Drop x and y indexes before zeroing data in DayNightCompositor

[pnuu]

This PR removes extra indexes from the data arrays before zeroing the data for DayNightCompositor.

• Closes DayNightCompositor does not work with eg overview_sun as the day part #584
• Tests added
• Tests passed
• Passes git diff origin/master -- "*py" | flake8 --diff

[TAlonglong]

I have tested the bugfix and now it works both when providng an overview or an overview_sun.

Thanks

[djhoese]

Is this the fault of the sunz modifier? Would it be possible to have the sunz modifier retain the x/y coords/indexes?

[pnuu]

The problem is, as far as I can see it, that sunz modifier adds the extra indexes. Maybe it's this line (349) in composites/__init__.py?

coszen = xr.DataArray(cos_zen(vis.attrs["start_time"], lons, lats), dims=['y', 'x'], coords=[vis['y'], vis['x']])

This seems to be copied (by me) also to DayNightCompositor if solar elevation data are not given and needs to be calculated. Otherwise all the compositors/modifiers use the input coordinates: dims=data.dims, coords=data.coords.

[pnuu]

Using dims=data.dims, coords=data.coords for sunz modifier won't work, as the shapes can be different ((z, y, x) for composite data, (y, x) for SZA data).

[coveralls]

Coverage increased (+0.03%) to 77.48% when pulling 8617761 on pnuu:bugfix_dnc_overview_sun into d83abd8 on pytroll:master.

[codecov]

Codecov Report

Merging #587 into master will increase coverage by 0.03%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #587      +/-   ##
==========================================
+ Coverage   77.44%   77.47%   +0.03%     
==========================================
  Files         136      136              
  Lines       19170    19196      +26     
==========================================
+ Hits        14847    14873      +26     
  Misses       4323     4323

Impacted Files	Coverage Δ
satpy/composites/__init__.py	64.51% <100%> (+0.35%)	⬆️
satpy/tests/compositor_tests/__init__.py	99.17% <100%> (+0.04%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update d83abd8...8617761. Read the comment docs.

[djhoese]

Using dims=data.dims, coords=data.coords for sunz modifier won't work, as the shapes can be different ((z, y, x) for composite data, (y, x) for SZA data).

But the sunz corrector should never be given RGB data, right? Or am I misunderstanding you.

Also, all SatPy DataArrays should have a y and x dimension, optionally with .coords. Is the problem that it is adding coords not necessarily that it has the dimensions?

[pnuu]

The data can have the bands dimension (e.g. 'R'), thus making it 3D and incompatible in dims with the strictly 2D SZA array. The problem - as far as I can understand - are the .indexes getting extra things like what is shown in my comment in #584 . Doing the .drop(['x', 'y']) removes those, and after that everything works.