Cesium Terrain Builder
This is a C++ library and associated command line tools designed to create
Cesium can create interactive 3D globes (à la Google Earth) in your web browser
whereby imagery is draped over a model of the underlying terrain. Cesium
provides a number of
different sources for
the terrain data, one of which is height map data for use with the
Cesium Terrain Builder can be used to create the tilesets that sit behind a
terrain server used by
CesiumTerrainProvider. Note that these tools do not
provide a way of serving up those tilesets to the browser: instead
Cesium Terrain Server is
designed to serve terrain tilesets. In particular the Docker
image is designed to simplify the visualisation of terrain tilesets.
Command Line Tools
The following tools are built on top of the C++
This creates gzipped terrain tiles from a GDAL raster representing a
Digital Elevation Model
(DEM), saving the resulting tiles to a directory. It calculates the maximum
zoom level concomitant with the native raster resolution and creates terrain
tiles for all zoom levels between that maximum and zoom level
0 where the
tile extents overlap the raster extents, resampling and subsetting the data as
ctb-tile --output-dir ./terrain-tiles dem.tif
The input raster should contain data representing elevations relative to sea
NODATA (null) values are not currently dealt with: these should be
filled using interpolation in a data preprocessing step.
Note that in the case of multiband rasters, only the first band is used as the
As well as creating terrain tiles, the tool can also be used for generating
tiles in GDAL supported formats using the
--output-format option. This
provides similar functionality to the
gdal2tiles.py script. Tiles can be
created in either Web Mercator or Global Geodetic projections using the
--profile option. e.g.
ctb-tile --output-format JPEG --profile mercator \ --output-dir ./jpeg-tiles RGB-image.tif
An interesting variation on this is to specify
--output-format VRT in order to
generate GDAL Virtual Rasters: these can be useful for debugging and are easily
Usage: ctb-tile [options] GDAL_DATASOURCE Options: -V, --version output program version -h, --help output help information -o, --output-dir <dir> specify the output directory for the tiles (defaults to working directory) -f, --output-format <format> specify the output format for the tiles. This is either `Terrain` (the default) or any format listed by `gdalinfo --formats` -p, --profile <profile> specify the TMS profile for the tiles. This is either `geodetic` (the default) or `mercator` -c, --thread-count <count> specify the number of threads to use for tile generation. On multicore machines this defaults to the number of CPUs -t, --tile-size <size> specify the size of the tiles in pixels. This defaults to 65 for terrain tiles and 256 for other GDAL formats -s, --start-zoom <zoom> specify the zoom level to start at. This should be greater than the end zoom level -e, --end-zoom <zoom> specify the zoom level to end at. This should be less than the start zoom level and >= 0 -r, --resampling-method <algorithm> specify the raster resampling algorithm. One of: nearest; bilinear; cubic; cubicspline; lanczos; average; mode; max; min; med; q1; q3. Defaults to average. -n, --creation-option <option> specify a GDAL creation option for the output dataset in the form NAME=VALUE. Can be specified multiple times. Not valid for Terrain tiles. -z, --error-threshold <threshold> specify the error threshold in pixel units for transformation approximation. Larger values should mean faster transforms. Defaults to 0.125 -m, --warp-memory <bytes> The memory limit in bytes used for warp operations. Higher settings should be faster. Defaults to a conservative GDAL internal setting. -R, --resume Do not overwrite existing files -q, --quiet only output errors -v, --verbose be more noisy
For performance reasons it is recommended that the input raster be in the same
spatial reference system as the output tile grid in order to bypass the need
to reproject the data. For terrain data this is
World Geodetic System
(WGS 84). If the source data is in another spatial reference system, however,
the tool will attempt to reproject the data but with an associated performance
For large rasters a tile based format (as opposed to scanline based) will
drastically speed up processing. A block size that is similar to the tile
output size (i.e. 65x65 for terrain tiles) should be chosen.
