Hydrographic feature extraction

>> Center of Excellence for Geospatial Information Science (CEGIS) | Science | Knowledge extraction | Feature  extraction

Machine learning techniques to automatically extract road and drainage valley intersections 

Semi-automated workflows to extract drainage lines from elevation data using hydrologic conditioning, flow-direction, and flow-accumulation modeling have been available for several years. 

However, some issues hinder these methods when implemented on high-resolution digital elevation models (DEMs) as collected through 3DEP lidar point-cloud data.

For instance, in a high-resolution DEM, roads act as barriers that alter the natural drainage patterns and impact extracted drainage lines. To overcome this problem, breaches (artificial lowering of elevation) can be added to the DEM at culvert and bridge locations, which is part of a process called hydro-conditioning. 

Current research involves development of machine learning techniques to automatically extract road and drainage valley intersections from DEM and other data to assist with hydro-conditioning of DEM data. Other work focuses on validation of extracted drainage lines through 3-dimensional analysis of drainage valleys and riparian zones using lidar-point cloud data.

The above image shows 1-m resolution elevation data for a section of the Rocky Brook watershed in New Jersey. Also shown are drainage lines derived from the DEM, which includes some obvious places where highways have been manually breached to allow flow modeling over the natural surface. The red point symbols indicate locations where hydrographic features were field verified.