National Open-Data Terrain

The built-in sources on Terrain and Basemap Sources are worldwide tilesets. This page covers the other half of the library: elevation services published directly by national mapping agencies, at resolutions global tilesets can't match — 0.5 m LiDAR for France and the Netherlands against ~30 m for a global DEM.

Load them from Sources → Terrain → Library, which opens a picker where each dataset is added or removed with one click (or all at once), graded against Mapterhorn and filterable by name, or open any of them directly with a ?terrainSourceA=<id> link. To see where each one has data before loading it, turn on its footprint in Coverage overlays.

Credit where it is due: Mapterhorn

This page only exists because Mapterhorn did the hard part first. It is an open, community-run project that solves the same problem from the other end: rather than talking to each agency's API live, it ingests the world's open elevation data — API endpoints, hundred-thousand-file bulk archives, S3 buckets, national download portals — normalises it, and republishes the result as one global, unified Terrarium raster tileset distributed as PMTiles, with a coverage map and full attribution for every contributing dataset.

Its source catalog is the best inventory of open elevation data anywhere: 153 sources across 32 countries, each with a licence, a resolution and a file list. Working through it makes the shape of the problem obvious — those 153 sources amount to roughly 690,000 individual files, overwhelmingly .tif, .asc and .zip bulk downloads. Only a handful are a single fetchable file, and only a minority of countries expose a live API at all.

That is the division of labour between the two projects. Mapterhorn gives you consistent global terrain with no per-country special cases. The sources below skip the ingest entirely and stream from the agency in real time, which keeps the native resolution — 0.5 m where the agency publishes 0.5 m — at the cost of only working country by country, at whatever speed that government's server manages. Mapterhorn is the sensible default; these are for when you want the sharpest possible data over one country.

The Mapterhorn issue tracking new data sources is also where several of the endpoints below were first discussed.

How these compare to Mapterhorn

vs Mapterhorn	Count	Sources
⬆️ Not in Mapterhorn	5	HKG 5 m vs 30 m global · HTI 1.5 m vs 30 m global · IDN 8 m vs 30 m global · MEX 15 m vs 30 m global · URY 30 m vs 30 m global
⬆️ Finer here	7	CZE 0.5 m vs 5 m · CZE 2 m vs 5 m · NLD 0.5 m vs 5 m · NOR 0.25 m vs 1 m
↔️ Same as Mapterhorn	9	CHE ? m vs 0.5 m · EST 1 m · FRA 0.5 m · GBR 1 m · NZL 1 m · USA 1 m
⬇️ Finer in Mapterhorn	11	CAN 30 m vs 2 m · CHE 10 m vs 0.5 m · ESP 5 m vs 0.5 m · FIN 10 m vs 2 m · FRA 1 m vs 0.5 m · FRO 25 m vs 10 m · ITA 10 m vs 1 m · JPN 5 m vs 1 m

Only a minority are an actual upgrade. That is worth stating plainly, because it is not obvious why a national agency's own API would ever be coarser than a re-published global tileset — and the reason is that the two get their data by different routes.

Mapterhorn ingests whatever an agency publishes in bulk: multi-terabyte download archives, S3 buckets, hundreds of thousands of tiles. Agencies routinely publish their finest LiDAR that way while exposing only a coarser, cheaper product over a live API — Italy publishes 1 m regional LiDAR for download but serves TINITALY at 10 m; Spain's PNOA LiDAR is 0.5 m in bulk while the IDEE tile service is ~5 m. So Mapterhorn can be finer.

It runs the other way too. Norway's hoydedata.no ImageServer resamples a 0.25 m grid on demand, four times finer than the 1 m product Mapterhorn downloaded. And where an agency publishes no bulk data at all, Mapterhorn has nothing to ingest and falls back to global 30 m, so any working API wins by default.

One caveat on both sides: Mapterhorn's delivered detail is also capped by its tileset's max zoom, whereas these APIs resample to whatever window is asked for.

The datasets

API resolution is the grid the live service streams (what the viewer renders and what the grading above uses); Bulk download is the finest grid the agency advertises for download, which is what Mapterhorn itself ingests. The sample picker in the app can regrade by either.

65 national datasets, generated at build time from lib/custom-sources.json.

