Impacts of biological parameterization, initial conditions, and environmental forcing on parameter sensitivity and uncertainty in a marine ecosystem model for the Bering Sea

We use a series of Monte Carlo experiments to explore simultaneously the sensitivity of the BEST marine ecosystem model to environmental forcing, initial conditions, and biological parameterizations. Twenty model output variables were examined for sensitivity. The true sensitivity of biological and environmental parameters becomes apparent only when each parameter is allowed to vary within its realistic range. Many biological parameters were important only to their corresponding variable, but several biological parameters, e.g., microzooplankton grazing and small phytoplankton doubling rate, were consistently very important to several output variables. Assuming realistic biological and environmental variability, the standard deviation about simulated mean mesozooplankton biomass ranged from 1 to 14 mg C m− 3 during the year. Annual primary productivity was not strongly correlated with temperature but was positively correlated with initial nitrate and light. Secondary productivity was positively correlated with primary productivity and negatively correlated with spring bloom timing. Mesozooplankton productivity was not correlated with water temperature, but a shift towards a system in which smaller zooplankton undertake a greater proportion of the secondary production as the water temperature increases appears likely. This approach to incorporating environmental variability within a sensitivity analysis could be extended to any ecosystem model to gain confidence in climate-driven ecosystem predictions.

Research highlights
► We explored the sensitivity of the BEST marine ecosystem model to environmental forcing, initial conditions, and biological parameterizations.
► True sensitivity of model output became apparent only when biological and environmental parameters were varied within a realistic range.
► Several biological parameters were consistently very important to multiple model output variables.
► The presented approach to sensitivity analysis enables confidence intervals to be drawn around model output.