Comparison and Validation of Satellite-Derived Digital Surface/Elevation Models over India

Abstract

India presents among the world's most topographically complex geomorphologies, with land elevations ranging from -2 m to +8586 m and terrain gradients sometimes exceeding 45°. Here we present an evaluation of four freely available digital surface models (DSMs) on a model-to-model basis, as well as a validation using independent ground-truth data from levelled benchmarks in India. The DSMs tested comprise SRTM1", SRTM3", ASTER1" and Cartodem1" [an India-only model]. Along with these four DSMs, the MERIT3" digital elevation model (DEM) is also tested with the ground-truth data. Our results for India indicate some mismatch of these DEMs/DSMs from their claimed accuracies/precisions. All DSMs/DEMs (except for ASTER) have >90% of pixels satisfying ±16 m at the one-sigma level, but only in the low-lying (<500 m) parts of India, i.e., the Gangetic plains and the Thar desert.

Validation of FABDEM, a global bare-earth elevation model, against UAV-lidar derived elevation in a complex forested mountain catchment

“…COP30 uses the Earth Gravitational Model 2008 (EGM2008) geoid model (Pavlis et al 2012) as compared to EGM1996 as used by the other datasets. EGM2008 is substantially improved over EGM1996 and Goyal et al (2021) reported differences between the two up to 12 m between the two models.…”

“…The strong striping in the SRTM dataset was exceptionally problematic. Such patterns have been described in various SRTM evaluations (Yamazaki et al 2017, Goyal et al 2021. The stripping pattern would have profound impacts on many processes' simulation and quantification, and these data should be used with care in steep mountain basins.…”

“…Complete information on the data collection, processing, and accuracy assessment, including data and code, for the UAV-lidar DEM generation, can be found in Harder et al (2020). The area of interest was mapped in a sequence of terrain following flights at m s 7 , Goyal et al (2021) reported differences between the two up to 12 m between the two models. The COP30 dataset is an ESA/Airbus product acquired between 2010 and 2015.…”

An experimental Indian gravimetric geoid model using Curtin University’s approach

“…Quantification of the differences in the topographical effect with the use of DEM and DSM has been investigated by Yang et al (2019). Since India does not have a national DEM, therefore, after a DEM/DSM analysis (Goyal et al 2021a), it was decided to work with the best available DEM over India, i.e., the MERIT 3"x3" DEM (Yamazaki et al 2017), for our computations. Though the accuracy of the MERIT DEM varies considerably (± 11.7 m to ± 47.3 m) over different landforms in India, an overall estimate for the whole of India is ± 17.3 m (Goyal et al 2021a).…”