M. urundeuva (Fr. All.) is a tree currently listed as endangered species of Brazilian Flora since 1992. However, it manifests monodominance in some regions of Minas Gerais state, especially in middle Rio Doce. The objective of this work was to compare supervised methods of classification of Rapideye images for forest fragments monodominated by Myracrodruon urundeuva mapping, in Tumiritinga, MG. The maximum likelihood (Maxlike) and the artificial neural networks (ANN) classification were assessed. Nineteen combinations involving different bands, the principal components and the normalized difference vegetation index (NDVI) were tested for classification of Rapideye image. The neural network training was conducted by varying the learning rate, the number of interactions and the number of neurons in the hidden layer. The thematic maps were evaluated by Kappa and conditional Kappa index for "aroeira" soil use class and by the analysis of the Confusion Matrix. The best performance was achieved by the Maxever algorithm using all bands of Rapideye image, getting a Kappa index of 80 and conditional Kappa index of 90. The thematic map presented user's accuracy equal to 93% and producer's accuracy equal to 90%. The greatest errors of classification for "aroeira monominant" class were related to the "native forest" and "managed pasture" classes. The thematic map produced shows that 22% of the town is under occupation of M. urundeuva in monodominance and this specie is not endangered in Tumiritinga, Minas Gerais.
Myracrodruon urundeuva; Mono-dominance; Supervised classification of images
