DEVELOPMENT AND APPLICATION OF AUTOMATED CHANNEL EXTRACTION FROM LIDAR IN CHESAPEAKE BAY WATERSHEDS
Accurate stream maps are critical for a variety of applications, yet despite high-resolution terrain, objective extraction of accurate hydrography remains challenging. Existing approaches approximate channel–forming processes with thresholds of contributing area or detection methods involving either local curvature or topographic openness, followed by flow tracing. Process-based methods are easily implemented, but typically suffer from substantial error, especially where multiple channel-forming mechanisms operate. Direct detection relies upon quantitative thresholds to identify depressions and reduce omission, but generates substantial commission without subsequent filtering, especially in human-modified landscapes. We present a new method for automated channel extraction based on line-of-sight concepts from computer vision to classify discrete geomorphic features. This approach eliminates the need for topographic thresholds by interpreting surrounding patterns of relative elevation, integrating and analyzing terrain beyond local curvature to detect channel features. Unlike approaches that filter by downslope tracing to reduce commission, our approach maps broader river valleys as interpretive context, employing ancillary information to overcome common routing challenges (e.g., road crossings). We summarize case studies from throughout the Chesapeake Bay watershed to compare delineation approaches across different physiographic and land-use contexts that demonstrate the efficiency of our approach.
Matthew Baker (Primary Presenter/Author), University of Maryland Baltimore County, email@example.com;
David Saavedra (Co-Presenter/Co-Author), The Chesapeake Concervancy, firstname.lastname@example.org;