A robust channel network extraction method combining discrete curve evolution and the skeleton construction technique

The automatic mapping of drainage networks from terrain representation has been an interesting topic in hydrological and geomorphological modeling. However, the existing methods often suffer from high sensitivity to terrain noise or lose significant stream branches and accurate channel paths. In this paper, we propose a contour-based framework in drainage network extraction. The proposed framework incorporates discrete curve evolution (DCE) to eliminate the noise influence by dynamically segmenting the contour lines (CLs) into valley bends, and to detect the valley feature points. The skeleton construction technique is then applied to distill more accurate channel paths in complex terrain. Finally, a linking step is undertaken to generate the channel network. The proposed method was tested on a series of elevation datasets, with varied resolution, region size, and local relief. The experiments verified that the proposed method can achieve highly accurate channel networks and is robust, even in regions with high-contrast relief, and/or in cases with significant terrain noise and irregularities.