Bridge Detection In Grid Terrains And Improved Drainage Enforcement
Proceedings Of The 18th SIGSPATIAL International Conference On Advances In Geographic Information Systems
Bare Earth gridded digital elevation models (DEMs) are often used to extract hydrologic features such as rivers and watersheds. DEMs must be conditioned by removing spurious sinks (or depressions) which impede water flow in the model, but are not true hydrologic barriers. This conditioning process is designed to enforce proper drainage and connect real hydrologic networks (rivers) that would otherwise be disconnected in the unconditioned DEM. Primary means of conditioning DEMs include filling sinks and cutting barriers. The availability of high resolution DEMs derived from lidar introduces new forms of false hydrologic barriers, primarily bridges. While attempts are made to automatically remove trees, buildings and bridges from bare Earth terrains, in practice many bridges remain in the final "cleaned" DEM. We present a supervised machine learning approach for detecting bridges and other hydrologic barriers in DEMs. Furthermore, we locally apply a simple cutting algorithm to condition DEMs in areas tagged as barriers by the machine learning step. After cutting, we use a filling technique to remove any remaining spurious depressions. Experimental results indicate that our approach accurately identifies a variety of bridge and bridge-like features. Our final conditioned DEM both modifies fewer grid cells and modifies cells to a lesser extent than other traditional conditioning approaches. The result is more realistic hydrologic models on high resolution terrains.
ACM SIGSPATIAL International Conference On Advances In Geographic Information Systems 2010
November 2-5, 2010
San Jose, CA
Ryan A. Carlson , '11 and Andrew Danner.
"Bridge Detection In Grid Terrains And Improved Drainage Enforcement".
Proceedings Of The 18th SIGSPATIAL International Conference On Advances In Geographic Information Systems.