Teaching: 20 min
Exercises: 10 min
  • What functionality does GDAL offer?

  • What raster dataset formats are supported?

  • How do I interact with raster data within a python program?

  • Understand the basic components of a raster dataset and how to access them with GDAL

  • Read from and write to raster datasets

  • Perform numpy operations on a raster’s values

What is GDAL (Geospatial Data Abstraction Library)?

Supported Formats:

As of the latest version of GDAL, 142 formats are supported, but builds for various platforms may omit support for some formats. The docker images for this tutorial include a GDAL build with support for 124 formats, including a few that have both raster and vector layers.

Not all formats are supported equally

GDAL’s drivers do not support all formats equally, and differences are listed in gdalinfo --formats. * ro - read-only support * rw - reading and writing to existing files (and making copies) supported * rw+ - reading, writing and creation of new files supported * v - the format supports streaming through virtual filesystem API. Streaming sources include compressed archives (such as .tar.gz) and remote file servers (such as HTTP, FTP) * s - the format support subdatasets

$ gdalinfo --formats
Supported Formats:
  VRT -raster- (rw+v): Virtual Raster
  GTiff -raster- (rw+vs): GeoTIFF
  NITF -raster- (rw+vs): National Imagery Transmission Format
  RPFTOC -raster- (rovs): Raster Product Format TOC format
  # There are lots more, results depend on your build

Specifics about each format can be found with the --format parameter. Let’s take a look at the GeoTiff format, which is well-supported and has many possible creation options.

$ gdalinfo --format GTiff

GDAL also has prose documentation available for each format, including detailed information about creation options available on their website at: http://www.gdal.org/formats_list.html.

GDAL CLI Utilities

GDAL provides a number of command-line utilities to automate common processes. To see the utilities available on your system:

$ ls -la /usr/local/bin/gdal*

A full listing of GDAL utilities is available on the GDAL website. A few that might be particularly useful to you include:

Each of these GDAL utilities automate certain pieces of commonly-used functionality. Python scripts can also be used as examples for how to use the GDAL python API for those writing software that interfaces directly with the SWIG API.

Access to GDAL libraries

GDAL is a C++ library, but you don’t need to write your software in C++ to use it. Official bindings are available for other languages: Python PyPI, C#, Ruby, Java, Perl, and PHP.

Of course, you can write your software in C++ if you like :) For the purposes of this tutorial, we’ll interact with the official python bindings.

A note about pythonic behavior

There are several ways that GDAL’s bindings behave that may catch you by surprise. This tutorial will avoid most of them, but it’s good to know that these exist if you end up writing software that uses GDAL. For the full list of gotchas, take a look at https://trac.osgeo.org/gdal/wiki/PythonGotchas

Sample datasets

We’ll use a couple of datasets for this tutorial, all of which are in /data on the geohackweek2016/raster docker image.

These datasets are projected in WGS84/UTM zone 11N, as they are located in the southern Sierra Nevada mountains in California.


ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) is a remote sensor operated as a collaboration between NASA and Japan’s Ministry of Economy, Trade and Industry (METI). Located on board the Terra satellite that was launched in 1999, raw ASTER data provides imagery across 14 bands at resolutions between 15 and 90 meters. The sample datasets we’ll be using today are from the ASTER Global DEM (GDEM) version 2, and have been projected into the WGS84/UTM11N coordinate system.

/data/N38W120.tif /data/N37W120.tif
ASTER GDEM is a product of METI and NASA.  

Let’s take a look at one of these rasters with gdalinfo:

$ gdalinfo /data/N38W120.tif

Note a few relevant details about the raster:

* Only 1 band in this raster
* Data type of raster is 16-bit integer
* The raster is compressed with ``DEFLATE`` mode
* Pixel values range from 1272 - 3762
* Note block size is a whole row of pixels.

Land-Use / Land-Cover

This Land-use/land-cover raster is a subset of a global climatology analysis based on Moderate-resolution imaging spectroradiometer (MODIS) datasets at a 1km scale. The MODIS sensor is on board the same satellite as the ASTER sensor, and records data at a scale between 250m and 1km across 36 bands. This raster dataset has been clipped to a region that includes the southern portion of the Sierra Nevada mountain range.

Source: USGS

Using GDAL

We’ll interact primarily with GDAL through their official python bindings.

Begin by importing these libraries

from osgeo import gdal

Open the dataset

First we open the raster so we can explore its attributes, as in the introduction. GDAL will detect the format if it can, and return a gdal.Dataset object.

ds = gdal.Open('/data/N37W120.tif')

You’ll notice this seemed to go very fast. That is because this step does not actually ask Python to read the data into memory. Rather, GDAL is just scanning the contents of the file to allow us access to certain critical characteristics.

Filepath encodings:

GDAL expects the path to the raster to be ASCII or UTF-8, which is a common filesystem encoding on linux and macs. Windows is often ISO-8859-1 (Latin-1), or else uses a codepage most similar to your locale. This will only be an issue with python 2.x, as the str type in python3 is assumed to be encoded as UTF-8.

