- 公益社団法人 日本地震学会
- 地震 第2輯 (ISSN:00371114)
- vol.57, no.4, pp.409-418, 2005-03-20 (Released:2010-03-11)
When a larger earthquake than the preceding ones occurs successively in a clustered activity, it seems a further larger earthquake is apt to be observed afterwards (Yoshikawa et al., 2000). We investigate alarm rate, success rate, and probability gain for the occurrence of an earthquake with M5 or larger in such sequential activities. We use the JMA catalogue during the period from 1995 to 2000. First, we select out clusters of shallow (depth≤30km) earthquakes withM≥2.5 in and around Japanese islands. We name the first earthquake in a cluster E0, and the one larger than E0 that occurs first in the succeeding activity E1, and term such clusters that contain E1 earthquake E1-class clusters. E2 earthquake and E2-class clusters are defined in the similar way. We calculate the alarm rate, the success rate and the probability gain, when M≥5 earthquakes are forecasted using E1-class clusters and E2-class clusters, respectively. We also examine how the alarm and success rates change when criterion of selecting clusters and upper limit of distance and time interval are varied. It is shown that, for E1-class clusters, the alarm rate is 24 to 30% and the success rate is about 5%. For E2-class clusters, the alarm rate is 10 to 15% and the success rate is 10 to 14%, and we get a value as large as 5702.2 for the probability gain in the best case that both the criterion and the upper limit of selecting clusters are set to be 5km-5days. The large values of the alarm and success rates and the probability gain indicate that the algorithm for forecasting M5 earthquakes proposed in this paper is practically effective as well as statistically significant.