TileDB -- TileDB vector

Driver short name


New in version 3.7.

Build dependencies

TileDB >= 2.7

GDAL can read and write TileDB sparse arrays containing vector data.

The driver relies on the Open Source TileDB library (MIT licensed).

Driver capabilities

Supports Create()

This driver supports the GDALDriver::Create() operation

Supports VirtualIO

This driver supports virtual I/O operations (/vsimem/, etc.)

Supported datasets

The driver can read TileDB sparse arrays that contain at least 2 dimensions of type Float64. By default dimensions names _X and _Y are looked up, but this can be customized with the DIM_X and DIM_Y open options.

Attributes (or extra dimensions) of the following TileDB data types are recognized as OGR fields: Bool, Int16, Int32, Int64, Float32, Float64, String_ASCII, String_UTF8, Blob, varying size UInt8 (as Binary), DateTime_Day, DateTime_MS, Time_MS. Attributes of other data types will be ignored.

If a wkb_geometry named attribute of type Binary is found, it is used as the geometry column, and is assumed to contain WKB encoded geometries. If there is no such column, Point geometries are assumed and the X, Y (and optional Z) dimensions are used to create the corresponding OGR point geometries.

Dataset connection string

Valid dataset connection strings are paths to a local or remote TileDB sparse array, or to a local or remote TileDB group that contains TileDB sparse arrays.


The bounding box of OGR spatial filters is forwarded to the TileDB query engine.

OGR attribute filters are translated to TileDB query conditions for the following elements:

  • AND

  • OR

  • NOT

  • comparisons of the form "attribute_name operator constant" where operator is =, <>, <, <=, >, >=

  • attribute_name IS NULL

  • attribute_name IS NOT NULL

  • attribute_name IN (value1, ... valueN)

  • attribute_name BETWEEN min_val AND max_val

Other OGR attribute filter elements may be used, but this may cause the filter to be fully or partially evaluated on OGR side. For example, given the filter "int_attribute = 5 AND string_attribute LIKE '%foo%'", the condition on int_attribute will be translated as a TileDB query condition, and the right side of the AND operation will be evaluated as a post-filter on OGR side.

Arrow C Stream data interface

The driver has an efficient implementation of the Arrow C Stream data interface

The INCLUDE_FID=YES/NO and MAX_FEATURES_IN_BATCH=number options of OGRLayer::GetArrowStream() are supported. If MAX_FEATURES_IN_BATCH is not specified, it defaults to the value of the BATCH_SIZE open option.

Creation issues

By default, when creating a layer, the driver will create a TileDB sparse array at the location of the dataset connection string. Consequently only one OGR layer can be created in that mode. If several layers need to be created and accessible through a OGR dataset connection, a TileDB group needs to be created to point to the different arrays (layers), by specifying the CREATE_GROUP=YES dataset creation option.

The driver supports appending features to exiting layers.

The driver does not support adding new fields to a layer where features have already been written.

Open options

The following open options are available:

  • TILEDB_CONFIG=<filename>: A local file with TileDB configuration options

  • TILEDB_TIMESTAMP=<integer>: Open array at this timestamp. The timestamp should be greater than 0.

  • BATCH_SIZE=<integer>: Defaults to 500000. Number of features to fetch/write at once.

  • DIM_X=value: Defaults to _X. Name of the X dimension.

  • DIM_Y=value: Defaults to _Y. Name of the Y dimension.

  • DIM_Z=value: Defaults to _Z. Name of the Z dimension.

Dataset creation options

The following dataset creation options are available:

  • TILEDB_CONFIG=<filename>: A local file with TileDB configuration options

  • CREATE_GROUP=[YES/NO]: Defaults to NO. (TileDB >= 2.9) Whether to create a group for multiple layer support. When set to YES, a TileDB group will be created in the directory of the dataset name, and layers will be created as members of the group, and written in subdirectories of a layers subdirectory.

Layer creation options

The following layer options are available:

  • COMPRESSION=[NONE/GZIP/ZSTD/LZ4/RLE/BZIP2/DOUBLE-DELTA/POSITIVE_DELTA]: Defaults to NONE. compression method for dimensions and attributes.

  • COMPRESSION_LEVEL=<integer>: compression level

  • BATCH_SIZE=<integer>: Defaults to 500000. Number of features to write at once.

  • TILE_CAPACITY=<integer>: Defaults to 10000. Number of non-empty cells stored in a data tile.

  • TILE_EXTENT=<float>: The square TileDB tile extents in the X and Y dimensions. Default is auto-calculated.

  • TILE_Z_EXTENT=<float>: The TIleDB tile extent in the Z dimension. Default is auto-calculated.

  • BOUNDS=<minx,miny,[minz,]maxx,maxy[, maxz]>: Specify bounds for sparse array. If not specified, the CRS passed at layer creation will be used to infer default values for bounds.

  • ADD_Z_DIM=[AUTO/YES/NO]: Defaults to AUTO. Whether to add a Z dimension. In the default AUTO mode, a Z dimension is only added if the layer geometry type has a Z component or is unknown. Setting it to YES or NO explicitly force or disable creation of a Z dimension.

  • FID=value: Defaults to FID. Feature id column name. Set to empty to disable its creation.

  • GEOMETRY_NAME=value: Defaults to wkb_geometry. Name of the geometry column that will receive WKB encoded geometries. Set to empty to disable its creation (only for point).

  • TILEDB_TIMESTAMP=<integer>: Timestamp at which to create the array. The timestamp should be greater than 0.

  • TILEDB_STRING_TYPE=[UTF8/ASCII]: Which TileDB type to create string attributes. Default is UTF8 starting with TileDB 2.14 (ASCII for earlier versions)

See Also