Adaptive model of plankton dynamics for the North Atlantic

Plankton ecosystems in the North Atlantic display strong regional and interannual variability in productivity and trophic structure, which cannot be captured by simple plankton models. Additional compartments subdividing functional groups can increase predictive power, but the high number of parameters tends to compromise portability and robustness of model predictions. An alternative strategy is to use property state variables, such as cell size, normally considered constant parameters in ecosystem models, to define the structure of functional groups in terms of both behaviour and response to physical forcing. This strategy may allow us to simulate realistically regional and temporal differences among plankton communities while keeping model complexity at a minimum. We fit a model of plankton and DOM dynamics globally and individually to observed climatologies at three diverse locations in the North Atlantic. Introducing additional property state variables is shown to improve the model fit both locally and globally, make the model more portable, and help identify model deficiencies. The zooplankton formulation exerts strong control on model performance. Our results suggest that the current paradigm on zooplankton allometric functional relationships might be at odds with observed plankton dynamics. Our parameter estimation resulted in more realistic estimates of parameters important for primary production than previous data assimilation studies. Property state variables generate complex emergent functional relationships, and might be used like tracers to differentiate between locally produced and advected biomass. The model results suggest that the observed temperature dependence of heterotrophic growth efficiency [Rivkin, R.B., Legendre, L., 2001. Biogenic carbon cycling in the upper ocean: effects of microbial respiration. Science 291 (5512) 2398-2400] could be an emergent relation due to intercorrelations among temperature, nutrient concentration and growth efficiency. A major advantage of using property state variables is that no additional parameters are required, such that differences in model performance can be directly related to model structure rather than parameter tuning.