Increasing model structural complexity inhibits the growth of initial condition errors

Abstract

One source of inaccuracy in non-linear deterministic ecological models is the growth of initial condition errors. A size-resolved pelagic ecosystem model is used to investigate the effects of changing model structural complexity on the growth rate of initial condition errors. Structural complexity is altered by (1) changing the number of biological size-classes; and (2) changing prey size-ranges which changes the number of linkages for the same number of size-classes. Ensembles of model runs with tiny variations in initial conditions are undertaken and member divergence used to estimate ensemble spread (a measure of the growth of initial condition errors). Increasing prey ranges and therefore the number of linkages greatly reduced the rate of growth of initial condition errors, but ecosystem behaviour is also altered, restricting the generality of the result. At more than 123 size-classes, increasing the number of size-classes while not changing either the model equations or parameters does not alter ecosystem behaviour for over 200 days. In this case, increasing structural complexity through increasing the number of size-classes did not alter the growth of initial condition errors for the first 30 days of the simulations, but afterwards reduced error growth. There are many advantages of parsimonious ecological models with small numbers of classes and linkages, but they are more likely to suffer from the growth of initial condition errors than structurally complex models.