The sigma coordinate, Princeton Ocean Model (POM) has been configured for t
he North Atlantic ocean between 5 degrees N and 50 degrees N as part of dat
a assimilation, model predictability and intercomparison studies. The model
uses a curvilinear orthogonal grid with higher resolution in the western N
orth Atlantic and lower resolution in the eastern North Atlantic. A series
of experiments, each one of a 10-year duration, are performed to evaluate t
he sensitivity of the ocean mean state and variability to model parameters
and model configuration; these experiments include open vs. closed boundary
conditions, low vs, high resolution grids, and different choices of diffus
ion and viscosity, The results show that the use of closed boundaries toget
her with near-boundary buffer zones where temperature and salinity are rela
xed towards the observed values give less realistic flows, weaker recircula
tion gyres and less realistic Gulf Stream separation than do open boundary
conditions. The experiments show that the sensitivity of the ocean variabil
ity in the model to the choice of the Smagorinsky diffusion and viscosity c
oefficients significantly differs from one region to another and largely de
pends on other attributes such as the mean position of the Gulf Stream in e
ach simulation. A 50% change in model resolution in the Gulf Stream region
has a larger effect on ocean variability than a change of diffusivity by a
factor of 10. In areas where either the high or the low resolution models h
ave sufficient resolution, as in the Gulf of Mexico, they are able to produ
ce variability comparable to that observed from altimeter data; elsewhere,
model variability is underestimated. (C) 2000 Elsevier Science B.V. All rig
hts reserved.