ABSTRACT
Land-cover information is essential for planning and studying the effects of changes from natural to disturbed landscapes. This study aimed at studying the dynamics of land cover from 1989 to 2011 in the Marombas River basin using a Decision Tree (DT) algorithm. Landsat-5 spectral bands, vegetation indices, and terrain attributes from elevation models were used as attributes for classification in the years of 1989, 1991, 1993, 1997, 2001, 2004, and 2011. DT classifier quality was assessed by a set of 500 independent random points, generated for each year; which allowed calculating Kappa index parameters and overall accuracy from the confusion matrices. The DT algorithm achieved a mean Kappa index of about 83% and a mean global accuracy of 86%. Therefore, it can be stated an excellent classification, from which we can securely associate anthropogenic influence with land-cover dynamics in this basin. We also observed an increase of agricultural and silvicultural activities at the expense of more natural land covers. Adding to that, the results showed that rapid fragmentation has occurred in the natural mixed ombrophilous forest along this interval of 22 years.
remote sensing; supervised classification; land use; data mining