2006 年 71 巻 607 号 p. 183-191
Improving the prediction of strong ground motion intensity is one of the interesting research subjects in the earthquake engineering. Site-specific attenuation relation based on the Bayesian approach is developed in this paper. The site factors of observatories in K-NET and KiK-NET are inferred as well as the uncertainty associated with the site with the observations. The parameters are inferred as random variables so that the all uncertainties can be expressed in terms of PDF, which is conceptually different from the classical regression analysis adopted to develop the empirical attenuation relations. The uncertainty is different with different sites in Bayesian framework and the result shows the uncertainty can be reduced, while it is identical with disregard with site in the empirical attenuation relation. The sequential use of Bayesian theorem makes it feasible to update the inference with new observations again.