Skip to content
DraconiticResearch
Search Sources API
197,768 studies31 sourcesUpdated 1 min ago
PreprintCC BY 4.0EarthArXiv

NepalDTM: A National 8-Metre Bare-Earth Digital Terrain Model of Nepal from Multi-Source Satellite Data Fusion

Abstract

NepalDTM is an open bare-earth elevation model of Nepal stored on an 8 m grid. Where the NASA High Mountain Asia 8 m stereo mosaic has a usable height, that mosaic is the surface. Elsewhere the surface is Copernicus GLO-30. A histogram gradient-boosting model, trained on 6,230,548 ICESat-2 points and 7,035,904 GEDI points, predicts how far the surface sits above the ground and subtracts that height. On 4,463,463 held-out ICESat-2 points inside Nepal, with whole 48 km blocks left out of each training fold, the root-mean-square error is 6.10 m (NMAD 2.61 m, bias -0.15 m). The same points give 9.07 m for Copernicus GLO-30, 8.47 m for GEDTM30, and 7.78 m for FABDEM. The error is 1.96 m below 300 m and 8.27 m between 2.5 and 4 km. Independent differential GPS on the Khumbu surface agrees to 2.49 m RMSE at glacier margins. Inside the delivered grid, 37% of the area uses the 8 m mosaic and 63% is Copernicus fill, so the posted spacing is not 8 m terrain detail on every cell. The tiles, the figures, and the code that rebuilds the model are public.

Related

  1. Applying satellite observations to improve bottom-up national emission inventories for methane: application to Colombia
    Sarah E. Hancock, Daniel J. Jacob, Rodrigo Jiménez +17
    Atmospheric chemistry and physicsPeer-reviewedCC BYFull text6 citedClimate scienceOpenAlex
  2. From plot to region: canopy height-mediated mangrove aboveground biomass mapping by synergizing UAV-LiDAR and multi-source satellite data
    Yunyan Xiang, Wen H. Cai, Hongxin Su +5
    Frontiers in Plant SciencePeer-reviewedCC BY 4.0Full textEnvironmental science & pollutionCrossrefEurope PMC
  3. Mapping soil particle fractions by training digital soil mapping models with surrogate measurements obtained from laboratory and satellite Vis-NIR spectral data
    S. Dharumarajan, Mlmaw Weerasekara, Philippe Lagacherie +1
    GeodermaPeer-reviewedCC BY 4.0Full textEnvironmental science & pollutionHAL
  4. Multi-source satellite remote sensing of harmful algal blooms in the East China Sea: cross-sensor comparison of GOCI, Sentinel-2 and Sentinel-3
    Chengxin Zhang, Chengxin Zhang, Honghua Shi
    Frontiers in Marine SciencePeer-reviewedCC BYFull textOceanography & cryosphereDOAJ
  5. Microbiological drinking water quality at source and consumption in Indonesia: An analysis of the national drinking water quality study
    Heru Prasetyo, D. Daniel
    Cleaner WaterPeer-reviewedCC BYFull textHydrology & soilsDOAJ
BibTeX
@misc{paudel2026nepaldtm,
  title = {{NepalDTM: A National 8-Metre Bare-Earth Digital Terrain Model of Nepal from Multi-Source Satellite Data Fusion}},
  author = {Achyut Paudel},
  howpublished = {EarthArXiv},
  year = {2026},
  month = oct,
  doi = {10.31223/x5dr5m},
  url = {https://eartharxiv.org/repository/view/15728/},
  note = {CC BY 4.0},
}
RIS
TY  - UNPB
TI  - NepalDTM: A National 8-Metre Bare-Earth Digital Terrain Model of Nepal from Multi-Source Satellite Data Fusion
AU  - Achyut Paudel
PY  - 2026
DA  - 2026/10/24
PB  - EarthArXiv
DO  - 10.31223/x5dr5m
UR  - https://eartharxiv.org/repository/view/15728/
AB  - NepalDTM is an open bare-earth elevation model of Nepal stored on an 8 m grid. Where the NASA High Mountain Asia 8 m stereo mosaic has a usable height, that mosaic is the surface. Elsewhere the surface is Copernicus GLO-30. A histogram gradient-boosting model, trained on 6,230,548 ICESat-2 points and 7,035,904 GEDI points, predicts how far the surface sits above the ground and subtracts that height. On 4,463,463 held-out ICESat-2 points inside Nepal, with whole 48 km blocks left out of each training fold, the root-mean-square error is 6.10 m (NMAD 2.61 m, bias -0.15 m). The same points give 9.07 m for Copernicus GLO-30, 8.47 m for GEDTM30, and 7.78 m for FABDEM. The error is 1.96 m below 300 m and 8.27 m between 2.5 and 4 km. Independent differential GPS on the Khumbu surface agrees to 2.49 m RMSE at glacier margins. Inside the delivered grid, 37% of the area uses the 8 m mosaic and 63% is Copernicus fill, so the posted spacing is not 8 m terrain detail on every cell. The tiles, the figures, and the code that rebuilds the model are public.
ER  -