Linear Mixture Model Applied to the Land-Cover Classification in an Alluvial Plain Using Landsat TM Data

Abstract

The accurate delineation of the different pixel information is required for many remote sensing applications. However, the complexity of land cover makes the classification process difficult when using traditional methods, especially in areas where the heterogeneity is pronounced. In this paper, a Linear Mixture Model (LMM) approach is applied to classify the land cover in an alluvial tract using Thematic Mapper (TM) imagery around Talakad, parts of Mysore, Mandya and Chamarajanagar districts, Karnataka, India, in respect of five classes, viz:, sand, sparse vegetation, settlements, vegetation, and water. Fraction images of these classes were generated from Landsat TM image by un-mixing the image using LMM. This study indicates that the LMM approach is a promising method for distinguishing successional land cover in alluvial plain, where thick vegetation is noticed, using TM data. It gave better classification accuracy than traditional techniques did. The outputs of fraction images showed the high capability of LMM to extract many features. This was not possible with maximum likelihood classification method in spite of an overall accuracy of 98.83%, which was particularly not so efficient in extracting the vegetation and water bodies. The land cover units contained in the area of this alluvial tract were not picked up properly in the maximum likelihood classification.