The objective of this work was to evaluate the performance of the enhanced vegetation index (EVI) and the normalized difference vegetation index (NDVI) - both from the moderate resolution imaging spectroradiometer (Modis) sensor - to discriminate soybean cultivated areas from sugarcane, pasture, cerrado, and forest ones in the state of Mato Grosso, Brazil. Images acquired during two periods were used: off-season and maximum soybean crop development. For each analyzed class, 31 samples were selected from reference maps, and the differences in the values of each soybean vegetation index were evaluated against the other classes using the Tukey‑Kramer test. Afterwards, the differences between the vegetation indices were assessed using the Wilcoxon paired test. NDVI performed best in discriminating soybean areas during the off-season period, particularly when using images acquired from day of year (DOY) 161 to 273, whereas EVI performed best during maximum crop development, particularly when using images from DOY 353 to 33. Therefore, best classification results for soybean in the state of Mato Grosso can be achieved by coupling Modis NDVI images acquired during off-season period and EVI images acquired during the maximum crop development period.
multi‑temporal image classification; Modis data; crop area estimates; satellite images; remote sensing
