Bayesian modeling for uncertainty quantification and ground model calibration #65
Description
This issue is intended as a feature suggestion and to share relevant literature, rather than a bug report
There is a body of research on using Bayesian model updating for ground-motion models (GMMs) by calibrating residual or regression parameters of empirical models. This approach could be incorporated into the current eGSIM framework, allowing users to input a dataset and apply Bayesian inference methods—such as Markov Chain Monte Carlo (MCMC)—to obtain posterior distributions of the model parameters.
A representative reference is:
Kowsari, M., Halldorsson, B., Hrafnkelsson, B., Snæbjörnsson, J. Þ., & Jónsson, S. (2019). Calibration of ground motion models to Icelandic peak ground acceleration data using Bayesian Markov Chain Monte Carlo simulation. Bulletin of Earthquake Engineering.
In addition, posterior residuals can be used for model ranking using Bayesian information criteria, such as the Deviance Information Criterion (DIC), as a complement to existing log-likelihood–based rankings. Posterior residuals and their summary statistics (e.g., posterior means) can be used both for uncertainty-aware residual diagnostics and for Bayesian model comparison using DIC.
A relevant reference is:
Kowsari, M., Halldorsson, B., Hrafnkelsson, B., & Jónsson, S. (2019). Selection of earthquake ground motion models using the deviance information criterion.
Expected inputs for implementation:
Ground-motion dataset (observed earthquake records used to update prior knowledge)
Selected GMMs for calibration and/or comparison
Prior distributions for residual and/or regression parameters
Expected outputs:
Posterior distributions of residual and/or regression parameters
Posterior distribution plots and summary statistics
Comparison tables and visualizations of DIC-based model rankings, presented alongside existing LLH-based rankings