DTM terrain model

The DTM is an elevation tile set for all of Austria: bare-earth height without buildings or vegetation, encoded as colour (Terrain-RGB), usable as raster-dem for hillshade and 3D terrain in MapLibre, Mapbox GL and compatible clients.

1. Key facts

Tilesetrelief-terrainrgb

File: relief-terrainrgb-20250915-ed2e4341.pmtiles
Reference date: 2025-09-15
Size: 39.1 GB (39,067,398,974 Bytes)
SHA-256: ed2e4341e6e10d0fa62880e2f1c91d9828fc3de84b7ddd541417110221c8121a
Format: webp, encoding mapbox, 512 px
Zoom levels: z5–z16
Bounds (WGS84): 9.5306 / 46.3697 / 17.1661 / 49.0235 (W/S/O/N)
Tiles: 664,979 addressed, 663,931 unique contents
License: CC BY 4.0

Generated at build time from viewer/tiles-manifest.json and the PMTiles header of the file (generated 2026-11-07).

2. Live map

3D terrain HillshadeClick the map: elevation and tile RGB.

A click reads the tile of the current zoom level (max z16) with a range request, decodes the pixel at the click point with the formula in section 4 and shows elevation and RGB.

3. Using it

import * as maplibregl from "maplibre-gl";
import { Protocol } from "pmtiles";

maplibregl.addProtocol("pmtiles", new Protocol().tile);
const map = new maplibregl.Map({
container: "map",
center: [12.694, 47.074], zoom: 11, pitch: 60,
style: {
  version: 8,
  sources: { dem: { type: "raster-dem", url: "pmtiles://https://tiles.mapx.at/relief-terrainrgb-20250915-ed2e4341.pmtiles", tileSize: 512, encoding: "mapbox", maxzoom: 16 } },
  terrain: { source: "dem", exaggeration: 1.2 },
  layers: [{ id: "hill", type: "hillshade", source: "dem" }]
}
});

# pip install pmtiles pillow numpy requests
import io, requests, numpy as np
from PIL import Image
from pmtiles.reader import Reader

URL = "https://tiles.mapx.at/relief-terrainrgb-20250915-ed2e4341.pmtiles"

def get_bytes(offset, length):
  r = requests.get(URL, headers={"Range": f"bytes={offset}-{offset + length - 1}"}, timeout=30)
  r.raise_for_status()
  return r.content

tile = Reader(get_bytes).get(12, 2192, 1439)
rgb = np.asarray(Image.open(io.BytesIO(tile)).convert("RGB")).astype(np.uint32)
elevation = -10000 + (rgb[..., 0] * 65536 + rgb[..., 1] * 256 + rgb[..., 2]) * 0.1

curl -s -r 0-126 https://tiles.mapx.at/relief-terrainrgb-20250915-ed2e4341.pmtiles | xxd | head -4
curl -s -o tile.webp https://tiles.mapx.at/v1/dtm/12/2192/1439.webp

Leaflet, OpenLayers and QGIS examples: see the German page.

4. Encoding

elevation_m = -10000 + (R * 65536 + G * 256 + B) * 0.1

Values are rounded to 0.1 m and clamped to the range. In Austria they span about 115 m to about 3798 m. Lossless WebP, 512 x 512 px, no border (no 1 px skirt): use tileSize: 512.

5. Resolution per zoom

At 47.5 N, 512 px tiles:

Zoom	m per pixel	Tile edge (km)	Note
z5	1,652	846
z6	826	423
z7	413	212
z8	207	106
z9	103	53
z10	52	26
z11	26	13
z12	13	6.61
z13	6.46	3.30
z14	3.23	1.65
z15	1.61	0.83
z16	0.81	0.41	highest stored level (≈ 1 m source)
z17	0.40	0.21	overzoom (client)

z16 (about 0.8 m per pixel) is the first level that does not coarsen the 1 m source. From z17 the client overzooms the z16 tile (maxzoom: 16); no new information appears. z15 to z5 are 2x2 averages in height space.

6. Provenance

BEV ALS-DTM 1 m (Geonetwork series 5ce253fc-b7c5-4362-97af-6556c18a45d9), 55 source tiles of 50 x 50 km as COG in EPSG:3035, reference date 15 Sep 2025, reprojected to EPSG:3857 bilinearly.

7. Accuracy and limits

Measured: quantisation 0.1 m; z16 versus the bilinearly interpolated source at three check points within +-0.1 m (-0.03, -0.04, +0.02 m); Rust render versus source on ten tiles at most 0.05 m (rounding).

Limits: bare earth only (no buildings, trees). No measurements outside the data: the tile set covers the rectangle 9.53-17.17 E, 46.37-49.02 N; pixels without source data are a smooth continuation of the terrain (mean error 11 m in a test outside the data near Melk). Deviation of overview levels z15-z5 from the source and country-wide RMSE: not measured.

8. Variants

DSM (surface incl. buildings and vegetation, same encoding): relief-dsm-terrainrgb-20250915-d19cb434.pmtiles, 82.1 GB.

9. Versions

Reference date	File	SHA-256
2025-09-15	`relief-terrainrgb-20250915-ed2e4341.pmtiles`	`ed2e4341e6e10d0fa62880e2f1c91d9828fc3de84b7ddd541417110221c8121a`

Files never change; a new reference date is a new file.

10. Attribution

Relief: BEV – ALS-DTM 1 m (Stichtag 15.09.2025), CC BY 4.0

11. FAQ

Why lossless WebP? Any colour change is a different elevation.

Why no nodata? Missing pixels are filled with a smooth continuation (area mean, push-pull) before encoding, so hillshade and 3D show no holes. Tiles without any source data do not exist.

How does it differ from Mapbox/MapTiler? Same encoding as mapbox-terrain-dem-v1 but without the 1 px border; Mapbox serves up to z14, mapx up to z16 (Austria only), key-free and downloadable as a whole file.