Common Spatial Support and Registration Networks for Reproducible ArcticDEM Elevation Change Analysis

Comparisons of elevation change from repeat digital elevation models (DEMs) depend on which pixels are evaluated and which DEM pairs are used for alignment. These effects were examined using ArcticDEM strips, four terrain windows with known perturbations, and four acquisitions at the Batagay megaslump. In the original single-realization tests, the maximum joint-gradient component and protected-change median errors were 0.087 and 0.027 m. In 480 additional paired perturbations, all joint-gradient estimates met the prescribed 0.50 m component tolerance; the median maximum component error was 0.026 m. On the observed Batagay terrain, joint-gradient, xDEM Nuth–Kääb, and historical internal Nuth–Kääb fits were accepted for 16, 17, and 14 of 18 pair-by-reference combinations, respectively, under matched candidate sampling. In the historical two-implementation comparison, restricting the combined-reference network to shared edges increased the July–September 2024 root-mean-square difference between the two change maps from 1.755 to 2.577 m on 184,120 identical target cells. Changing from pair-specific to fixed evaluation support had little effect on that interval. These results distinguish sensitivity to graph selection from sensitivity to the terrain evaluated. Independent elevations were unavailable at Batagay, so the outputs cannot be ranked by field accuracy.