A New Framework for Vertical Accuracy Assessment and Spatial Error Mapping of Global Digital Elevation Models Using ICESat-2

Abstract Although Global Digital Elevation Models (GDEMs) are widely used, their spatial error distributions over extensive regions remain insufficiently understood. Using ICESat-2 Reference Control Points (RCPs) (~ 30 points/km²), this paper not only evaluates GDEMs with point-based metrics such as Root Mean Square Error (RMSE) and Mean Error (ME), but also proposes a novel raster-independent Spatial Error Mapping (SEM) approach. The proposed SEM leverages dense RCPs to generate two complementary maps: the Spatial Accuracy Map (SAM) and the Spatial Bias Map (SBM). Results show that FABDEM consistently exhibits the highest vertical accuracy (RMSE = 2.06 m; ME = -1.01 m), followed by AW3D30 (3.47 m; -2.31 m), NASADEM (3.59 m; -1.33 m), SRTM (3.90 m; -1.22 m), and ASTER GDEM (6.58 m; -0.89 m). The results reveal three key patterns: (1) accuracy generally decreases with increasing elevation and slope; (2) most GDEMs exhibit higher errors on north- and northwest-facing slopes; and (3) forested and urban areas show the lowest overall accuracy. The SEM analysis uncovered continuous spatial error patterns in GDEMs such as localized biases and processing- or acquisition-geometry-related artifacts in the SBM and SAM that were previously unreported in the literature and are not captured by traditional point-based metrics.