GDAL Data Model

This document attempts to describe the GDAL data model. That is the types of information that a GDAL data store can contain, and their semantics.


A dataset (represented by the GDALDataset class) is an assembly of related raster bands and some information common to them all. In particular the dataset has a concept of the raster size (in pixels and lines) that applies to all the bands. The dataset is also responsible for the georeferencing transform and coordinate system definition of all bands. The dataset itself can also have associated metadata, a list of name/value pairs in string form.

Note that the GDAL dataset, and raster band data model is loosely based on the OpenGIS Grid Coverages specification.

Coordinate System

Dataset coordinate systems are represented as OpenGIS Well Known Text strings. This can contain:

For more information on OpenGIS WKT coordinate system definitions, and mechanisms to manipulate them, refer to the osr_tutorial document and/or the OGRSpatialReference class documentation.

The coordinate system returned by GDALDataset::GetProjectionRef() describes the georeferenced coordinates implied by the affine georeferencing transform returned by GDALDataset::GetGeoTransform(). The coordinate system returned by GDALDataset::GetGCPProjection() describes the georeferenced coordinates of the GCPs returned by GDALDataset::GetGCPs().

Note that a returned coordinate system strings of "" indicates nothing is known about the georeferencing coordinate system.

Affine GeoTransform

GDAL datasets have two ways of describing the relationship between raster positions (in pixel/line coordinates) and georeferenced coordinates. The first, and most commonly used is the affine transform (the other is GCPs).

The affine transform consists of six coefficients returned by GDALDataset::GetGeoTransform() which map pixel/line coordinates into georeferenced space using the following relationship:

    Xgeo = GT(0) + Xpixel*GT(1) + Yline*GT(2)
    Ygeo = GT(3) + Xpixel*GT(4) + Yline*GT(5)

In case of north up images, the GT(2) and GT(4) coefficients are zero, and the GT(1) is pixel width, and GT(5) is pixel height. The (GT(0),GT(3)) position is the top left corner of the top left pixel of the raster.

Note that the pixel/line coordinates in the above are from (0.0,0.0) at the top left corner of the top left pixel to (width_in_pixels,height_in_pixels) at the bottom right corner of the bottom right pixel. The pixel/line location of the center of the top left pixel would therefore be (0.5,0.5).


A dataset can have a set of control points relating one or more positions on the raster to georeferenced coordinates. All GCPs share a georeferencing coordinate system (returned by GDALDataset::GetGCPProjection()). Each GCP (represented as the GDAL_GCP class) contains the following:

typedef struct
    char        *pszId;
    char        *pszInfo;
    double      dfGCPPixel;
    double      dfGCPLine;
    double      dfGCPX;
    double      dfGCPY;
    double      dfGCPZ;

The pszId string is intended to be a unique (and often, but not always numerical) identifier for the GCP within the set of GCPs on this dataset. The pszInfo is usually an empty string, but can contain any user defined text associated with the GCP. Potentially this can also contain machine parsable information on GCP status though that isn't done at this time.

The (Pixel,Line) position is the GCP location on the raster. The (X,Y,Z) position is the associated georeferenced location with the Z often being zero.

The GDAL data model does not imply a transformation mechanism that must be generated from the GCPs ... this is left to the application. However 1st to 5th order polynomials are common.

Normally a dataset will contain either an affine geotransform, GCPs or neither. It is uncommon to have both, and it is undefined which is authoritative.


GDAL metadata is auxiliary format and application specific textual data kept as a list of name/value pairs. The names are required to be well behaved tokens (no spaces, or odd characters). The values can be of any length, and contain anything except an embedded null (ASCII zero).

The metadata handling system is not well tuned to handling very large bodies of metadata. Handling of more than 100K of metadata for a dataset is likely to lead to performance degradation.

Some formats will support generic (user defined) metadata, while other format drivers will map specific format fields to metadata names. For instance the TIFF driver returns a few information tags as metadata including the date/time field which is returned as:

TIFFTAG_DATETIME=1999:05:11 11:29:56

Metadata is split into named groups called domains, with the default domain having no name (NULL or ""). Some specific domains exist for special purposes. Note that currently there is no way to enumerate all the domains available for a given object, but applications can "test" for any domains they know how to interpret.

