Unexpected time resampling leads to extremely large dask graphs in default installation

[Hem-W]

Setup Information

• Xclim version: 0.55.0
• Python version: 3.11.11
• Operating System: Ubuntu 22.04

Description

When inputting dask datasets, setting freq in standardized indices (SPI and SPEI) could lead to unexpected extremely large graphs, which chokes the following computations. However, when the user use freq=None, this could be avoided.

Besides, this issue only occurs when flox is not installed (which is an optional package). So by default the user does not install this package (but would be installed in the development environment) and may input the same freq as the input dataset to accidentally trigger this large graph layers instead of using freq=None or installing flox to avoid this. Also, it would raise a UserWarning that is confusing to the user as the user may have complied with non-chunk time dimensions (UserWarning: The input data is chunked on time dimension and must be fully rechunked to run fit on groups . Beware, this operation can significantly increase the number of tasks dask has to handle.).

Steps To Reproduce

Please note that this only occurs when user uses the default installation (i.e. without flox installed)

import xarray as xr
from xclim.testing.utils import open_dataset
import xclim.indices as xci

da = open_dataset("sdba/CanESM2_1950-2100.nc").pr.resample(time="MS").mean().chunk({"time": -1, "location": 1})

spi_wFreq = xci.standardized_precipitation_index(pr=da, freq="MS", window=1, dist="gamma", method="ML")
spi_wFreq

Additional context

Using freq=None or having flox installed will not lead to this issue but I guess this is not intuitive to users.

spi_woFreq = xci.standardized_precipitation_index(pr=da, freq=None, window=1, dist="gamma", method="ML")
spi_woFreq

Contribution

• I would be willing/able to open a Pull Request to address this bug.

Code of Conduct

• I agree to follow this project's Code of Conduct