Regional datasets

Some of the best data on this page covers a state or a province rather than a country. These sit below the divider in the Sample picker and are left out of Load all so the default list stays national, but each is one click away there and loads from a ?terrainSourceA=<id> link.

Beyond those, a second tier was verified to the same standard but deliberately left out of the library, because a national list is more useful than an exhaustive one. Each is a normal endpoint you can paste into Add Custom Source:

Two are worth singling out. Queensland serves a best-available mosaic of 0.5–1 m drone and aircraft LiDAR — ten times finer than the 5 m national grid, and Geoscience Australia publishes no national equivalent at all. And Aguada Fénix is not a mapping agency product: it is the airborne survey that revealed the largest known Maya monumental construction (Inomata et al., Nature 2020), a platform invisible from the ground and unmistakable in a hillshade. That is the case for LiDAR archaeology in one dataset.

Global and polar datasets

Not every good source belongs to a country. These are worldwide, polar or sea-floor products from research consortia, listed under Global in the sample library:

Two things to know. ArcticDEM and REMA are ellipsoidal: the Polar Geospatial Center publishes heights above the WGS84 ellipsoid, so summits sit tens of metres off their map values and open sea reads +28 m off Greenland. And the two ML terrain models, GEDTM30 and ANADEM, are explained in the next section.

Copernicus GLO-30, the supra-national fallback

Most of the world has no national elevation API at all. What fills that gap is Copernicus GLO-30 — a free, global 30 m DEM from ESA and Airbus, and the single most important dataset on this page: it is the base layer of Mapterhorn, the source behind Digital Earth Africa, and the fallback anywhere a country publishes nothing.

Several organisations re-serve the same GLO-30 pixels in different shapes:

Provider	Scope	Served as	Endpoint	Status here
Digital Earth Africa	Africa + Arabia	WCS 2.1, raw Float32	ows.digitalearth.africa	Shipped as a sample source
Mapterhorn	Worldwide	Terrarium PMTiles	tiles.mapterhorn.com	Built in as the default terrain source
OpenTopography	Worldwide	Single global .vrt over COGs	opentopography.s3.sdsc.edu	Not shipped — VRT header is 11.7 MB per read
AWS Open Data	Worldwide	Per-1° COGs	copernicus-dem-30m.s3.amazonaws.com	Not usable — bucket sends no CORS header

ANADEM and GEDTM30 are a newer idea worth understanding, both distributed through OpenTopography. Rather than re-serving GLO-30 unchanged, they apply machine learning to correct it towards bare earth — GLO-30 is a surface model, so over the Amazon it measures the canopy, not the ground, and can sit tens of metres high. ANADEM covers South America; GEDTM30 is global and calibrated against GEDI spaceborne LiDAR. Neither is finer than 30 m — the gain is vertical accuracy under forest, not detail. See OpenTopography's note on ML-derived terrain models and the GEDTM30 discussion.

Digital Earth Africa is the one worth loading directly: it is the only keyless live raw-elevation service covering any African country, and it turns a continent of nothing into a working source. It is not, however, finer than Mapterhorn — it is literally the same GLO-30 data, just fetched live from an API instead of a tileset.

Why these and not others

Surveying national mapping agencies worldwide turns up two consistent traps.

No agency anywhere publishes Terrarium tiles. Only two national agencies publish a pre-rendered elevation tile pyramid at all (Japan's GSI and Mexico's INEGI), and both invented their own encoding — Mexico's is close enough to Terrain-RGB to be described with custom factors, Japan's is not, and Japan only loads because the Geological Survey of Japan re-serves GSI's tiles as genuine Terrain-RGB. Everything else serves raw rasters on demand.

FORMAT=image/geotiff on a WMS is usually a lie. Germany's BKG, the UK Environment Agency, Portugal, Croatia, Ecuador and Canada's MRDEM all advertise it and all return an 8-bit RGB colour ramp — a picture of elevation, not elevation. The raw values live on WCS or on an ArcGIS ImageServer, never on the WMS. A related trap: ArcGIS estates usually expose MapServer (rendered) rather than ImageServer (raw), which is what rules out Slovakia, Colombia, Argentina and Australia.

