NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI · #4855 most downloaded on PyPI
Python interface to map GRIB files to the NetCDF Common Data Model following the CF Convention using ecCodes.
Last release 1 years ago
30 Sep 2025
Release timing varies
gaps range from 3 weeks to 11 months
Most releases are documented
notes for 49 of 60 stable releases
Nothing withdrawn
no release was ever pulled
8 years old
61 releases · first in 2018
Set default dtype for step to timedelta64[ns] for compatibilty with future versions of xarray. See #427 _.
Added values_dtype argument to open_dataset() to allow control over the type of numpy array used for the values array (default is np.dtype("float32"))
Added values_dtype argument to open_dataset() to allow control over the type of numpy array used for the values array (default is np.dtype("float32")). Usage: ds = xr.open_dataset("data.grib", engine="cfgrib", backend_kwargs={"values_dtype": np.dtype("float64")},) See #407.
One column per quarter.
Fixed compatibility with xarray 2024.09.0 See #401 _.
Fixed compatibility with xarray 2024.09.0 See #401.
Feature/dependabot alerts by @shahramn in https://github.com/ecmwf/cfgrib/pull/389
Full Changelog: https://github.com/ecmwf/cfgrib/compare/0.9.13.0...0.9.14.0
Added coords_as_attributes argument to open_dataset() to allow selected dimensions to be stored as attributes rather than dimensions, allowing more heterogeneous data to be encoded as an xarray dataset. See #394.
Added valid_month dimension if monthlyVerificationDate and validityTime are available. See #393.
Added uvRelativeToGrid to list of GRIB keys read by default. See #379.
Allow users to pass of list of values to filter a key by. See #384 _.
fixed issue where GRIB messages with non-hourly steps could not be read See #370 _.
fixed issue where GRIB messages with non-hourly steps could not be read See #370.
added automatic caching of geographic coordinates for improved performance See #341 _.
added automatic caching of geographic coordinates for improved performance See #341.
fixed issue where to_grib() could crash if given a dataset with a single-valued dimension See #347.
fixed issue where values could not be extracted when alternativeRowScanning=1 and grid is not represented as 2D See #358.
fixed issue where the grib_errors parameter was not being handled correctly. This parameter has now been renamed to errors. See #349.
dropped support for Python 3.6. See #363.
added --var-encoding-json (or -v) option to the to_netcdf tool, e.g. cfgrib to_netcdf -v '{"dtype": "float", "scale_factor": 0.1}' -o $OUTFILE $INFILE
added --var-encoding-json (or -v) option to the to_netcdf tool, e.g. cfgrib to_netcdf -v '{"dtype": "float", "scale_factor": 0.1}' -o $OUTFILE $INFILE See #334.
fix issue where xarrays derived from Gaussian grids did not have the correct geometry when written back out as GRIB See #330.
fix issue where open_datasets() could merge different GRIB fields that have the same data values See #336.
large reduction in memory leak See #320 _.
added --netcdf_kwargs_json option to 'cfgrib to_netcdf' See #294 _.
Fix failure to read index files. See #292 _.
Fix failure to read index files. See #292.
Allow backend kwargs to be provided in the to_netcdf executable, either via a json format string, or a path to a json file via -b. See #288.
Fixed issue where the use of relpath() could cause a problem on Windows. See #284.
Fix passing of pathlib.Path. See #282.
Fixed issue where writing an ensemble number into a GRIB file caused an error. See #278.
Big internal refactor to add support for a generic Fieldset similar to Metview. See #243 _.
Big internal refactor to add support for a generic Fieldset similar to Metview. See #243.
Fix the plugin interface that was missing extra_coords. See #231 _.
Depend on the ECMWF eccodes python package _ to access the low level ecCodes C-library, dropping all other GRIB decoding options. See: #95 _, #14 _. #
Depend on the ECMWF eccodes python package to access the low level ecCodes C-library, dropping all other GRIB decoding options. See: #95, #14. #204, #147 and #141.
