithms by Smith and Wessel (1990). Horizontal ampli.cation factors are almost homogeneous at 0.1 Hz, gradually becoming large toward the southeast of the valley at 0.2 and 0.5 Hz. At 1 Hz, the ampli.cation factors tend to concentrate at the southeast of valley (NRD, TKD, and CTS). Those of 2 Hz show a rather complex pattern of two dominating areas near TKD and KNS. This ampli.cation pattern roughly corresponds to the surface geology, which is predominantly backmarsh at both sites. Spatial distribu-tions for the UD component show large ampli.cation factors at the rock sites AKD and KHZ and at the sediment sites of southeast valley at 0.2 Hz. This pattern is different from the Figure 3. Comparison of velocity waveforms and Fourier spectra observed at YGS and JIZ. These data were obtained from event L5 (2000 Torishima-Kinkai event) in Table 1. Waveforms of YGS were integrated from the original acceleration data. Fourier spectra shown by solid and broken lines are data for JIZ and YGS, respectively. The upper-left map shows the locations of YGS and JIZ stations. distribution of young sediments. Details for the low fre-quency up.down motion is outside the scope of this article. Horizontal to Vertical Spectral Ratio The method using horizontal-to-vertical spectral ratios (HVRs) of earthquake motions has been proposed as a sub-stitute for site response (e.g., Nakamura, 1988; Lermo and Chavez-Garcia, 1993; Seekins etal., 1996), which employs a simple interpretation that surface vertical motions are P-waves converted from S-waves at a certain depth and prop-agated vertically as P-waves; therefore, less ampli.cation is expected in a sediment site, whereas surface horizontal mo-tions are ampli.ed because of large changes of S-wave ve-locity in sediments. If the method is valid everywhere, the advantage will be immense for assessing the site effects by using a single station or in an area where a rock reference site cannot be found. However, caution in the use of the method has been indicated by Field and Jacob (1995), Bon-illa etal.(1997), and Satoh etal.(2001), because HVRs tend to underestimate or lose correlation with horizontal spectral ratios (HHRs). Most of the studies analyzed records by dividing them into their S-wave and coda parts to discriminate the wave types that are concerned with the interpretation of the cor-respondence of HVRs to site responses. However, we ob-tained HVRs for large events using the same time window to compare with HHRs, which are representative of empirical ampli.cation factors, as shown in Figure 9. The geometrical average of the north.south and east.west components is used as a horizontal value. Spectral ratios are stable in the frequencies higher than roughly 1.0 Hz irrespective of events and site classi.cations. However, their deviations are very large at frequencies lower than 0.5 Hz, similar to Lermo and T. Uetake and K. Kudo Figure 4. Velocity seismograms of large remote events observed in the Ashigara Valley network. (a) 1993 Kushiro-Oki earthquake (MJMA 7.5). (b) 1994 Near Vladi-vostok earthquake (MJMA 7.6). (c) 1994 Hokkaido Toho-Oki earthquake (Shikotan) (MJMA 8.2). (d) 1994 Sanriku Haruka-Oki earthquake (MJMA 7.6). (e) 2000 Trishima-Kinkai earthquake (MJMA 7.2). Each trace was integrated from the original acceleration data in the frequency domain. The letters written on the left side of waveforms indicate station code, and the numbers written on the right side of the of waveforms are the PGV values of the waveforms. Station codes in roman and italic letters represent rock and sediment sites, respectively. (continued) Figure 4. Continued. T. Uetake and K. Kudo Figure 4. Chavez-Garcia (1993). The geometrical averages of HVRs are compared with those of HHRs in Figure 10. The large deviations of HVRs in the low-frequency range may be at-tributed, in part, to SV-wave incidences with different an-gles, even if the simple interpretation of HVRs (Nakamura, 1988) is valid. HVRs at sediment sites are similar to HHRs at frequencies higher than 0.5 Hz. However, the HVR at NRD underestimates by a factor of 2, and those at CTS, SHS, and TKD are also smaller than HHRs, even at higher frequencies. On the contrary, HVRs at rock sites strongly overestimate HHRs without exception. The HVRs for KNO have a peak at about 0.2 Hz and there are no signi.cant troughs at about 0.2 Hz in HVRs at the sediment sites despite the signi.cant peak in spectral ratios of the up.down com-ponent as shown in Figure 7c. These results suggest that the frequency response of up.down component at KNO have a trough as shown in Figure 6 (panel 3) and the trough makes the peaks at about 0.2 Hz as shown in Figure 7c. Satoh etal.