Lidar DEM error analyses and topographic depression identification in a hummocky landscape in the prairie region of Canada

Topographic depressions are abundant in topographically complex landscapes. A common practice with earlier, low resolution Digital Elevation Models (DEMs) was to remove all depressions to ensure that water flowed continuously to the edge of the DEM domain. The assumption was that most depressions were created due to errors in the DEMs. This practice is no longer justified with the increasing availability of high accuracy DEMs. However, very few studies have addressed how DEM processing options such as smoothing and coarsening and setting area and depth thresholds can affect depression identification.In this study, a site located in the Prairie Region of Canada was examined. The site is a hummocky glaciated landscape with many in-field wetlands. Lidar topographic data were collected and were used to generate a 1 m by 1 m square-grid DEM. Detailed error analyses of the lidar DEM were conducted. A set of DEMs were generated after different degrees of smoothing and coarsening. FlowMapR, an established terrain analysis tool, was used to identify depressions in each DEM with various user-defined area and depth thresholds. The results were validated against a field wetland survey.We determined that the problems associated with depression identification using a lidar DEM are two-fold. On one hand, artefactual depressions created due to DEM errors need to be eliminated, for which the raw lidar DEM need to be smoothed. On the other hand, it is also desirable to remove those topographic depressions that do not function as closed basins at the spatial or temporal scale of the processes of interest. Setting area and depth thresholds appeared to be the preferred choice for this. We suggested using the un-autocorrelated lidar DEM error as the criterion for DEM smoothing and considering depression connections in the selection of area and depth thresholds. Using lidar data on a hummocky landscape with loamy soils in the Prairie Region of Canada, 10 to 20 times smoothing operations with an area threshold of 200 m2 and a depth threshold of 0.1 m were recommended as guidelines for depression identification.

Research highlights► We analyze the unautocorrelated and autocorrelated errors in lidar DEM. ► We examine depression identification on a hummocky landscape using lidar DEM. ► We illustrate depression connections and their effects on depression identification.