Conserving arid habitats in many parts of the world requires effective management at scales commensurate with threats impacting them. Implementing such a landscape-scale management paradigm has been impeded by difficulty mapping from field data and remotely-sensed images all but the coarsest patterns of plant communities resulting from processes interacting at different scales. A relatively new remote sensing method, object-based image analysis (OBIA), has been proposed as a method for extracting multiple scales of patterns inherent in imagery. OBIA segments an image into polygons, or objects, based on similarity of neighboring pixels, and the objects, not pixels, form the basic unit for analyzing the image. A scale parameter controls allowable heterogeneity of pixel values within an object. By varying this parameter, collections of different-sized objects can be created to represent different information scales. Little research has been done, however, on how to pick scale levels from OBIA that are ecologically meaningful. We developed a method for using OBIA to identify ecologically-relevant scales in arid ecosystems by comparing within-object pixel variance to variance between objects. We created synthetic landscape images from realizations of neutral landscape models varying contagion to generate multi-scale patterns, and assembled hierarchies of objects representing a wide range of scales.
Results/Conclusions
Plotting the ratio of mean within-object variance to variance of object means against scale level showed regions of rapid change in variance and segments where variance changed little as scale increased. Image objects corresponding to scale-invariant regions closely matched patterns in the synthetic landscapes. Areas of rapid variance change corresponded to transitions between landscape patterns. We found similar results using this method to identify scales present in ASTER and IKONOS satellite imagery for a portion of the Craters of the
