Accuracy of land use/land cover in Tres Pontas municipality, Minas Gerais State, Brazil

One of the most relevant questions for the use of information obtained from remote sensing systems, refers to its capacity to represent adequately those data obtained during field survey. The exactness of a map is based on a quantitative measurement of a thematic map that has been correctly classified. Under this assumption, the objective of this study is to analyze four statistics: Kappa, Global, Tau and Sampling by Significance Level. Therefore the authors used a TM-Landsat composite (RGB channels 4,5,3) that was manually analyzed. Samples for the evaluation of ground truth were obtained using a random procedure. Through SAL it was possible to determine the confidence of a land use/land cover map. Sampling by significance level indicated the need to express in advance statistically the truth of the reference data (the estimated trustworthy of the map), instead of using an index of agreement to estimate the exactitude. Referring to the estimators of agreement G, K and T the authors concluded that K did express best the estimate for accuracy of a thematic map obtained, when comparing to the other estimators