PARENT SESSION
Symposium 9: Spatial heterogeneity in urban ecosystems: Integration and scaling in coupled natural-human systems
Organized by: ML Cadenasso, M Grove, and STA Pickett
Tuesday, August 9, 1:30 PM - 5:00 PM, Meeting Room 517 C, Level 5, Palais des congrès de Montréal
Automating the measurement of spatial landscape heterogeneity using object-oriented classification.
O'Neil-Dunne, Jarlath*,1, Troy, Austin1, Grove, Morgan2, 1 University of Vermont, Burlington, VT, 054052 Northeastern Research Station, South Burlington, VT, USA
ABSTRACT- The proliferation of high-resolution remotely sensed imagery provides an opportunity to characterize fine-scale spatial patterns in urban ecosystems. Humans are adept at manually identifying these patterns from imagery because of their ability to incorporate contextual information. Automated image classification techniques can be more cost-efficient and consistent when compared with manual interpretation, but they are not as well suited to the highly heterogeneous landscapes of urban ecosystems. Because they operate by assigning each image pixel to a unique land cover class based on spectral reflectance, automated techniques will fail to discriminate functionally different features whose pixels have similar reflectance. As part of the Baltimore Ecosystem Study, we attempt to classify high-resolution multi-spectral aerial imagery from 1999 and 2004 so as to characterize the spatial distribution and typology of patches of built and natural capital in the Gwynn's Falls watershed. We do so using object-oriented image classification, a novel automated technique that incorporates spectral, spatial and contextual information into a rule-based classification process. Object oriented classification first segments images into discrete objects and then allows the user to hierarchically classify those objects based on their form and function. Classification rules can be based on reflectance, shape metrics, and spatial context, including coincidence with features in other thematic layers. We explore the extent to which this could be used to supplement manual methods for characterizing urban patches based on their mix of human and natural features. We also discuss its limits in replicating manual image interpretation techniques.
Key words: GIS, Imagery, Spatial Analysis, Urban Ecosystems