Comparison and Validation of Digital Elevation Models Derived from InSAR for a Flat Inland Delta in the High Latitudes of Northern Canada

Abstract

Inaccuracies in topographic data can be a significant source of error and uncertainty for hydrologic, atmospheric-land exchange and climate change modeling. This study was conducted to examine how different data sources and techniques used to generate digital elevation models (DEM) influence DEM accuracy. Using the interferometric synthetic aperture radar (InSAR) technique, the system acquisition parameters and the topographic characteristics are important factors in determining DEM quality. ICESat elevation footprints and highly accurate ground control points (based on GPS measurements) are used as independent references to validate the accuracy of DEMs. The TanDEM-X DEM shows the highest accuracy among all studied DEM products, with the root mean square error (RMSE) of 2.9 m, followed by 3.3 m for Canadian digital elevation data, 5.2 m for ASTER-GDEM, and 5.5 m for ALOS-PALSAR DEM. The RMSEs of the RADARSAT-2 based InSAR DEM and global GTOPO30 are considerably higher than the other DEMs, 24 m and 36 m, respectively. The interferograms of RADARSAT-2 and ALOS-PALSAR data acquired in early winter resulted in higher DEM accuracy compared with the summer image pairs, even with their lower InSAR coherence. This is attributed to the higher importance of perpendicular baseline and height of ambiguity and potentially the significant influence of atmospheric phase delay in DEM generation. We also compared the accuracy of DEMs of different land cover types and their representation of the land-water surface. The results show that vegetated areas of coniferous and broadleaf trees (i.e., white and black spruce, trembling aspen, balsam poplar) seem to be the main sources of error in all DEMs. The river slopes derived from the TanDEM-X, ASTER, and ALOS-PALSAR DEMs show patterns consistent with the river bed slope from our river bathymetry survey. Further studies are necessary to evaluate the performance of those different DEMs in earth surface modeling as well as to improve the quality of high error DEMs such as RADARSAT-2 data in this study. The Slave River Delta in the Northwest Territories of Canada was used as a test case.