CFD is a primary tool used to assess the in-flight effects of atmospheric icing on aircraft. In-flight ice accretion codes use CFD computed quantities, such as shear stress and heat transfer, to predict ice shape formation over rough surfaces. The equivalent sandgrain roughness approach is the model commonly used in icing codes for the prediction of skin friction and heat fluxes over iced surfaces. Additional turbulent Prandtl number corrections can be added to the Reynolds Averaged Navier-Stokes (RANS) equations turbulence model to refine the heat transfer. Still, uncertainties persist in identifying the roughness parameters to input into the thermal correction, leaving the characterization of rough surfaces incomplete in terms of research. This chapter develops a methodology for the estimation of roughness input parameters based on the observation of experimental ice accretion. Metamodeling involving Polynomial Chaos Expansion (PCE) and calibration with a Bayesian inversion are employed. The methodology is applied to a NACA0012 airfoil, yielding a glaze ice cross-sectional area and maximum thickness with less than a 6% error from experiments. The approach opens perspectives for the estimation of appropriate case-dependent roughness parameters for RANS-based ice shape predictions.