We constructed a prototype of the basin and crustal structure model for the Kinki area, southwest of Japan, for the simulation of strong ground motions of hypothetical crustal and subduction earthquakes. We collected results of the deep seismic velocity profiles obtained by the reflection experiments and seismic imaging results, which were conducted in the Kinki area. The obtained profiles give underground velocity structures of the crust, from the surface to the subducting slab. We also gather the basin velocity structure information of the Osaka, Kyoto, Nara, and Ohmi basins. To examine the applicability of the constructed velocity structure model to the ground motion simulation, we simulated waveforms of an intermediate size event occurred near the source area of the hypothetical subduction earthquakes. Simulated ground motions using the basin and crustal velocity structure model are fairly well reproducing the observations at most of stations, and the constructed basin and crustal velocity structure model is applicable for the long-period ground motion simulations.

The 8-10 May 1997 eruption of Bezymianny volcano began with extrusion of a crystallized plug from the vent in the upper part of the dome. Progressive gravitational collapses of the plug caused decompression of highly crystalline magma in the upper conduit, leading at 13:12 local time on 9 May to a powerful, vertical Vulcanian explosion. The dense pyroclastic mixture collapsed in boil-over style to generate a pyroclastic surge which was focused toward the southeast by the steepwalled, 1956 horseshoe-shaped crater. This surge, with a temperature <200°C, covered an elliptical area >30 km 2 with deposits as much as 30 cm thick and extending 7 km from the vent. The surge deposits comprised massive to vaguely laminated, gravelly sand (Md -1.2 to 3.7φ; sorting 1.2 to 3φ) of poorly vesiculated andesite (mean density 1.82 g cm -3 ; vesicularity 30 vol%; SiO 2 content ~58.0 wt%). The deposits, with a volume of 5-15×10 6 m 3 , became finer grained and better sorted with distance; the maximal diameter of juvenile clasts decreased from 46 to 4 cm. The transport and deposition of the surge over a snowy landscape generated extensive lahars which traveled >30 km. Immediately following the surge, semi-vesiculated block-and-ash flows were emplaced as far as 4.7 km from the vent. Over time the juvenile lava in clasts of these flows became progressively less crystallized, apparently more silicic (59.0 to 59.9 wt% SiO 2 ) and more vesiculated (density 1.64 to 1.12 g cm -3 ; vesicularity 37 to 57 vol%). At this stage the eruption showed transitional behavior, with mass divided between collapsing fountain and buoyant column. The youngest pumice-and-ash flows were accompanied by a sustained sub-Plinian eruption column ~14 km high, from which platy fallout clasts were deposited (~59.7% SiO 2 ; density 1.09 g cm -3 ; vesicularity 58 vol%). The explosive activity lasted about 37 min and produced a total of ~0.026 km 3 dense rock equivalent of magma, with an average discharge of ~1.2×10 4 m 3 s -1 . A lava flow 200 m long terminated the eruption. The evolutionary succession of different eruptive styles during the explosive eruption was caused by vertical gradients in crystallization and volatile content of the conduit magma, which produced significant changes in the properties of the erupting mixture.

SUMMAR YSince the installation of three limited-aperture strong-motion networks in the Himalayan region in 1986, six earthquakes with M w =5.2±7.2 have been recorded up to 1991. The data set of horizontal peak accelerations and velocities consists of 182-component data for the hypocentral distance range 10±400 km. This data set is limited in volume and coverage and, worst of all, it is highly inhomogeneous. Thus, we could not determine regional trends for amplitudes by means of the traditional approach of empirical multiple regression. Instead, we perform the reduction of the observations to a ®xed distance and magnitude using independently de®ned distance and magnitude trends. To determine an appropriate magnitude-dependent distance attenuation law, we use the spectral energy propagation/random function approach of Gusev (1983) and adjust its parameters based on the residual variance. In doing so we con®rm the known, rather gradual mode of decay of amplitudes with distance in the Himalayas; this seems to be caused by the combination of high Qs and crustal waveguide effects for high frequencies.The data are then reduced with respect to magnitude. The trend of peak acceleration versus magnitude cannot be determined from observations, and we assume that it coincides with that of abundant Japanese data. For the resulting set of reduced log 10 (peak acceleration) data, the residual variance is 0.37 2 , much above commonly found values. However, dividing the data into two geographical groups, western with two events and eastern with four events, reduces the residual variance to a more usual level of 0.27 2 (a station/site component of 0.22 2 and an event component of 0.16 2 ). This kind of data description is considered acceptable. A similar analysis is performed with velocity data, and again we have to split the data into two subregional groups. With our theoretically grounded attenuation laws we attempt a tentative extrapolation of our results to small distances and large magnitudes. Our minimum estimates of peak acceleration for the epicentral zone of M w =7.5±8.5 events is A peak =0.25±0.4 g for the western Himalayas, and as large as A peak =1±1.6 g for the eastern Himalayas. Similarly, the expected minimum epicentral values of V peak for M w =8 are 35 cm s x1 for the western and 112 cm s x1 for the eastern Himalayas. To understand whether our results re¯ect the properties of the subregions and not of a small data set, we check them against macroseismic intensity data for the same subregion. The presence of unusually high levels of epicentral amplitudes for the eastern subregion agrees well with the macroseismic evidence such as the epicentral intensity levels of X±XII for the Great Assam

