Intraevent Spatial Correlation Characteristics of Stochastic Finite-Fault Simulations by T. J. Liu, Gail M. Atkinson, H. P. Hong, and Karen Assatourians Abstract Spatially correlated strong ground motions can cause severe seismic da- mage to spatially-distributed buildings and infrastructure. Empirical spatial correlation models for ground-motion measures such as peak ground acceleration and spectral acceleration have been developed using strong ground-motion records from California, Taiwan, and Japan. In this paper, we compare an empirical spatial correla- tion model for the 1999 Chi-Chi, Taiwan, earthquake with that obtained from stochas- tic finite-fault simulations; the stochastic finite-fault method is a widely used technique to generate synthetic ground-motion records for engineering applications and hazard assessment. We show that the stochastic records do not reproduce the ob- served intraevent spatial correlation characteristics and comment on what would be required to make them do so. Introduction Spatially correlated strong ground motions can cause severe accumulation and concentration of seismic damage and loss to spatially-distributed buildings and infrastructure. The spatial correlation of peak ground motions and response spectra for a given event (i.e., the intraevent spatial correla- tion) has recently been studied by several researchers (Boore et al., 2003; Kawakami and Mogi, 2003; Wang and Takada, 2005; Goda and Hong, 2008; Jayaram and Baker, 2009; Goda and Atkinson, 2009; Goda and Atkinson, 2010). Com- mon findings are that, for a given earthquake, ground-motion measures at different sites are spatially correlated; the corre- lation decreases as the separation distance increases. Goda and Hong (2008) developed an empirical spatial correlation model for peak ground acceleration (PGA) and spectral acceleration (SA) at various natural vibration periods, using ground-motion records from the Next Generation Attenua- tion (NGA) database of the Pacific Earthquake Engineering Research (PEER) Center. Goda and Atkinson (2010) com- pared the spatial correlation of strong ground-motion data from SK-net, a regional network in Japan having dense sta- tion coverage, with the Goda–Hong model developed from the NGA data (Goda and Hong, 2008); they concluded that the extended dataset, which includes the Japanese data, is also consistent with this model. The stochastic finite-fault method is a widely-used ground-motion simulation technique that has been shown to be simple and effective in the generation of synthetic ground-motion records (Beresnev and Atkinson, 1997, 1998; Motazedian and Atkinson, 2005; Atkinson et al., 2009; Boore, 2009). Stochastic simulations do a good job in match- ing response spectra and peak motions observed in earth- quakes over a broad frequency range but are particularly applicable at high frequencies (> 1 Hz). More detailed broadband simulation methods use deterministic waveform modeling at low frequencies but also revert to stochastic methods at high frequencies (e.g., Hartzell et al., 1999; Fran- kel, 2009; Halldorsson et al., 2011). Thus the performance of stochastic simulation methods is an important issue. In this paper, we assess whether the spatial correlation structure of ground motions simulated by the stochastic finite-fault method match that of real records using the 1999 M 7.6 Chi-Chi earthquake as a case study. To evaluate the intraevent spatial correlation characteristics of stochastic finite-fault simulations, we perform simulations for a test event, that of the Chi-Chi earthquake, and compare the obtained spatial correlation to that calculated from the records. For the simulations, we use the EXSIM stochastic finite-fault program of Motazedian and Atkinson (2005). This is a freely-available and widely-used program that has been validated under a range of conditions (Motazedian and Atkinson, 2005; Atkinson et al., 2009; Boore, 2009). The Chi-Chi earthquake is a well-recorded event, with more than 400 records from the mainshock available, and its spatial cor- relation characteristics are already available (Wang and Ta- kada, 2005; Goda and Hong, 2008). In this study, we develop a simulation model for the Chi-Chi earthquake using EXSIM and assess the intraevent spatial correlation coefficient using the simulated records. We compare the obtained spatial cor- relation to that of the actual records. Our results indicate that the former does not match the latter satisfactorily. To over- come this deficiency, we believe a postprocessing fitting approach aimed at improving this aspect of the synthetic records generated using the stochastic finite-fault method could be valuable. 1740 Bulletin of the Seismological Society of America, Vol. 102, No. 4, pp. 1740–1747, August 2012, doi: 10.1785/0120110266