Ice roughness, which has a major influence on in-flight icing heat transfer and, hence, ice shapes, is generally input from empirical correlations to numerical simulations. It is given as uniform in space, while sometimes being varied in time. In this paper, a predictive model for roughness evolution in both space and time during in-flight icing is presented. The distribution is determined mathematically via a Lagrangian model that accounts for the stochastic process of bead nucleation, growth, and coalescence into moving droplets and/or rivulets and/or water film. This general model matches well the spatial and temporal roughness distributions observed in icing tunnel experiments and is embedded in FENSAP-ICE, extending its applicability outside the range of airfoil types for which correlations exist. Thus, an additional important step has been taken toward removing another empirical aspect of in-flight icing simulation.= temperature u, v = velocity = collection efficiency = viscosity = density of water = shear stress