(2001) examined the differences of empir-ical site ampli.cation using different time windows, such as Swaves, Pwaves, coda, and microtremors. They showed that HVRs for the S-wave portions are different from HHRs, whereas HVRs are consistent with theoretical HVRs for obliquely incident (30) SVwaves. They also indicated that the S-coda portions at soft-sediment sites are mixed with surface waves and that the S-coda HVRs in frequencies lower than 3 Hz may be interpreted by the fundamental mode of Rayleigh waves. Because of the long time window used in our analysis, HHRs in the present results re.ect not the ampli.cation fac-tor of Swaves alone, but the empirical-site factors, including secondary generated surface waves. Judging from the sta-bility and similarity of HVRs with HHRs, HVRs may be used as substitutes for HHRs in frequencies higher than 0.5 Hz for the empirical-site ampli.cation factors. However, at the same time, the deviations of HVR at sediment sites are very large at frequencies lower than 0.5 Hz and, at rock sites, overestimate the site ampli.cation factor. The large devia-tions of HVRs at low frequencies may be attributed to the variations of incident angles of seismic motion; therefore, the average of HVRs at low frequency may not be meaning-ful. However, the average has a certain meaning if plural incident angles have to be considered because of multipath incoming waves from distant events. Site Factors Evaluated by the GeneralizedInversion Method In this section, we compared the site factors obtained by the .rst method (HHRs) with those estimated by the gen-eralized inversion method (Iwata and Irikura, 1988) using small local events. We analyzed moderate ground motions Figure 5. North.south components of bandpass-.ltered velocity seismograms. The top plots are wave-forms for L3, and the bottom plots are for L5. The letters written on the left side of the waveforms indicate station code, and the numbers written on the right side of waveforms are the PGV values of the waveforms. T. Uetake and K. Kudo Figure 6. Fourier spectral ratios of each rock site with respect to the average of all sites. Panels 1, 2, and 3 show the results for the north.south (NS), east.west (EW), and up.down (UD) components, respectively. Thick lines and broken lines represent the ratio to the average spectra of events L4 and L5 in Table 1, respectively. at 31 sites, from 9 events that occurred near Ashigara valley, The observed S-wave Fourier amplitude spectrum is ex-as shown in Table 4. These earthquakes are mostly shallow, pressed by, with magnitudes from 4.0 to 5.5, as shown in Table 3. The location of observation sites and epicenters are shown in Figure 2. ij i ij ij ssj O( f) .S( f) .R.1 .exp(.pRf/ Q( f)v) .G( f), (3) Figure 7. Fourier spectral ratios or ampli.cation factors of the strong-motion sites with respect to KNO. (a), (b), and (c) show the result for the north.south (NS), east.west (EW), and up.down components, respectively. Only stations with more than three events retrieved are displayed. The spectral ratio for each event and their average are displayed. These spectral ratios were smoothed by Parzen windows of 0.1-Hz widths. (continued) where Oij(f) is the observed S-wave Fourier amplitude spec-by taking its logarithm. Source spectra, Qs-value, and site trum of the ith event at the jth station; Si(f) is the source ampli.cation factor at each station are obtained in a least-amplitude spectrum of the ith event; Gj(f) is the site ampli-squares sense by the linear inversion method. As a constraint .cation factor at the jth station; Rijis the source distance condition, we assigned the KNO rock site as the reference between the ith event and the jth station; Qs(f) is the average site and set the site ampli.cation factor for KNO to be 2, Qs-value along the wave propagation path; and vsis the irrespective of frequency. We analyzed the S-wave portions average S-wave velocity along the wave propagation path of