Close Window

A Landsat TM scene from July, 1991 was analyzed for an area of the Amazon estuary in: Ponta de Pedras, Marajo Island, Brazil. Distinctive spectral signatures were determined for 14 land-cover classes, including upland and floodplain forest, three stages of secondary succession, palm forest, mangrove, pasture, and three types of savanna. Image classification (unsupervised and supervised using a hybrid maximum-likelihood/texture algorithm) of the study area was conducted using the 14 class spectral statistics informed by 1992 vegetation inventories and field studies documenting historical land use. The use of field-based information supportive of classification resulted in individual rest field class results which ranged from 81 to 100 percent individual class accuracy. Historically, attainment of good accuracy for many of these classes using satellite data has been difficult, but this research indicates suitable accuracy can be obtained using TM data when carefully integrated with detailed ground surveys. Elements of the classification were focused on addressing the difficult problem of identifying the conversion of \'\'natural\'\' to \'\'managed\'\' floodplain forest. The combination of feature classification using computer-analyzed TM data in conjunction with detailed ground measurements/surveys permitted identification of subtle changes in natural forest that was associated with conversion to managed floodplain forest

Close Window