|
Richard M Allen and Hiroo Kanamori California Institute of Technology _______________________________________ The rapid increase in the number of seismic stations in earthquake prone regions, combined with the implementation of near real time data transmission technologies, provides the potential for earthquake early warning. In the absence of earthquake prediction methodologies in the foreseeable future, the rapid detection and analysis of a seismic event on its initiation, allowing the issuance of a ground motion warning of the order of seconds, is appealing. We present our efforts to design and implement such a system in southern California. The early warning systems currently operating in Mexico and Taiwan rely on significant distances (> 100 km) between the source and populated regions. In this scenario an early warning system can wait for several stations to detect an event, allowing the application of standard location and magnitude determination algorithms (a process that may take tens of seconds), and still issue a warning tens of seconds in advance of associated ground motion. The close proximity of fault zones to metropolitan areas in southern California precludes such an approach. Instead, we develop a system more similar to the UrEDAS warning system in Japan. The two event parameters needed are location and magnitude. The high density of seismic stations in southern California (~25 km spacing in populated areas) allows for an adequate location of events based solely on the first station to detect a P-arrival. The classical use of amplitude to determine magnitude is problematic due to its relatively high sensitivity to epicentral distance close to the source. Instead, we utilize the frequency dependence of the P-arrival to magnitude, which is less sensitive to epicentral distance. With this approach we estimate event magnitude with an accuracy of +/- 1 magnitude unit using the P-arrival at one station only. As the P-arrival is recorded at additional stations, we average the magnitude estimates, which reduces the uncertainty. The event location and magnitude may then be used to estimate ground motion throughout the region using attenuation relations. Using the current TriNet infrastructure we expect to be able to reduce data transmission and analysis time sufficiently to be able to give zero to a few seconds warning prior to the onset of peak, damaging ground motion in the epicentral region. The warning time improves for locations further from the epicenter and, as the time-since-event initiation increases, the uncertainty in ground motion predictions decreases and warning messages can be updated. |