Allen CV
Seismo Lab
Earth & Planetary
UC Berkeley

Testing ElarmS in Japan

Holly M Brown, Richard M Allen
Seismological Laboratory, University of California, Berkeley
Veronica F. Grasso
Group on Earth Observations, Geneva, Switzerland

Seismo. Res. Lett., 80 (5) p727-739, doi: 10.1785/gssrl.80.5.727, 2009.

Part of a special issue on Earthquake Early Warning
Seismo. Res. Lett., 80 (5) 2009.

Download a reprint: BrownAllenGrassoSRL2009.pdf

Earthquake Alarm Systems, or ElarmS, is a network-based earthquake early warning methodology developed in California. It is currently processing seismic data in realtime from stations throughout California. Here we test the methodology's ability to process large-magnitude events, using a dataset of 84 Japanese earthquakes recorded by the K-NET strong motion seismic network including 43 with magnitude greater than 6. Using the first few seconds of the P-wave arrival we determine regional scaling relations between magnitude and peak displacement, and between magnitude and maximum predominant period. We consider the effect of the number of seconds of P-wave data used and number of stations reporting data, and examine ElarmS' ability to process large-magnitude events. Estimates improve as additional stations trigger and more station data is incorporated into the estimate. For the entire dataset the average magnitude error and standard deviation is 0.0±0.4. For events with JMA magnitude of 6.0 or greater the error is 0.0±0.5, for >=7 events it is -0.2±0.5. This indicates a slight "saturation" effect for the M>=7 earthquake, which is partly due to saturation in P-wave amplitude and partly due to difficulty in rapid and accurate location of the large events, which are all offshore. From the observed errors in magnitude, location, and ground acceleration estimates, we develop an error model for ElarmS. Having estimated the distribution of errors at each stage of the ElarmS algorithm, the model can be used to predict the error in any ElarmS estimate, based on the quantity of station observations included in the estimate. The model separates out error due to magnitude estimation, location estimation, and attenuation relations. Using Monte Carlo simulations we explore the full range of errors in PGA predictions from the first estimate using 1 sec of P-wave data at the first station to trigger, to using P-wave data and PGA observations at 5 stations. The error distributions have mean errors of -0.2 to 0.2 (median 0.0) and standard deviations of 0.3 to 0.6 (median 0.4). A factor of two difference in the predicted PGA relative to the observed corresponds to an error of 0.7.

Download a reprint: BrownAllenGrassoSRL2009.pdf