Change log

version 3.16.2


version 3.16.1


version 3.16.0


version 3.15.4


version 3.15.3


version 3.15.2


version 3.15.1


version 3.15.0


version 3.14.1


version 3.14.0 (first Dask release)


version 3.13.1 (last LAMA release)


version 3.13.0


version 3.12.0


version 3.11.0


version 3.10.0


version 3.9.0


version 3.8.0


version 3.7.0


  • Python 3.5 support deprecated (3.5 was retired on 2020-09-13)

  • New method: cf.Field.del_domain_axis

  • New method: cf.Field._docstring_special_substitutions

  • New method: cf.Field._docstring_substitutions

  • New method: cf.Field._docstring_package_depth

  • New method: cf.Field._docstring_method_exclusions

  • New keyword parameter to cf.Field.set_data: inplace

  • New keyword parameter to cf.write: coordinates (

  • New keyword parameter to cf.aggregate: ignore (

  • Fixed bug that caused a failure when reading a dataset with incompatible bounds units. Now a warning is given (controllable by the logging level) and the offending bounds are returned as a separate field construct.

  • Fixed bug in cf.aggregate that caused it to error if either the equal_all or exist_all parameter were set to True.

  • Fixed bug in Data.percentile that caused it to error for non-singular ranks if the squeeze parameter was set to True.

  • cfa now prints error messages to the stderr stream rather than stdout.

  • Changed dependency:<=cfdm<

  • Changed dependency: cfunits>=3.3.0

version 3.6.0


version 3.5.1


  • Changed dependency: 1.8.5<=cfdm<1.9.0

  • Fixed bug (emerging from the cfdm library) that prevented the reading of certain netCDF files, such as those with at least one external variable.

version 3.5.0


version 3.4.0


version 3.3.0


version 3.2.0


version 3.1.0


version 3.0.6


version 3.0.5


version 3.0.4


version 3.0.3


version 3.0.2


version 3.0.1


  • Updated description in

version 3.0.0 (first Python 3 release)


  • Complete refactor for Python 3, including some API changes.

    Scripts written for version 2.x but running under version 3.x should either work as expected, or provide informative error messages on the new API usage. However, it is advised that the outputs of older scripts be checked when running with Python 3 versions of the cf library.

  • Deprecated cfdump (its functionality is now included in cfa).

version 2.3.8 (last Python 2 release)


  • In cf.write, can set single=False to mean double=True, and vice versa.

  • Fixed bug in cf.aggregate - removed overly strict test on dimension coordinate bounds.

  • Fixed bug in that set the climatology attribute to True when there are no bounds.

  • Fixed bug in cf.write when writing missing values (set_fill was off, now on)

version 2.3.5


  • Changed calculation of chunksize in parallel case to avoid potential problems and introduced a new method cf.SET_PERFORMANCE to tune the chunksize and the fraction of memory to keep free.

version 2.3.4


  • Fix bug in creating a during cell method during a field collapse.

version 2.3.3


  • Allow failure to compile to go through with a warning, rather than failing to install. if this happens, reading a PP/UM file will result in “Exception: Can’t determine format of file test2.pp”

  • Fixed bug in cf.Field.convolution_filter giving false error over units.

version 2.3.2


  • cf.Field.regridc now compares the units of the source and destination grids and converts between them if possible or raises an error if they are not equivalent.

version 2.3.1


  • Fixed bug in cf.Field.regridc that caused it to fail when regridding a multidimensional field along only one dimension.

  • Fixed bug which in which the default logarithm is base 10, rather than base e

version 2.3.0


  • The collapse method can now be parallelised by running any cf-python script with mpirun if mpi4py is installed. This is an experimental feature and is not recommended for operational use. None of the parallel code is executed when a script is run in serial.

version 2.2.8


  • Bug fix: better handle subspacing by multiple multidimensional items

version 2.2.7


  • Bug fix: correctly set units of bounds when the cf.Data object inserted with insert_bounds has units of ‘’. In this case the bounds of the parent coordinate are now inherited.

version 2.2.6


  • Improved error messages

  • Changed behaviour when printing reference times with a calendar of 'none' - no longer attempts a to create a date-time representation

version 2.2.5


  • Fixed bug with HDF chunk sizes that prevented the writing of large files

version 2.2.4


  • Interim fix for with HDF chunk sizes that prevented the writing of large files

version 2.2.3


  • During writing, disallow the creation of netCDF variable names that contain characters other than letters, digits, and underscores.

version 2.2.3


  • During writing, disallow the creation of netCDF variable names that contain characters other than letters, digits, and underscores.

version 2.2.2


  • Fix for removing duplicated netCDF dimensions when writing data on (e.g.) tripolar grids.

version 2.2.1


  • Fix for calculating are weights from projection coordinates

version 2.2.0


  • Updated for netCDF4 v1.4 cftime API changes

version 2.1.9


  • Allowed invalid units through. Can test with cf.Units.isvalid.

version 2.1.8


version 2.1.7


  • Fixed bug in cf.Field.collapse when doing climatological time collapse with only one period per year/day

version 2.1.6


  • Fixed bug in Variable.mask

version 2.1.4


  • Added override_calendar method to coordinates and domain ancillaries that changes the calendar of the bounds, too.

