The objective of this work was to evaluate a new methodology to map soybean crop area in Mato Grosso State, Brazil, using Modis imagery and different image classification approaches. Single-day and 16-day images were used. The single-day images were classified using the Isoseg algorithm. Two series of 16-day composite images, covering the full and the half soybean crop cycles, were transformed using principal component analysis (PCA) prior to the classification. A reference data set, achieved by visual interpretation of TM/Landsat-5 images, was used to evaluate the accuracy of the classifications. The best results were reached using the image classification of soybean full cycle, transformed by PCA: overall accuracy of 0.83, and Kappa of 0.63. The best single-day classification showed an overall index of 0.80, and 0.55 Kappa. PCA applied to the images of the full cycle allowed for the mapping of soybean crop areas with better accuracy indices than those obtained by the single-day classification.
Glycine max; image classifications; principal components; agricultural statistics; digital processing; remote sensing
