| Popis: |
Land cover products derived from remotely sensed data are now widely used in forest ecology and management as environmental layers for predictive modeling of wildlife distribution and abundance and as inputs to the design of bio-monitoring programmes. In the boreal forest, there is an urgent need to quantify the effects of industrial activity on wildlife habitat (e.g., songbirds and woodland caribou), in part to meet Environment Canada’s mandates under the Species at Risk Act. Remote sensing data is the most likely spatially extensive data on habitat that can potentially meet such mandates. The purpose of this study was to evaluate the relative efficacy of three recently developed land cover products to describe forest habitat, by systematic comparison against a common vegetation data layer at resolutions of 250m and 1km. Specifically, we evaluated the: GLC 2000 North American Land Cover (NALC) 1km, 250m MODIS 2005 Land Cover Classification (LCC05) and the 25m Earth Observation for Sustainable Development of Forests (EOSD LC 2000) products. As ground truth data, we used an extensive suite of georeferenced vegetation relevé data, pre-classified according to a standardized taxonomy of plant communities (Canadian National Vegetation Classification). The study provisionally corrected ground truth data for temporal changes in land cover due to fire and forest harvesting over the sampling period. Relations between the classified relevé data and the land cover products are reported by means of user, producer and overall accuracy of a six common class legend. Overall the MODIS 2005 LC product showed the most consistency in agreement with independent reference data. The highest accuracy for all LC products was achieved with open to closed coniferous forest that had accuracies as high as 87.24 +/- 4.28% |
