Preprocessor multimodel_statistics fails when data have no horizontal dimension

[stefsmeets]

Describe the bug
Hi all, I'm developing some tests for the multimodel statistics using real data (#856), and I'm coming accross this bug:

Running multimodel statistics with a list of cubes with no horizontal dimension, will result in
iris.exceptions.CoordinateNotFoundError: 'Expected to find exactly 1 depth coordinate, but found none.'

The bug can be reproduced by:

cubes = [cube[:, :, 0, 0] for cube in cubes]
multi_model_statistics(cubes, span='full', statistics=['mean'])

See the full stack trace below.

timeseries_cubes_month = [<iris 'Cube' of air_temperature / (K) (time: 14; air_pressure: 2; latitude: 3; longitude: 2)>, <iris 'Cube' of air_te... 3; longitude: 2)>, <iris 'Cube' of air_temperature / (K) (time: 14; air_pressure: 2; latitude: 3; longitude: 2)>, ...]

    @pytest.mark.functional
    # @pytest.mark.xfail('iris.exceptions.CoordinateNotFoundError')
    def test_multimodel_no_horizontal_dimension(timeseries_cubes_month):
        """Test statistic without horizontal dimension using monthly data."""
        span = 'full'
        cubes = timeseries_cubes_month
        cubes = [cube[:, :, 0, 0] for cube in cubes]
        # Coordinate not found error
        # iris.exceptions.CoordinateNotFoundError:
        # 'Expected to find exactly 1 depth coordinate, but found none.'
>       multimodel_test(cubes, span=span, statistic='mean')

tests/functional/test_multimodel.py:207:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
tests/functional/test_multimodel.py:125: in multimodel_test
    output = multi_model_statistics(cubes, span=span, statistics=statistics)
esmvalcore/preprocessor/_multimodel.py:439: in multi_model_statistics
    statistic_cube = _assemble_full_data(cubes, statistic)
esmvalcore/preprocessor/_multimodel.py:355: in _assemble_full_data
    stats_cube = _put_in_cube(cubes[0], stats_dats, statistic, time_axis)
esmvalcore/preprocessor/_multimodel.py:180: in _put_in_cube
    plev = template_cube.coord('depth')
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

self = <iris 'Cube' of air_temperature / (K) (time: 14; air_pressure: 2)>, name_or_coord = 'depth', standard_name = None, long_name = None
var_name = None, attributes = None, axis = None, contains_dimension = None, dimensions = None, coord_system = None, dim_coords = None

    def coord(self, name_or_coord=None, standard_name=None,
              long_name=None, var_name=None, attributes=None, axis=None,
              contains_dimension=None, dimensions=None, coord_system=None,
              dim_coords=None):
        """
        Return a single coord given the same arguments as :meth:`Cube.coords`.

        .. note::

            If the arguments given do not result in precisely 1 coordinate
            being matched, an :class:`iris.exceptions.CoordinateNotFoundError`
            is raised.

        .. seealso::

            :meth:`Cube.coords()<iris.cube.Cube.coords>` for full keyword
            documentation.

        """
        coords = self.coords(name_or_coord=name_or_coord,
                             standard_name=standard_name,
                             long_name=long_name, var_name=var_name,
                             attributes=attributes, axis=axis,
                             contains_dimension=contains_dimension,
                             dimensions=dimensions,
                             coord_system=coord_system,
                             dim_coords=dim_coords)

        if len(coords) > 1:
            msg = 'Expected to find exactly 1 coordinate, but found %s. ' \
                  'They were: %s.' % (len(coords), ', '.join(coord.name() for
                                                             coord in coords))
            raise iris.exceptions.CoordinateNotFoundError(msg)
        elif len(coords) == 0:
            _name = name_or_coord
            if name_or_coord is not None:
                if not isinstance(name_or_coord, six.string_types):
                    _name = name_or_coord.name()
            bad_name = _name or standard_name or long_name or ''
            msg = 'Expected to find exactly 1 %s coordinate, but found ' \
                  'none.' % bad_name
>           raise iris.exceptions.CoordinateNotFoundError(msg)
E           iris.exceptions.CoordinateNotFoundError: 'Expected to find exactly 1 depth coordinate, but found none.'

../../miniconda3/envs/esmvaltool/lib/python3.8/site-packages/iris/cube.py:1497: CoordinateNotFoundError

[valeriupredoi]

you mean vertical dimension, right? Yeah, this has been a long-standing issue that I somehow never managed to proper fix-fix, see also #687 - I'll see if I can just nip it in the bud and fix it this week, good find while doing real-data tests, mate! 🍺

[stefsmeets]

Hi @valeriupredoi , I meant horizontal dimension (notice that the last 2 dimensions are reduced to 0). There is also a big PR to refactor and simplify the multimodel stats here: #685 . We are hoping that once we have a better grip on testing the multimodel stats that we can move forward with that one.

[valeriupredoi]

oh yeah forgot about that PR - good call, we should accelerate the merge of that one, I'd be happy to see it in before the 2.2 release 👍

[bouweandela]

@valeriupredoi See #41 for a description of the problem, I think that e.g. the ocean variable thetaoga has no horizontal dimensions.

[bouweandela]

@stefsmeets @Peter9192 Can this issue be closed, now that #1150 has been merged?

[stefsmeets]

We implemented a test for this case, so that will tell you if this can be closed or not 😁

ESMValCore/tests/sample_data/multimodel_statistics/test_multimodel.py

Line 260 in 9e27908

def test_multimodel_no_horizontal_dimension(timeseries_cubes_month):