Prediction of a subsequent large earthquake

Many large earthquakes come in pairs separated by relatively small times and distances. The subsequent earthquake may be a large aftershock of the first one, or another main shock. The prediction of a subsequent large earthquake is important for reducing the great hazard caused by destabilization of buildings, lifelines, and other constructions, mountain slopes, etc., after the first large earthquake. An algorithm for prediction of a subsequent large earthquake is based on analysis of the aftershock sequence following a first shock and of local seismic activity preceding it. The algorithm was found by analysis of 21 large earthquakes in the California-Nevada region; six of them were followed by subsequent large earthquakes. In application to these earthquakes, the algorithm made one error (a failure-to-predict). Then this algorithm with all parameters fixed was tested retrospectively on independent data in different regions of the world. It was applied to 98 large earthquakes, not considered in its development; 10 of them were followed by a subsequent large shock. There were two failures-to-predict and four false alarms. Since 1989, 20 advance predictions have been made, including the 1991 Rachi earthquake in Georgia (Caucasus), and three Californian earthquakes: Loma-Prieta, 1989; Joshua Tree, 1992; and Northridge, 1994. Among three mistakes, two were false alarms and one was a failure-to-predict. Both the advance and retrospective predictions were made by the fixed algorithm, previously developed and unambigously defined in publications. The statistical significance of advance predictions exceeds 98%. (C) 1999 Elsevier Science B.V. All rights reserved.