Abstract
An attempt was made to predict seismic activity in the Izu Islands region, Japan, between 15 July and 23 October 2000. In spite of a simple assumption that the occurrence of a pair of earthquakes (i. e. signal pair) with similar magnitudes may signal an impending major earthquake, 13 out of 25 imminent predictions were successful. Among these, 4-hour alarms using one of the empirical rules (i. e. hypothesis A: when earthquakes with similar magnitudes were observed at intervals of less than 2 hours, a larger event would occur within about 4 hours) were presented 11 times, resulting in 5 successful predictions. Such a success rate is expected only to have a probability of 1%, if the seismic activity did not change before and after the time of prediction. Other empirical rules were also useful, at least to some extent, although the significance was not necessarily clear because of a lack of sufficient experience. In addition, based on the present method, 4 out of 5 major earthquakes with a magnitude of 6 could be predicted, including those during the test period from 27 June to 14 July. Considering the total length of the alarm period, there was only a probability of about 1% if the occurrence of events with a magnitude of 6 was random and independent of the present rules.
