Towards ice-thickness inversion: an evaluation of global digital elevation models (DEMs) in the glacierized Tibetan Plateau

Abstract

Abstract. Accurate estimates of regional ice thickness, which are generally produced by ice-thickness inversion models, are crucial for assessments of available freshwater resources and sea level rise. A digital elevation model (DEM) derived from surface topography of glaciers is a primary data source for such models. However, the scarce in situ measurements of glacier surface elevation limit the evaluation of DEM uncertainty. Hence the influence of DEM uncertainty on ice-thickness modeling remains unclear over the glacierized area of the Tibetan Plateau (TP). Here, we examine the performance of six widely used and mainly global-scale DEMs: AW3D30 (ALOS – Advanced Land Observing Satellite – World 3D – 30 m; 30 m), SRTM-GL1 (Shuttle Radar Topography Mission Global 1 arc second; 30 m), NASADEM (NASA Digital Elevation Model; 30 m), TanDEM-X (TerraSAR-X add-on for Digital Elevation Measurement, synthetic-aperture radar; 90 m), SRTM v4.1 (Shuttle Radar Topography Mission; 90 m), and MERIT (Multi-Error-Removed Improved-Terrain; 90 m) over the glacierized TP by comparison with ICESat-2 (Ice, Cloud and land Elevation Satellite) laser altimetry data while considering the effects of glacier dynamics, terrain factors, and DEM misregistration. The results reveal NASADEM to be the best performer in vertical accuracy, with a small mean error (ME) of 0.9 m and a root mean squared error (RMSE) of 12.6 m, followed by AW3D30 (2.6 m ME and 11.3 m RMSE). TanDEM-X also performs well (0.1 m ME and 15.1 m RMSE) but suffers from serious errors and outliers on steep slopes. SRTM-based DEMs (SRTM-GL1, SRTM v4.1, and MERIT) (13.5–17.0 m RMSE) had an inferior performance to NASADEM. Errors in the six DEMs increase from the south-facing to the north-facing aspect and become larger with increasing slope. Misregistration of the six DEMs relative to the ICESat-2 footprint in most glacier areas is small (less than one grid spacing). In a next step, the influence of six DEMs on four ice-thickness inversion models – GlabTop2 (Glacier bed Topography), Open Global Glacier Model (OGGM), Huss–Farinotti (HF), and Ice Thickness Inversion Based on Velocity (ITIBOV) – is intercompared. The results show that GlabTop2 is sensitive to the accuracy of both elevation and slope, while OGGM and HF are less sensitive to DEM quality and resolution, and ITIBOV is the most sensitive to slope accuracy. NASADEM is the best choice for ice-thickness estimates over the whole TP.