QuantileDeltaMapping adjustment in Dask fails if training dataset has been loaded

[saschahofmann]

Setup Information

• Xclim version: 0.47.0
• Python version: 311.6
• Operating System: Ubuntu 18.04

Description

When I modify the example from the docs to run the QuantileDeltaMapping in Dask, it fails during adjustment if I load the dataset after training.

As you can see in the example below, all I did is add chunking to ref and sim. I then load the qdm.ds to make sure that training is performed. In my real world example, I want to do this because I will adjust multiple simulations with the same training dataset. In fact, the same xclim example recommends doing this.

The error message (TypeError: object of type 'NoneType' has no len()) points to xclim/sdba/base.py:606.
These are the lines

                badchunks.update(
                    {
                        dim: chunks.get(dim)
                        for dim in reduced_dims
                        if len(chunks.get(dim)) > 1
                    }

reduced_dims are groups and quantiles which aren't in chunks after I called load on qdm.ds. That's why chunks.get(dim) return NoneType which explains the error message.

In fact, just 2 lines above xclim seems to anticipate this case by returning an empty list by default (L599 if len(chunks.get(dim, [])) > 1).

I suggest to use the same approach to fix this issue. Happy to make a PR.

Steps To Reproduce

import xarray as xr
import numpy as np
from xclim.sdba import QuantileDeltaMapping

t = xr.cftime_range("2000-01-01", "2030-12-31", freq="D", calendar="noleap")
ref = xr.DataArray(
    (
        -20 * np.cos(2 * np.pi * t.dayofyear / 365)
        + 2 * np.random.random_sample((t.size,))
        + 273.15
        + 0.1 * (t - t[0]).days / 365
    ),  # "warming" of 1K per decade,
    dims=("time",),
    coords={"time": t},
    attrs={"units": "K"},
).chunk({"time": -1})
sim = xr.DataArray(
    (
        -18 * np.cos(2 * np.pi * t.dayofyear / 365)
        + 2 * np.random.random_sample((t.size,))
        + 273.15
        + 0.11 * (t - t[0]).days / 365
    ),  # "warming" of 1.1K per decade
    dims=("time",),
    coords={"time": t},
    attrs={"units": "K"},
).chunk({"time": -1})

ref = ref.sel(time=slice(None, "2015-01-01"))
hist = sim.sel(time=slice(None, "2015-01-01"))

qdm = QuantileDeltaMapping.train(ref, hist, group="time", kind="+")
qdm.ds.load()
adjusted = qdm.adjust(sim)

Additional context

No response

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