Ecological non-linear state space model selection via adaptive particle Markov chain Monte Carlo (AdPMCMC)

摘要: We develop a novel advanced Particle Markov chain Monte Carlo algorithm that is capable of sampling from the posterior distribution non-linear state space models for both unobserved latent states and unknown model parameters. apply this methodology to five population growth models, including with strong weak Allee effects, test if it can efficiently sample complex likelihood surface often associated these models. Utilising real also synthetically generated data sets we examine extent which observation noise process error may frustrate efforts choose between Our involves an Adaptive Metropolis proposal combined SIR MCMC (AdPMCMC). show AdPMCMC samples complex, high-dimensional spaces efficiently, therefore superior standard Gibbs or Hastings algorithms are known converge very slowly when applied ecological considered in paper. Additionally, how be used recursively estimate Bayesian Cram\'er-Rao Lower Bound Tichavsk\'y (1998). derive expressions Bounds them considered. results demonstrate number important features common most notably their multi-modal surfaces dependence static dynamic conclude by each use Bayes factors highlight significantly diminishes our ability select among some particularly those designed reproduce effect.