Many performance improvements during the generation of the index and during data access. See: #142 and #197.
filter_by_keys now can select on all keys known to ecCodes without the need to add non default ones to read_keys explicitly. See: #187.
Include support for engine="cfgrib" using xarray 0.18+ new backend API. See: #216.
Fixed issue where could not load a GRIB message that has only one grid point. See: #199.
Decode level coordinates as float in all cases, fixed issue with non-int levels. See: #195.
Simpler and clearer messages in the event of errors.
Simpler and clearer messages in the event of errors.
Use ECCODES_DIR environment variable if present. Ported from eccodes-python by xavierabellan. See: #162.
Fix using current ecCodes bindings when setting CFGRIB_USE_EXTERNAL_ECCODES_BINDINGS=1.
Use ecmwflibs if present to find the *ecCodes* installation.
Use ecmwflibs if present to find the ecCodes installation.
Added support for indexingDate, indexingTime time coordinates.
Added support for indexingDate, indexingTime time coordinates.
lambert_azimuthal_equal_area grids are now returned as 2D arrays. See: #119.
Add support for MULTI-FIELD messages used in some GRIB products to store u and v components of wind (e.g. GFS, NAM, etc). This has been the single mos
Always open GRIB files in binary mode, by @b8raoult
Always open GRIB files in binary mode, by @b8raoult
Add support of experimental pyeccodes low-level driver by @b8raoult
Add support of experimental pyeccodes low-level driver by @b8raoult
Add support for forecastMonth in cf2cdm.translate_coords.
Add support for forecastMonth in cf2cdm.translate_coords.
- Fix the README.
Fix the README.
Deprecate ensure_valid_time and the config option preferred_time_dimension that are now better handled via time_dims.
Deprecate ensure_valid_time and the config option preferred_time_dimension that are now better handled via time_dims.
Add more options to time_dims forecasts products may be represented as ('time', 'verifying_time') or ('time', 'forecastMonth'). See: #97 _.
Add more options to time_dims forecasts products may be represented as ('time', 'verifying_time') or ('time', 'forecastMonth'). See: #97.
Add support for selecting the time coordinates to use as dimensions via time_dims. Forecasts products may be represented as ('time', 'step') (the defa
Add support for selecting the time coordinates to use as dimensions via time_dims. Forecasts products may be represented as ('time', 'step') (the default), ('time', 'valid_time') or ('valid_time', 'step'). See: #97.
Reduce the in-memory footprint of the FieldIndex and the size of .idx files.
Add support to read additional keys from the GRIB files via read_keys, they appear in the variable attrs and you can filter_by_keys on them. This is a
Fix a bytes-in-the-place-of-str bug when attempting to write a GRIB on Windows. See: #91 _.
Much improved cfgrib.open_datasets heuristics now reads many more heterogeneous GRIB files. The function is now a supported API. See: #63 _, #66 _, #7
Drop support for Python 2, in line with *xarray* 0.12.0. The 0.9.6.x series will be supported long term for Python 2 users. See: #69 _.
Drop support for Python 2, in line with xarray 0.12.0. The 0.9.6.x series will be supported long term for Python 2 users. See: #69.
Sync internal ecCodes bindings API to the one in eccodes-python. See: #81.
Source code has been formatted with black -S -l 99.
Added initial support for spectral coordinates.
Improve merging of variables into a dataset. See: #63 _.
Improve merging of variables into a dataset. See: #63.
Fix an issue in the README format.
Fix an issue in the README format.
Fixed (for real) MULTI-FIELD messages, See: #45 _.
Fixed (for real) MULTI-FIELD messages, See: #45.
Added a protocol version to the index file. Old *.idx files must be removed.
Fix an important typo in the README. See: #64 _.
Fix an important typo in the README. See: #64.
Add support for *Windows* by installing *ecCodes* via *conda*. See: #7 _.
