Autoria: ADAMI, M.; GOMES, A. R.; BELUZZO, A.; COELHO, A. dos S.; VALERIANO, D. de M.; RAMOS, F. de S.; NARVAES, I. da S.; BROWN, I. F.; OLIVEIRA, I. D. de; SANTOS, L. B.; MAURANO, L. E. P.; WATRIN, O. dos S.; GRAÇA, P. M. L. de A.

Resumo: PRODES is completing almost thirty years of uninterrupted monitoring of clear-cut deforestation over the Brazilian Amazon. Until now, no estimate of its mapping accuracy has been made. In this sense, this article brings a first approximation of mapping accuracy estimation of PRODES deforested areas, taking as example the state of Mato Grosso for the year 2014. For this, a random sampling panel was constructed, stratified with two strata, the deforestation of 2014 and the remaining forest. The sample size was calculated using the binomial function. In addition, a web platform was built to evaluate the points drawn by three independent evaluators. The global accuracy of the mapping of deforestation for the state of Mato Grosso, for the year 2014 was 94.5%, and may vary between 92.4% and 96.5%, in the evaluated scenario there was no class discordance to be found. Regarding the Forest class, the user accuracy was 90.5% and the producer's accuracy was 88.4%, this imbalance between user accuracy and producer accuracy indicates that there is a tendency for the forest class area to be underestimated for this mapping, in this year.

Ano de publicação: 2017

Tipo de publicação: Artigo em anais e proceedings

Unidade: Embrapa Amazônia Oriental

Palavras-chave: Acurácia, Desmatamento