two horizontal components (north.south and east.west) (.3.5 km/sec). Equation (3) is modi.ed into a linear form using the time windows (cosine tapered) of 5 sec and 10 sec T. Uetake and K. Kudo Figure 7. after the onset of the Swaves. Velocity spectra obtained by vectorial summation of the two horizontal components are then used. We restrict our discussion in the over-0.5-Hz fre-quency range because of the low S/N for frequencies lower than 0.5 Hz. Results of the analysis are summarized as follows. The Qs-value of basement rock in the 0.5-to 1-Hz frequency range is about 20, whereas Qs(f).20f.ts the data in the frequency range from 1 to 15 Hz. This value is very small relative to the results from shallow earthquakes occurring in the Kanto or Tohoku district, Japan (e.g., Kato etal., 1992). However, in this article, our concern is to determine the site factor and not the Qsitself. Because the bedrock Qs-value does not affect relative site factors, we will not go into detail on this in this article. The site ampli.cation factors evaluated by the inversion method are shown in Figure 11 for the two time-window durations, together with two components average of HHRs of large events (Fig. 7) obtained by geometrical averaging of the horizontal components from large, distant events. Site factors by the two different methods are similar for both rock and sediment sites; however, peak levels of inversion results at about 1 Hz are smaller than HHRs at several sediment sites (TKD, KNS, CTS, and SHS). These peak levels of inversion results, however, tend to increase slightly with the length of the time window. Thus, in the time window of 10 sec, they approach the HHRs as shown by the broken lines in Figure 11. These results are similar to those of previous studies at different sites (e.g., Field and Jacob, 1995; Satoh etal., 2001). This may be because of the effects of later arrivals of surface waves in the case of a long time window. The spectral ratio method using large, distant events (HHRs) is advantageous in empirically assessing site effects, includ-ing the effects of secondary generated waves in a basin, ow-ing to the use of long time windows. In frequencies higher Continued. than 2 Hz, however, no signi.cant change of ampli.cation factors are found, irrespective of the length of time windows except for TKD. This implies that 1D analysis or estimation is applicable for high-frequency motions over 2 Hz. On the contrary, site factors at about 1 Hz motion in sediment sites have to be carefully assessed considering 2D/3D effects. Discussion As mentioned in the preceding section, 1D analysis may be able to explain the empirical-site factor in frequencies T. Uetake and K. Kudo Figure 8. Spatial variations of spectral ratios with respect to KNO or site ampli.-cation factors. Filled squares indicate the observation stations. GMT (Wessel and Smith, 1998) was used for spatial interpolation and for drawing these .gures. higher than 2 Hz. We conducted 1D analysis for four sedi-ment sites using the underground structure models based on PS-logging data shown in Table 5. The model for CTS is based on PSlogging (e.g., Kudo and Shima, 1988) to a depth of 467 m but is modi.ed by Saito etal.(1995) with the method for optimizing the spectral ratio of vertical array observation data. At the same time, the Qs-model for CTS was also obtained. The KNO located on outcrop of rock has a Vs.1200 m/sec (e.g., Sawada, 1992), which is almost equal to the bottom of the borehole at CTS. For other sites, the data within the bold lines in Table 5 are determined by PSlogging, whereas other parameters are assumed using the CTS model. The different estimates of HHR for the sites in Table 5 are compared in Figure 12. Those obtained from large-event data are shown by thick solid lines, site factors estimated by generalized inversion using a time window of 5 sec are shown by thin lines, and amplitude factors calcu-lated from the basement layer with Vs.1260 m/sec are shown by dotted lines. Because Swaves propagate almost vertically toward the low-velocity surface layers, we as-sumed vertically incident SHwaves in the calculation. Al-though 1D SHresponses cannot represent some peaks, the Figure 9. Fourier spectral ratios of horizontal to vertical components at each station. Only the stations with more than three events retrieved are displayed. The spectral ratio for each event and their average are displayed. spectral shapes of site factors roughly match with HHRsby the generalized inversion in frequencies higher than 2 Hz. We will not go into detail, but tuning of the model param-eters, especially for the Qs-values, may give better agree-ment. In general, empirical-site factors are affected by the reference-site responses. Kato etal.