Super-sensitivity to structure in biological models

Applied scientific disciplines use mathematical models to make predictions. In the majority of cases these models are constructed using plausible mathematical characterizations of various component processes of the modelled system, rather than being based entirely on exact mathematical descriptions of proven mechanisms. We use general arguments and a specific example from applied ecology to demonstrate that model predictions can show alarming sensitivity to apparently tiny changes in model specification, in a manner that is counter-intuitive and entirely invisible to conventional model sensitivity analysis. This result has serious implications for practical prediction using biological models.