Classification of land cover in the Amazon based on satellite imagery and characterization of the classes according to ground surface

Abstract

In this work was made the classification of land cover in an area of Caxiuanã National Forest, Pará State, based on satellite imagery using a supervised approach by regions. It was started with a fit between the means and variances of the bands, followed by application of median filter. Subsequently, it was proceeded to image segmentation by region growing. The next step was to define the classes of coverage and the selection of training samples. Finally, the image classification was done. It was extracted information from elevation and slope to the generated classes. As a result, it was found that in the studied area predominates the Upland Forest - Heterogeneous Canopy class (67.6%), and that together with the Upland Forest - Homogeneous Canopy class (8.3%), occupy the higher areas (?46m). The areas of Upland Forest - Hill (8.0%) and Secondary Vegetation (1.6%) classes are in intermediate altimetric position (?42 m). Finally, the Floodplain Forest class (12.0%) occupies a lower altimetric position (?39m). About 80% of the area is on gently undulating relief. The Upland Forest - Canopy Heterogeneous and Upland Forest - Homogeneous Canopy classes differ in slope terms but not in elevation terms. It is believed that phytosociological studies (structure and/or floristic) showing differences between these classes.