ABSTRACT

Dislocation models of the 1995 Hyogo-ken Nanbu (Kobe) earthquake derived by Yoshida et al. (1996) show substantial changes in direction of slip with time at specific points on the Nojima and Rokko fault systems, as do striations we observed on exposures of the Nojima fault surface on Awaji Island. Spudich (1992) showed that the initial stress, i.e. the shear traction on the fault before the earthquake origin time, can be derived at points on the fault where the slip rake rotates with time if slip velocity and stress change are known at these points. From Yoshida's slip model we calculated dynamic stress changes on the ruptured fault surfaces. To estimate errors we compared the slip velocities and dynamic stress changes of several published models of the earthquake. The differences between these models had an exponential distribution, not gaussian. We developed a Bayesian method to estimate the probability density function of initial stress from the striations and from Yoshida's slip model. Striations near Toshima and Hirabayashi give initial stresses of about 13 and 7 Mpa, respectively. We obtained initial stresses of about 7 to 17 Mpa at depths of 2 to 10 km on a subset of points on the Nojima and Rokko fault systems. Our initial stresses and coseismic stress changes agree well with post-earthquake stresses measured by hydrofracturing in deep boreholes near Hirabayashi and Ogura on Awaji Island. Our results indicate that the Nojima fault slipped at very low shear stress, and fractional stress drop was complete near the surface and about 32% below depths of 2 km. Our results at depth depend on the accuracy of the rake rotations in Yoshida's model, which are probably correct on the Nojima fault but debatable on the Rokko fault. Our results imply that curved or crosscutting fault striations can be formed in a single earthquake, contradicting a common assumption of structural geology.