Accuracy assessment of the global TanDEM-X digital elevation model in a mountain environment

Abstract

The past two decades have been prolific in production of global or near-global Digital Elevation Models (DEMs) derived from satellite data. The most recent addition to the family of global DEMs is the TanDEM-X DEM with resolution of 0.4 arc sec. DEMs are essential for a wide range of environmental applications, many of which are related to mountains including studies on natural hazards, forestry or glacier mass changes. However, synthetic aperture radar interferometry used for acquisition of TanDEM-X DEM is especially challenging over steep and irregular mountain surfaces due to shadowing and foreshortening effects. In this study, we assessed the absolute vertical accuracy of TanDEM-X DEM in European mountains. We compared it with both a Digital Terrain Model (DTM) and a Digital Surface Model (DSM) derived from airborne laser scanning data. Our results indicate that the height error of TanDEM-X DEM expressed as absolute deviation at the 90% quantile is consistent with the 10 m mission specification benchmark. We further concentrated on the absolute height error with respect to environmental characteristics (i.e. forested and non-forested areas, slope, and aspect). The comparison of TanDEM-X DEM with a reference DTM showed a positive vertical offset; however, the mean error differed greatly between forested and non-forested areas. When compared to reference DSM, our results showed a slight underestimation. We observed the highest underestimation in deciduous forests, followed by coniferous forests and non-forested areas. A significant decrease in accuracy was observed with increasing slope, especially for slopes above 10°. In mountains where the imagery was acquired only in one orbit direction (i.e. ascending for Northern hemisphere), the largest TanDEM-X DEM error when compared to DSM was recorded for the west-facing slopes (i.e. slopes facing the sensor); however, the association with terrain orientation diminished in mountains, the imagery of which was acquired from both the ascending and descending orbit. Finally, we evaluated the effect of data acquisition characteristics provided with TanDEM-X DEM as auxiliary data. Our results show that two coverages might not be sufficient in mountain environment. Additional acquisitions, especially those with different acquisition geometry, improved the absolute vertical accuracy of TanDEM-X DEM and eliminated areas of inconsistency. We discourage from using the Height Error Map (HEM) to estimate the error magnitude. On the other hand, auxiliary data (COM, COV) provide valuable information that should be always used in pre-analyses to identify possible problematic areas.