Simplifying fish population dynamics for ecosystem-based modeling

The presence of complex, nonlinear dynamics in fish populations, and uncertainty in the structure (functional form) of those dynamics, pose challenges to our ability to predict how these populations change over time.  Here, we describe a set of simple nonlinear forecasting models (requiring only one or two free parameters) that test for the hallmarks of complex behavior, avoid problems of structural uncertainty, and produce accurate short-term forecasts of changes in fish populations.  The concept of co-predictability, in which the dynamics of one species predicts the dynamics of another, is used to identify groupings of species that form ecologically relevant and interacting units.  We use a variety of time series data, including landings, catch per unit effort, and model estimates of biomass and recruitment, from both the California Current and Georges Bank.  In doing so, we compare the dynamics of fish communities in the two systems and explore the usefulness of different datasets in describing population dynamics.  These low-dimensional forecasting models can complement whole-ecosystem models containing hundreds of estimated parameters.  The identification of coupled species units can be the foundation of ecosystem-based modeling by defining a minimum level of ecological organization that is relevant to describing and forecasting community dynamics.