期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2006
卷号：XXXVI Part 7
出版社：Copernicus Publications
摘要：Among the alluvial communities of the cerrado (Brazilian savanna), palm swamps represents an important and fragile ecosystem that develop under very specific geomorphologic conditions associated with a river system and a mostly .at topography. In the cerrado, the term veredas describes palm swamps that are mainly associated with Mauritia .exuosa. Veredas are in fact vegetation complexes composed of a succession of vegetation forms that range from grassland to forest. Veredas are protected by law in Brazil but their extent and condition are mainly unknown. As a humid environment in a semiarid context and being associated with a very specific topographic and hydrographic context, veredas are easily detected on remotely sensed data but are also often mistaken for riparian forests when using optical data. Being very sensitive to humidity, synthetic aperture radar data from RADARSAT offers good potential for delineating veredas. In this article, a methodology is described for the delineation of veredas using a segmentation approach limited to a buffer based on the hydrographic network. The segmentation technique is based on a "modified adaptive Markov random field" (MAMSEG) algorithm that solely uses the SAR backscattering data. Segmentation is performed exclusively within the buffer area and only segments connected to the hydrographic network are kept. Preliminary results suggest a correspondance of about 70% between the segments extracted and the actual localization of the veredas. Additionally, it appears that filtered RADARSAT data offers some potential for separating vegetation forms within the veredas ecosystem