  • Fixed bug in cf.Data.where when the condition is a cf.Query object.

  • Fixed bug in cf.Variable.mask

version 2.1.3


version 2.1.2


version 2.1.1


  • Disallowed raising offset units to a power (e.g. taking the square of data in units of K @ 273.15).

  • Removed len() of cf.Field (previously always, and misleadingly, returned 1)

  • Fixed setting of cell methods after climatological time collapses

  • Added printing of ncvar in cf.Field.__str__ and cf.Field.dump

  • Added user stash table option to cfa script

version 2.1


  • Misc. bug fixes

version 2.0.6


  • Removed error when finds no fields - an empty field list is now returned

  • New method cf.Field.count

version 2.0.5


  • Bug fix when creating wrap-around subspaces from cyclic fields

  • Fix (partial?) for memory leak when reading UM PP and fields files

version 2.0.4


  • submodel property for PP files

  • API change for cf.Field.axis: now returns a cf.DomainAxis object by default

  • Bug fix in cf.Field.where

  • Bug fix when initialising a field with the source parameter

  • Changed default output format to NETCDF4 (from NETCDF3_CLASSIC)

version 2.0.3


version 2.0.1.post1


  • Bug fix for reading DSG ragged arrays

version 2.0.1


version 2.0


  • First release with full CF data model and full CF-1.6 compliance (including DSG)

version 1.5.4.post4


version 1.5.4.post1


  • removed errant scikit import

version 1.5.4


  • Tripolar regridding

version 1.5.3


  • Updated STASH code to standard_name table (with thanks to Jeff Cole)

  • Fixed bug when comparing masked arrays for equality

version 1.5.2


  • Fixed bug when accessing PP file whose format/endian/word-size has been specified

version 1.5.1


  • Can specify ‘pp’ or ‘PP’ in um option to

version 1.5


  • Changed weights in calculation of variance to reliability weights (from frequency weights). This not only scientifically better, but faster, too.

version 1.4


  • Rounded datetime to time-since conversions to the nearest microsecond, to reflect the accuracy of netCDF4.netcdftime

  • Removed import tests from

  • New option –um to cfa, cfdump

  • New parameter um to

version 1.3.3


  • Rounded datetime to time-since conversions to the nearest microsecond, to reflect the accuracy of netCDF4.netcdftime

  • Fix for netCDF4.__version__ > 1.2.4 do to with datetime.calendar handle with care

version 1.3.2


  • Added –build-id to LDFLAGS in umread Makefile, for sake of RPM builds (otherwise fails when building debuginfo RPM). Pull request #16, thanks to Klaus Zimmermann.

  • Improved test handling. Pull request #21, thanks to Klaus Zimmermann.

  • Removed udunits2 database. This removes the modified version of the udunits2 database in order to avoid redundancies, possible version incompatibilities, and license questions. The modifications are instead carried out programmatically in Pull request #20, thanks to Klaus Zimmermann.

version 1.3.1


version 1.3


  • Removed asreftime, asdatetime and dtvarray methods

  • New method: convert_reference_time for converting reference time data values to have new units.

version 1.2.3


version 1.2.2


  • Fixed bug in binary operations to do with the setting of Partition.part

  • Added cf.TimeDuration functionality to get_bounds cellsizes parameter. Also new parameter flt (“fraction less than”) to position the coordinate within the cell.

version 1.2


  • Added HDF_chunks methods

version 1.1.11


  • Added cellsize option to cf.Coordinate.get_bounds, and fixed bugs.

  • Added variable_attributes option to cf.write

  • Added cf.ENVIRONMENT method

version 1.1.10


  • Added reference_datetime option to cf.write

  • Fixed bug in which incorrectly ordered vertical coordinates

version 1.1.9


  • New methods cf.Variable.files and cf.Data.files, cf.Field.files which report which files are referenced by the data array.

  • Fix to stop partitions return numpy.bool_ instead of numpy.ndarray

  • Fix to determining cyclicity of regridded fields.

