Constraining parameters in state-of-the-art marine pelagic ecosystem models – is it actually feasible with typical observations of standing stocks?

In a changing climate, marine pelagic biogeochemistry may modulate the atmospheric concentrations of climate-relevant species such as CO2 and N2O. To-date, projections rely on earth system models featuring simple pelagic biogeochemical model components, embedded into 3-D-ocean circulation models. Typically, the nucleus of these biogeochemical components are ecosystem models (i.e., a set of partial differential equations) which describe the interaction between nutrients, phytoplankton, zooplankton, and sinking detritus. Most of these models rely on the hyperbolic Michaelis–Menten (MM) formulation which specifies the limiting effect of light and nutrients on carbon assimilation by autotrophic phytoplankton. The respective MM constants, along with other model parameters, are usually tuned by trial-and-error exercises where the parameters are changed until a "reasonable" similarity with observed standing stocks is achieved. Here, we explore with twin experiments (or synthetic "observations") the demands on observations that allow for a more objective estimation of model parameters. We start with parameter retrieval experiments based on "perfect" (synthetic) observations which we, step by step, distort to approach realistic conditions and finally confirm our findings with real-world observations. In summary, we find that MM constants are especially hard to constrain because even modest noise (10%) inherent to observations may hinder the parameter retrieval already. This is of concern since the MM parameters are key to the model's sensitivity to anticipated changes of the external conditions. Further, we illustrate problems associated with parameter estimation based on sparse observations which reveals (additional) parameter dependencies. Somewhat counter to intuition we find, that more observational data can degrade the ability to constrain certain parameters.