Explicitly linking field- and satellite- derived measurements for improved vegetation quantification and disturbance detection



Journal Title

Journal ISSN

Volume Title



Arid and semi-arid ecosystems have been recognized as critical in supporting over one-third of the world's populations, notably those more dependent on the natural resource base for their livelihoods. These systems, and especially savannas within them, are highly vulnerable to predicted fluctuations in climatic change, disturbances, and management regimes. This research posits these areas in a social-ecological system (SES) framework that encompasses human, governance, and recourse units. A challenge in both SES and CHANS (coupled human and natural systems) research is how to explicitly and empirically link the social and the ecological, and further how to extrapolate from sets of case studies to the greater region, supra-system, or SES / CHANS theory and practice. This work leverages Landsat and IKONOS imagery as well as field-based vegetation sampling (structure and species) through the use of IDL (interactive data language) visualizations, both pixel- and object-based classifications, and CART (classification and regression tree) analysis. The longer term goal of this work is to produce a protocol and classification scheme modified from the 1976 Anderson scheme to include both structure and disturbance explicitly in processing, mapping, monitoring, and management. In creating SVCs (Structural Vegetation Categories) built from field data there is strong potential for extracting 3-D data from 2-D imagery once the protocol produces robust results with high enough accuracies. As hypothesized, the object-based classifications produced higher overall accuracy (70.83%), though the pixel-based classification performed better in the detection of woodlands (90.91%). Given the spatial scales of the imagery as compared to the size of the field plots and transect spacing, it is important to remember that when extrapolating to other areas a critical part of spatial scale is extent (not just grain). That is, the inherent clumping of trees versus shrubs may be driving the better performance of pixel-based for woodlands but not so for shrublands. Sensitivity to placement of plots and especially plot sizes across future sites will help explore this question and move SES research into a realm whereby remote sensing and vegetation sampling can provide improved empirical linkages among the subsystems and their feedbacks.