Note on Error Handling:

While GDAL’s python bindings have become much more pythonic, errors are not automatically raised as python exceptions. Instead, the default error behavior is to print a message to your stdout console and return None from the operation.

If you experience AttributeError: 'NoneType' object has no attribute 'foo', this may be why.

Pythonic Error Handling


Inspecting the Dataset:

Since rasters can be very, very large, GDAL will not read its contents into memory until we need to access those values. Various formats support different amounts of data (GeoTiff, should support exabytes-large files), so GDAL provides various methods to access attributes of the dataset without reading in all of the pixel values.

As with any python library, the methods available from the ds object can be read through with help().



Since GDAL will happily open any dataset it knows how to open, one of the attributes we can query is the driver that was used:


You’ll recognize these names from when we previously looked into gdalinfo --format(s). Either of these format names are acceptable for internal GDAL or if we’re creating our own raster datasets.

Coordinate Reference System:

Each dataset can have a coordinate reference system defined, which we can retrieve as Well-Known Text (WKT).


This isn’t especially easy to read, but it is a valid projection string. We can also print this nicely with a couple of extra steps, using the osgeo.osr module for manipulating spatial references.

from osgeo import osr

raster_wkt = ds.GetProjection()
spatial_ref = osr.SpatialReference()
print spatial_ref.ExportToPrettyWkt()

If you’re familiar with GIS software, the PROJ.4 projection string may be more useful:

print spatial_ref.ExportToProj4()


Dimensions are accessed as attributes of the Dataset class. X size represents the number of columns, Y size represents the number of rows.


Block Sizes

Rasters are stored on disk in contiguous chunks called blocks, which become very useful when trying to optimize your application for speed of execution. We’ll cover more on that later, but for now, you can access the block size like so:


In this case, the blocksize is one row at a time, but different rasters can be laid out differently on disk (represented by different block sizes). The landcover raster, for example, has a very different blocksize:

lulc_ds = gdal.Open('/data/landcover.tif')

LULC blocksizes

Reading Raster Values

Several GDAL objects have ReadAsArray() methods:

array = ds.ReadAsArray()

In our case, the raster /data/N37W120.tif only contains a single band, so both ds.ReadAsArray() and band.ReadAsArray() will return the same data. The number of bands can be checked before loading the array:


Even if there is only 1 band in the raster, you can still retrieve the band object before accessing the raster’s array. Note that some attributes, especially nodata value and band-specific metadata. For a full listing of band attributes, see the GDAL Data Model documentation and the python API docs.

band = ds.GetRasterBand(1)
array = band.ReadAsArray()
nodata = band.GetNoDataValue()
metadata = band.GetMetadata()

While ReadAsArray() can be used to read the whole array into memory, you can also specify a subset of the raster or band to open with a few optional parameters to ReadAsArray().

band = ds.GetRasterBand(1)
full_array = band.ReadAsArray()

# Start at index (100, 100) and read in an array 250 pixels wide
array_part = band.ReadAsArray(

Note that while the size of array_part is exactly what we requested, GDAL has to read in an enormous number of pixels in order to obtain the small, requested subset.

Blocks read in and discarded for a small window of values

Copying Raster Datasets

Copying files without GDAL:

import os

Copying files with GDAL CLI utilities

gdalmanage copy /data/N37W120.tif /tmp/N37W120_copy.tif

Copying files with GDAL SWIG bindings

from osgeo import gdal
driver = gdal.GetDriverByName('GTiff')
new_ds = driver.CreateCopy('/path/to/new_raster.tif', ds)
new_ds = None  # flush the dataset to disk and close the underlying objects.

Creating new files with GDAL

from osgeo import gdal
driver = gdal.GetDriverByName('GTiff')
new_ds = driver.Create(
    400,  # xsize
    600,  # ysize
    1,    # number of bands
    gdal.GDT_Float32,  # The datatype of the pixel values
    options=[  # Format-specific creation options.
        'BLOCKXSIZE=256',  # must be a power of 2
        'BLOCKYSIZE=256'   # also power of 2, need not match BLOCKXSIZE

# fill the new raster with nodata values

# When all references to new_ds are unset, the dataset is closed and flushed to disk
new_ds = None

Writing to a raster

Unlike reading to an array, writing only happens at the band level.

writeable_ds = gdal.Open('/path/to/raster.tif', gdal.GA_Update)
band = writeable_ds.GetRasterBand(1)

array = band.ReadAsArray()
array += 1

band = None
ds = None

Other libraries that make use of GDAL:

Two libraries in particular extend GDAL to provide a more pythonic interface.

For building geospatial web applications based on Django, be sure to check out GeoDjango:

How many square meters does /data/landcover.tif cover?

lulc_ds = gdal.Open('/data/landcover.tif')
geotransform = lulc_ds.GetGeoTransform()
coverage_in_m = (lulc_ds.RasterXSize * math.abs(geotransform[1]) +
  lulc_ds.RasterYSize *  math.abs(geotransform[5]))

Key Points