Author(s):
Yongxin Deng* - Western Illinois University
Chad E. Sperry - Western Illinois University
Abstract:
Spatial structure of DEM errors is crucial for error propagation in most DEM applications. However, various approaches - including empirical, simulation, and analytical ones – to such structure have been frustrated by the lack of ground truth data, hence unable to evaluate the possible, or realistic, effects of DEM errors on the terrain analysis outputs. For instance, previous empirical efforts have been discouraged by survey difficulty and loss of spatial autocorrelation, and the simulation approach can only provide the probable, rather than possible, impact of error with the lack of a "true" error surface. This presentation reports how we used truck-loaded continuous and linearly-distributed GPS elevations along rural road networks to evaluate the spatial character of DEM errors and DEM-based catchment boundaries (e.g. NHD, or National Hydrography Dataset). A one-dimensional statistical (geostatistical) analysis was carried out to represent the spatial structure of GPS-DEM differences to characterize the error structure. The horizontal error of catchment boundaries was represented as their distances to the nearest ridge points – if they are not on the ridge themselves – along these linear elevation profiles, which are perpendicular to the general drainage directions. The summary statistics of inaccuracies in catchment sizes and coverage locations – as a result of elevation errors – were subsequently evaluated based on the boundary (horizontal) error characterization. In an ensuing research, the above method is used to evaluate DEM error and propagation in four types of landscapes following a qualitative landform/landscape classification of the study area in western Illinois.
