The Effect of Stratified Topographic Correction on Land Cover Classification in Tropical Mountainous Regions

Topographic correction is an important data preprocessing step when land cover classification and quantitative analysis of multispectral data are carried out in mountainous regions. Solar illumination effects may cause variations in reflectance of similar ground features, leading to a possible misclassification due to different topographic positions. Since nearly all surfaces exhibit a varying degree of anisotropy in bidirectional reflectance, the terrain normalization methodology should ideally be tailored to each vegetation type and account for the variability of land cover types within the research area. This study evaluated different topographic correction algorithms using statistical analysis and quantified their impact on classification accuracy. A pre-stratification of the dataset using continuous parameters was evaluated in order to model the degree of non-lambertian reflectance behaviour separately for the most prevalent vegetation types of the study area. The results suggest that the method used to extract semiempirical constants has a strong influence on correction results and classification accuracy. The derivation of pixel-specific Minnaert constants based on a stratification according to terrain slope and vegetation index thresholds improved classification accuracies while not requiring a priori knowledge.