Stream Fusion between Consecutive Watersheds

Keywords: stream extraction, DEM, data fusion
Abstract Type: Poster Abstract

Abstract

Extracting streams from high-resolution DEMs is computation and memory-intensive. Over larger areas, some algorithms employ a divide-and-conquer strategy. For example, in the Chesapeake Bay Watershed high-resolution stream mapping project, basin terrain is divided into computational units. Each computational unit includes a twelve-digit USGS HUC (hydrologic unit code) and a 100-meter buffer to reduce boundary effects and to create overlap for stream fusion. Despite the buffer overlap, mapped streams in consecutive watersheds may not connect cleanly or consistently, especially in flat terrain, due to discrepancies in flow direction continuity. Here, we summarize stream fusion problems encountered in the project and provide feasible automated solutions.

Based on our observations, two scenarios are possible during stream network fusion between consecutive watersheds: 1) upstream and downstream channels have intersections but are not connected cleanly; and 2) upstream and downstream networks have no intersection and a connection must be built.

We propose an automated approach that recognizes different scenarios by analyzing spatial relations and provides connection solutions. In the first scenario, we distinguish mapped streams as valid and redundant channel classes based on their intersection and spatial relations. Then, we delete redundant streams. In the second scenario, we use the A* algorithm to connect the upstream and downstream networks based on terrain and relative proximity. After successful stream fusion, hierarchical hydrologic relations and stream ordering between consecutive watersheds are constructed, redundant streams are deleted, and missing connections are built. Through this process, streams extracted from small watershed segments are fused together rapidly and successfully.