Comparison of Gridded DEMs by Buffering

Abstract

Comparing two digital elevation models (DEMs), S1 (reference) and S2 (product), in order to get the S2 quality, has usually been performed on sampled points. However, it seems more natural, as we propose, comparing both DEMs using 2.5D surfaces: applying a buffer to S1 (single buffer method, SBM) or to both S1 and S2 (double buffer method, DBM). The SBM and DBM approaches have been used in lines accuracy assessment and, in this paper, we generalize them to a DEM surface, so that more area of the S2 surface (in the case of the SBM), or the area and volume (in the case of the DBM) that are involved, more similarly are S1 and S2. The results obtained show that across both methods, SBM recognizes the presence of outliers and vertical bias while DBM allows a richer and more complex analysis based on voxel intersection. Both methods facilitate creating observed distribution functions that eliminate the need for the hypothesis of normality on discrepancies and allow the application of quality control techniques based on proportions. We consider that the SBM is more suitable when the S1 accuracy is much greater than that of S2 and DBM is preferred when the accuracy of S1 and S2 are approximately equal.