A new procedure for forestry database updating with gis and remote sensing

Abstract

The aim of this study was to develop an automated, simple and flexible procedure for updating raster-based forestry database. Four modules compose the procedure: (1) location of changed sites, (2) quantification of changed area, (3) identification of the new land cover, and (4) database updating. Firstly, a difference image is decomposed with wavelet transforms in order to extract changed sites. Secondly, segmentation is performed on the difference image. Thirdly, each changed pixel or each segmented region is assigned to the land cover class with the highest probability of membership. Then, the output is used to update the GIS layer where changes took place. This procedure was less sensitive to geometric and radiometric misregistration, and less dependent on ground truth, when compared with post classification comparison and direct multidate classification.