Global land cover mapping is becoming increasingly important to earth system science studies of climate change, carbon dynamics and biodiversity. However, the discipline of global land cover mapping is still a nascent one. This study involves demonstrating improvements to global land cover methodologies in three topic areas: algorithm choice, thematic content, and validation. For these studies, the tested algorithm, decision trees, was shown to be robust, easily implemented and easily interpretable. These features are displayed in the development of the University of Maryland 1km global land cover product and in a fully automated procedure used in creating a global 1km continuous field of percent tree cover. This second global product represents an advance in global land cover mapping as a map of proportional cover offers several advantages in terms of thematic content over traditional discrete classifications. Percent cover maps capture landscape heterogeneity by better depicting mixed pixels. As such, they have greater potential to be used in change detection studies. They also allow users to develop, through thresholding, their own classification systems. The final part of this study involved demonstrating the use of very high resolution satellite data with limited field work to create a calibration/validation data set for global land cover mapping. This exercise was focused on percent tree cover estimates, but can be translated to other vegetation cover types. In this study, coarse resolution estimates of percent tree crown cover were related to crown delineations on the ground and in very high resolution satellite imagery. Results show the feasibility of directly relating coarse resolution cells to the in situ variables of interest and point the way to a possible multi-resolution approach to global land cover validation.
展开▼
