On the Estimation of Ground-Motion Duration Models with an Application to the M9 Simulations
Keywords:Significant Duration, Ground-Motion Model, Bayesian Inference, M9 simulations
Different parameterizations, related to mixed-effects structure, of ground-motion models for the duration of strong shaking are revisited. A new parameterization is proposed which can better account for the separation of source and path duration in additive ground-motion duration models. In addition, different distributions for the likelihood of ground-motion duration given the predictor variables are investigated. Traditionally, duration has been modeled as lognormally distributed, which makes sense for a multiplicative model, but less so for an additive model. Models using a lognormal and Gamma likelihood are compared, using the M9 simulations (Frankel et al., 2018) as an underlying data set. In general, differences between models in terms of predictions are small, but the models using a Gamma distribution perform slightly better than the ones employing a lognormal distribution. Furthermore, the new parameterization outperforms traditional models. All models are estimated via Bayesian inference, accounting for epistemic uncertainty through the posterior distribution of the parameters.
Copyright (c) 2022 Nicolas Kuehn, Melanie Walling
This work is licensed under a Creative Commons Attribution 4.0 International License.