Fixed a serious bug in the computation of the suggested filter_by_keys for non-cubic GRIB files. As a result cfgrib.xarray_store.open_datasets was not
Fixed a serious bug in the computation of the suggested filter_by_keys for non-cubic GRIB files. As a result cfgrib.xarray_store.open_datasets was not finding all the variables in the files. See: #54.
Fixed a serious bug in variable naming that could drop or at worse mix the values of variables. Again see: #54.
Re-opened #45 as the fix was returning wrong data. Now we are back to dropping all variable in a MULTI-FIELD except the first.
Do not set explicit timezone in units to avoid crashing some versions of *xarray*. See: #44 _.
Do not set explicit timezone in units to avoid crashing some versions of xarray. See: #44.
Enable ecCodes implicit MULTI-FIELD support by default, needed for NAM Products by NCEP. See: #45 _.
Enable ecCodes implicit MULTI-FIELD support by default, needed for NAM Products by NCEP. See: #45.
Added support for depthBelowLand coordinate.
Add support for building valid_time from a bad time-step hypercube.
Add support for building valid_time from a bad time-step hypercube.
Also convert is valid_time can index all times and steps in translate_coords.
Also convert is valid_time can index all times and steps in translate_coords.
Set valid_time as preferred time dimension for the CDS data model.
Set valid_time as preferred time dimension for the CDS data model.
Fall back to using the generic GRIB2 ecCodes template when no better option is found. See: #39.
Fix the crash when using cf2cdm.translate_coords on datasets with non-dimension coordinates. See: #41 _.
Drop support for *xarray* versions prior to *v0.11* to reduce complexity. (This is really only v0.10.9). See: #32 _.
Drop support for xarray versions prior to v0.11 to reduce complexity. (This is really only v0.10.9). See: #32.
Declare the data as CF-1.7 compliant via the Conventions global attribute. See: #36.
Tested larger-than-memory and distributed processing via dask and dask.distributed. See: #33.
Promote write support via cfgrib.to_grib to Alpha. See: #18.
Provide the cf2cdm.translate_coords utility function to translate the coordinates between CF-compliant data models, defined by out_name, units and store_direction. See: #24.
Provide cfgrib.__version__. See: #31.
Raise with a better error message when users attempt to open a file that is not a GRIB. See: #34.
Make 2D grids for rotated_ll and rotated_gg gridType's. See: #35.
Nothing published for this version
- Fix formatting for PyPI page.
Fix formatting for PyPI page.
Saves one index file per set of index_keys in a much more robust way.
Saves one index file per set of index_keys in a much more robust way.
Refactor CF-encoding and add the new encode_cf option to backend_kwargs. See: #23.
Refactor error handling and the option to ignore errors (not well documented yet). See: #13.
Do not crash on gridType not fully supported by the installed ecCodes See: #27.
Several smaller bug fixes and performance improvements.
Assorted README fixes, in particular advertise index file support as alpha.
Assorted README fixes, in particular advertise index file support as alpha.
Big performance improvement: add alpha support to save to and read from disk the GRIB index produced by the full-file scan at the first open. See: #20
Big performance improvement: add alpha support to save to and read from disk the GRIB index produced by the full-file scan at the first open. See: #20.
Rename coordinate air_pressure to isobaricInhPa for consistency with all other vertical level coordinates. See: #25 _.
Rename coordinate air_pressure to isobaricInhPa for consistency with all other vertical level coordinates. See: #25.
- Fix PyPI description.
Fix PyPI description.
Change the usage of cfgrib.open_dataset to allign it with xarray.open_dataset, in particular filter_by_key must be added into the backend_kwargs dicti
Change the usage of cfgrib.open_dataset to allign it with xarray.open_dataset, in particular filter_by_key must be added into the backend_kwargs dictionary. See: #21.
Beta release with read support.
Beta release with read support.
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Your coding agent can read these notes before it upgrades. Set up the MCP server →