SPECIAL SECTION-ASSESSMENT OF SCHEMES FOR EARTHQUAKE PREDICTION 2 SEP YB ’ Assessment of schemes for earthauake mediction :

S U M M A R Y This contribution introduces a series of papers on the assessment of schemes for earthquake prediction, originally presented at an RAS/JAG Discussion Meeting. The papers also address the social and administrative implications of earthquake prediction and the question whether this objective is feasible in either a strictly scientific or a socially useful sense. Contributions on the assessment process itself stress the importance of clearly defining procedures and anticipated outcomes in advance of any prediction experiment. After-the-fact searches for correlations drastically weaken the statistical significance of any association which is found. The record of a proposed prediction scheme can be made to appear more significant by comparing it with an inadequate or unsuitable model of seismicity. Because the distribution of earthquakes is clustered in both time and space, simple models (e.g. Poisson distribution in time) do not fully characterize it, and statistical tests which rely on such assumptions are inappropriate. The history of attempts to predict earthquakes suggests that little if any progress is being made towards the goal of specifying-in advance and within reasonably tight limits-the time and place of occurrence and the size of the forthcoming event. The view that prediction may be impossible is regaining support following a change in perception regarding the nature and origin of earthquakes and faulting. Reid’s elastic rebound model and its associated conceptual framework of a nearhomogeneous Earth are being superseded by a fundamentally heterogenous model employing the stochastically oriented concept of self-organized criticality. Within the latter model, both the initiation and the eventual progress of rupture on a fault depend on a multitude of factors related to both local stress and strength, which are inter-related and heterogeneous beyond practical measurement. Thus, the course of a fracture towards large-scale failure cannot practically be determined in advance. Other contributors show that an unusually high degree of scientific confidence is required in order for a prediction to be of significant social value. Therefore, in terms of saving both lives and property, statistical approaches to hazard assessment, combined with appropriate programmes of risk reduction, continue to offer the most cost-effective means of reducing earthquake-related losses.