(1992) showed that the difference between the constraint condition and the reference-site response gave a systematic discrepancy be-tween the site factors and 1D analysis results using vertical array data. In fact, some discrepancies of HHR exist between 1D SHresponses and inversion results, as shown in Figure 12. The peaks at about 1 Hz by the inversion results are systematically larger than those of the 1D analysis, whereas discrepancies in the high-frequency range are not so large and systematic. The discrepancy at about 1 Hz may be pro-duced by the site effects of KNO and some factors that can-not be modeled by 1D analysis. HHRs estimated by large and remote events tend to overestimate those from 1D re-sponses and inversion. This will be attributed to the effects T. Uetake and K. Kudo HHR, respectively. of secondary generated waves in the basin. We have to con-sider these factors; however, we may say that it is adequate to select KNO as a reference site for evaluating site re-sponses. This is because the response of KNO shows nearly average response among the rock sites, as shown in Figure 6, and the deviation among the rock site responses is smaller than the deviation between rock and sediment sites, as shown in Figure 7. In detail the rock sites surrounding Ashigara Valley show the different site responses; therefore, it is im-portant for precise assessments or detailed discussions to select a proper reference site. Conclusions The site effects in and around Ashigara Valley were evaluated using records from large, distant events. Site am-pli.cation factors for both rock and sediment sites were eval-uated by taking spectral ratios with respect to one rock site (KNO). A long time window of 81.92 sec was used for ob-taining the spectra of earthquake motions, so that the spectral ratios relative to KNO give empirical ampli.cation factors that include the effects of secondary generated waves in the basin. These ampli.cation factors do not discriminate be-tween wave types or ampli.cation mechanism, although between 0.03 and 0.04 Hz, and such low-frequency waves they should be regarded as essential in earthquake hazard are not affected by the heterogeneity of surface and under-assessment, at least in cases where earthquake motions of ground structures near Ashigara Valley. On the other hand, almost vertical incidence can be modeled. In the analysis, the higher-frequency motions from the shallow events dom-two shallow events were included; therefore, the incidence inated in the early arrivals of Swaves; thus, the assumption of surface waves to the valley should also be taken into of quasivertical incidence of waves is allowable. We mapped account. In fact, surface waves were clearly found in the data the obtained site ampli.cation factors as a function of fre-(L3), but they are predominantly in the low-frequency range quency. The spectral ratios tend to be large toward the south- T. Uetake and K. Kudo Table 5Underground Velocity Structure Models for SeveralSediment Sites Thickness Vp Vs Density Qs.Q0 *f**n (m) (m/sec) (m/sec) (g/cm3) Q0 n CTS 10 800 110 1.2 2.0 0.8 15 1650 220 1.7 3.0 0.7 25 1650 250 1.7 4.0 0.6 15 1650 330 1.9 6.0 0.6 25 1960 560 2.0 7.0 0.6 150 2250 790 2.1 10.0 0.8 180 2350 970 2.2 10.0 0.8 . 2520 1260 2.3 10.0 0.8 NRD 8 1300 120 1.2 2.0 0.8 10 1300 180 1.2 2.0 0.8 4 1660 300 1.9 6.0 0.6 20 1530 210 1.7 4.0 0.6 15 1850 630 2.0 10.0 0.6 150 2250 790 2.1 10.0 0.8 180 2350 970 2.2 10.0 0.8 . 2520 1260 2.3 10.0 0.8 TKD 7 600 65 1.5 2.0 0.8 2 350 110 1.2 2.0 0.8 12 900 110 1.2 2.0 0.8 6 1550 240 1.7 3.0 0.7 6 1550 330 1.9 6.0 0.6 5 1550 250 1.7 6.0 0.6 45 1640 350 1.9 6.0 0.6 25 1960 560 2.0 7.0 0.6 150 2250 790 2.1 10.0 0.8 180 2350 970 2.2 10.0 0.8 . 2520 1260 2.3 10.0 0.8 KNS 5 1500 170 1.5 3.0 0.7 16 2100 690 2.0 4.0 0.6 48 1650 400 1.9 6.0 0.6 6 2300 750 2.0 6.0 0.6 12 1800 400 1.9 7.0 0.6 80 2300 700 2.2 10.0 0.8 180 2350 970 2.2 10.0 0.8 . 