Open data · CC BY 4.0
The height of the ground itself, with trees and buildings removed, on an 8 m grid for the whole country. Built from satellite stereo imagery and 13.3 million laser measurements from space.
Shaded relief: NepalDTM
What it is
Free global elevation models such as Copernicus GLO-30 measure the surface: the top of the forest canopy and the roofs of buildings. In Nepal, where forests cover much of the hills and the Terai, that puts the "ground" metres too high.
NepalDTM estimates the bare earth underneath. Flood models, road and hydropower design, landslide studies, and irrigation planning all need the ground, not the canopy.
Every tile comes with a surface model, a bare-earth model, and a per-pixel error estimate, so you can see how far to trust each value.
How it is built
The sharpest available surface is stitched together, then a model trained on laser ground returns predicts how much vegetation and buildings add to it, and subtracts that.
The NASA High Mountain Asia 8 m stereo mosaic is used wherever it exists. Copernicus GLO-30 fills the gaps.
Laser altimeters on ICESat-2 and GEDI see through gaps in the canopy to the ground. A model learns the canopy and building offset from them.
The offset is subtracted from the surface. Whole 48 km blocks were left out of training and used only for scoring.
Accuracy
RMSE against ICESat-2 ground points from 48 km blocks that the model never saw. Lower is better. The gain is largest in the Terai, where flood modelling needs it most.
| Group | Points | NepalDTM | Copernicus | FABDEM | GEDTM30 |
|---|---|---|---|---|---|
| All of Nepal | 4,463,463 | 6.10 | 9.07 | 7.78 | 8.47 |
| Terai <300 m | 1,251,581 | 1.96 | 4.53 | 2.46 | 2.12 |
| Siwalik 300-1000 m | 487,932 | 5.18 | 8.53 | 6.16 | 5.87 |
| Hills 1-2.5 km | 1,076,095 | 6.24 | 8.69 | 7.33 | 7.57 |
| Mountains 2.5-4 km | 776,214 | 8.27 | 12.88 | 10.27 | 10.45 |
| High himal >4 km | 2,636,007 | 5.33 | 7.55 | 7.50 | 8.98 |
| Bagmati | 423,665 | 5.92 | 9.64 | 8.23 | 8.86 |
| Gandaki | 745,860 | 5.99 | 8.63 | 7.99 | 9.14 |
| Karnali | 1,325,510 | 6.86 | 9.88 | 8.55 | 9.39 |
| Koshi | 640,238 | 6.30 | 9.42 | 8.55 | 9.39 |
| Lumbini | 547,250 | 5.13 | 7.57 | 5.74 | 5.76 |
| Madhesh | 234,163 | 1.78 | 3.88 | 2.04 | 1.98 |
| Sudurpashchim | 546,777 | 6.28 | 9.67 | 7.54 | 7.68 |
| Forest | 1,616,752 | 6.82 | 11.20 | 7.95 | 7.84 |
| Non-forest | 4,611,077 | 4.94 | 6.89 | 6.78 | 7.82 |
| Slope <20 | 3,163,548 | 3.26 | 5.31 | 4.22 | 4.50 |
| Slope 20-40 | 2,547,624 | 6.07 | 8.89 | 7.56 | 8.26 |
| Slope >=40 | 516,657 | 10.81 | 15.91 | 14.71 | 16.67 |
| On HMA 8 m | 3,199,580 | 4.00 | 6.82 | 6.35 | 7.39 |
| Copernicus-filled | 3,028,249 | 6.71 | 9.49 | 7.82 | 8.26 |
RMSE in metres on the 6,227,829 held-out points shared by all models (the national row is the 4,463,463 points inside Nepal). GEDTM30 was trained on ICESat-2, so its score here is optimistic. Independent check: Khumbu differential GPS agrees with the surface model to 2.49 m RMSE at glacier margins (83 points).
Know the limits
The file spacing is 8 m across the whole country, but where the stereo mosaic has gaps the terrain detail comes from 30 m data. The aux file in every tile says which source each pixel came from.
RMSE where the 8 m stereo mosaic is the base, against 6.71 m in Copernicus-filled areas.
RMSE on slopes of 40° and steeper, where every model struggles.
Median and 90th-percentile expected error across the country, from the per-pixel error band.
Data
Free to use with attribution under CC BY 4.0. Click any point on the map to get the tile that covers it.
324 tiles of 24 × 24 km, each as _dtm (bare earth), _dem (surface) and _aux
(band 1 source: 1 = HMA 8 m, 2 = Copernicus fill; band 2 expected error in decimetres).
Bare earth as Terrarium-encoded terrain tiles, zoom 5–13, for MapLibre and other web maps.
PMTiles · 1.7 GB# Read one tile straight from the web with GDAL, no download needed
gdalinfo /vsicurl/https://nepaldtm-data.achyutpdl.com/cog/E0360N3072_dtm.tif
# Or download it
curl -O https://nepaldtm-data.achyutpdl.com/cog/E0360N3072_dtm.tifPaper
Cite
@article{paudel2026nepaldtm,
author = {Paudel, Achyut},
title = {{NepalDTM}: A National 8-Metre Bare-Earth
Digital Terrain Model of {Nepal} from
Multi-Source Satellite Data Fusion},
year = {2026},
url = {https://github.com/achyut/NepalDTM}
}