Started working on unit tests for the parallax correction.  Not much
content yet.  Everything is likely to change completely.

Added a dummy implementation, failing, just so that the unit test gets
beyond the initial import.

…rection

Merge history of monti pytroll parallax branch into satpy to preserve
history of parallax.py

Merge first version of parallax.py into satpy, adding (some) docstrings
and fixing pep8 issues.  From here we can try to make a composite out of
it and match it up with the unit test already added.

…rection

Adapt a dependency from the parallax module adopted from meteoswiss.  I
expect adding this module is temporary.  I don't really know what it's
for.

To AUTHORS.md, add Jussi Leinonen due to his parallax contributions.
Also add affiliation for DWD authors (request from DWD).

Add unit tests for the low-level parallax_correct function, according to
an interface I believe it should have, but that it doesn't have.  Most
of the unit tests are failing.

The higher level class is not unit tested yet because I don't understand
what it's supposed to do exactly.  Probably much of it needs to be
deleted.

Add more unit tests for the parallax correction.  Started with unit
tests for the class.  They're failing, but I don't know why.

Remove some unused code.  Will remove more later.

Add another parallax correction unit test that uses a much tighter area.

Merge the simplification and improvement from the previous commit by
Jussi back into parallax.py, and delete the unused and unneedded
projection.py as well as an unused convolve import.

Parameterise some of the fixtures used in the parallax unit tests.
Still needs some more testing but we're making progress.

Add more unit tests for the parallax correction, or rather more
parameters to use in the unit tests.  The parallax correction fails at
the antimeridian.

Started a partly cloudy unit test.  Needs more tests for actual content.

Switch to nearest neighbour resampling and add some tolerance to
floating point comparisons.  Now only 3 rather than 8 tests fail.

In order to try to understand where the serious loss of precision
occurs, add more tests.  Haven't found it yet.

Make unit test pass by fixing a swapping of lon/lat.

Add a parameterised fixture for generating a cloud with certain
parameters.  Will be used for to-be-written unit tests.

Add more detailed tests to understand why (corrdist < uncorrdist).all()
succeeds for me but fails on GitHub CI::

    (Pdb) p corr_dist
    array([[7847.61343724, 6029.28528333, 4715.82906281, 4386.94953032,
            5231.74248224],
           [6029.28528332, 3923.80691175, 2357.91453878, 2615.87129834,
            4386.94953031],
           [4715.8290628 , 2357.91453878,    0.        , 2357.91453878,
            4715.8290628 ],
           [4386.94953031, 2615.87129834, 2357.91453878, 3923.80691175,
            6029.28528332],
           [5231.74248224, 4386.94953032, 4715.82906281, 6029.28528333,
            7847.61343724]])
    (Pdb) p uncorr_dist
    array([[7862.66446501, 6040.84891332, 4724.87359889, 4395.36330457,
            5241.77649481],
           [6040.84891332, 3931.33241962, 2362.43680545, 2620.88830285,
            4395.36330457],
           [4724.87359889, 2362.43680545,    0.        , 2362.43680545,
            4724.87359889],
           [4395.36330457, 2620.88830285, 2362.43680545, 3931.33241962,
            6040.84891332],
           [5241.77649481, 4395.36330457, 4724.87359889, 6040.84891332,
            7862.66446501]])

For the uncorrected / corrected distance comparison, add a tolerance.
Although tests pass for me, they fail on GitHub CI.  Maybe adding a
tolerance will make them pass.

Add tests that CTH that fails to overlap with the base area raise an
error, and if it overlays only partly raises a warning.  Implementation
of this functionality is awaiting fixes in pyresample.

Allow user to customise the method used for the parallax correction.
Test for nearest and bilinear.  Mark bilinear tests as failing since
they are.

Add parallax correction tests where the areas are the same.  Also enable
dask for one version of the partly cloudy tests.

In the docstring of CompositeBase, point out that all modifier classes
/also/ derive from CompositeBase.

Started work on the ParallaxCorrectionModifier class integrating the new
parallax correction into the modifier interface.

Add a modifier interface for the parallax correction, as well as a unit
test testing this implementation.

Mark the partly/no overlap tests as xfail, awaiting fixes in pyresample.

Changed dictionary merging syntax to support Python 3.8

Add documentation on the parallax correction modifier.  NB: this commit
relies that pytroll#1969 is already merged.

Remove unused empty function from geometry.py.
Replace a TODO by a catch for the unlikely scenario that the satellite
elevation is exactly zero, and add a unit test for this case.

Repair the forward parallax function after introducing a bug with the
elevation change.  Rename X to x.

In the parallax correction unit tests, mark only particular combinations
of parameters as failing.  Until now the pytest.mark.xfail was marked
too widely.

On different architecture another variant of bilinear fails, such as on
GitHub.  Mark as fail because bilinear fails anyway.

