Land cover classification based on multi-date JERS-1 imagery as a basis for deforestation detection

Abstract

Deforestation detection is a key issue on tropical environment monitoring. It has been done in Brazil based mainly on visual interpretation of optical images. Cloud coverage, however, is an impediment to have reliable estimates over several Amazonian areas. L-band SAR data is a promising information source to monitor those areas. One possible approach, used in this work, is to analyse land use/cover change between successive dates to spot deforestation, being each consecutive land cover map obtained through JERS-1 land use/cover classification. Initially, each JERS-1 image is speckle filtered and a standard segmentation routine is then applied to each filtered channel. The result is an image in which each segment is represented by the average backscatter level within that segment. After segmentation stage, the segments are classified into four land use/cover classes of interest: pasture+bare soil, dirty pasture, secondary and primary forest, producing a land cover map for each year. Analysing the changes on the 1996 land cover maps related to the 1995 map, it was possible to point out areas of deforestation and other change classes. An assessment is done over an well known area near the Tapajos National Forest (Flona), in Para State, Brazil. The land use/cover maps and the change map are compared to reference areas defined by visual interpretation.