- Ground-motion model is: where Y is in m∕s2, a0 = -2.154894158, a1 = 2.015257331, a2 = -0.160029327, a3 = -2.819451336,
a4 = 0.227083419, a5 = 12.11797779, a6 = -0.448080444, a7 = -0.141680182, a8 = 0.128978526,
τ = 0.37718 (inter-event), ϕS2S = 0.40838 (inter-station) and ϕ = 0.49836 (within-event). Here the mean
of the posterior distributions are reported.
- Use V s,30 to characterise sites.
- Use 3 mechanisms:
- SN = 1 and SR = 0.
- SR = 1 and SN = 0.
- SN = SR = 0.
- Use same subset of RESORCE (Akkar et al., 2014d) as Hermkes et al. (2014).
- Use Bayesian inference via Markov Chain Monte Carlo sampling that jointly estimates coefficients and
correlations between periods and also outputs various components of σ. Use broad normal distributions as
priors for the coefficients so that they are fairly unconstrained. For covariance matrices use inverse-Wishart
distributions as priors to ensure they are positive definite. Check convergence of chains using the potential
scale reduction factor.
- Data from 251 stations.
- Formulate model as a multi-level model with levels for earthquake, station and record, which allows
simultaneous estimation of coefficients and correlations for all parameters.
- Choose functional form because relatively simple but it is still able to capture general characteristics of
- Provide full covariance matrices.
- Examine residuals w.r.t. Mw, rjb and V s,30 and find no clear trends.
- Present predictions from model by sampling from the posterior distributions of the coefficients,
demonstrating the epistemic uncertainty in the median predictions.