ABSTRACT
Partially nonergodic ground‐motion models (GMMs) are used in probabilistic seismic hazard analysis for calculating site‐specific hazards for soil sites. In calculating soil hazard as a part of the partially nonergodic methodology (also called single‐station sigma), the ergodic sigma is reduced by modeling the median site‐specific site term relative to the ergodic GMM. The standard deviation of the median site‐specific site term is removed from the ergodic standard deviation, but the global value of the site‐amplification (AF) variability about the median site term remains in the single‐station sigma. It has been common practice to include the aleatory variability of site amplification from analytical modeling of the site response in the soil hazard calculation combined with the single‐station sigma for the input motion variability, but this double counts the site‐amplification variability by including both the global AF variability and the site‐specific AF variability. Two approaches for avoiding this double counting are presented: (1) remove the global AF variability and add the site‐specific AF variability for both linear and nonlinear behavior; and (2) assume that at low strain, the site‐specific AF variability is equal to the global AF variability. This second approach eliminates the need to estimate global AF variability for the linear range, which is currently not well calibrated. A second issue related to aleatory variability for site response is the treatment of the model error from a 1D site response analysis. In some approaches, the model error is combined with aleatory variability for a simplified approach to estimating the mean hazard, but the model error should be modeled as an epistemic uncertainty as the model error can be reduced with improvements to the model.