| Digital elevation models (DEMs) are frequently used for hydro-geomorphic applications involving overland flow routing. These applications are sensitive to the occurrence of elevation errors that result in spurious depressions because flow directions are indeterminate at depression bottoms. Therefore, in effect, simulated overland flow becomes "trapped" within digital depressions. It is a common practice to remove all depressions (i.e., artifact and actual) and enforce uninterrupted flow paths. This dissertation considers the appropriateness of this practice by (1) comparing methods for distinguishing artifact from actual depressions, (2) determining the prevalence of artifact depressions in DEMs, (3) evaluating the impacts of the removal of artifact depressions to the DEM and its derived attributes, and (4) examining the importance of actual depressions in DEMs used for hydrological purposes.;The comparison of methods for distinguishing actual and artifact depression indicated that the stochastic modeling method was the most promising in terms of automation and its ability to handle uncertainty in topography.;Artifact depressions were found to be inherent in DEMs across a range of data sources, data scales, and landscape types suggesting that the removal of depressions will always be necessary regardless of improvements in the source or scale of DEMs. Significant differences were observed in the impacts of different methods for removing artifact depressions to the DEM (i.e., the number of modified cells and mean absolute elevation difference) and in the spatial and statistical distributions of derivatives of the DEM. Depression filling, the most common method for removing depressions, performed worse than all other methods, while depression breaching and the impact reduction approach, a new method, performed best in terms of impacts to the DEM.;Preserving actual depressions in DEMs was found to be important for hydrogeomorphic applications involving simulated overland flow. Simple measures of size, position, and connectivity of depressions were found to explain a significant amount of variance in runoff from 12 headwater catchments on the Canadian Shield. Depression metrics were superior predictors of runoff compared with simple basin metrics, e.g. slope and catchment area. Therefore, use of depressionless DEMs for hydro-geomorphic applications is not prudent. |
