We examine sensitivity of simulated sea ice states to scale dependence of model parameters controlling ice dynamics, thermodynamics, clouds and coupling of sea ice with the atmosphere and ocean. For this study we use a fully coupled and a subset of the Regional Arctic System Model (RASM), where the atmospheric and land components are replaced with prescribed realistic atmospheric reanalysis data for 1948-2009. The latter approach allows direct comparison of model results with observations and reduces computational costs. The parameters examined here include the sea ice strength, surface roughness length scale, ice-ocean drag, oceanic mixing and cloud parameterization. Model outputs from multi-decadal ensembles are compared against each other and against basin-wide estimates of sea ice extent and volume. Our results confirm that many parameterizations of sub-grid physical processes currently used in climate models are scale-dependent and we provide further details on fine-tuning required when changing model spatial resolution. We also show that while sea ice extent in many runs compares well against observations, sea ice thickness distribution is acceptable only in a few cases. Hence, we conclude that the use of observed sea ice extent only to validate the skill of sea ice models is not a sufficient model constraint.