The main goal of this study was to evaluate the information produced from Landsat/TM5 images using Principal Component Analysis (PCA) and Illumination Factor built from Digital Elevation Model from ASTER images for coffee areas mapping in complex relief. Three Landsat images were used to monitor the crop cycle. The Principal Component Analysis was applied to the Landsat images and the two first components were chosen, responsible for 94% of the initial information, and used as a sample set for the supervised classification of those images. That classification was compared with a conventional supervised classification (sampled from Landsat reflectance images) and multitemporal conventional supervised classification (using the three images). The accuracies of the classifications were calculated by Kappa index of agreement and Global Accuracy, using a coffee mask as reference. The results have shown that PCA was very efficient in illumination class definition as well as in sample choice, despite the samples had not represented the area classified. Due to that, the accuracy has increased, specially the one considering all the pixels classified as coffee in each image using PCA samples, demonstrating the importance of the multitemporal aspect.
ASTER; atmospheric correction; radiometric normalization; digital image processing
