Because of the limited number of strong‐motion records that have measured ground response at large strains, any statistical analyses of seismic site‐response models subject to strong ground motions are severely limited by a small number of observations. Recent earthquakes in Japan, including the Mw 9.0 Tohoku earthquake of March 2011, have substantially increased the observations of strong‐motion records that can be used to compare alternative site‐response models at large strains and can subsequently provide insight into the accuracy and precision of site‐response models. Using the Kiban‐Kyoshin network (KiK‐net) downhole array data in Japan, we analyze the accuracy (bias) and variability (precision) resulting from common site‐response modeling assumptions, and we identify critical parameters that significantly contribute to the uncertainty in site‐response analyses. We perform linear and equivalent‐linear site‐response analyses at 100 KiK‐net sites using 3720 ground motions ranging in amplitude from weak to strong; 204 of these records have peak ground accelerations greater than at the ground surface. We find that the maximum shear strain in the soil profile, the observed peak ground acceleration at the ground surface, and the predominant spectral period of the surface ground motion are the best predictors of where the evaluated models become inaccurate and/or imprecise. The peak shear strains beyond which linear analyses become inaccurate in predicting surface pseudospectral accelerations (PSA; presumably as a result of nonlinear soil behavior) are a function of vibration period and are between 0.01% and 0.1% for periods <0.5 s. Equivalent‐linear analyses become inaccurate at peak strains of ∼0.4% over this range of periods. We find that, for the sites and ground motions considered, site‐response residuals at spectral periods >0.5 s do not display noticeable effects of nonlinear soil behavior.
Online Material: Site‐specific information and model residuals at 100 KiK‐net stations.