An evaluation of partially nonergodic PGA ground-motion models for Japanese megathrust earthquakes

We used a ground-motion database of Japanese subduction interface (megathrust) earthquakes from the NGA-Subduction project to develop a series of ground-motion models (GMMs) for peak ground acceleration (PGA) that explore the effects of including random-effect event and station terms and differences in attenuation between forearc and backarc tectonic regions. The GMMs with the lowest standard deviations and the best goodness-of-fit metrics are those that include event and station terms and that differentiate between forearc and backarc attenuation using fractional travel paths within each region as distance metrics. In the best models, anelastic attenuation is found to be much greater in the backarc than in the forearc. The inclusion of station terms does not result in a geographic bias in anelastic attenuation as has been found in past studies. Inclusion of a rake-angle term is found to be marginally significant, whereas the addition of a volcanic-arc-crossing term is most important when station terms are excluded. We provide model coefficients, statistical significance of coefficients, standard deviations, and goodness-of-fit metrics for all of the GMMs explored in this study to show the impact of model assumptions as well as allow users the ability to select a model consistent with their application.