Vertical Accuracy Assessment and Bias Correction of Freely Available Global DEMs

Accurate digital elevation models (DEMs) are essential for hydrological modelling and floodplain analysis, particularly in low-relief floodplains where small elevation errors can significantly affect flow routing and inundation extent. This study evaluated the vertical accuracy of six freely available global DEMs across the Flinders River catchment, North Queensland, Australia, using 30,916,100 quality-filtered ICESat-2 ATL06 elevation points for regression-based bias correction and airborne LiDAR datasets from five benchmark regions for independent validation. The evaluated DEMs included TANDEM-X, Copernicus DEM, ALOS AW3D30, SRTM, ASTER GDEM, and the Hydrological DEM. Vertical accuracy was assessed using mean error (ME), root mean square error (RMSE), and residual dispersion before and after calibration. Results showed substantial pre-calibration bias in the Hydrological DEM (ME = −2.93 m) and SRTM (ME = −2.66 m), whereas Copernicus DEM showed minimal initial bias (ME = −0.01 m). Regression-based correction reduced mean errors to within ±0.13 m across all DEMs. SRTM showed the largest improvement, with RMSE decreasing from 3.20 m to 0.55 m, while TANDEM-X achieved the highest post-calibration accuracy (RMSE = 0.14 m). Independent LiDAR validation confirmed improved vertical accuracy while preserving terrain morphology and river gradients.