Rupture Process Analysis of the 1995 Hyogo-ken Nanbu Earthquake
Haruko Sekiguchi
This study aims to clarify the physics of the source process by detailed source process analysis and to understand the mechanism of ground motion generation during large earthquakes.
In Part #I, I investigated the source process of the 1995 Hyogo-ken Nanbu earthquake by the waveform inversion especially focusing on the fault geometry in the rupture termination. The branching possibility of Okamoto fault was suggested both by static displacement distribution and damage extension towards the east of Kobe (Nishinomiya area). In order to exclude data contaminated by the basin-edge diffracted wave into the waveform inversion process, I examined the spatio-temporal variation of its influence, and from comparison to the flat model, determined the appropriate windows for the data. Therefore, the present inversion procedure includes the largest amount of available instrumented information in the near-source region so far. The obtained source process showed that the rupture progressed upward in Awaji side; southeast of the hypocenter and went at depths larger than about 7km in Kobe side ; northeast of the hypocenter. Three subevents were identified: the largest one in the shallow part of the Awaji side, whereas the smaller two occurred at large depths (>7km) in the Kobe side. I found relatively large slips at the deep part of the branch. Total variance reduction is larger when the branch is assumed than when not. Resolution checks showed that my data set can discriminate the slips on the two branching segments, and that the slips on proposed branched portion is physical; it is not the effect of random data noise neither due to systematic errors of the Green's functions arising from misestimation of velocity structures. I conclude that branching possibility of the Okamoto fault during the Hyogo-ken Nanbu earthquake is very likely. Near-source ground motion is simulated to see the effect of slip on the branched fault using 3D FDM. Characteristic distribution of ground motion in the near-source region indicated by the damage distribution is well reproduced by modeling of both source process and wave propagation in the realistic 3-D velocity structure. The slip on the branched fault affected the ground motion in eastern part of Kobe (Nada and Higashi-Nada Ward), Ashiya and Nishinomiya City. Although its contribution is not dominant even in those regions; about 30 to 50 % in maximum velocity in frequency 0.1 to 1.0 Hz.
In Part #II, I introduced a convolution method (Ben-Menahem, 1961) which incorporate the effect of a bi-directional moving dislocation over a rectangular subfault with a point source synthetic waveform in calculation of element source waves for the waveform inversion. Each of such element source waves includes the effect of rupture directivity itself. Therefore, the rupture directivity effect on the entire fault area is fully considered in the waveform inversion. The slip distribution inverted by these element source waves is expected to be different from that inverted by point source element waves and maybe closer to the true source process, because it seems more reasonable to expect continuous rupture distribution. For the same purpose, Wald and Heaton (1994) calculated many point sources over each subfault and summed them up considering rupture time delay (SM). Advantage of CM compared to SM is smaller computation. Only one or several Green's functions' calculation is needed for one subfault for most of the subfault-staion pairs, and a function of a finite, moving dislocation is convolved to each Green's function together with a sip time function. Appropriateness of CM is tested numerically. I examined the variation of the approximation error due to size of subfault, hypocenter distance, rupture propagation velocity and its direction by numerical tests, and showed that a ratio of right and left side of the Fraunhofer inequality condition is useful to estimate the approximation error of CM. Obtained moment release history for each subfault using this technique directly reflects the slip time function on the subfault. The inverted rupture process has a smooth slip distribution even inside the subfaults. Such a model, used as a source model in near-source ground motion simulation, escapes the problem of artificial, and large ground motion near the point sources due to the use of sparsely distributed point sources. I applied CM in the waveform inversion for the source process of the 1995 Hyogo-ken Nanbu earthquake. The global characteristics of moment release distributions are similar between two inversion results; one inverted with point source element waveforms (in Part #I) and the other with element source waves considering the finite moving dislocation effects inside each subfault. But details like places of peak moment releases differ between these two inversion results. Rise time shows a large spatial variation without remarkable depth dependence. Using the inverted source process for the Hyogo-ken Nanbu earthquake as an input, the near-source ground motions were simulated using 3-dimensional finite difference method to examine the effect of continuous rupture on the near-source ground motion simulation. Smoother distribution of the ground motion was obtained than that obtained by the source model with the sparsely distributed point sources (Part #II). Characteristics of ground motion distribution in the near-source region expected by the damage distribution are better reproduced with continuously distributed source model. Comparing the synthetic spectra at rock sites in Awaji Island and in Kobe, I found that the input ground motions at frequencies which was most effective to the damage generation (around 1Hz) were comparable in spite of the large and shallow slip in the Awaji side.