COG – Cloud Optimized GeoTIFF generator

New in version 3.1.

Driver short name


Driver built-in by default

This driver is built-in by default

This driver supports the creation of Cloud Optimized GeoTIFF (COG)

It essentially relies upon the GTiff – GeoTIFF File Format driver with the COPY_SRC_OVERVIEWS=YES creation option, but automatically does the needed preprocessing stages (reprojection if asked and creation of overviews on imagery and/or mask) if not already done, and also takes care of morphing the input dataset into the expected form when using some compression types (for example a RGBA dataset will be transparently converted to a RGB+mask dataset when selecting JPEG compression)

Driver capabilities

Supports CreateCopy()

This driver supports the GDALDriver::CreateCopy() operation

Supports Georeferencing

This driver supports georeferencing

Supports VirtualIO

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

Creation Options

General creation options

  • BLOCKSIZE=n: Sets the tile width and height in pixels. Defaults to 512.


    • JPEG should generally only be used with Byte data (8 bit per channel). But if GDAL is built with internal libtiff and libjpeg, it is possible to read and write TIFF files with 12bit JPEG compressed TIFF files (seen as UInt16 bands with NBITS=12). See the “8 and 12 bit JPEG in TIFF” wiki page for more details. For the COG driver, JPEG compression for 3 or 4-band images automatically selects the PHOTOMETRIC=YCBCR colorspace with a 4:2:2 subsampling of the Y,Cb,Cr components.

    • LZW, DEFLATE and ZSTD compressions can be used with the PREDICTOR creation option.

    • ZSTD is available when using internal libtiff and if GDAL built against libzstd >=1.0, or if built against external libtiff with zstd support.

    • LERC is available when using internal libtiff.

    • LERC_ZSTD is available when LERC and ZSTD are available.

  • LEVEL=integer_value: DEFLATE/ZSTD/LERC_DEFLATE/LERC_ZSTD compression level. A lower number will result in faster compression but less efficient compression rate. 1 is the fastest. For DEFLATE, 9 is the slowest/higher compression rate (the default is 6) For ZSTD, 22 is the slowest/higher compression rate (the default is 9)

  • MAX_Z_ERROR=threshold: Set the maximum error threshold on values for LERC/LERC_DEFLATE/LERC_ZSTD compression. The default is 0 (lossless).

  • QUALITY=integer_value: JPEG/WEBP quality setting. A value of 100 is best quality (least compression), and 1 is worst quality (best compression). The default is 75. For WEBP, QUALITY=100 automatically turns on lossless mode.

  • NUM_THREADS=number_of_threads/ALL_CPUS: Enable multi-threaded compression by specifying the number of worker threads. Default is compression in the main thread. This also determines the number of threads used when reprojection is done with the TILING_SCHEME or TARGET_SRS creation options. (Overview generation is also multithreaded since GDAL 3.2)

  • PREDICTOR=[YES/NO/STANDARD/FLOATING_POINT]: Set the predictor for LZW, DEFLATE and ZSTD compression. The default is NO. If YES is specified, then standard predictor (Predictor=2) is used for integer data type, and floating-point predictor (Predictor=3) for floating point data type (in some circumstances, the standard predictor might perform better than the floating-point one on floating-point data). STANDARD or FLOATING_POINT can also be used to select the precise algorithm wished.

  • BIGTIFF=YES/NO/IF_NEEDED/IF_SAFER: Control whether the created file is a BigTIFF or a classic TIFF.

    • YES forces BigTIFF.

    • NO forces classic TIFF.

    • IF_NEEDED will only create a BigTIFF if it is clearly needed (in the uncompressed case, and image larger than 4GB. So no effect when using a compression).

    • IF_SAFER will create BigTIFF if the resulting file *might* exceed 4GB. Note: this is only a heuristics that might not always work depending on compression ratios.

    BigTIFF is a TIFF variant which can contain more than 4GiB of data (size of classic TIFF is limited by that value). This option is available if GDAL is built with libtiff library version 4.0 or higher. The default is IF_NEEDED.

    When creating a new GeoTIFF with no compression, GDAL computes in advance the size of the resulting file. If that computed file size is over 4GiB, GDAL will automatically decide to create a BigTIFF file. However, when compression is used, it is not possible in advance to known the final size of the file, so classical TIFF will be chosen. In that case, the user must explicitly require the creation of a BigTIFF with BIGTIFF=YES if the final file is anticipated to be too big for classical TIFF format. If BigTIFF creation is not explicitly asked or guessed and the resulting file is too big for classical TIFF, libtiff will fail with an error message like “TIFFAppendToStrip:Maximum TIFF file size exceeded”.

