Programmatic Detection of Spatial Behaviour in an Agent-Based Model

Abstract

The automated detection of aspects of spatial behaviourin an agent-based model is necessary for model testingand analysis. In this paper we compare four predictors of herdingbehaviour in a model of a grazing herbivore. We find that a) the mean number of neighbours adjustedto account for population variation and b) the mean Hammingdistance between rows of the two-dimensional environment can beused to detect herding. Visual inspection of the model behaviourrevealed that herding occurs when the herbivore mobility reachesa threshold level. Using this threshold we identify a limits forthese predictors to use in the program code. These results apply only to one set of parameters and environmentsize; future research will involve a wider parameterspace.