Pasture mapping by classification of Landsat TM images. Analysis of the spectral behaviour of the pasture class in a real medium-scale environment: The case of the Piracicaba Catchment (12 400 km 2, Brazil)

This is the authors' abstract. We don't add key points for this paper.

Attempts to map vegetation types, especially pasture, from satellite sensor data in tropical and sub-tropical regions very often have limited success. This study analyses the accuracy of two classifications of Landsat Thematic Mapper (TM), with the aim of distinguishing the pastures from other vegetation classes in a meso-scale basin (12 400 km 2, Piracicaba basin, Brazil). The initial classification is based on non-supervised clustering of the images. The delimited classes are interpreted and merged by comparison with standard spectra from NASA. The second classification is a parallelepiped partition based on the merged clusters issued from the first one. The two classifications are compared by validation with 287 random field observations selected within the whole catchment. The results are discussed, analysing the spectral behaviour variability of the pasture class.