  • RESAMPLING=[NEAREST/AVERAGE/BILINEAR/CUBIC/CUBICSPLINE/LANCZOS]: Resampling method used for overview generation or reprojection. For paletted images, NEAREST is used by default, otherwise it is CUBIC.

  • OVERVIEW_RESAMPLING=[NEAREST/AVERAGE/BILINEAR/CUBIC/CUBICSPLINE/LANCZOS]: (since GDAL 3.2) Resampling method used for overview generation. For paletted images, NEAREST is used by default, otherwise it is CUBIC. This overrides, for overview generation, the value of RESAMPLING if it specified.

  • WARP_RESAMPLING=[NEAREST/AVERAGE/BILINEAR/CUBIC/CUBICSPLINE/LANCZOS]: (since GDAL 3.2) Resampling method used for reprojection. For paletted images, NEAREST is used by default, otherwise it is CUBIC. This overrides, for reprojection, the value of RESAMPLING if it specified.

  • OVERVIEWS=[AUTO/IGNORE_EXISTING/FORCE_USE_EXISTING/NONE]: Describe the behavior regarding overview generation and use of source overviews.

    • AUTO (default): source overviews will be used if present (even if the dimension of the smallest level is not < 512 pixels), and, if not present, overviews will be automatically generated in the output file.

    • IGNORE_EXISTING: potential existing overviews on the source dataset will be ignored and new overviews will be automatically generated.

    • FORCE_USE_EXISTING: potential existing overviews on the source will be used (even if the dimension of the smallest level is not < 512 pixels). If there is no source overview, this is equivalent to specifying NONE.

    • NONE: potential source overviews will be ignored, and no overview will be generated.


      When using the gdal_translate utility, source overviews will not be available if general options (i.e. options which are not creation options, like subsetting, etc.) are used.

  • GEOTIFF_VERSION=[AUTO/1.0/1.1]: Select the version of the GeoTIFF standard used to encode georeferencing information. 1.0 corresponds to the original 1995, GeoTIFF Revision 1.0, by Ritter & Ruth. 1.1 corresponds to the OGC standard 19-008, which is an evolution of 1.0, which clear ambiguities and fix inconsistencies mostly in the processing of the vertical part of a CRS. AUTO mode (default value) will generally select 1.0, unless the CRS to encode has a vertical component or is a 3D CRS, in which case 1.1 is used.


    Write support for GeoTIFF 1.1 requires libgeotiff 1.6.0 or later.

  • SPARSE_OK=TRUE/FALSE ((GDAL >= 3.2): Should empty blocks be

    omitted on disk? When this option is set, any attempt of writing a block whose all pixels are 0 or the nodata value will cause it not to be written at all (unless there is a corresponding block already allocated in the file). Sparse files have 0 tile/strip offsets for blocks never written and save space; however, most non-GDAL packages cannot read such files. On the reading side, the presence of a omitted tile after a non-empty one may cause optimized readers to have to issue an extra GET request to the TileByteCounts array. The default is FALSE.

File format details

High level

A Cloud optimized GeoTIFF has the following characteristics:

  • TIFF or BigTIFF file

  • Tiled (512 pixels by default) for imagery, mask and overviews

  • Overviews until the maximum dimension of the smallest overview level is lower than 512 pixels.

  • Compressed or not

  • Pixel interleaving for multi-band dataset

  • Optimized layout of TIFF sections to minimize the number of GET requests needed by a reader doing random read access.

Low level

A COG file is organized as the following (if using libtiff >= 4.0.11 or GDAL internal libtiff. For other versions, the layout will be different and some of the optimizations will not be available).

  • TIFF/BigTIFF header/signature and pointer to first IFD (Image File Directory)

  • “ghost area” with COG optimizations (see Header ghost area)

  • IFD of the full resolution image, followed by TIFF tags values, excluding the TileOffsets and TileByteCounts arrays.

  • IFD of the mask of the full resolution image, if present, followed by TIFF tags values, excluding the TileOffsets and TileByteCounts arrays.

  • IFD of the first (largest in dimensions) overview level, if present

  • IFD of the last (smallest) overview level, if present

  • IFD of the first (largest in dimensions) overview level of the mask, if present

  • IFD of the last (smallest) overview level of the mask, if present

  • TileOffsets and TileByteCounts arrays of the above IFDs

  • tile data of the smallest overview, if present (with each tile followed by the corresponding tile of mask data, if present), with leader and trailer bytes

  • tile data of the largest overview, if present (interleaved with mask data if present)

  • tile data of the full resolution image, if present (interleaved with corresponding mask data if present)

Header ghost area

To describe the specific layout of COG files, a description of the features used is located at the beginning of the file, so that optimized readers (like GDAL) can use them and take shortcuts. Those features are described as ASCII strings “hidden” just after the 8 first bytes of a ClassicTIFF (or after the 16 first ons for a BigTIFF). That is the first IFD starts just after those strings. It is completely valid to have ghost areas like this in a TIFF file, and readers will normally skip over them. So for a COG file with a transparency mask, those strings will be:



  • A newline character \n is used to separate those strings.

