地震災害を考える 予測と対策 大地震発生の長期的予測

Abstract

After the devastating Kobe earthquake of 1995, the Headquarters of Earthquake Research Promotion was established in the prime minister's office. Among the mandates given to the headquarters are collection, analysis, and evaluation of the results of surveys and observations related to earthquakes. The headquarters set up a plan to survey 98 major active faults in Japan for studies of paleoearthquakes on those faults. Mainly on the basis of the survey results and investigations on historical earthquakes, the Earthquake Research Committee of the headquarters evaluated earthquake potential and made public long-term earthquake forecasts in many source regions. These will be a basis for a new probabilistic estimate of seismic hazards throughout Japan, which will be completed by March 2005.The size of a future earthquake is estimated from the empirical relationship between earthquake magnitude and fault length/source area. The location of the event is fairly precisely known from active fault data and historical earthquake catalogs, except for a certain types of earthquake such as deep events. For the occurrence time, we can only give a probabilistic estimate.Several renewal models have been tested against available recurrence data in Japan by the Sub-committee for Long Term Evaluation. They are log normal, gamma, Weibull, double exponential, and BPT (Brownian passage time) distributions. Because none of the distributions can fit the data significantly better than the other models, the BPT model has been chosen because of the clear physical meaning of its model process and of the stability of parameters. The BPT model consists of a regular loading process and irregular Brownian motion disturbances. We found that the common relative standard deviation explains the data set better than different relative standard deviations for each sequence, so far as the four examples of Japanese shallow crustal earthquake sequence are concerned. By extrapolating the results, we use the common relative standard deviation to evaluate shallow crustal events. When sufficient data on successive events of co-seismic slip are available, the time-predictable models is used for evaluations of subduction-zone earthquakes.