HR: 1330h
AN: S42B-0155    [PDF]
TI: Artificial Neural Networks for Earthquake Early-Warning
AU: * Boese, M
EM: maren.boese@gpi.uni-karlsruhe.de
AF: Institute of Geophysics, Karlsruhe University, Hertzstrasse 16, Karlsruhe, 76187 Germany
AU: Erdik, M
AF: Dept. of Earthquake Engineering, Bogazici University, Kandilli Observatory and Earthquake Engineering, Cengelkoy, 81220 Turkey
AU: Wenzel, F
AF: Institute of Geophysics, Karlsruhe University, Hertzstrasse 16, Karlsruhe, 76187 Germany
AB: The rapid urbanization and industrial development in areas of high seismic hazard increase the threat to human life and the vulnerability of industrial facilities by earthquakes. As earthquake prediction is elusive and, most likely, will not be achievable in near future, early-warning systems play a key role in earthquake loss reduction. Seismic waves propagate with significant lower velocity than information on these waves can be passed along to a vulnerable area or facility using modern telemetry systems. Within shortest time an earthquake early-warning system estimates the ground motion that will be caused by the oscillating seismic waves in the endangered area. Dependent on the predicted possible damage appropriate automatisms for loss reduction (such as the stoppage of trains or the interruption of gas pipelines) are triggered and executed some seconds to minutes before the devastating waves actually arrive. The Turkish megacity Istanbul faces a seismic hazard of particular severity due to its proximity to the complex fault system in the Marmara region. The likelihood for an seismic event of moment magnitude above 7.2 to occur within the next 30 years is estimated to be 70%. The Istanbul Earthquake Rapid Response and Early-Warning System (IERREWS) is an important contribution to be prepared for future earthquakes in the region. The system is operated by the Kandilli Observatory and the Earthquake Research Institute of the Bogazici University in cooperation with other agencies. The early-warning part of IERREWS consists of ten strong motion stations with 24-bit resolution, communication links and processing facilities. The accelerometers are installed on the shoreline of the Marmara Sea and are operated in on-line mode for continuous and near-real time transfer of data. Using the example of the IERREWS station configuration and seismic background of the Marmara region we present an approach that considers the problem of earthquake early-warning as a pattern recognition task. The seismic patterns are defined by the shape and frequency content of the parts of seismograms that are available at each time step. From these, parameters that are relevant to seismic damage, such as peak ground acceleration (PGA), peak ground velocity (PGV), response spectral amplitudes at certain periods and macroseismic intensity, are estimated using Artificial Neural Networks (ANN). We combine pattern recognition with an additional rule-based system in order to detect inconsistencies between ground motion estimations and measurements. This combination provides a reliable and accurate system for early-warning that is demanded by its huge social and economic impact.
DE: 7212 Earthquake ground motions and engineering
DE: 7215 Earthquake parameters
DE: 7223 Seismic hazard assessment and prediction
DE: 7299 General or miscellaneous
DE: 9335 Europe
SC: Seismology [S]
MN: 2003 Fall Meeting