  • Functionality to recursively read directories in, cfa and cfump

  • Print warning but carry on when ESMF import fails

  • Fixed bug in cf.Field.subspace when accessing axes derived from UM format files

version 1.1.8


  • Slightly changed the compression API to cf.write

  • Added compression support to the cfa command line script

  • Added functionality to change data type on writing to cf.write and cfa - both in general and for with extra convenience for the common case of double to single (and vice versa).

  • Removed annoying debug print statements from

version 1.1.7


  • Added fix for change in numpy behaviour (numpy.number types do not support assignment)

  • Added capability to load in a user STASH to standard name table:

version 1.1.6


  • Added –reference_datetime option to cfa

  • Bug fix to cf.Field.collapse when providing cf.Query objects via the group parameter

  • Added auto regridding method, which is now the default

version 1.1.5


  • Bug fix in cf.Field.where when using cf.masked

  • conda installation (with thanks to Andy Heaps)

  • Bug fix for type casting in cf.Field.collapse

  • Display long_name if it exists and there is no standard_name

  • Fix for compiling the UM C code on certain OSs (with thanks to Simon Wilson)

  • Fixed incorrect assignment of cyclicity in cf.Field.regrids

  • Nearest neighbour regridding in cf.Field.regrids

version 1.1.4


  • Bug fix to cf.Field.autocyclic

  • Bug fix to cf.Field.clip - now works when limit units are supplied

  • New methods: cf.Data.round, cf.Field.Round

  • Added lbtim as a cf.Field property when reading UM files

  • Fixed coordinate creation for UM atmosphere_hybrid_height_coordinate

  • Bug fix to handling of cyclic fields by cf.Field.regrids

  • Added nearest neighbour field regridding

  • Changed keyword ignore_dst_mask in cf.Field.regrids to use_dst_mask, which is false by default

version 1.1.3


  • Bug fixes to cf.Field.collapse when the “group” parameter is used

  • Correct setting of cyclic axes on regridded fields

  • Updates to STASH_to_CF.txt table: 3209, 3210

version 1.1.2


  • Updates to STASH_to_CF.txt table

  • Fixed bug in decoding UM version in

  • Fixed bug in cf.units.Utime.num2date

  • Fixed go-slow behaviour for silly BZX, BDX in PP and fields file lookup headers

version 1.1.1


  • Fixed bug in decoding UM version in

version 1.1


  • Fixed bug in cf.Units.conform

  • Changed cf.Field.__init__ so that it works with just a data object

  • Added cf.Field.regrids for lat-lon regridding using ESMF library

  • Removed support for netCDF4-python versions < 1.1.1

  • Fixed bug which made certain types of coordinate bounds non-contiguous after transpose

  • Fixed bug with i=True in cf.Field.where and in cf.Field.mask_invalid

  • cyclic methods now return a set, rather than a list

  • Fixed bug in _write_attributes which might have slowed down some writes to netCDF files.

  • Reduced annoying redirection in the documentation

  • Added cf.Field.field method and added top_level keyword to

  • Fixed bug in calculation of standard deviation and variance (the bug caused occasional crashes - no incorrect results were calculated)

  • In items method (and friends), removed strict_axes keyword and added axes_all, axes_superset and axes_subset keywords

version 1.0.3


  • Added default keyword to fill_value() and fixed bugs when doing delattr on _fillValue and missing_value properties.

version 1.0.2


  • PyPI release

version 1.0.1


  • Fixed bug in when using the select keyword to

version 1.0


  • Mac OS support

  • Limited Nd functionality to cf.Field.indices

  • Correct treatment of add_offset and scale_factor

  • Replaced -a with -x in cfa and cfdump scripts

  • added ncvar_identities parameter to cf.aggregate

  • Performance improvements to field subspacing

  • Documentation

  • Improved API to match, select, items, axes, etc.

  • Reads UM fields files

  • Optimised reading PP and UM fields files

  • cf.collapse replaced by cf.Field.collapse

  • cf.Field.collapse includes CF climatological time statistics



  • Fixed bug for changes to netCDF4-python library versions >= 1.1.2

  • Miscellaneous bug fixes

version 0.9.9


  • Added netCDF4 compression options to cf.write.

  • Added __mod__, __imod__, __rmod__, ceil, floor, trunc, rint methods to cf.Data and cf.Variable

  • Added ceil, floor, trunc, rint to cf.Data and cf.Variable

  • Fixed bug in which array cf.Data.array sometimes behaved like cf.Data.varray

  • Fixed bug in which affected reading fields with formula_terms.