2520 1260 2.3 10.0 0.8 The model for CTS was determined by Saito etal.(1995). The other three site models are basically the same as the CTS model, except that original PS-logging data were used for the shallow part (shaded areas). east area of the valley for the low-frequency range below 0.5 Hz. The area showing large ampli.cation factors at about 1 Hz and at higher frequencies is localized near the southeast of the valley. HVRs, using the same time windows as the reference-site method, were compared with the spectral ratios of hor-izontal motions (HHRs). HVRs at rock sites are not always equal to unity and vary between 1 and 3 against frequency. Moreover, the deviation from event to event is large, and the averages of HVRs, in general, are larger than HHRs. HVRs in the frequency range higher than 0.5 Hz coincide with HHRs at many sediment sites, but no systematic relationship could be seen at lower frequencies. By restricting the scope to the frequency range higher than 0.5 Hz, HVRs may be used as substitutes for HHRs in determining the empirical-site ampli.cation factors of sediment sites. However, this method should be used with caution, because the applica-bility of HVRs may be valid only for soft sediments and the reliable frequency range depends on the site conditions. Site effects were also evaluated by the generalized in-version method using data from local small events, restricted in the frequency range higher than 0.5 Hz. Site factors es-timated by the inversion method coincide with HHRs from large events in the frequency range higher than 2 Hz. How-ever, the spectral peaks at about 1 Hz of sediment sites by the inversion method are lower than those of HHRs. The horizontal spectral ratio to a reference site (HHR), horizontal-to-vertical (HVR), and generalized inversion methods give similar results if restricted in the higher fre-quency of over 2 Hz, with the ampli.cation factors explained by 1D responses. For the low-frequency range, basin-induced or transduced surface waves may affect site re-sponses as basin effects. Acknowledgments We sincerely thank Mr. Takahashi and Mr. Sakaue of ERI for their efforts to maintain the Ashigara Valley observation system and to retrieve the strong-motion data. We are grateful to NIED, CRIEPI, and TEPCO for providing the observation data. We also thank Dr. T. Satoh, the anonymous reviewer, and associate editor Dr. H. Kawase for their valuable comments in improving this article. Many .gures were prepared with Generic Map-ping Tools developed by Wessel and Smith (1998). References Andrews, D. J. (1982). Separation of source and propagation spectra of seven Mammoth Lake aftershocks, in ProceedingsofWorkshop16,DynamicCharacteristicsofFaulting,1981, U.S.Geol.Surv.Open-FileRept.82-591, 437. Boatwright, J., J. B. Fletcher, and T. E. Fumal (1991). A general inversion scheme for source, site, and propagation characteristics using multiply recorded sets of moderate-sized earthquakes, Bull.Seism.Soc.Am.81,1754.1782. Bonilla, L. F., J. H. Steidl, G. T. Lindley, A. G. Tumarkin, and R. J. Ar-chuleta (1997). Site ampli.cation in the San Fernando Valley, Cali-fornia: variability of site-effect estimation using the S-wave, coda, and H/V methods, Bull.Seism.Soc.Am.87,710.730. Borcherdt, R. D. (1970). Effects of local geology on ground notion near San Francisco Bay, Bull.Seism.Soc.Am.60,29.81. Field, E. H., and K. H. Jacob (1995). A comparison test of various site-response estimation techniques, including three that are not reference-site dependent, Bull.Seism.Soc.Am.85,1127.1143. Imamura, A. (1929). On the earth-vibrations induced in some localities at the arrival of seismic waves, Bull.EarthquakeRes.Inst.TokyoUniv.7,489.494. Ishimoto, M. (1932). Comparaison accelerometrique des secousses sis-miques dans deux parties de la ville de Tokyo, Bull.EarthquakeRes.Inst.TokyoUniv.10,171.187. Iwata, T., and K. Irikura (1988). Source parameters of the 1983 Japan Sea earthquake sequence, J.Phys.Earth36,155.184. Figure 12. Comparison between spectral ratios from the horizontal components with reference to KNO, the generalized inversion results, and the calculated transfer functions from the layer with Vs.1260 m/sec. Bold lines show the geometrical average of two horizontal spectral ratios. Thin lines show the site factors