Résumé : The formation and accretion of ice on the leading edge of an airfoil can be detrimental to aerodynamic performance. Furthermore, the geometric shape of leading edge ice profiles can vary significantly depending on a wide range of physical parameters, which can translate into a wide variability in aerodynamic performance. The purpose of this work is to explore the variability in airfoil aerodynamic performance that results from variability in leading edge ice shape profile. First, we demonstrate how to identify a low-dimensional set of parameters that governs ice shape from a database of ice shapes using Proper Orthogonal Decomposition (POD). Then, we investigate the effects of uncertainty in the POD coefficients. This is done by building a global response surface surrogate using Polynomial Chaos Expansions (PCE). To construct this surrogate efficiently, we use adaptive sparse grid sampling of the POD parameter space. We then analyze the data from a statistical standpoint.