Techniques for quantifying the accuracy of gridded elevation models and for mapping uncertainty in digital terrain analysis

We first provide a critical review of statistical procedures employed in the literature for testing uncertainty in digital terrain analysis, then focus on several aspects of spatial autocorrelation that have been neglected in the analysis of gridded elevation data. When applied to first derivatives of elevation such as topographic slope, a spatial approach using Moran’s I and the LISA (Local Indicator of Spatial Association) allows: (1) georeferenced data patterns to be generated; (2) error hot- and coldspots to be located; and (3) error propagation during DEM manipulation to be evaluated. In a worked example focusing on the Wasatch mountain front, Utah, we analyse the relative advantages of six DEMs resulting from different acquisition modes (airborne, optical, radar, or composite): the LiDAR (2 m), CODEM (5 m), NED10 (10 m), ASTER DEM (15 m) and GDEM (30 m), and SRTM (90 m). The example shows that (apart from the LiDAR) the NED10, which is generated from composite data sources, is the least error-ridden DEM for that region. Knowing error magnitudes and where errors are located determines where corrections to elevation are required in order to minimize error accumulation or propagation, and clarifies how they might affect expert judgement in environmental decisions. Ground resolution issues can subsequently be addressed with greater confidence by resampling the preferred grid to terrain resolutions suited to the landscape attributes of interest. Source product testing is an essential yet often neglected part of DEM analysis, with many practical applications in hydrological modelling, for predictions of slope- to catchment-scale mass sediment flux, or for the assessment of slope stability thresholds.