Zarr Encoding Specification

In implementing support for the Zarr storage format, Xarray developers made some ad hoc choices about how to store NetCDF data in Zarr. Future versions of the Zarr spec will likely include a more formal convention for the storage of the NetCDF data model in Zarr; see Zarr spec repo for ongoing discussion.

First, Xarray can only read and write Zarr groups. There is currently no support for reading / writing individual Zarr arrays. Zarr groups are mapped to Xarray Dataset objects.

Second, from Xarray’s point of view, the key difference between NetCDF and Zarr is that all NetCDF arrays have dimension names while Zarr arrays do not. Therefore, in order to store NetCDF data in Zarr, Xarray must somehow encode and decode the name of each array’s dimensions.

To accomplish this, Xarray developers decided to define a special Zarr array attribute: _ARRAY_DIMENSIONS. The value of this attribute is a list of dimension names (strings), for example ["time", "lon", "lat"]. When writing data to Zarr, Xarray sets this attribute on all variables based on the variable dimensions. When reading a Zarr group, Xarray looks for this attribute on all arrays, raising an error if it can’t be found. The attribute is used to define the variable dimension names and then removed from the attributes dictionary returned to the user.

Because of these choices, Xarray cannot read arbitrary array data, but only Zarr data with valid _ARRAY_DIMENSIONS or NCZarr attributes on each array (NCZarr dimension names are defined in the .zarray file).

After decoding the _ARRAY_DIMENSIONS or NCZarr attribute and assigning the variable dimensions, Xarray proceeds to [optionally] decode each variable using its standard CF decoding machinery used for NetCDF data (see decode_cf()).

Finally, it’s worth noting that Xarray writes (and attempts to read) “consolidated metadata” by default (the .zmetadata file), which is another non-standard Zarr extension, albeit one implemented upstream in Zarr-Python. You do not need to write consolidated metadata to make Zarr stores readable in Xarray, but because Xarray can open these stores much faster, users will see a warning about poor performance when reading non-consolidated stores unless they explicitly set consolidated=False. See Consolidated Metadata for more details.

As a concrete example, here we write a tutorial dataset to Zarr and then re-open it directly with Zarr:

In [1]: import os

In [2]: import xarray as xr

In [3]: import zarr

In [4]: ds = xr.tutorial.load_dataset("rasm")

In [5]: ds.to_zarr("rasm.zarr", mode="w")
Out[5]: <xarray.backends.zarr.ZarrStore at 0x7f68c0515000>

In [6]: zgroup = zarr.open("rasm.zarr")

In [7]: print(os.listdir("rasm.zarr"))
['time', 'xc', 'yc', 'zarr.json', 'Tair']

In [8]: print(zgroup.tree())
---------------------------------------------------------------------------
ModuleNotFoundError                       Traceback (most recent call last)
File /usr/lib/python3/dist-packages/zarr/core/_tree.py:9
      8 try:
----> 9     import rich
     10     import rich.console

ModuleNotFoundError: No module named 'rich'

The above exception was the direct cause of the following exception:

ImportError                               Traceback (most recent call last)
Cell In[8], line 1
----> 1 print(zgroup.tree())

File /usr/lib/python3/dist-packages/zarr/core/group.py:2300, in Group.tree(self, expand, level)
   2281 def tree(self, expand: bool | None = None, level: int | None = None) -> Any:
   2282     """
   2283     Return a tree-like representation of a hierarchy.
   2284 
   (...)
   2298         A pretty-printable object displaying the hierarchy.
   2299     """
-> 2300     return self._sync(self._async_group.tree(expand=expand, level=level))

File /usr/lib/python3/dist-packages/zarr/core/sync.py:208, in SyncMixin._sync(self, coroutine)
    205 def _sync(self, coroutine: Coroutine[Any, Any, T]) -> T:
    206     # TODO: refactor this to to take *args and **kwargs and pass those to the method
    207     # this should allow us to better type the sync wrapper
--> 208     return sync(
    209         coroutine,
    210         timeout=config.get("async.timeout"),
    211     )

File /usr/lib/python3/dist-packages/zarr/core/sync.py:163, in sync(coro, loop, timeout)
    160 return_result = next(iter(finished)).result()
    162 if isinstance(return_result, BaseException):
--> 163     raise return_result
    164 else:
    165     return return_result

File /usr/lib/python3/dist-packages/zarr/core/sync.py:119, in _runner(coro)
    114 """
    115 Await a coroutine and return the result of running it. If awaiting the coroutine raises an
    116 exception, the exception will be returned.
    117 """
    118 try:
--> 119     return await coro
    120 except Exception as ex:
    121     return ex

File /usr/lib/python3/dist-packages/zarr/core/group.py:1550, in AsyncGroup.tree(self, expand, level)
   1531 async def tree(self, expand: bool | None = None, level: int | None = None) -> Any:
   1532     """
   1533     Return a tree-like representation of a hierarchy.
   1534 
   (...)
   1548         A pretty-printable object displaying the hierarchy.
   1549     """
-> 1550     from zarr.core._tree import group_tree_async
   1552     if expand is not None:
   1553         raise NotImplementedError("'expand' is not yet implemented.")

File /usr/lib/python3/dist-packages/zarr/core/_tree.py:13
     11     import rich.tree
     12 except ImportError as e:
---> 13     raise ImportError("'rich' is required for Group.tree") from e
     16 class TreeRepr:
     17     """
     18     A simple object with a tree-like repr for the Zarr Group.
     19 
     20     Note that this object and it's implementation isn't considered part
     21     of Zarr's public API.
     22     """

ImportError: 'rich' is required for Group.tree

In [9]: dict(zgroup["Tair"].attrs)
Out[9]: 
{'units': 'C',
 'long_name': 'Surface air temperature',
 'type_preferred': 'double',
 'time_rep': 'instantaneous',
 'coordinates': 'yc xc',
 '_FillValue': 'AAAAAAAAnkc='}