Data Assimilation Meetings at Reading

30 Nov 2011 Prof. Richard Sibly (School of Biological Sciences, Reading)
Evaluation of Agent Based Models, ABMs.
ABMs are becoming widely used in Ecology and Biology and in Economics and the human sciences. Methods with which to evaluate ABMs are urgently needed, but it is hard to get funding to develop and evaluate techniques. In one of our applications individual woodpigeon move around realistic landscapes making decisions about where to feed, how much to eat, when to reproduce etc in relation to the state of the environment and internal state (e.g., energy reserves). The models are based on the best available ecological data. We also have available datasets showing how ecological variables (e.g., numbers of birds in a studied landscape that were eating brassicas) vary over time. We want to know how well the model predictions fit the data. We want to compare the performance of alternative models. And we may sometimes need to find best fitting parameter values.
So far we have identified Approximate Bayesian Computation as a suitable approach. This has two key advantages over MCMC. Runs do not have to be done sequentially: this is important because individual runs may take 0.5 h. And it is not necessary to write down a likelihood function, which we cannot do.

Page navigation