[gdal-dev] Support for Terrain-RGB

[Lars Ahlzen]

Good points.

Given these and other comments, it sounds to me like we should support 
different RGB raster formats, as the storage format really is 
independent from the encoding. Also, supporting Mapbox Terrain RGB, 
Terrarium RGB, and custom encodings would be desirable. I believe these 
are all variants of the same thing anyway, with a base (offset) and a 
scale for each of the R, G and B channels.

Even, you probably know GDAL internals quite well, what are your 
thoughts? If we were to support arbitrary RGB raster formats, would an 
implementation elsewhere (like gdaldem) be more suitable than a new 
raster driver, or could a raster driver support different actual file 
formats?

If the latter, could the existing MBTiles driver be used for direct DEM 
-> RGB tiles (.png/.webp/...) -> .mbtiles generation?

- Lars

On 06/01/2026 08:40, Andrey VI wrote:
> Hi all.
>
> It would be great to have RGB DEM support in GDAL.
> Some considerations.
> 1. It makes sense to add support not only for Mapbox Terrain RGB, but 
> also for Mapzen Terrarium RGB [1][2] encoding. Mapbox and Maplibre 
> support both encodings, the latter also supports custom encoding 
> [3][4]. Therefore, support for custom encoding will also be useful.
> 2. I don't quite understand why only PNG format is being discussed 
> here. It is possible to encode a DEM into any raster RGB image 
> format. Moreover, PNG isn't the most optimal of them [5]. In my 
> opinion, the WEBP format is preferable. For example, Maptiler uses it 
> for Terrain RGB and Ocean RGB datasets [6].
> 3. Ideally, the output should be not just an encoded image, but tiles 
> ready for use in Mapbox/Maplibre/etc., including in the form of data 
> sets (MBTiles, PMTiles).
> [1] https://www.mapzen.com/blog/terrain-tile-service/
> [2] https://github.com/tilezen/joerd/blob/master/docs/formats.md#terrarium
> [3] 
> https://docs.mapbox.com/style-spec/reference/sources/#raster-dem-encoding
> [4] https://maplibre.org/maplibre-style-spec/sources/#encoding_1
> [5] 
> https://medium.com/@frederic.rodrigo/optimization-of-rgb-dem-tiles-for-dynamic-hill-shading-with-mapbox-gl-or-maplibre-gl-55bef8eb3d86
> [6] https://www.maptiler.com/on-prem-datasets/dataset/terrain-rgb/ 
> <https://www.maptiler.com/on-prem-datasets/dataset/terrain-rgb/#0.22/0/0>
>
>     Tuesday, January 6, 2026 0:46 AM +03:00 from Lars Ahlzen via
>     gdal-dev <gdal-dev at lists.osgeo.org>:
>     On 05/01/2026 19:47, rayg wrote:
>     > Strange encoding, -10000 meters isn't deep enough for the Marianas
>     > Trench. You could borrow another 10 km from the 1.6 million meters
>     > still available on the positive side.
>     >
>     > Scaling by 0.1 also means a 10 cm vertical resolution., which is
>     > useless for some use cases. The 24 bits per pixel could be much
>     better
>     > allocated.
>
>     I kind of agree with both points, but I guess the format is what
>     it is.
>
>     I don't think it's supposed to be a general-purpose storage
>     format. It's
>     encoding just enough data to do good enough real-time visualization of
>     elevation, like hillshading or hypsometric tints, client side.
>
>     Looks like at least rio-rgbify [1] allows setting a custom base value
>     and interval. That might be something we want to emulate as well,
>     perhaps as creation options (if a raster format driver) or command
>     line
>     options (if part of gdaldem).
>
>     - Lars
>
>     [1] https://github.com/mapbox/rio-rgbify
>
>     _______________________________________________
>     gdal-dev mailing list
>     gdal-dev at lists.osgeo.org
>     https://lists.osgeo.org/mailman/listinfo/gdal-dev
>
>
> --
> Andrey
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/gdal-dev/attachments/20260108/87f80f7d/attachment-0001.htm>