- Ground-motion model is:
- Uses V s,30 to characterise sites.
- Uses 3 mechanisms:
- FNM = FRV = 0.
- FNM = 1, FRV = 0.
- FRV = 1, FNM = 0.
- Uses 2 earthquake types:
- FAS = 0.
- FAS = 1.
- Response parameter is acceleration
for 5% damping.
- Develops model to demonstrate use of implicitly nested, partially crossed, mixed-effects regression, which
addresses limitations of multistage additive random-effects approach to obtain single-station σs. Also
considers the influence of uncertainties in input variables (Mw and V s,30) by using Bayesian influence to
derive model. Focus is on estimation of variance components. Selects a sufficiently flexible functional form
to enable reasonable estimates of these components to be made.
- Uses data from the NGA West database (Chiou et al., 2008). Data from 15 countries but heavily
dominated by earthquakes in California and Taiwan. Uses random effects for each country not because
ground motions recognize politcal borders but as a demonstration of mixed-effect regressions accounting
for geographical differences. Approach does not try to solve directly for the country random effects but to
quantify their variance.
- Finds that some of the coefficients are not significant (e.g. quadratic magnitude-scaling and mechanism
terms), by applying traditional approaches, but retains all coefficients so that same functional form applies
to all periods.
- Finds that coefficients vary greatly when considering different sets of random effects, which demonstrates
difficulty in removing ergodic assumption in a series of steps. Although notes for (overparameterized)
models that predictions from models may be similar, except at edges of data.
- Finds country-to-country (or regional) variance is nontrivial.
- Finds that when accounting for uncertainties in Mw and V s,30 that magnitude dependence of model
changes and the variance components reduce.
- Develops model that allows removal of ergodic assumption by including additional country-dependent
random effects for magnitude-dependence of geometric spreading, anelastic attenuation and V s,30 scaling.
Finds that many of the random effects are not well constrained and hence could remove them.
- Finally proposes model for intra-event site-corrected σ (ϕSS) based on combining analyses.