Distinguishing actual and artefact depressions in digital elevation data

Topographic depressions in digital elevation models (DEMs) are frequently a combination of artefacts and actual features. It is common practice to remove all digital depressions, from DEMs that are used in hydrogeomorphic applications. This practice is inappropriate because actual depressions affect many of the environmental phenomena at study. Nonetheless, indiscriminate depression removal persists because of an inability to distinguish artefacts from actual depressions. Five potential approaches for distinguishing artefacts from actual depressions in DEMs are described in this paper: ground inspection, examining the source data, classification approaches, knowledge-based approaches, and modelling approaches. Of the five methods, ground inspection was the only approach that actually confirms the existence of digital depressions. The other four methods that were identified operate by establishing justification for why a digital depression is likely to be an artefact or actual depression. A comparison of the depression validation approaches for a small subcatchment on the Canadian Shield showed that the modelling approach performed slightly better than the other methods. While being highly automated and applicable to all landscape types, this approach also explicitly handles DEM uncertainty. By applying the Monte Carlo method, this approach estimates the likelihood of a digital depression actually occurring in the landscape given the degree of uncertainty in local topography. After artefact and actual depressions are identified, it is then possible to remove the artefacts and to preserve the real features for incorporation into modelling. r 2005 Elsevier Ltd. All rights reserved.