Concatenating zarr groups with xr.open_datatree results in bad output

What happened?

I tried to load and then concatenate groups of a zarr file that was loaded with xr.open_datatree. The groups of the data are years in an ERA5 time series. The concatenated result seems to be more of a mean seasonal cycle than the time series.

What did you expect to happen?

When looping through the groups with xr.open_zarr, the data is as expected, showing interannual variability.

Minimal Complete Verifiable Example

# %pip install s3fs

import xarray as xr
from matplotlib import pyplot as plt

# data set up for bug report
s3_uri = 's3://spi-greenfjord-public/era5_t2m-test_data_for_bug_report.zarr'

## specifications for zarr + s3 bucket ####################################
kwargs = dict(
    consolidated=True, 
    chunks={},
    storage_options=dict(
        anon=True, 
        endpoint_url='https://os.zhdk.cloud.switch.ch'))

## xr.open_datatree #######################################################
datatree = xr.open_datatree(s3_uri, engine='zarr', **kwargs)
ds_treecat = xr.combine_nested([datatree[year].ds for year in datatree], concat_dim="time")  # same behaviour with xr.concat
ds_treecat = ds_treecat.compute()

## xr.open_zarr #######################################################
ds_zarrlist = [xr.open_zarr(s3_uri, group=year, **kwargs) for year in range(1980, 2023)]
ds_zarrcat = xr.combine_nested(ds_zarrlist, concat_dim='time')
ds_zarrcat = ds_zarrcat.compute()

## Plotting #######################################################
def plot_t2m_time_series(da_hourly, label='', **kwargs):
    if 'ax' not in kwargs:
        fig, ax = plt.subplots(figsize=(12, 3), dpi=140)
        kwargs['ax'] = ax

    da_daily = da_hourly.resample(time='1D').mean()
    da_yearly = da_hourly.resample(time='1YS').mean()

    props = dict(lw=0.2) | kwargs
    da_daily.plot(**props)

    props = props | dict(lw=5, label=label, c=ax.get_lines()[-1].get_color())
    da_yearly.plot(**props)
    return fig, ax


_, ax0 = plot_t2m_time_series(ds_treecat.t2m, label="xr.open_datatree(s3_uri, engine='zarr', consolidated=True, chunks={{}})")
_, ax1 = plot_t2m_time_series(ds_zarrcat.t2m, label="xr.open_zarr(s3_uri, group=year, consolidated=True) ...", c='C1')

for ax in [ax0, ax1]:
    ax.set_title('ERA5 2m temperature for area in Greenland (1980-2022)', loc='left')
    ax.legend(ncol=2, frameon=True, edgecolor='none')

MVCE confirmation

• Minimal example — the example is as focused as reasonably possible to demonstrate the underlying issue in xarray.
• Complete example — the example is self-contained, including all data and the text of any traceback.
• Verifiable example — the example copy & pastes into an IPython prompt or Binder notebook, returning the result.
• New issue — a search of GitHub Issues suggests this is not a duplicate.
• Recent environment — the issue occurs with the latest version of xarray and its dependencies.

Relevant log output

No response

Anything else we need to know?

No response

Environment

Start by running the provided MVCE against the linked public Zarr store and compare the results from xr.open_datatree with xr.open_zarr for each year. Trace the group iteration and concatenation path used by xr.combine_nested or xr.concat; done means the datatree-based result preserves the interannual time series like the open_zarr result.