Ice accretion codes depend on models of roughness parameters to account for the enhanced heat transfer during the ice accretion process. While mitigating supercooled large droplet (SLD or Appendix O) icing is a significant concern for manufacturers seeking future vehicle certification due to the pending regulation, historical ice roughness studies have been performed using Appendix C icing clouds which exhibit mean volumetric diameters (MVD) much smaller than SLD clouds. Further, the historical studies of roughness focused on extracting parametric representations of ice roughness using multiple images of roughness elements. In this study, the ice roughness developed on a 21-in. NACA 0012 at 0 angle of attack exposed to short duration SLD icing events was measured in the Icing Research Tunnel at the NASA Glenn Research Center. The MVD's used in the study ranged from 100 µm to 200 µm, in a 67 m/s flow, with liquid water contents of either 0.6 gm/m 3 or 0.75 gm/m 3 . The ice surfaces were measured using a Romer Absolute Arm laser scanning system. The roughness associated with each surface point cloud was measured using the two-dimensional self-organizing map approach developed by McClain and Kreeger (2013) resulting in statistical descriptions of the ice roughness. Nomenclature Ac = accumulation parameter AOA = angle of attack b = codebook vectors h(i,j) = neighborhood function of i to j codebook vectors j = codebook vector index LWC = liquid water content [gm/m 3 ] M = number of codebook vectors MVD = median volumetric diameter [µm] N = airfoil or mean ice shape surface normal coordinate direction R 2 = coefficient of determination (regression) R d = high-dimensional data space RMH = roughness maximum height R q = the root-mean-square or "standard deviation" roughness height r a = leading edge radius of curvature SOM = Self-Organizing Map 2 S = airfoil or mean ice shape surface tangential coordinate direction SEE = Standard error of the estimate for regression x = element of data set  = local direction angle of manifold through a codebook vector β = manifold  = direction angle of surface point relative to manifold direction through winning codebook vector  = scaling parameter governing neighborhood size  = learning rate