Technical Note on Quality Assessment for Copernicus DEM

This document presents the quality assessment of the three Copernicus DEMs: Copernicus DEM EEA-10, Copernicus DEM GLO-30 and Copernicus DEM GLO-90. A first major section focuses on evaluating the maturity of the Copernicus DEMs with respect to the EDAP Quality Assessment Guidelines. The second major section of the document encompasses two quality assessment methods:- Qualitative assessments - the VisioTerra VtWeb tool is used to compute the difference between Copernicus DEM GLO-30 and ALOS World 3D. This tool allows to compute the difference between two DEMs at any scale, immediately highlighting their anomalies.- Quantitative assessments - Copernicus DEM EEA-10, GLO-30 and GLO-90 are compared to ICESat-1 / GLAS, ICESat-2 / ATLAS and GEDI LiDAR reference data. Five quantitative studies are followed per DEM, respectively using ICESat-1, ICESat-2 terrain, ICESat 2 terrain with canopy, GEDI lowest mode (ground) and GEDI highest return (top of canopy) heights as vertical references.Illustrations and results of the qualitative assessment, consisting of comparing ALOS World 3D to Copernicus DEM GLO-30, are available in section 4.1. This section provides both global views of this comparison and a case study of the Lake Garda in Italy, highlighting an anomaly of ALOS World 3D.Regarding the quantitative assessment, this technical note aims to answer to the following questions:- Considering ICESat-1, ICESat-2 and GEDI products, which reference data is the most suitable for DEM quantitative assessments?- Which instance (among EEA 10, GLO 30 and GLO 90) of the Copernicus DEM has the best overall height error statistics?- Taking into account results of previous quantitative assessments of ALOS World 3D, ASTER GDEM and SRTMGL1, and the results of the quantitative assessment of Copernicus DEM GLO 30, which global DEM at 1'' is the most accurate?The quantitative studies show that the best results are achieved with ICESat-1, retrieving an arithmetical mean of 0.033 metres and a RMSE of 0.628 metres among all height differences for the assessment of Copernicus DEM GLO 30 (LE95 statistics). Similar results are achieved with ICESat-2 terrain height only, with an arithmetical mean of 0.195 metres and a RMSE of 0.999 metres. Worse results are obtained using ICESat 2 terrain with canopy height as a vertical reference, retrieving an arithmetical mean of 1.124 m and a RMSE of 2.907 m. Considering GEDI lowest mode reference data gives an arithmetical mean of 1.088 m, but a high RMSE of 3.639 m. The worst statistics are obtained considering GEDI highest return, retrieving an arithmetical mean of -5.840 m and a RMSE of 6.710 m. According to these studies, ICESat-1 seems to be the best vertical reference for DEM quantitative assessments. ICESat-2 terrain heights give similar results and is still recommended for DEM quantitative assessments. ICESat 2 terrain with canopy heights is not recommended, as the canopy height is only an estimation and not the actual canopy height retrieved at each geolocation indicated in the ICESat 2 products. GEDI lowest mode and highest return are not recommended as the statistics show very high error arithmetic mean and RMSE.Among Copernicus DEM EEA 10, GLO 30 and GLO 90, the Copernicus DEM GLO 30 obtains the best height error statistics. Considering ICESat 1 reference data, a mean of 0.033 metres, a standard deviation of 0.627 metres and a RMSE of 0.628 m are obtained for the height errors of Copernicus DEM GLO 30. Copernicus DEM GLO 90 height error statistics show similar results, with a mean of 0.066 metres, a standard deviation of 0.706 metres and a RMSE of 0.709 metres. Copernicus DEM EEA-10 height error statistics are the worse, considering a mean of 0.276 metres, a standard deviation of 1.389 metres and a RMSE of 1.415 metres.The quality of ALOS World 3D, ASTER GDEM, SRTMGL1 and Copernicus DEM GLO 30 was assessed using ICESat 1 reference data. These DEMs are available at global scale and share the same spatial resolution of 1''. In previous quantitative studies, ALOS World 3D was assessed to be the best DEM, as opposed to SRTMGL1 and ASTER GDEM. According to the quantitative studies of this technical note, Copernicus DEM GLO 30 is assessed to be the best DEM, showing a significative improvement with regard to the three other global DEMs. The height error statistics of Copernicus DEM GLO 30 are even better than those of ALOS World 3D, retrieving a mean of 0.033 metres, a standard deviation of 0.627 metres and a RMSE of 0.628 metres, as opposed to ALOS World 3D, which obtained a mean of -0.151 m, a standard deviation of 1.660 metres and a RMSE of 1.653 metres.