Restrict size 1 limitation to source field in regridding

[ThibHlln]

resolves #250

Hi David, Hi Sadie,

With the simple change contained in this PR, the code below would work, which is not currently the case (i.e. a regridding towards a single grid cell domain).

import cf
import numpy

# source field (2x3 grid)
fld1 = cf.Field()
fld1.set_construct(
    cf.DimensionCoordinate(
        properties={'standard_name': 'longitude',
                    'units': 'degrees_east',
                    'axis': 'X'},
        data=cf.Data([5, 15, 25]),
        bounds=cf.Bounds(data=cf.Data([[0, 10], [10, 20], [20, 30]]))
    ),
    axes=fld1.set_construct(cf.DomainAxis(size=3))
)
fld1.set_construct(
    cf.DimensionCoordinate(
        properties={'standard_name': 'latitude',
                    'units': 'degrees_north',
                    'axis': 'Y'},
        data=cf.Data([55, 65]),
        bounds=cf.Bounds(data=cf.Data([[50, 60], [60, 70]]))
    ),
    axes=fld1.set_construct(cf.DomainAxis(size=2))
)

# destination field (1x1 grid)
fld2 = cf.Field()
fld2.set_construct(
    cf.DimensionCoordinate(
        properties={'standard_name': 'longitude',
                    'units': 'degrees_east',
                    'axis': 'X'},
        data=cf.Data([15]),
        bounds=cf.Bounds(data=cf.Data([[0, 30]]))
    ),
    axes=fld2.set_construct(cf.DomainAxis(size=1))
)
fld2.set_construct(
    cf.DimensionCoordinate(
        properties={'standard_name': 'latitude',
                    'units': 'degrees_north',
                    'axis': 'Y'},
        data=cf.Data([60]),
        bounds=cf.Bounds(data=cf.Data([[50, 70]]))
    ),
    axes=fld2.set_construct(cf.DomainAxis(size=1))
)

# set some data
fld1.set_data(numpy.arange(6).reshape(2, 3), axes=('Y', 'X'))
fld2.set_data(numpy.ones((1, 1)), axes=('Y', 'X'))

# test regridding on all existing methods
for m in ['linear', 'conservative', 'conservative_2nd', 
          'patch', 'nearest_stod', 'nearest_dtos']:
    
    fld3 = fld1.regrids(fld2, method=m)
    
    print(fld3.data)

[davidhassell]

Hi Thibault,

Would you mind copying the above into an issue and referencing that issue from this PR? (https://ncas-cms.github.io/cf-python/contributing.html#preparing-pull-requests) - I'll reply over there ...

Thanks, David

[ThibHlln]

Hi David,

As discussed in #250, it seems that the source grid can be with size 1 dimensions for some of the methods.

My latest commit reflects that and also un-indent the check in regrid_get_latlon to cover both coords_2D = False and coords_2D = True. However, for it to work, I also needed to remove the squeeze on src_data in regrids, I am not sure whether this is a problem?

The testing script below should work now (same fields as above):

for m in ['conservative_1st', 'conservative_2nd', 'nearest_stod', 'nearest_dtos']:
    fld3 = fld2.regrids(fld1, method=m)
    print(fld3.array)

[davidhassell]

Hi @ThibHlln, I wasn't sure that anything would happen here, but I've managed to push to this PR of yours. This commit (7dfb455) tidies up the removal of the "squeeze", which needs to go back in sometimes to get rid of size 1 dimensions that are not being regridded. Hope that's OK! (this commit can always be nuked, of course).

[davidhassell]

Hi @ThibHlln - I think we can merge this, now. I'll await your OK, though ...

[ThibHlln]

Hi @davidhassell,

I was not sure what the squeezing was for, but I was guessing it would require more than just removing it.
Thank you for your help with this.

Yes, I think it can be merged now, I ran the tests above successfully after your commits.
So, it looks all good to me.

[davidhassell]

Hi @ThibHlln - I think the squeezing is all about making sure we pass the correctly shaped numpy array down to the ESMF layer. Thanks for your help with this. We expect a release of 3.11.0 in a couple of weeks (ish). Merging now ...