Publication Name: AEGC 2021
Authors: Laurence Davies, Christopher Drovandi, Matthew Sutton, Alan Yusen Ley-Cooper
Date Published: September 2021
Number of Pages: 7
Abstract:
Exact methods of Bayesian model selection require exhaustive computation of normalising constants or sampling of the joint posterior of parameters and models via reversible jump Markov chain Monte Carlo (RJMCMC). Until recently, the latter has been favoured in geophysics applications where there is a mid-to-high cardinality of the set of candidate models. However, RJMCMC algorithms alone do not easily translate to parallel computing architectures. Motivated by the detection of induced polarisation (IP) effects in airborne electromagnetic (AEM) data, this contribution explores a Bayesian approach to the IP-detectability problem using model selection methods, and also employs an approach novel to geophysics whereby cross-dimensional proposals are used within the embarrassingly parallelisable static Sequential Monte Carlo (SMC) class of algorithms as applied to parameter and model inference.