This study presents a vegetation map of the Amazon River floodplain between the towns of Parintins (AM) and Almeirim (PA), based on Landsat-MSS scenes from 1975 to 1981. Digital processing involved the transformation of multispectral images into fraction-images of vegetation, soil and dark water (shadow), followed by the application of segmentation and region-classification techniques. The resulting map was organized four classes of land cover types: floodplain forest, non-forest floodplain vegetation, bare soil, and open water. Map accuracy was estimated from two types of ground data 1) sample points describing ground cover classes not subjected to major changes, such as permanent water bodies, and identifying indicators of the 30 year old vegetation type landscape (72 points); 2) interviews with community early residents for memory recovery of information on the vegetation cover existing in the 1970 (44 interviews). Altogether, 116 information points was collected along the study area. These points were used to calculate the Kappa Index for agreement between the four field-verified classes and the automatic classification, with value (0.78) indicates the good quality of the floodplain vegetation cover map. The region had 8650 km2 coverage of floodplain forest at the time of image acquisition.
remote sensing; automatic mapping; floodplain forest; Landsat-MSS
