Classification par arbre de decision des milieux humides de la Vallee du Saint-Laurent a partir d'images RADARSAT et Landsat TM

摘要

The Canadian Wildlife Service, Quebec region, has initiated a project oriented toward the production of a Wetland atlas covering the agricultural landscapes in the St-Lawrence Valley, Quebec. This cartography is based on the combined use of the 1999 spring acquired RADARSAT data and Landsat TM images in complement with digital topographic and hydrologic maps. A classification method that integrates a series of RADARSAT ortho-images and their derived products (homogeneity and contrast local textural indices), the Landsat TM 3,4,5 decorrelated images and the digital elevation model has been developed. Data from several field campaigns on wetlands have been introduced in the tree based algorithm. Natural peat bogs, emergent marshlands, flooded hardwood forests, swamps and humid prairies have been considered. The decision tree classification approach is an alternative to conventional image classification techniques commonly used in remote sensing. Its ability to get the better of radar and optical imagery has guided our choice. In fact, some classes can be identified solely on optical images (type of crops, forest types) while others are best delineated on radar images ( both flooded agricultural fields and hardwood forests). The classification results, reported on a vector format, will constitute the Wetland Conservation Atlas designed to provide guidance to each regional county municipality of southern Quebec of the St-Lawrence Valley ecoregion.