Numerical modelling of the Halten Bank area: a validation study

Abstract

From observations and simple model simulations, it has been established that a proper model of the Halten Bank area must, as a minimum, (i) include the barotropic component and at least one baroclinic component, (ii) allow shear instabilities to properly grow into eddies, (iii) include vertical mixing to allow for entrainment of lower layer water masses into the upper layer, (iv) allow for advection of density (temperature and salinity) in the upper water masses, and (v) allow the inclusion of tides. All of these physics factors are included in the mathematical formulation of fully three-dimensional numerical ocean models. Two such models, arbitrarily named the POM model and the SINMOD model, have been used to simulate the current, salinity and temperature structure in the Halten Bank area for March 1988 and then validated against the Norwegian Continental Shelf Experiment 1988 (NORCSEX’88) data set. Although the motion in the two models is governed by the same set of continuous equations, an inspection of the parameterizations and numerical methods employed to solve the governing equations reveals that the SINMOD model should be expected to provide smoother and less energetic results than the POM model. It is also speculated that eddy growth is inhibited in both models due to (i) the tendency of level models to create fictitious diapycnal mixing in areas of sloping density surfaces, and (ii) too coarse resolution. These conclusions are partially supported by the validation analysis. In order to facilitate a meaningful model-model comparison and validation, specific products were provided by careful analyses of both data and model results in accordance with a preselected hierarchic set of comparison criteria. The criteria focus on mean circulation patterns and horizontal and vertical distributions. The analyses reveal that the most probable limitation inhibiting the models in reproducing the observed structures is the lack of horizontal resolution. The chosen grid mesh size (4 km) only barely resolves the eddy scale dictated by the Rossby radius of deformation, thus inhibiting eddy growth due to shear instabilities. Further, both this and earlier studies conclude that it is an open question how complex a model needs to be to be able to simulate the observed structure to a satisfactory degree.