  • A space character is inserted after the newline following KNOWN_INCOMPATIBLE_EDITION=NO

  • For a COG without mask, the MASK_INTERLEAVED_WITH_IMAGERY item will not be present of course.

The ghost area starts with GDAL_STRUCTURAL_METADATA_SIZE=XXXXXX bytes\n (of a fixed size of 43 bytes) where XXXXXX is a 6-digit number indicating the remaining size of the section (that is starting after the linefeed character of this starting line).

  • LAYOUT=IFDS_BEFORE_DATA: the IFDs are located at the beginning of the file. GDAL will also makes sure that the tile index arrays are written just after the IFDs and before the imagery, so that a first range request of 16 KB will always get all the IFDs

  • BLOCK_ORDER=ROW_MAJOR: (strile is a contraction of ‘strip or tile’) the data for tiles is written in increasing tile id order. Future enhancements could possibly implement other layouts.

  • BLOCK_LEADER=SIZE_AS_UINT4: each tile data is preceded by 4 bytes, in a ghost area as well, indicating the real tile size (in little endian order). See Tile data leader and trailer for more details.

  • BLOCK_TRAILER=LAST_4_BYTES_REPEATED: just after the tile data, the last 4 bytes of the tile data are repeated. See Tile data leader and trailer for more details.

  • KNOWN_INCOMPATIBLE_EDITION=NO: when a COG is generated this is always written. If GDAL is then used to modify the COG file, as most of the changes done on an existing COG file, will break the optimized structure, GDAL will change this metadata item to KNOWN_INCOMPATIBLE_EDITION=YES, and issue a warning on writing, and when reopening such file, so that users know they have broken their COG file

  • MASK_INTERLEAVED_WITH_IMAGERY=YES: indicates that mask data immediately follows imagery data. So when reading data at offset=TileOffset[i] - 4 and size=TileOffset[i+1]-TileOffset[i]+4, you’ll get a buffer with:

    • leader with imagery tile size (4 bytes)

    • imagery data (starting at TileOffsets[i] and of size TileByteCounts[i])

    • trailer of imagery (4 bytes)

    • leader with mask tilesize (4 bytes)

    • mask data (starting at mask.TileOffsets[i] and of size mask.TileByteCounts[i], but none of them actually need to be read)

    • trailer of mask data (4 bytes)


The content of the header ghost area can be retrieved by getting the GDAL_STRUCTURAL_METADATA metadata item of the TIFF metadata domain on the datasett object (with GetMetadataItem())

Tile data leader and trailer

Each tile data is immediately preceded by a leader, consisting of a unsigned 4-byte integer, in little endian order, giving the number of bytes of payload of the tile data that follows it. This leader is ghost in the sense that the TileOffsets[] array does not point to it, but points to the real payload. Hence the offset of the leader is TileOffsets[i]-4.

An optimized reader seeing the BLOCK_LEADER=SIZE_AS_UINT4 metadata item will thus look for TileOffset[i] and TileOffset[i+1] to deduce it must fetch the data starting at offset=TileOffset[i] - 4 and of size=TileOffset[i+1]-TileOffset[i]+4. It then checks the 4 first bytes to see if the size in this leader marker is consistent with TileOffset[i+1]-TileOffset[i]. When there is no mask, they should normally be equal (modulo the size taken by BLOCK_LEADER and BLOCK_TRAILER). In the case where there is a mask and MASK_INTERLEAVED_WITH_IMAGERY=YES, then the tile size indicated in the leader will be < TileOffset[i+1]-TileOffset[i] since the data for the mask will follow the imagery data (see MASK_INTERLEAVED_WITH_IMAGERY=YES)

Each tile data is immediately followed by a trailer, consisting of the repetition of the last 4 bytes of the payload of the tile data. The size of this trailer is not included in the TileByteCounts[] array. The purpose of this trailer is forces readers to be able to check if TIFF writers, not aware of those optimizations, have modified the TIFF file in a way that breaks the optimizations. If an optimized reader detects an inconsistency, it can then fallbacks to the regular/slower method of using TileOffsets[i] + TileByteCounts[i].


gdalwarp src1.tif src2.tif out.tif -of COG
gdal_translate world.tif world_webmerc_cog.tif -of COG -co TILING_SCHEME=GoogleMapsCompatible -co COMPRESS=JPEG