French-Japanese Workshop on Earthquake Source, Paris-Orléans, France 5-9 October 2009
C. Satriano, M. Lancieri, G. Festa and A. Zollo
In the framework of experimenting earthquake early warning in Southern Italy, we developed an integrated methodology called PRESTo (PRobabilistic and Evolutionary early warning SysTem; Zollo et al., 2009; Iannaccone et al., 2009).
PRESTo is a regional approach to earthquake early warning (EEW), i.e. it relies on the availability of a seismic network deployed in the hypocentral area, in order to rapidly detect a potentially destructive earthquake and estimate its location and size. The final goal of such a system is to provide a reliable estimate of expected peak ground motion at one or more target sites, a few seconds after the earthquake occurrence and before the shaking actually reaches the target.
Regional EEW is an inherently time-dependent problem, since the quantity of available information (the number of recording stations and the amount of signal recorded at each station) increases with time. This leads to a typical trade-off between the quality of the information and the time at which it is available. In general, the earlier we get information, the less reliable it is.
PRESTo tries to address this problem through a fully probabilistic and evolutionary framework. The estimates provided by our approach are continuously updated, as new data is available. Moreover, earthquake location, size and expected ground motion are computed as probability density functions (pdf), so that it is always possible to evaluate the uncertainty associated to these estimates, and to set an appropriated level of reliability for each early warning action, depending on how critical it is.
The real-time approach for earthquake location combines at each time step after the first trigger, the P-wave arrivals from triggered stations with the information that, at that time, other stations have not yet triggered (Satriano et al., 2008). When the first station detects the P wave, the hypocentral volume is defined by the Voronoi cell associated to that station. At any time after the first P-arrival, the information from triggered and not-yet-triggered stations is combined using a generalization of the equal differential time formulation (Font et al., 2004). The procedure for real-time magnitude estimation (Lancieri and Zollo, 2008) leverages the empirical correlation laws between low-pass-filtered P and S peak displacement (PD), measured in the first seconds after the phase arrival, and the final magnitude (Zollo et al., 2006).
Its formulation, founded on the Bayes theorem, allows evaluation of the conditional pdf of magnitude, at any time step after the first event detection. The information from different stations is expressed by the likelihood product, and then combined with prior information, given by the pdf retrieved in the previous time step, and by the Gutenberg-Richter distribution.
Finally, PRESTo makes use of ground motion attenuation laws in order to predict the expected shaking at distant targets.
We will illustrate the principles of the system and discuss an application to a Japanese earthquake, using the K-Net and Kik-Net networks.