Selecting the most suitable rupture model for the stochastic simulation of the 1999 Izmit earthquake and prediction of peak ground motions

In this study, we use a stochastic finite-fault technique based on a dynamic corner frequency to investigate how the fault and slip models affect the high frequency simulations of the 1999 Izmit (Turkey) earthquake. Seven different rupture models, one of them generated using common fault parameters and random slip distribution are tried to obtain the best matching with the observations. The synthetic seismograms computed in the frequency band 0.1-25 Hz are compared with the observations both in time and frequency domain. Six accelerometric stations located close to the observed surface rupture are chosen for the comparisons considering the fact that the slip contributions are visible at these station records better than the other stations. We also estimated average H/V spectral ratios using the available accelerometric recordings to take into account site amplification at each site. We also acquired ambient noise data at some stations that lack sufficient earthquake records. The results show that none of the rupture models fully simulate the observations at all the stations. Most of the rupture models underestimate the Fourier amplitudes at frequencies lower than 0.4 Hz, whereas overestimate them at higher frequencies. The underestimation may result from a directivity effect which likely causes higher amplitudes on the observed ground motions in low frequency band. While all the rupture models display similar average bias functions, the minimum average error for spectral amplitudes is obtained for the rupture model of Bouchon et al. (2002). The achievement of the random slip distribution model also yields satisfactory results. After we have optimized the rupture model, we have simulated the strong ground motions on a regular grid (0.2 degrees x 0.2 degrees) covering the study area for bedrock conditions. In total, we have estimated peak accelerations and peak velocities at 135 points. The results show that the maximum acceleration values during the lzmit earthquake reached 785 cm/s(2), and the largest velocity values were around 75 cm/s. The peak ground accelerations are still smaller than those predicted by the empirical relations but are well above the observed ones. The low accelerations might be attributed to the low stress drop that might be due to the large rupture length. (c) 2012 Elsevier Ltd. All rights reserved.