  • Refactored the test suite to use the unittest package

  • Cyclic axes functionality

  • Documentation updates



  • Implemented multiple grid_mappings (CF trac ticket #70)

  • Improved functionality and speed of field aggregation and cfa and cfdump command line utilities.

  • Collapse methods on cf.Data object (min, max, mean, var, sd, sum, range, mid_range).

  • Improved match/select functionality



  • Copes with PP fields with 365_day calendars

  • Revamped CFA files in line with the evolving standard. CFA files from PP data created with a previous version will no longer work.

version 0.9.8


  • Improved API.

  • Plenty of speed and memory optimisations.

  • A proper treatment of datetimes.

  • WGDOS-packed PP fields are now unpacked on demand.

  • Fixed bug in for numpy v1.7. Fixed bug when deleting the ‘id’ attribute.

  • Assign a standard name to aggregated PP fields after aggregation rather than before (because some stash codes are too similar, e.g. 407 and 408).

  • New subclasses of cf.Coordinate: cf.DimensionCoordinate and cf.AuxiliaryCoordinate.

  • A cf.Units object is now immutable.



  • Fixed endian bug in CFA-netCDF files referring to PP files

  • Changed default output format to NETCDF3_CLASSIC and trap error when when writing unsigned integer types and the 64-bit integer type to file formats other than NETCDF4.

  • Changed unhelpful history created when aggregating

version 0.9.7


  • Read and write CFA-netCDF files

  • cf.Field creation interface

  • New command line utilities: cfa, cfdump

  • Redesigned repr, str and dump() output (which is shared with cfa and cfdump)

  • Removed superseded (by cfa) command line utilities pp2cf, cf2cf

  • Renamed the ‘subset’ method to ‘select’

  • Now needs netCDF4-python 0.9.7 or later (and numpy 1.6 or later)



  • Fixed bug in cf/ which caused the creation of incorrect latitude coordinate arrays.



  • Fixed bug in cf/ which caused a failure when a file with badly formatted units was encountered.

version 0.9.6


  • Assignment to a field’s data array with metadata-aware broadcasting, assigning to subspaces, assignment where data meets conditions, assignment to unmasked elements, etc. (setitem method)

  • Proper treatment of the missing data mask, including metadata-aware assignment (setmask method)

  • Proper treatment of ancillary data.

  • Ancillary data and transforms are subspaced with their parent field.

  • Much faster aggregation algorithm (with thanks to Jonathan Gregory). Also aggregates fields transforms, ancillary variables and flags.

version 0.9.5 (first LAMA release)


  • Restructured documentation and package code files.

  • Large Amounts of Massive Arrays (LAMA) functionality.

  • Metadata-aware field manipulation and combination with metadata-aware broadcasting.

  • Better treatment of cell measures.

  • Slightly faster aggregation algorithm (a much improved one is in development).

  • API changes for clarity.

  • Bug fixes.

  • Added ‘TEMPDIR’ to the cf.CONSTANTS dictionary

  • This is a snapshot of the trunk at revision r195.



  • Loads of exciting improvements - mainly LAMA functionality, metadata-aware field manipulation and documentation.

  • This is a snapshot of the trunk at revision r185. A proper vn0.9.5 release is imminent.



  • General bug fixes and code restructure

version 0.9.4


  • A proper treatment of units using the Udunits C library and the extra time functionality provided by the netCDF4 package.

  • A command line script to do CF-netCDF to CF-netCDF via cf-python.



  • Objects renamed in line with the CF data model: cf.Space becomes cf.Field and cf.Grid becomes cf.Space.

  • Field aggregation using the CF aggregation rules is available when reading fields from disk and on fields in memory. The data of a field resulting from aggregation are stored as a collection of the data from the component fields and so, as before, may be file pointers, arrays in memory or a mixture of these two forms.

  • Units, missing data flags, dimension order, dimension direction and packing flags may all be different between data components and are conformed at the time of data access.

  • Files in UK Met Office PP format may now be read into CF fields.

  • A command line script for PP to CF-netCDF file conversion is provided.

version 0.9.3


  • A more consistent treatment of spaces and lists of spaces (cf.Space and cf.SpaceList objects respectively).

  • A corrected treatment of scalar or 1-d, size 1 dimensions in the space and its grid.

  • Data stored in cf.Data objects which contain metadata need to correctly interpret and manipulate the data. This will be particularly useful when data arrays spanning many files/arrays is implemented.

version 0.9.2


  • Created a script for easier installation (with thanks to Jeff Whitaker).

  • Added support for reading OPeNDAP-hosted datasets given by URLs.

  • Restructured the documentation.

  • Created a test directory with scripts and sample output.

  • No longer fails for unknown calendar types (such as '360d').

version 0.9.1


  • First release.