Adding overviews to the source dataset will speed up tile generation. The
overview will be chosen whose resolution most closely matches that of the zoom
level being rendered. Overviews will only be downsampled, never upsampled.
As such, it is recommended to use rolutions corresponding to the
Global Geodetic Profile
in the Tile Mapping Service specification. See the
gdaladdotool for creating overviews.
GDAL runtime configuration
options will also affect Cesium Terrain Builder. Specifically the
environment variable should be set to a relatively high value, in conjunction
with the warp memory, if required (see next recommendation).
If warping the source dataset then set the warp memory to a relatively high
value. The correct value is system dependent but try starting your benchmarks
from a value where the combined value of
GDAL_CACHEMAXand the warp memory
represents about 2/3 of your available RAM.
ctb-tilewill resample data from the source dataset when generating
tilesets for the various zoom levels. This can lead to performance issues and
datatype overflows at lower zoom levels (e.g. level 0) when the source dataset
is very large. To overcome this the tool can be used on the original dataset
to only create the tile set at the highest zoom level (e.g. level 18) using
--end-zoomoptions. Once this tileset is generated
it can be turned into a GDAL Virtual Raster dataset for creating the next zoom
level down (e.g. level 17). Repeating this process until the lowest zoom
level is created means that the resampling is much more efficient (e.g. level
0 would be created from a VRT representation of level 1). Because terrain
tiles are not a format supported by VRT datasets you will need to perform this
process in order to create tiles in a GDAL DEM format as an intermediate step.
VRT representations of these intermediate tilesets can then be used to create
the final terrain tile output.
This provides various information on a terrain tile, mainly useful for
Usage: ctb-info [options] TERRAIN_FILE Options: -V, --version output program version -h, --help output help information -e, --show-heights show the height information as an ASCII raster -c, --no-child hide information about child tiles -t, --no-type hide information about the tile type (i.e. water/land)
This exports a terrain tile to GeoTiff
format for use in GIS software. Terrain tiles do not contain information
defining their tile location, so this must be specified through the command
Note that the tool does not normalise the terrain data to sea level but
displays it exactly as it is found in the terrain data.
Usage: ctb-export -i TERRAIN_FILE -z ZOOM_LEVEL -x TILE_X -y TILE_Y -o OUTPUT_FILE Options: -V, --version output program version -h, --help output help information -i, --input-filename <filename> the terrain tile file to convert -z, --zoom-level <int> the zoom level represented by the tile -x, --tile-x <int> the tile x coordinate -y, --tile-y <int> the tile y coordinate -o, --output-filename <filename> the output file to create
Sometimes it is useful to see the extent of coverage of terrain tilesets that
would be produced from a raster. This tool does this by outputting each zoom
level as a GeoJSON file containing the tile extents for
that particular zoom level.
Usage: ctb-extents GDAL_DATASET Options: -V, --version output program version -h, --help output help information -o, --output-dir <dir> specify the output directory for the geojson files (defaults to working directory) -p, --profile <profile> specify the TMS profile for the tiles. This is either `geodetic` (the default) or `mercator` -t, --tile-size <size> specify the size of the tiles in pixels. This defaults to 65 for terrain tiles and 256 for other GDAL formats -s, --start-zoom <zoom> specify the zoom level to start at. This should be greater than the end zoom level -e, --end-zoom <zoom> specify the zoom level to end at. This should be less than the start zoom level and >= 0
libctb is a library implemented in standard C++11. It is capable of creating
terrain tiles according to the
heightmap-1.0 terrain format. It
does not provide a way of serving up or storing the resulting tiles: this is
application specific. Instead its aim is simply to take a
GDAL supported raster format representing a Digital
Terrain Model (DTM) and convert this to terrain tiles.
