A New Algorithm for Extracting Drainage Networks from Gridded DEMs

Abstract

Drainage networks are important abstract features in terrain modeling and play functional roles in hydrological, geomorphologic and biological analyzing models. Watershed indices based on drainage networks are crucial in flood predicting models. Huge efforts have been made on automatic extraction of drainage networks. However, there are no effective methods to extract threshold-insensitive and noise-free drainage networks from gridded elevation data. This chapter proposes an algorithm to extract complete and reasonable drainage networks from gridded digital elevation models (DEMs) by integrating global and local methodologies. First, the flow routing algorithm is employed to derive primary drainage segments, which includes depression removal, flow direction computation, flow accumulation and threshold value setting. Threshold values, which are used to filter out meaningful drainage segments from accumulation information, are set based on experts’ experience and terrain types. The value heavily influences lengths of individual drainage segments and geometric forms of extracted drainage networks and watersheds, whereas the results are sensitive to threshold values and can introduce uncertainties to further analysis. In order to incorporate the missing drainage segments filtered out by given threshold values, the second step utilizes a moving-kernel method to flag morphometrically characteristic points, which are then integrated into the initial result by downward and upward connecting processes based on flow direction information produced in the first step. Both above mentioned two steps introduce congested drainage segments, which are taken as noises for constructing geometrically clear and topologically consistent drainage networks. In the third step, noisy drainage segments and parallel drainage segments are classified into different types and handled by inductive analysis and rule-based treatment. The topological consistency of drainage networks is maintained in every step. The final results include single-pixel-width drainage networks and correspondent watershed sub-divisions. Two datasets covering different geomorphological areas are used to test our new algorithm. The Strahler ordering scheme, length and structure of extracted drainage networks are analyzed. The quantitative analysis demonstrates that extracted drainage networks based on the new algorithm are insensitive to threshold values. Visual inspection by overlaying extracted drainage networks with contour lines shows that drainage segments are consistent with the contour curvature.