In order to maintain a backlog of relevant issues, we automatically label them as stale after 500 days of inactivity.

If this issue is still important to you, then please comment on this issue and the stale label will be removed.

Otherwise this issue will be automatically closed in 28 days time.

I just discovered a very similar problem: if the input cubes for merging have a scalar coordinate that uses a masked array, the scalar coordinate of the merged cube is not a masked array anymore.

MWE:

import numpy as np
import iris
import iris.coords
import iris.cube

# Setup cubes
scalar_coord = iris.coords.AuxCoord(np.ma.masked_array([0]), var_name='x')
merge_coord_1 = iris.coords.AuxCoord([1], var_name='merge_coord')
merge_coord_2 = iris.coords.AuxCoord([2], var_name='merge_coord')
cube_1 = iris.cube.Cube(1, var_name='cube',
                        aux_coords_and_dims=[(scalar_coord, ()), (merge_coord_1, ())])
cube_2 = iris.cube.Cube(2, var_name='cube',
                        aux_coords_and_dims=[(scalar_coord, ()), (merge_coord_2, ())])

# Create merged cube
cubes = iris.cube.CubeList([cube_1, cube_2])
merged_cube = cubes.merge_cube()

# Print scalar coordinates
print("Scalar coordinate of input cubes:")
print(cube_1.coord('x'))
print(cube_2.coord('x'))
print("")
print("Scalar coordinate of merged cube:")
print(merged_cube.coord('x'))

gives

Scalar coordinate of input cubes:
AuxCoord(masked_array(data=[0],
             mask=False,
       fill_value=999999), standard_name=None, units=Unit('unknown'), var_name='x')
AuxCoord(masked_array(data=[0],
             mask=False,
       fill_value=999999), standard_name=None, units=Unit('unknown'), var_name='x')

Scalar coordinate of merged cube:
DimCoord(array([0]), standard_name=None, units=Unit('unknown'), var_name='x')

As you can see, the scalar coordinate of the merged cube is not a masked_array anymore.

In order to maintain a backlog of relevant issues, we automatically label them as stale after 500 days of inactivity.

If this issue is still important to you, then please comment on this issue and the stale label will be removed.

Otherwise this issue will be automatically closed in 28 days time.

Still relevant

In order to maintain a backlog of relevant issues, we automatically label them as stale after 500 days of inactivity.

If this issue is still important to you, then please comment on this issue and the stale label will be removed.

Otherwise this issue will be automatically closed in 28 days time.

Still relevant

I just discovered a very similar problem: if the input cubes for merging have a scalar coordinate that uses a masked array, the scalar coordinate of the merged cube is not a masked array anymore.
(...snip...)
As you can see, the scalar coordinate of the merged cube is not a masked_array anymore.

Hi @schlunma

I've looked into this and it's a slightly tricky edge-case.

When aux coords are built from their coordinate data, Iris first attempts to construct them as a DimCoord, then if that fails it will try to construct them as an AuxCoord:

Your x scalar coord in cube_1 and cube_2 can successfully be constructed with as a DimCoord, so that's what Iris does (note that it is a DimCoord in cube_merged, but an AuxCoord in cube_1 and cube_2).

The problem is that masked data is NOT ALLOWED in a DimCoord, so if any masked values exist it will fail. However, if your points are a masked array instance, but do not have any masked values, then the mask is simply dropped as part of the DimCoord._values setter:

This is what is happening here and why your mask is lost.

The fix I am working on for the original problem mentioned in this issue will mitigate this slightly (although not fully) by allowing you to create AuxCoords with masked data. However, as Iris tries to make a DimCoord first, you will still loose the mask in your example; at lease one element of your array will need to be actually masked to make it an AuxCoord (and retain the mask).