Iris-grib v0.20 (unreleased)

The library iris-grib provides functionality for converting between weather and climate datasets that are stored as GRIB files and Iris Cubes. GRIB files can be loaded as Iris cubes using iris-grib so that you can use Iris for analysing and visualising the contents of the GRIB files. Iris cubes can also be saved to GRIB edition-2 files using iris-grib.

Simple GRIB Loading and Saving with Iris

You can use the functionality provided by iris-grib directly within Iris without having to explicitly import iris-grib, as long as you have both Iris and iris-grib installed in your Python environment.

This is the preferred route if no special control is required.

For example, to load GRIB data :

>>> cube = iris.load_cube('testfile.grib')

Similarly, you can save cubes to a GRIB file directly from Iris :

>>>, 'my_file.grib2')


As the filename suggests, only saving to GRIB2 is currently supported.

Phenomenon translation

iris-grib attempts to translate between CF phenomenon identities (i.e. ‘standard_name’ and possibly ‘long_name’ attributes), and GRIB parameter codes, when converting cubes to or from the GRIB format.

A set of tables define known CF translations for GRIB1 and GRIB2 parameters, and can be interrogated with the functions in iris_grib.grib_phenom_translation.

Parameter loading record

All cubes loaded from GRIB have a GRIB_PARAM attribute, which records the parameter encodings present in the original file message.

Examples :

  • "GRIB2:d000c003n005" represents GRIB2, discipline=0 (“Meteorological products”), category=3 (“Mass”) and indicatorOfParameter=5 (“Geopotential height (gpm)”).

    • This translates to a standard_name and units of “geopotential_height / m”

  • "GRIB1:t002c007n033" is GRIB1 with table2Version=2, centre=7 (“US National Weather Service - NCEP (WMC)”), and indicatorOfParameter=33 (“U-component of wind m s**-1”).

    • This translates to a standard_name and units of “x_wind / m s-1”.

Parameter saving control

When a cube has a GRIB_PARAM attribute, as described above, this controls what the relevant message keys are set to on saving. (N.B. at present applies only to GRIB2, since we don’t support GRIB1 saving)

Iris-grib Load and Save API

In addition to direct load and save with Iris, as described above, it is also possible to load and save GRIB data using iris-grib functions.

Loading and saving Cubes


To load from a GRIB file with iris-grib, you can call the load_cubes() function :

>>> cubes_iter = iris_grib.load_cubes('testfile.grib')
>>> print(cubes_iter)
<generator object load_cubes at ...>

As we can see, this returns a generator object. The generator object may be iterated over to access all the Iris cubes loaded from the GRIB file, or converted directly to a list:

>>> cubes = list(cubes_iter)
>>> print(cubes)
[<iris 'Cube' of air_temperature / (K) (projection_y_coordinate: 200; projection_x_coordinate: 247)>]

In effect, this is the same as using iris.load_raw(...). So, in most cases, that is preferable.


To use iris-grib to save Iris cubes to a GRIB file we can make use of the save_grib2() function :

>>> iris_grib.save_grib2(cube, 'my_file.grib2')

In effect, this is the same as using, ...). So, in most cases, that is preferable.

Working with GRIB messages

Iris-grib also provides lower-level functions which allow the user to inspect and adjust actual GRIB encoding details, for precise custom control of loading and saving.

These functions use intermediate objects which represent individual GRIB file “messages”, with all the GRIB metadata.

For example:

  • correct loading of some messages with incorrectly encoded parameter number

  • save messages with adjusted parameter encodings

  • load messages with an unsupported parameter definition template : adjust them to mimic a similar type which is supported by cube translation, and post-modify the resulting cubes to correct the Iris metadata

You can load and save messages to and from files, and convert them to and from Cubes.


at present this only works with GRIB2 data.


Messages are not represented in the same way for loading and saving : the messages generated by loading from files are represented by iris_grib.message.GribMessage objects, whereas messages generated from cubes, for saving to files, are represented as message handles from the Python eccodes library .


The key functions are load_pairs_from_fields() and messages_from_filename(). See those for more detail.

You can load data to ‘messages’, and filter or modify them to enable or correct how Iris converts them to ‘raw’ cubes (i.e. individual 2-dimensional fields).

For example:

>>> from iris_grib.message import GribMessage
>>> fields_iter = GribMessage.messages_from_filename('testfile.grib')
>>> # select only wanted data
>>> selected_fields = [
...   field
...   for field in fields_iter
...   if field.sections[4]['parameterNumber'] == 33
... ]
>>> cube_field_pairs = iris_grib.load_pairs_from_fields(selected_fields)

Filtering fields can be very useful to speed up loading, since otherwise all data must be converted to Iris before selection with constraints, which can be quite costly.


The key functions are save_pairs_from_cubes() and save_messages(). See those for more detail.

You can convert Iris cubes to eccodes messages, and modify or filter them before saving.


The messages here are eccodes message “ids”, essentially integers, and not GribMessages. Thus, they must be inspected and manipulated using the eccodes library functions.

For example:

>>> # translate data to grib2 fields
>>> cube_field_pairs = list(iris_grib.save_pairs_from_cube(cube_height_2m5))
>>> # adjust some of them
>>> for cube, field in cube_field_pairs:
...   if cube.coords('height') and cube.coord('height').points[0] == 2.5:
...     # we know this will have been rounded, badly, so needs re-scaling.
...     assert eccodes.codes_get_long(field, 'scaleFactorOfFirstFixedSurface') == 0
...     assert eccodes.codes_get_long(field, 'scaledValueOfFirstFixedSurface') == 2
...     eccodes.codes_set_long(field, 'scaleFactorOfFirstFixedSurface', 1)
...     eccodes.codes_set_long(field, 'scaledValueOfFirstFixedSurface', 25)
>>> # save to file
>>> messages = [msg for (cube, msg) in cube_field_pairs]
>>> iris_grib.save_messages(messages, 'temp.grib2')
>>> # check result
>>> print(iris.load_cube('temp.grib2').coord('height').points)

Getting Started

To ensure all iris-grib dependencies, it is sufficient to have installed Iris itself, and ecCodes .

The simplest way to install is with conda , using the package on conda-forge , with the command:

$ conda install -c conda-forge iris-grib

Pip can also be used, to install from the package on PyPI , with the command:

$ pip install iris-grib

Development sources are hosted at .


For recent changes, see Release Notes .

Indices and tables


See also: