This samples from the prior and posterior distributions and performs a
oct/ directory contains a few functions for plotting these
results (GNU Octave and OctBi required).
This package is based on a single-box NPZD (nutrient, phytoplankton, zooplankton and detritus) marine biogeochemical model. It consists of 15 parameters and 15 state variables. Four of the state variables (the $N$, $P$, $Z$ and $D$) interact via a system of differential equations, with flux between them determined by various nonlinear processes computed from the remaining state variables, each of which is allowed to vary over time following a first-order stochastic autoregressive process.
This version of the model was introduced in Parslow et al. (2013). Its behaviour under sampling with the particle marginal Metropolis-Hastings (PMMH) sampler was studied in Murray, Jones & Parslow (2013), then further studied using a bridge particle filter in Del Moral & Murray (2014).
A real data set is provided from the site of Ocean Station Papa (OSP), taken from Matear (1995).
Parslow, J.; Cressie, N.; Campbell, E. P.; Jones, E. & Murray, L. M. Bayesian Learning and Predictability in a Stochastic Nonlinear Dynamical Model. Ecological Applications, 2013, 23, 679-698.
Murray, L. M.; Jones, E. M. & Parslow, J. On collapsed state-space models and the particle marginal Metropolis-Hastings sampler, 2013.
Del Moral, P. & Murray, L. M. Sequential Monte Carlo with Highly Informative Observations, 2014. [arXiv]
Matear, R. Parameter optimization and analysis of ecosystem models using simulated annealing: A case study at Station P. Journal of Marine Research, 1995, 53, 571-607.