Earthquake Warning Systems: Characterizing Prediction Uncertainty

Veronica Grasso
University of Naples

Richard M. Allen
University of California Berkeley

AGU Fall Meeting 2005

Uncertainty associated with the real-time prediction of seismic parameters such as magnitude and location of an event by an earthquake warning system (EWS) reprents a critical aspect of its application. Here we consider the causes and effects of prediction uncertainty and develop a strategy for effective use of the information provided by EWS. Our uncertainty analysis has been applied to the ElarmS methodology for seismic protection of Southern California (Allen and Kanamori, 2003) in order to define and quantify the uncertainty associated with the magnitude and location prediction, as time variant parameters, and the total error associated with the parameter on which decision-making is based. We find that the total error associated with peak ground acceleration (PGA) prediction, evaluated by a Monte-Carlo simulation, follows a Gaussian distribution and decreases with time as more data become available and more stations are triggered. We have carried out a sensitivity analysis in order to determine the principal cause of the uncertainty. The analysis provides the errors associated with each component of the system, including the magnitude and location estimates, and the attenuation relation used to determine PGA. The results show that the overall error is mainly due to the attenuation model. EWS performance could be improved with the use of single station error characterization allowing for the total error to be expressed as a product of the single station errors. This would allow users to estimate the advantage of waiting for additional information during the course of an event.