Some excellent data sits behind a login. NASA's High Mountain Asia 8 m DEM (HMA_DEM8m_MOS, stereo from commercial imagery, 2014–2016, 26–46 N by 67–103 E) is the natural pre-event reference for the Himalaya, and Mapterhorn does not carry it (its Nepal coverage is Copernicus GLO-30). It needs an Earthdata account, which the app cannot supply, so it is not in the library; download it and pick it as a local COG, or serve it yourself.

The two best global commercial DEMs are in the same position, one step further out. Vantor Precision3D (1 m DTM/DSM, the maxar-sdx collection) and Airbus WorldDEM Neo (5 m DSM, airbus-dem) are both browsable through NASA's CSDA catalog in STAC search — the catalog is open and CORS-open, so you can see exactly what exists where — but the assets are s3:// URIs behind an Earthdata login, a EULA and per-request approval, so nothing streams. Getting proper API-key access to both is on the list: Airbus runs a OneAtlas elevation API whose free trial includes the WorldDEM layer, and Vantor sells Precision3D through its own delivery platform. Either would become an ordinary custom source with a key, the way Mapbox and MapTiler already are — the app has nowhere to put those credentials yet because there has been nothing to point them at.

Everything listed above was verified by decoding a real tile and checking it against a known summit or sea-level point before being added.

How they are served

Three different transports end up as the same raster-dem source in the app:

• WMS/WCS raw Float32 — the app requests a GeoTIFF per tile and re-encodes it to Terrarium in the browser, via the float32dem:// protocol. WCS 1.0 works unmodified because its BBOX= is shaped exactly like a WMS GetMap; WCS 2.0 needs a rewrite to subset= per axis, which a &__wcs2subset=X,Y marker on the URL performs.
• ArcGIS ImageServer — exportImage with format=tiff, which returns raw Float32 and takes the same bbox= as a WMS, so it reuses the identical code path (USGS 3DEP, Czechia).
• Single COG — Switzerland and ANADEM are each one national file. Neither is in Web Mercator (LV95 and geographic EPSG:4674), and the in-browser COG reader does not reproject, so these are pinned to titiler with the source's Always serve via titiler switch, whatever the global COG setting says.
• XYZ tiles — a normal tile pyramid. Spain, New Zealand and Japan are standard Terrain-RGB. Mexico is not: its layer is named "terranium" but packs elevation with Terrain-RGB's factors and a baseShift of 1000 instead of 10000, so it is configured with custom RGB encoding and reads ~9 km too low without it.

Surveyed, but not API-usable

These countries do publish a national elevation product — often an excellent one — that still cannot be consumed here. Each row is a reproduced failure, not an assumption.

Two patterns dominate. Auth at the data step: Portugal and Sweden both publish modern, public STAC catalogs under CC BY 4.0 where the index is open and CORS-enabled but every raster byte sits behind OAuth2 or HTTP Basic. Rendered, not raw: the product is a hillshade or a hypsometric colour ramp, sometimes mislabelled as an elevation coverage.

Beyond the table, a long tail of countries publishes nothing retrievable at all. Across Central Asia, the Caucasus, the western Balkans, the Caribbean, Central America, the Middle East, South-East Asia and the Pacific, the national geoportal is typically a dead host, a single-page app with no service directory, or an ArcGIS estate exposing only rendered MapServer layers. Two small countries need no entry of their own: Monaco is inside France's IGN LiDAR coverage and San Marino inside Italy's TINITALY, so both already work through the sources above.

This is exactly the gap Mapterhorn fills. It ingests several of these under licence — Portugal's 0.5 m and Sweden's 1 m are both in its source catalog — so those countries already render at national resolution in this app through the Mapterhorn tileset, just not by talking to the agency live. Greece is the one case where the data isn't open at all, and falls back to global 30 m.

Coverage edges and no-data

A national dataset stops at its border, and services disagree about how to say so. Some return a sentinel (-9999 for France and Finland, -3.4e38 for England, positive +3.4e38 for the Netherlands), some return a service exception, and some return a GeoTIFF that cannot be parsed at all.

Sentinels are handled per-source by a no-data floor/fill, so out-of-coverage areas sit flat at sea level instead of decoding as a 10 km pit. Where a service answers with an error instead, those tiles simply don't render.

Because of this, selecting a national dataset constrains the map to that country's extent by default — see Settings → Map Bounds to change or disable it.