| Forest inventory is the fundamental base information for most decision-making processes in today's forest management planning. Recently in Ontario, with increasing industrial involvement, new environmental and multiple use issues, and rapidly developing technology, the requirement and opportunity for investigation into new inventory methods has increased. The method developed in this thesis focuses on the inventory requirements for large scale, strategic-level forest management for the boreal forest region. With recent improvement in satellite sensors and computer tools, the process of acquiring the imagery and analyzing the information has become significantly cheaper and faster. A multisource approach is used in this project to improve upon current forest classification attempts using satellite imagery. By merging the superior multispectral properties of Landsat 7 ETM+ (30 m multispectral) with the spatially detailed IRS-1 D panchromatic (5 m) imagery, an attempt is made to derive a species-level classification scheme. Image data merging techniques are explored and the utilization of image segmentation procedures is evaluated. Principle component substitution is used to integrate the imagery, and a nearest neighbour algorithm is used in an object-based classification system. Area-based accuracy assessment is used to test the success of the methods with reference derived from interpreted aerial photography. Accuracy assessments show satisfactory agreement between the thematic product and reference data, with overall accuracies reaching 72%. Pure species groups such as black spruce, jack pine and trembling aspen exhibited producer's accuracies of 90%, 83%, and 87%, respectively, with user's accuracies as high as 73%, 75%, and 61% respectively. |
