A regional ASTER GDEM error model for the Chinese Loess Plateau

• Published in: International journal of remote sensing Vol. 40; no. 3; pp. 1048 - 1065
• Main Authors: Dong, Youfu, Shortridge, Ashton M.
• Format: Journal Article
• Language: English
• Published: London Taylor & Francis 01-02-2019; Taylor & Francis Ltd
• Subjects: Accuracy; ASTER (radiometer); Digital Elevation Models; Elevation; Error analysis; Error reduction; Forests; Geomorphology; Imaging techniques; Land cover; Model accuracy; Radiometers; Regression models; Root-mean-square errors; Roughness; Spectroradiometers; Terrain; Thermal emission; Loess Plateau
• ISSN: 0143-1161; 1366-5901
• DOI: 10.1080/01431161.2018.1524171

Accuracy of the global ASTER GDEM (Advanced Space-borne Thermal Emission and Reflection Radiometer Global Digital Elevation Model) version 2 (v2) elevation data product is highly variable regionally, as are its empirical correlations with landscape variables. This paper investigates GDEM error along a 49-site geomorphologic gradient within the core region of the Chinese Loess Plateau, notable for its heterogeneous terrain. The error is modelled using its associations with MODIS (Moderate Resolution Imaging Spectroradiometer) composite forest cover percentage, GlobeLand30 land cover, and key elevation derivatives, including two indices, terrain roughness index (TRI) and topographic position index (TPI), not previously evaluated in GDEM accuracy studies. Overall root mean squared error (RMSE) is 20.33 m, in excess of the GDEM v2 accuracy specifications, while RMSE at each site varies substantially, from 10.67 m for a low relief area to 21.84 m for the most rugged site. Strong associations between covariates, especially slope, aspect, TRI, and forest cover are identified. A regression model using these variables is developed to formally characterize and predict GDEM error. External validation with independent checkpoints across all sites demonstrates that this model can reduce mean error by about 4 m.