Model error varieties and their relative importance in ocean state estimation.

Model uncertainty and controllability impact the possibility of fitting a model to data. Different varieties of model error need to be considered. Uncertain model components that are selected through discrete choices and switches (e.g. between advection schemes) are categorized as `structural model error’. Such model error may limit the possibility to control a model trajectory. In contrast, ‘external model error’ (e.g. in uncertain forcing fields) and `parametric model error’ (e.g. in uncertain diffusivity coefficients) are controlled by continuous variables that allow smooth state transitions. They can therefore be estimated under the constraint of fitting observations using, e.g., an adjoint model. A first assessment of the relative importance of external, parametric and structural model errors is presented based upon the ECCO v4 (Estimating the Circulation & Climate of the Ocean, version 4) global ocean state estimate that covers 1992-2011. Parametric and external model uncertainty generally appear to dominate over structural model uncertainty. Turbulent transport parameter adjustments are thus in fact much larger than what would be expected if they were merely compensating for unrelated structural model errors. Parametric and external model uncertainty appear to be of comparable magnitude in general, but depending on the ocean state characteristic of interest, one can predominate over the other. In particular, turbulent transport parameter adjustments are instrumental to the close fit of ECCO v4 to observed hydrographic profiles.