Abstract/Summary

Equations used to describe the main biological processes determine the dynamics of biogeochemical models. From previous studies, altering the form of these process ‘structure functions’ has been shown to produce larger differences than changing the values of the parameters used in the models. This study explores the effect of this structural sensitivity in a marine biogeochemical model by generating an ensemble of runs. We use a 1-D Model of Ecosystem Dynamics, nutrient Utilisation, Sequestration, and Acidification (MEDUSA) ensemble, where each member has a different combination of key biogeochemical process equations, each of which tuned to be as similar as possible to the default functions. The model is run at five oceanographic stations spanning three different biogeochemical regimes or provinces: oligotrophic, coastal, and abyssal plain. Marine biogeochemical models are also sensitive to the physical environment, so we also explore the relative impact of altering the physical input and biogeochemical process equations, separately and together. The impacts of perturbing the biogeochemistry and physics are quantified using statistical metrics, chlorophyll depth distributions, and phytoplankton bloom phenology. We explored the signature characteristics of the different ensembles by examining the anomaly correlations between different ensemble members and also the nitrogen fraction in phytoplankton across different ensemble members. We found that even small perturbations in model structure can produce a large ensemble spread in many metrics that then mostly easily encompasses the in situ observations. This perturbed biogeochemistry ensemble (PBE) also has an improved RMSE between observations and the ensemble mean, compared to a single deterministic model default run. Perturbing the physics does not generate as large an ensemble range in many of the metrics studied, and cannot always encompass the in situ chlorophyll observations. From exploring the signature characteristics of the different ensembles, very different characteristics are produced from the two ensembles. Perturbing biogeochemistry alters exchange fluxes between biogeochemical compartments, whereas perturbing the physics only alters the nutrient supply to the biological compartments. Therefore, the perturbed biogeochemistry ensemble provides better representations of uncertainty. We discuss how this might be useful for interpreting discrepancies against observational data.