estimated by generalized inversion. Dotted lines show the response from vertical incident SHwaves for the underground structure models in Table 5. Kato, K., M. Takemura, T. Ikeura, K. Urao, and T. Uetake (1992). Prelim-inary analysis for evaluation of local site effects from strong motion spectra by an inversion method, J.Phys.Earth40,175.191 Kawase, H., and T. Sato (1992). Simulation analysis of strong motions in the Ashigara valley considering one-and two-dimensional geological structures, J.Phys.Earth40,27.56. Kinoshita, S. (1998). Kyoshin Net (K-NET), Seism.Res.Lett.69,309.332. Kudo, K. (1992). Earthquake motions: given and blinded data, Proc.Intern.Symp.ontheEffectsofSurfaceGeologyonSeismicMotion,Odawara,Japan2,53.60. Kudo, K., and E. Shima (1988). Installation of strong motion seismographs, plan for geotechnical measurements, at Ashigara valley site effects test area, Japan, in ProceedingsoftheIASPEI/IAEEJointWorkingGrouponEffectsofSurfaceGeologyonSeismicMotion,SecondWorkshop,Tokyo,Japan, Japanese Working Group on Effects of Sur-face Geology on Seismic Motion, II-1.II-11. Kudo, K., E. Shima, and M. Sakaue (1988). Digital strong motion acceler-ograph array in Ashigara valley, Proc.9thWorldConf.EarthquakeEng.,Tokyo,Japan8,119.124. Kudo, K., and Y. Sawada (1998). A brief review on the Ashigara Valley blind prediction test and some follow-up studies, in TheEffectsofSurfaceGeologyonSeismicMotion, K. Irikura, K. Kudo, H. Okada, and T. Sasatani (Editors), Balkema, Rotterdam, The Netherlands, 305.312. Lermo, J., and F. J. Chavez-Garcia (1993). Site effects evaluation using spec-tral ratios with only one station, Bull.Seism.Soc.Am.83,1574.1594. Nakamura, Y. (1988). Inference of seismic responses of surface layer based on Microtremor measurement, QuarterlyReportonRailwayResearch4,18.27 (in Japanese). Riepl, J., P.-Y. Bard, D. Hatzheld, C. Papaionnou, and S. Necht-schein (1998). Detail evaluation of site response estimation methods across and along the sedimentary valley of Volvi (EURO-SEISTEST), Bull.Seism.Soc.Am.88,488.502. Saito, S., T. Sasatani, and K. Kudo (1995). Attenuation characteristics of S waves in the sedimentary layers at the Ashigara Valley, Geophys.Bull.HokkaidoUniv.,Sapporo,Japan, 58,39.61 (in Japanese). Sasatani, T., M. Ikeda, and N. Sakajiri (1992). A study of site effects by means of strong motion seismograms from near by, intermediate depth earthquake, J.Phys.Earth40,85.98. Sato, K., S. Higashi, H. Yajima, and S. Sasaki (1998). Ashigara valley test site, 1D or 2D-3D? in TheEffectsofSurfaceGeologyonSeismicMotion, K. Irikura, K. Kudo, H. Okada, and T. Sasatani (Editors), Balkema, Rotterdam, The Netherlands, 319.340. Satoh, T., H. Kawase, and S. Matsushima (2001). Differences between site characteristics obtained from microtremors, S-waves, P-waves, and Codas, Bull.Seism.Soc.Am.91,313.334. Sawada, Y. (1992). Geotechnical data, Proc.Intern.Symp.ontheEffectsofSurfaceGeologyonSeismicMotion,Odawara,Japan2,29.42. Seekins, L. C., L. Wennerberg, L. Margheriti, and H.-P. Liu (1996). Site ampli.cation at .ve locations in San Francisco, California: a com-parison of S waves, codas, and microtremors, Bull.Seism.Soc.Am.86,627.635. Smith, W. H. F., and P. Wessel (1990). Gridding with continuous curvature splines in tension, Geophysics55,293.305. Steidl, J. H., A. G. Tumarkin, and R. J. Archulata (1996). What is a ref-erence site?, Bull.Seism.Soc.Am.86,1733.1748. Uetake, T., and K. Kudo (1998). Evaluation of site effects in a wide fre-quency band at Ashigara Valley, Japan, using strong motion records from remote and large events, Zisin,2ndSer.50,397.414 (in Japa-nese). Wessel, P., and W. H. F. Smith (1998). New, improved version of the Genetic Mapping Tools released, EOSTrans.AGU79,579. Yamazaki, H. (1992). Tectonics of a plate collision along the northern margin of Izu peninsula, central Japan, Bull.Geol.Surv.Jpn.43,603. 657. Yamazaki, F., and M. A. Ansary (1997). Horizontal-to-vertical spectrum ratio of earthquake ground motion for site characterization, Earth-quakeEng.Struct.Dyn.26,671.689. Seismic Design Group, R&D Center The Tokyo Electric Power Company, Inc. 4-1, Egasaki-cho, Tsurumi-ku Yokohama 230-8510, Japan (T.U.) Earthquake Research Institute The University of Tokyo 1-1-1, Yayoi, Bunkyo-ku Tokyo 113-0032, Japan (K.K.) Manuscript received 8 April 2004.