Delineating sandhill communities: the use of advanced techniques to extract features from aerial photographs and satellite imagery

S. J. Harper and R. R. Sharitz

Savannah River Ecology Laboratory, University of Georgia

High-resolution satellite images and aerial photographs are now routinely acquired and used by diverse agencies and organizations, and these remotely-sensed data often form the foundation many natural resource layers within GIS databases. While the mapping of environmental and ecological features (e.g., those related to vegetation, land use, and disturbance) can provide valuable information to natural resource managers, maintaining up-to-date databases requires a major investment of time and labor. Historically, only highly-trained and experienced personnel could extract useful information from remotely-sensed imagery, which resulted in a bottleneck that prevented widespread utilization. Advanced software applications have been developed recently that provide users with ready access to powerful statistical techniques for extracting object-specific features from high-resolution panchromatic and multi-spectral imagery. For example, machine learning algorithms (e.g., neural networks, nearest neighbor, decision trees) allow the efficient extraction of user-defined features by utilizing spatial context in addition to spectral signatures. Similarly, hierarchical learning methods support improved image classification through iterative feedback provided by users. To demonstrate the utility of these approaches to forestry and resource management, an example is presented in which sandhills (xeric scrub communities that support sensitive flora and fauna) are extracted from surrounding habitats located along the Piedmont and Coastal Plain of the southeastern U.S. Results highlight the importance of federal lands in supporting this community throughout the region. Further development of feature extraction tools will allow up-to-date GIS data to be produced efficiently with reduced labor which, in turn, will help resource managers make effective decisions despite limited budgets and time constraints.

Keywords: sandhills, feature extraction, image analysis

(presentation)