Seismic Early Warning Systems: Procedure for Automated Decision Making

Abstract

An Early Warning System potentially allows mitigation measures to be carried out from the moment in which a seismic event is detected. Examples of such measures are evacuation of buildings, shut-down of critical systems (nuclear reactors, industrial chemical processes, etc.) and stopping of high-speed trains. The type of mitigation measures that can be effectively activated depends on the amount of warning time available, but timeliness is often in conflict with the reliability of the predictions, which become more accurate as more seismic sensor data is collected. There is therefore an inevitable trade-off between the amount of warning time available and the reliability of the predictions provided by the Early Warning System. To investigate this trade-off, the consequences of the two alternatives of taking mitigation actions or not acting must be analyzed, accounting for significant uncertainty in the predictions. In this report, we present a decision-making procedure based on the real-time evaluation of the consequences of taking no action and of activating mitigation measures which is based on the probabilities of false and missed alerts. The threshold at which mitigating actions should be taken is quantified based on a cost-benefit analysis. The method is applied to two recent seismic events in Southern California, an M 4.75 event in Yorba Linda and an M 6.5 event in San Simeon. Also, a feasibility assessment of any proposed regional Early Warning System is of critical importance, and it should involve an examination of whether the requirements, in terms of warning time available and the probability of making wrong decisions, are met. A useful tool in this assessment of an Early Warning System is a seismic hazard map to provide the probability of exceedance of ground shaking intensity, given a site and time interval of interest, and a corresponding map of the probability of making a wrong decision. In this report, a methodology is presented for estimating the probabilities of making wrong decisions that can be incorporated in a feasibility assessment of proposed Early Warning System.