Dr Elske van der Vaart on calibrating IBMs using ABC

Individual-based models, or IBMs, are becoming increasingly popular within ecology: they are computer simulations that explicitly simulate individual organisms interacting with each other and their surroundings, as defined by their own characteristics and decision rules.

However, due to their complexity, IBMs are difficult to systematically parametrise and evaluate. One promising way forward is to use Approximate Bayesian Computation, or ABC, for this purpose. Originally developed within population genetics, ABC is a technique for selecting the most likely parameter values and model structures given the available data. It involves running models a large number of times, with parameters drawn randomly from their prior distributions, and then retaining the simulations closest to the observations. 

In this talk, Dr van der Vaart will demonstrate the use of ABC with example IBMs concerning the energy budgets of earthworms and cockles. She will show that ABC is capable of calibrating both models, and also reveals parameters that are not estimable given the data available.

She will illustrate how these findings can inform future modelling cycles, and conclude that ABC makes the often complex process of optimizing an IBM’s structure and parameters more transparent and objective.