S. R. Hale

Impact of Topographic Normalization on Land-Cover Classification Accuracy

For pixel classifying algorithms to perform effectively, effects of topographic relief must be minimized or removed. In areas of high topographic relief, problems arise when spectral variations in ground target illumination and radiance, owing to differences in incident radiation and nonLambertian reflectance behavior, respectively, cause identical land-cover types to reflect differently, or different cover types to reflect similarly. A Landsat Enhanced Thematic Mapper image was processed using band ratios, the Minnaert Correction, aspect partitioning, and combinations of these treatments to generate independent land-cover classifications. Treatment classification accuracy was determined using error matrices and the Kappa statistic. Producer’s and User’s Accuracies were examined to determine if treatments were superior at producing greater classspecific accuracy. None of the treatments produced a significantly more accurate classification; however, assessment of class-specific accuracies indicated accuracy gains using aspect partitioning alone or in combination with the Minnaert Correction.