Use of remote sensing for vegetation and landuse mapping in mountainous areas: The case of Central Nepal

Abstract

Among the specific difficulties encountered while mapping vegetation and landuse utilizing remote sensing data (LANDSAT M.S.S.) in mountainous areas like the Himalaya of Central Nepal, one has to deal with the different lighting conditions of the slopes at the moment of data collection. The original MSS data have not been corrected; they have been clustered and treated separately according to sun illumination. An illumination model elaborated from a topographical map has been worked out and utilized to identify 3 types of homogeneously lightedslopes. In order to treat separately these 3 types of slopes, we have utilized an automatic classification from the 4 MSS bands and a sample of representative ground-truths (Bayesien non parametric discrimination) and a fuzzy thresholding of the topographical model according to ground-truth data. We use a stepwise numerical process which allows to treat each landscape unit not properly identified at the previous step. The parameters utilized at each step as well as the order of their utilization are stored by the system. At each step, the system allows to store the 3 following sets: - The units to be recognized (pixels, groups of pixels) which have not been identified at the previous step. - The variables describing the units to be identified (MSS bands, illumination, topography, texture indexes) - The numerical clustering method called for (multivariable classification, thresholding of a function, image segmentation) The final output is a map of vegetation (forest types and densities) and land use. According to its conception, this system provides automatically a complete description of the algorithm having produced the map.