Earthquake early warning applications using downhole arrays in Istanbul

The Kocaeli (Mw 7.5) and Düzce (Mw 7.2) earthquakes that occurred in 1999 revealed the fact that the earthquakes in Marmara Region should be considered especially in terms of the damage that may result in Istanbul. An Emergency Response and Early Warning System was established to reduce possible losses after a damaging earthquake in Istanbul and to produce Rapid Loss Maps to assist rescue teams with emergency response. The main element of the Earthquake Early Warning is the rapid and reliable estimation of the magnitude of the earthquake. In order to calculate the magnitude of the earthquake, it is necessary to determine whether the earthquake fracture will continue or not. This is generally understood from the characteristics of the initial movement (P waves). For this purpose, the characteristics of P waves have been determined by using the τc - Pd method and the waveforms of earthquakes recorded by the vertical component acceleration sensors located on the bottom of the wells and on the surface. Most of the downhole array data is from ATK station. The surface records used are from the early warning stations operated by KOERI and the strong motion network operated by AFAD. As a result, two models have been developed to predict the magnitude of an impending large earthquake and the peak ground velocity (PGV) amplitudes associated with it. A verification process of the models has been applied to predict the moment magnitude of September 26, 2019 Mw=5.7 offshore Silivri earthquake. The models developed from the ATK downhole array predicted the size of the earthquake as Mw=5.8 within 7 seconds of the origin time which in turn yield about 15 seconds early warning time for most of the Istanbul Metropolitan area. The models developed using waveforms recorded at surface have predicted the size of the earthquake as Mw=6.1 which is comparable to the Mw=6.0 prediction of Wu and Kanamori (2005) model. The size and amplitude of the prediction models obtained in this study have been considerably improved compared to the models published a decade ago.