We performed stochastic finite‐fault simulations for the moment magnitude (M) 9.0 11 March 2011 Tohoku earthquake at KiK‐net sites using the EXSIM algorithm (Motazedian and Atkinson, 2005; Atkinson et al., 2009; Boore, 2009). The average response spectra of the simulated ground motions were calculated and compared with observed ground motions to determine whether stochastic finite‐fault modeling can reproduce key features of the recorded motions. To account for variations in source characteristics, we examined three rupture‐process models: a single‐event model with random slip distribution, a single‐event model with prescribed slip distribution (Yagi, 2011; see Data and Resources), and a multiple‐event model with five strong‐motion generation areas (Kurahashi and Irikura, 2011).

The analysis results indicate that the single‐event models can model the high‐frequency response spectra values well but result in significant model bias in the low‐frequency range. By contrast, the multiple‐event model produces ground motions that are in good agreement with the observations in both high‐ and low‐frequency ranges. Furthermore, the multiple‐event model can capture temporal characteristics of observed ground motions, including multiphase arrivals of major seismic‐wave groups. Importantly, the calibrated EXSIM model for the 2011 Tohoku earthquake can be used to produce predicted ground motions in other regions, such as the Cascadia subduction region of North America, by suitable modifications of regional attenuation and site parameters. The suitability of the simple stochastic simulation method to prediction of motions for future events, the details of which are poorly known, makes its calibration to Tohoku a useful exercise.