INVESTIGATIONS INTO THE EFFECT OF TOPOGRAPHY ON CLASSIFICATION OF REMOTE SENSING DATA

Abstract

Remotely sensed imagery is an ideal data source for monitoring land cover in mountainous areas, as it minimizes the accessibility problems. However, information extraction from remotely sensed imagery in areas of rugged terrain is impaired due to the variations in spectral response. The variations are functions of the exitance angle and the surface cover characteristics. The illumination differences are often referred to as the "Topographic Effect". The topographic effect can be minimized considering the topographic information in some form or the other. This may be achieved by topographic normalization using Lambertian or non-Lambertian reflectance model, band ratioing and incorporating products generated from Digital Elevation Model (DEM) in the classification process. In the present study, the three approaches have been investigated to assess the topographic effect on data acquired over a typical hilly area. Classifications have been performed using the products generated as above. The most widely used Maximum Likelihood Classifier has been used. Slope, aspect and relative relief images have been generated from DEM. The incorporation of these images along with IRS LISS-II data, resulted in an improvement in the accuracy of classification. The ratio and NDVI images clearly showed the reduction in the shadow due to topographic effect in LISS-II data. However, the incorporation of these transformed images resulted in marginal improvement in the accuracy. The topographic normalization was also tried but this process could not be adequately applied on the LISS-II data considered, as the extent of atmospheric errors in the data was not known. Therefore this approach was not considered further. Nevertheless, the present study shows that the incorporation of topographic information in the classification process may improve the accuracy and thus more appropriate land cover classifications may be obtained.