Digital Elevation Model Resolution. The DEM for this project was provided by the LIDAR vendor, Watershed Sciences Inc. (now Quantum Spatial), at a three-foot resolution. The process for creating a DEM from LIDAR data can be complicated, and it is commonly assumed that a DEM provided by the vendor will have a higher accuracy. The first significant decision to be made when developing stream channel locations is the resolution of the DEM. The maximum resolution of most DEMs is around one meter, and is limited by the density of returns in the LIDAR acquisition. Higher return density allows for higher DEM resolution. A common return density for many acquisitions is eight returns per square meter. Producing a DEM with a resolution higher than one meter at this return density would result in many areas having few or no returns, resulting in greater inaccuracy due to cells in the model being interpolated. It is often assumed that higher resolution for a DEM is better, but very high resolution ground models introduce complications. It is not possible to know if the texture in a high resolution DEM is actually the ground, or if it is low vegetation, stumps and logs, or data processing artifacts, etc. Additionally, where streams cross roads in high resolution ground models, the roads often block the streams and redirect flow down the roads. It is possible, but again unknowable from the LIDAR alone, that culverts or cross drains are present, which would allow streams to flow under the road. These problems seem to occur less frequently in lower resolution DEMs, which may argue for their use (Figure 2). Figure 2. Comparison of DEMON flow accumulation results using 3 ft. high resolution (left) and 30 ft. low resolution (right) DEMs for the same location. The yellow circles highlight differences in flow accumulation behavior. For the purposes of testing, the vendor provided DEM was down-sampled to four lower resolutions: six, nine, fifteen, and thirty feet. All resolutions produced different stream channel locations when all other processing was the same. Final processing was performed at the three-foot resolution as that was deemed the most visually accurate model.
