Wind surface drag over sea-ice is a primary control on sea-ice flow patterns and deformations at scales that are important for climate and weather prediction models. Here, we perform a series of Large Eddy Simulations (LES) of fully developed flow over high-resolution snow-ice surfaces of Antarctic sea ice floes to study surface drag and roughness parameters at process scales from 1 cm to 100 m. Snow/ice surface morphology was obtained using a Terrestrial Laser Scanner during the SIPEX II (Sea Ice Physics and Ecosystem experiment II) research voyage to East Antarctica (September-November 2012). LES are performed on a regular domain adopting a mixed pseudo-spectral/finite difference spatial discretization. A scale-dependent dynamic subgrid-scale model based on Lagrangian time averaging is adopted to determine the eddy-viscosity in the bulk of the flow. The effects of large-scale features of the surface on the wind flow (those features that can be resolved in LES) are accounted for through an immersed boundary method. Conversely, the drag forces caused by subgrid-scale features of the surface should be accounted for through a parameterization. However the effective aerodynamic surface roughness parameter z0 for snow ice is not known. Hence, a dynamic approach is utilized, in which this parameter is determined using the first-principles based constraint that the total momentum flux (drag) must be independent on grid-filter scale. This dynamic surface roughness model is inspired by the Germano identity, traditionally used to determine model parameters for closing subgrid-scale stresses in the bulk of a turbulent flow. This type of model has been previously applied to flow over multi-scale terrain and ocean waves but never for snow-ice surfaces. The resulting dynamic parameter will be compared with values obtained for solid terrain and ocean sea surface values, and the implications on overall drag forces on snow ice surfaces will be discussed.