Sea ice thickness and surface morphology obtained from airborne electromagnetic induction sounding have been investigated in order to improve SAR ice type classification. The stochastic properties of the surface profiles have been analysed using the parameters mean elevation, RMS height, skewness, kurtosis, fractal dimension and correlation length. A clustering algorithm has been applied to the roughness parameters, and the analysis has been iterated in order to find roughness parameters that are characteristic for different ice thickness classes. The set of best parameters was found to consist of mean elevation, RMS height, skewness and kurtosis, and the optimal number of clusters was found to be 6. The thickness of the profiles belonging to the same roughness clusters were analysed. It was found that there exists some relation between similar roughness parameters and similar ice thickness. In addition, the roughness parameters have been compared to normalized backscatter coefficients obtained from Radarsat-1 images.