Statistical methods for process-based models

Parameter estimation for stochastic process-based models is an area of ongoing development which forms a bridge between complex systems models and statistical approaches. Recent developments in computational statistics, such as reversible-jump Markov chain Monte Carlo, enable statistically correct parameter estimation for more realistic models than ever before. Relative to regression techniques, stochastic spatio-temporal models are more transparent in their representation of biological processes facilitating communication with subject scientists. On the other-hand regressive techniques outstrip stochastic spatio-temporal models in their ability to handle covariates. In this talk I consider Bayesian parameter estimation for such models formulated as time-homogeneous Markov processes. The approach is initially applied to an agent-based model, and data, describing the foraging behaviour of dairy cows. As time permits I will further illustrate these techniques by application to data describing the spatial temporal spread of disease at the plot-scale and an invasive plant across the UK.