Evaluation of ASTER GDEM2, SRTMv3.0, ALOS AW3D30 and TanDEM-X DEMs for the Peruvian Andes against highly accurate GNSS ground control points and geomorphological-hydrological metrics

Abstract We present an analysis for Peru of the vertical accuracy and derived geomorphological parameters of three, freely available Digital Elevation Models (DEMs) with 30-m grid sizes and one commercially available 12-m DEM: ASTER GDEM2, SRTMv3.0 (hereafter called SRTM3), ALOS World 3D (AW3D30) and 12-m TanDEM-X. Their elevation values were compared against values measured with a dual-frequency Trimble 5800 Global Navigation Satellite Systems (GNSS) receiver with two antennas in static relative positioning. A total of 139 ground control points (GCPs) with elevations of up to 4334 m above sea level (asl). were obtained from four topographically contrasting regions along the flat, arid coast, the semi-dry Andes and the Amazon rainforest. The combined results of the four areas showed that the residuals of ASTER GDEM2 had a root mean square error (RMSE) of 6.907 m, SRTM3 5.113 m and AW3D30 6.246 m. For the two areas where TanDEM-X was investigated, a RMSE of 1.666 m was obtained. When considering each area individually, of the three 30-m DEMs, SRTM3 performed best in all areas, while ASTER GDEM2 was always performing worst. The standard deviation (SD) of the residuals of ASTER GDEM2 and AW3D30 varied much more between the four study areas than the SD of SRTM3 and TanDEM-X. It was possible to establish linear correlations with R2 > 0.95 between the GNSS-measured and DEM-derived elevation values for all areas and all DEMs, which can be used to correct ASTER GDEM2 and AW3D30 for Peru. No relationship was found between absolute elevation and residuals. Slope angle of >15°, however, correlated to much higher residuals for SRTM3 and AW3D30, whereas a constant increase in residual correlated to a constant increase in slope angle for TanDEM-X. In general, the highest residuals corresponded to the highest slope classes. A comparison of absolute elevation values and slope angles of selected areas between ASTER GDEM2 and SRTM3, as well as between AW3D30 and SRTM3, showed in all cases more deviation from true values in areas with steeper slopes. This effect seems independent of vegetation. The presence of buildings however, severely affected performance of ASTER GDEM2. For some areas, AW3D30 suffered from mismatch between adjacent scenes and horizontal and oblique stripping. An analysis of drainage networks showed that all three 30-m DEMs were capable of producing a similar number of Strahler orders, total number of channels, minimum and maximum channel length and accumulative channel length. All DEMs produced accurate networks in the upper reaches of the catchments, but suffered in flat areas such as wide fluvial valleys and sink-filled areas. ASTER GDEM2 in particular produced parallel river channels and unnatural short cuts. Our results agreed well with those of other studies elsewhere around the world and showed that i) the vertical error is much less than those officially reported for all DEMs; ii) slope is the most determinant factor in DEM accuracy. We conclude that SRTM3 delivers the most stable performance in almost all tests and is therefore the DEM of choice for landscape analysis in Peru. TanDEM-X 12 m performed very well in our elevation accuracy test and could become a worthy successor to SRTM3.