This 5.2.1 release supersedes 5.2.0 which had unpolished finishes.
As usual the package is available on pypi and conda-forge.

Breaking changes

None

Bug fixes

slice_mode seasonal samplings was giving wrong results for quite a few indices. This has been fixed and the performances should also be improved by the fix.
However, now seasonal slice_mode does not allow to use xclim missing values mechanisms.
user_index ExtremeMode config was not properly parsed when a string was used.
user_index Anomaly operator was not properly using the ref_time_range to setup a reference period as it should.
user_index Sum and Mean operators were broken due to a previous refactoring and a lack of unit tests, it is now fixed and tested.
create_optimized_zarr_store would throw an error when creating the first temp store if the chunks were not unified.

Internal changes

Made Frequency part of SliceMode union.
Changed how rechunker dependency is pinned to add flexibility. We want a version above '0.3' but not the '0.4'.
For the newly generate API, on custom_index function, the parameter user_index is now mandatory.
Update release process.

Enhancements

Improve create_optimized_zarr_store to accept a chunking schema instead of a single dim.
Make use of fsspec to generalize the storages where create_optimized_zarr_store can create its zarr stores.
Make CSDI and WSDI threshold configurable using the threshold parameter of icclim.index.
Add a function in icclim namespace for each ECA&D index for convenience.
Improved documentation about chunking.
Improved dev documentation display on readthedocs.