Marine ecosystems represent a rich assemblage of co-evolved species that have complex, non-linear dynamics. This has made them difficult to manage and the recent record of exploitation of marine ecosystems suggests that the mechanisms currently in place for their management are inappropriate for sustained and intensive exploitation (Pauly et al. 2002). Fisheries science has developed sophisticated single- and multispecies approaches to modelling resource dynamics but these have shown mixed success when used to advise about the regulation of exploitation levels. However, it is commonly acknowledged that attempting to model whole or partial ecosystems also has limited utility because the demands this has for data and knowledge about the system far outweigh the financial, logisitical and intellectual resources available (Yodzis 1998). Although some computer-intensive approaches are currently being attempted, their ability to improve predictions of the dynamics of marine ecosystems appears to be quite limited.

This whole- or partial-systems approach to modelling marine ecosystems is driven by a belief in the connectivity of predator–prey processes within ecosystems and the conviction that, with appropriate parameterization, the behaviour of these systems can be predicted within bounds of confidence that are sufficiently narrow to convince us that the investment in the modelling effort has been useful. However, to date the cost–benefit analysis of these approaches has not been computed and the few simple systems in which the approach has been applied soon run into trouble.