Hyperspectral imagery as a data source for floristic classification in southeastern Labrador

Predictive vegetation mapping utilizing digital data is the only practicable means by which spatial data can be collected at reasonable cost over relatively large and inaccessible areas (Markton 1995; Ustin et al. 1994). Hyperspectral data collected over a boreal landscape in southeastern Labrador using the CASI sensor were classified into floristic theme classes.; The study landscape lay at the western end of Lake Melville between Goose Bay and the northern slopes of the Mealy Mountains and is dominated by plateau bogs, ribbed fens and Black Spruce-Lichen forests. Three study areas representing the dominant floristic associations were classified. Training data were collected in each study area using a stratified random method and the phytosociological cover abundance ratings of the Braun-Blanquet method (Mueller-Dombois and Ellenberg 1974). Training data were classified into floristic theme classes based on established floristic classification schemes (Wells 1996; Treter 1995; Foster 1984; Foster 1983). Five peatland, four forest and two freshwater classes were recognized.; The usefulness of visual and computer-aided methods of image analysis for the optimization of image classification is described. Image classification was carried out using the Maximum Likelihood classification algorithm with a Null class. An overall classification accuracy of >85% was achieved in each study area. Class accuracies of >85% were achieved for all but two classes (plateau bog study area). Accuracies varied with end class and the number and combination of classes in the study area. In two study areas, two of the proposed thematic classes had to be modified to obtain a classification accuracy to >85%. Sensor spatial and spectral resolution restricted resolution at which floristic theme classes could be labeled.; Classifications encompassing parts of the study landscape are discussed and recommendations are made for future predictive vegetation studies in this landscape.