ALBERT

Description: Pelagic, coupled ocean circulation-ecosystem models, are widely used in climate research. These tools aim to quantify fluxes of nutrients and carbon in the ocean and are, increasingly, the base of future projections. For this purpose it is crucial to quantify and identify the sources of uncertainties. In contrast to physical models, the underlying equations for ecosystem models are derived from empirical relationships rather than based on first principles. This resulted in the development of a multitude of different ecosystem models – different in respect to both, underlying principles and complexity. Clearly, the question arises, to what extent the sensitivities of these models are comparable. This study focuses on the intrinsic dynamics of some widely used, simple (containing 2–3 prognostic variables) ecosystem models in a 0-D framework (i.e., comprising only the well-mixed oceanic surface layer). A suite of differing model approaches is tuned such that their behavior is similar. The setup resembles the well-mixed oceanic surface layer in the Baltic proper. It is illustrated that strong differences between the model approaches appear due to exemplary, anticipated changes in the external nutrient and light conditions. Herewith, we demonstrate the well-known, but rarely demonstrated fact that, apparent consistency between modeled prognostic variables with today's data bases is not necessarily a good measure of forecast skill. The causes which lead to the different sensitivities are illustrated by considering the steady state solutions. It is pointed out, that apparently small changes in the model formulations can result in very different dynamical behavior and an enormous spread between the model approaches, despite the feasibility to tune a common behavior in a limited range of light and nutrient supply. In our examples, the sensitivity is mainly a function of the formulation of the loss rate of phytoplankton. It is thus, in particular, the formulation of highly unknown heteorotrophic processes that determines the model sensitivity.