-Average envelope shapes (mean square amplitude time histories) of small earthquakes represent a convenient basis for the construction of semi-empirical stochastic ''Green's functions,'' needed for prediction of future strong ground motion. At the same time, they provide crucial evidence for verification of the theories of scattering of high-frequency seismic waves in the lithosphere. To determine such shapes in the Kamchatka region we use the records of near (R = 50-200 km) shallow earthquakes located around the broadband station PET. On these records, we select the S-wave group and determine its root-mean-square duration T rms , separately for each of the five octave frequency bands. We determine the empirical T rms vs. distance dependence and find it to be very close to a linear one. At the reference distance R = 100 km, average T rms decreases from 5.4 sec for the 0.75 Hz band to 3.9 sec for the 12 Hz band. To analyze average envelopes, we assume that the functional form of the envelope shape function is independent of distance, and stretch each of the observed envelopes along the time axis so as to reduce it to a fixed distance. Through averaging of these envelopes we obtain characteristic envelope shape functions. We qualitatively analyze these shapes and find that around the peak they are close to the shapes expected for a medium with power-law inhomogeneity spectrum, with the spectral exponent 3.5-4. From onset-topeak delay times we derive the values of transport mean free path and of scattering Q for a set of distances.

The Q-value is one of the parameters controlling attenuation of seismic waves with distance. Attenuation relations in crust used in the earthquake engineering applications usually consider models with uniform Q and geometrical spreading. In this work we try to estimate a nonuniform Q-value based on the ray geometrical spreading in a nonuniform velocity model. We estimate Q-values in the seismogenic and aseismic zones of the Kinki region (Japan) using Hi-net data. The Hi-net network consists of high-sensitivity seismometers in 100-200 m boreholes. We assumed a two-layer model of Q(f ) (seismogenic and aseismic zones), with uniform Q in each layer, and we applied a method for the separation of source, path, and site effects. Path lengths in the layers were calculated using raytracing. A geometrical spreading term was calculated for a realistic 1D velocity model (consisting of three layers over the Moho). Inversion was performed in two steps. (1) The Q value in the seismogenic layer was estimated using shallow earthquake data (depth Ͻ 20 km), assuming a one-layer Q model. (2) Data from subduction zone earthquakes covering the aseismic zone (with depths 20-70 km) and two-layer Q model (0-20 and 20-70 km) were used to calculate Q in the aseismic zone, where the Q-value for the upper layer was constrained by results of step 1. The total number of records used was 628. Only direct S-wave data were used to calculate Fourier amplitude spectra in the highfrequency range 1-10 Hz. Validation of the method and inversion results were made by inversion of synthesized data. We discuss in detail several possible sources of errors of the estimation of Q-values. The results of inversion showed a higher Q in the upper layer, Q(f ) ‫ס‬ 180f 0.7 for the seismogenic layer, than that in the lower, Q(f ) ‫ס‬ 90f 0.8 for the aseismic zone. This result supports the model of the crust containing a brittle seismogenic layer and a ductile aseismic zone. We proposed amplitude versus distance attenuation model for Kinki region, Japan, based on estimated Q-values and geometrical spreading.