The following metadata items have well defined semantics in the default domain:


The SUBDATASETS domain holds a list of child datasets. Normally this is used to provide pointers to a list of images stored within a single multi image file.

For example, an NITF with two images might have the following subdataset list.

  SUBDATASET_1_DESC=Image 1 of multi_1b.ntf
  SUBDATASET_2_DESC=Image 2 of multi_1b.ntf

The value of the _NAME is the string that can be passed to GDALOpen() to access the file. The _DESC value is intended to be a more user friendly string that can be displayed to the user in a selector.

Drivers which support subdatasets advertize the DMD_SUBDATASETS capability. This information is reported when the --format and --formats options are passed to the command line utilities.

Currently, drivers which support subdatasets are: ADRG, ECRGTOC, GEORASTER, GTiff, HDF4, HDF5, netCDF, NITF, NTv2, OGDI, PDF, PostGISRaster, Rasterlite, RPFTOC, RS2, TileDB, WCS, and WMS.


Metadata in the default domain is intended to be related to the image, and not particularly related to the way the image is stored on disk. That is, it is suitable for copying with the dataset when it is copied to a new format. Some information of interest is closely tied to a particular file format and storage mechanism. In order to prevent this getting copied along with datasets it is placed in a special domain called IMAGE_STRUCTURE that should not normally be copied to new formats.

Currently the following items are defined by RFC 14 as having specific semantics in the IMAGE_STRUCTURE domain.

RPC Domain

The RPC metadata domain holds metadata describing the Rational Polynomial Coefficient geometry model for the image if present. This geometry model can be used to transform between pixel/line and georeferenced locations. The items defining the model are:

These fields are directly derived from the document prospective GeoTIFF RPC document ( which in turn is closely modeled on the NITF RPC00B definition.

The line and pixel offset expressed with LINE_OFF and SAMP_OFF are with respect to the center of the pixel.

IMAGERY Domain (remote sensing)

For satellite or aerial imagery the IMAGERY Domain may be present. It depends on the existence of special metadata files near the image file. The files at the same directory with image file tested by the set of metadata readers, if files can be processed by the metadata reader, it fill the IMAGERY Domain with the following items:

xml: Domains

Any domain name prefixed with "xml:" is not normal name/value metadata. It is a single XML document stored in one big string.

Raster Band

A raster band is represented in GDAL with the GDALRasterBand class. It represents a single raster band/channel/layer. It does not necessarily represent a whole image. For instance, a 24bit RGB image would normally be represented as a dataset with three bands, one for red, one for green and one for blue.

A raster band has the following properties:

Color Table

A color table consists of zero or more color entries described in C by the following structure:

typedef struct
    /- gray, red, cyan or hue -/
    short      c1;
    /- green, magenta, or lightness -/
    short      c2;
    /- blue, yellow, or saturation -/
    short      c3;
    /- alpha or black band -/
    short      c4;
} GDALColorEntry;

The color table also has a palette interpretation value (GDALPaletteInterp) which is one of the following values, and indicates how the c1/c2/c3/c4 values of a color entry should be interpreted.

To associate a color with a raster pixel, the pixel value is used as a subscript into the color table. That means that the colors are always applied starting at zero and ascending. There is no provision for indicating a pre-scaling mechanism before looking up in the color table.


A band may have zero or more overviews. Each overview is represented as a "free standing" GDALRasterBand. The size (in pixels and lines) of the overview will be different than the underlying raster, but the geographic region covered by overviews is the same as the full resolution band.

The overviews are used to display reduced resolution overviews more quickly than could be done by reading all the full resolution data and downsampling.

Bands also have a HasArbitraryOverviews property which is TRUE if the raster can be read at any resolution efficiently but with no distinct overview levels. This applies to some FFT encoded images, or images pulled through gateways (like OGDI) where downsampling can be done efficiently at the remote point.

Generated for GDAL by doxygen 1.8.8.