See the source code for the tools provided with the library
ctb-tile) for examples on how the library is used to achieve
Doxygen based documentation is available for the C++
code: run the
doxygen command in the
doc/ directory and point your browser
Although the software has been used to create a substantial number of terrain
tile sets currently in production use, it should be considered beta quality
software: it needs broader testing, a comprehensive test harness and the API is
liable to change.
The software has primarily been developed and deployed on a Linux OS, and this
is the only officially supported platform. However, it has been reported as
compiling and running on:
Windows using Visual Studio 2010 and 2013 (see
Mac OS X Mavericks using clang (see
Porting it to other systems should be relatively painless as the library
dependencies have been ported to numerous systems and the code itself is
Ensure GDAL >= 2.0.0 is installed. At the time of
writing this is not a stable release so you may need to use a nightly build or
to build the source directly from version control. Specifically, you will need
a version of trunk that has added the min,max,med,q1 and q3 resampling
algorithms. In the subversion repository this is commit 28717 and on the GitHub
mirror this is 0a90a34.
In addition to ensuring the GDAL library is installed, you will need the GDAL
source development header files. You will also need
CMake to be available.
Ensure your system meets the requirements above.
unpack the source.
In the root package directory, assuming you are on a UNIX system, type
mkdir build && cd build && cmake .. && make install.
On a UNIX system you may need to run
ldconfigto update the shared library
Alternatively in step 3 above you can create a debug build by running
-DCMAKE_BUILD_TYPE=Debug ... You can also install to a different location by
CMAKE_INSTALL_PREFIX directive e.g.
Note that if you have GDAL installed in a custom location (e.g under
/home/user/install) it will likely not be found by running
cmake ... In this
case you will need to provide the
GDAL_INCLUDE_DIR directives e.g.
cmake -DGDAL_LIBRARY_DIR=/home/user/install/lib \ -DGDAL_LIBRARY=/home/user/install/lib/libgdal.so \ -DGDAL_INCLUDE_DIR=/home/user/install/include \ ..
The only requirement to getting up and running with Cesium Terrain Builder is
having docker available on your system: all software dependencies, build and
installation issues are encapsulated in the image.
In addition, the
Docker image provides a way of visualising the tilesets created by
Limitations and TODO
Create a comprehensive test harness (possibly using
Bandit, including code coverage and valgrind
Better coordination between threads in
ctb-tileto enable graceful exits if
there is a fatal error or other interrupt.
Add support for the new
quantized-mesh-1.0 terrain format.
ctb-tilecommand currently only outputs files to a directory and
as such is subjected to filesystem limits (e.g. inode limits): it should be
able to output tiles in a format that overcomes these limits and which is
still portable and accessible. SQLite would appear
to be a strong contender.
Provide hooks into the GDAL error handling mechanism to more gracefully
intercept GDAL errors.
Expose tilers using a standard container api (map and/or vector).
Enable more options to be passed to the VRT warper by parameterising
Encapsulate the multithreading tile generation functionality currently
ctb-tilewithin the library to make it more widely
One of the
ctb-tilerecommendations above illustrates a process for
efficiently creating tilesets at lower zoom levels by resampling an already
generated tileset at the next highest zoom level. This could be built
directly into the
ctb-tiletool. An implementation could create a
TiledDatasetdriver (or use a VRT, if it efficiently supports
the large number of tile files) which accesses the already generated tileset;
this dataset could then be used as an input to the tiler.
Add support for interpolating out
NODATAvalues. This could be done using
Adding support for creating water masks to tiles could be useful: at the
moment all tiles are flagged as being of type 'land' (see
Issues and Contributing
Please report bugs or issues using the
GitHub issue tracker.
Code and documentation contributions are very welcome, either as GitHub pull
requests or patches. If you cannot do this but would still like to improve the
software, particularly overcoming the limitations listed above, then please
consider funding further development.
Software developed by GeoData through the
University of Southampton Open Source Geospatial Laboratory.
Thanks to everyone in the community who has contributed to the code base.
Homme Zwaagstra email@example.com