Technical Note on Quality Assessment for Global DEM

This document assesses the quality of three global DEMs widely used around the world: SRTM, ASTER GDEM and ALOS World 3D. In addition to a classic method of controlling heights using reference altimetry data (here the LiDAR ICESat-1), innovative methods are used to compare DEMs with each other, their heights but also derivative measurements such as slopes, azimuths or curvatures. Finally, a section is dedicated to the comparison of the uses of DEMS and more specifically to the impact of the characteristics of DEMS on the quality of orthorectification on a same scene (here a Sentinel-1 IW scene).The evaluation of each DEM is carried out using a template designed for the evaluation of missions carrying out multi-date acquisitions. It follows an interpretation not always adapted to the analysis of a single product having often undergone several editions. These tensions are particularly noticeable in the maturity matrices and in the "mission assessment overview".The evaluation of the three DEMs could have motivated the publication of three separate documents. It was however chosen to treat the three DEMs in a single document for severalreasons:- The explanation of the methods, the description of the ICESat-1 reference mission or the motivation of the choices would have been duplicated three times, making maintenance difficult for future editions of three documents in parallel.- The intercomparison of DEMs provides interesting results. This intercomparison motivates the description of the three DEMs together in a single document.- The document is organized by separating whenever necessary the description, analysis and results of each of the three DEMs.For each DEM, numerous documents are available relating to product design, their control, their versions, their uses, etc. These documents were used to design maturity matrixes that reflect the maturity of the documentation, the compliance with the best practices in term of Calibration/Validation, and the quality of the data with regards to a reference dataset and in-situ data.The VisioTerra VtWeb tool has been adapted to allow viewing and comparison of DEMs. It is thus possible to evaluate at any scale the differences between two DEMs and to immediately locate their anomalies. On a small scale, it is easy to spot the footprints of the acquisition segments used for optical photogrammetry and a little less those of the shuttle of the SRTM radar interferometry mission. On large scale, large structures, such as a 300 m raised drainage basin in SRTM or an under-elevated tile in ALOS World 3D. For each of the three DEMS, we find singularities in different places that we do not observe in the other two. All these defects reveal the difficulties inherent in radar or optical acquisition on (often uniform) surfaces of ice, snow, desert, open water...A more conventional study is carried out by calculating the difference between the elevations of the LiDAR instrument (GLAS) on board the ICESat-1 satellite and the elevations interpolated at the same points from each DEM. According to the quality masks of the DEMs, this statistical study concerns up to 92 million points distributed over the entire surface of the globe. This numerical study shows that the DEM "ALOS World 3D" is the best one both in terms of quadratic mean error (RMSE) of 4.5 m and an arithmetic mean (bias) of 0.05 m. Second place is held by SRTM with roughly the same RMSE of 4.7 m but a larger bias of 0.59 m. Finally, ASTER GDEM obtains the worst results with a huge RMSE of 27.8 m and a large bias of 2.3 m. The originality of the study also concerns the evaluation of the quality for the use of DEMs. Here, we estimate the quality of orthorectification by comparing two spacemaps of a same Sentinel-1 radar product but orthorectified by two different DEMs. Examples of altimetric differences between SRTM and ASTER GDEM produce amazing deformations in the Alps or in the Himalayas. The singularities identified by comparing the DEMs with each other produce large differences between the orthorectified products. The study shows that an error of only one (1) meter can produce a planimetric error up to 1.80 m.