Effets de la résolution spatiale et des méthodes de classification d'images satellites pour améliorer la carte forestière étude de cas près de la rivière Watopeka
For many years now, professionals have experimented several approaches to extract forest characteristics with various level of automated algorithms applied to digital images. Currently, the main source of information for forest harvest planning is extracted from forestry maps and samples plots from forest inventory programs. Digital image processing can be a complement to the current operational approaches. With the wide variety of available images on the market, it is relevant to define the optimal spatial resolution for forest planning. For fine image resolutions, processing can become complex and hard to integrate in forestry practices. For coarse image resolutions, information can be lost resulting in a loss of reliability for the product. The current project is based on the premise that optimal spatial resolution is at the approximate size of typical tree crown. The objective was to identify infra-polygonal areas from satellite image (raster) processing in a forest map (vector) for sugar maple stands, clear areas and pure conifer patches. Two methods were used in this project for the image processing: pixel and object-based (segmentation) classifications. The latter uses fuzzy logic algorithms. Overall, 4 classifications were produced on images with spatial resolution ranging from 1 to 30 meters with IKONOS-2, SPOT-5 and LANDSAT-7 satellite images. The best results were those from 10 meters images which is near from an individual tree crown size. The integration of statistical information on chosen parameters in the forest map enabled us to add useful descriptive information in each polygon."--résumé abrégé par UMI.