Write a log message when performing the parallax correction.
Add a unit test to test that this log message is written.

Add a backup calculation of orbital parameters if those are missing from
the metadata.

Introduces soft dependencies on skyfield and astropy.

Downgrade syntax to Python 3.7 as this is still used for GitHub CI.

The latest functionality needs skyfield and astropy.  Install those via
continuous_integration/environmentm.yaml.

The previous commit added an astropy dependency.  The experimental test
fails as the latest released astropy doesn't work with the latest main
numpy.  Try latest main astropy to see if this resolves the problem.

Improve the documentation including the example on how to create a
composite that applies a parallax correction modifier.  Also point out
that composites can themselves not be modified, so that RGBs with
parallax correction applied will have to apply the correction to each
channel seperately.

Add a (failing) test corresponding to the confetti bug obtained when
using CTH from NWCSAF-GEO data.

Add an optional debug mode to parallax correction.  Among other things,
this will retain the lat/lons so that users can measure the distance
objects have been moved due to parallax correction.

In parallax correction, consistent use metres above the Earth's surface.
No km, and no distances above the centre of the Earth.

Remember that satellite altitude is actually reported in km.  Test it
accordingly.

The preprocess method was making things kaputt.

Add a unit test that a cloud moves toward the satellite after applying
parallax correction using the modifier interface.

This test is currently failing, it's not moving at all ☹

Improve the test that confirms the cloud moves south.  Now captures two
failure modes: one where it actually moves north, and one where it moves
so far out of the field of view that the original cloud isn't touched at
all.  Add some more diagnostics that I may remove later.

Fix the test for the cloud direction movement.  The origin of satpy
areas is in the northwest, so higher y actually means lower latitude.
To be sure, and for easier interpretation, compare the actual latitudes.

The test with the cloud movement in the modifier interface fails either
way, but it fails differently if the radius of influence is 10 km rather
than 25 km.  But for nearest neighbour it shouldn't matter?

Try to use bucket resampler in parallax correction.  Still a draft that
can't possibly work in the general case.

Change the parallax resampling approach.  Rather than assigning an area,
use resample_nearest with source and destination area directly.  By
subtracting rather than adding inv_lon_diff and inv_lat_diff, I can
resample to the resulting SwathDefinition directly.

This is the first version where the unit test confirming that parallax
correction moves the cloud equatorward and that the old cloud now has
fill values, passes.  However, some problems do remain.  Some cloud
pixels disappear entirely, and when the cloud moves out of the
destination area, the unit test still fails.  A later commit will fix
these problems and add corresponding unit tests.  My fixes have also
caused failure in other (weaker) unit tests.

Add a test confirming that the new destination cloud area is fully
cloudy.
This test is currently failing.

Fix bug in get_abs_max that was getting the wrong longitudes.  Uses the
new method in the bucket resampler implemented at
pytroll/pyresample#418

Resample CTH to target area first and clean up resampling in parallax
correction.  This improves the functionality when CTH is on a different
resolution.  It does mean yet more resampling steps.

Test case was having an invalid area when centred at the pole.  Correct
the EPSG code so that it's valid.

Also use EPSG 3575 at the antimeridian.  Not ideal but will do for now.

Fix tests for 15 km high cloud, where the could gets deformed.  Also
point out in documentation and comments that this does happen.

Add test to check that parallax correction works with dask arrays.

This test is currently failing.

Make parallax correction more dask-friendly by using resample_dataset
rather than resample_nearest.

Rearrange and extend unit tests for the parallax correction.  The unit
tests pass, but a real life test produces huge holes, which I haven't
reproduced in any unit test yet.

In the debug mode, store the counts as a dataset that can be resampled,
not as a basic ndarray without metadata.

The radiuf of influence in the final resampling step was too low.
Increase it!

Merge branch main while fixing one merge conflict

Refactor the tests in test_parallax.py into three classes, and remove
some dead code.  Hopefully this will satisfy the worries by CodeScene.

It is important to call apply_modifier_info in the modifier, including
the parallax correction modifier.  Otherwise Satpy may get confused
between the original and the corrected dataset.

Added a unit test that reproduces the bug fixed in the previous commit,
in which the original and the parallax corrected dataset would share the
same dask array and thus be both parallax corrected.

Add test confirming that ``sc.load(...)`` with the parallax correction
does not trigger any computation.

This test currently fails awaiting fixes in pyresample, and is therefore
marked with xfail.

Fix the examlpe and test YAML config, by not passing a standard_name and
using the SingleBandCompositor, so that the standard_name is inherited
correctly and the same enhancement is configured as for the single
channel data.

refactor unit test test_modifier_interface_cloud_moves_to_observer.
hopefully this satisfies the CodeScene PR check.

Refactor test_correct_area_clearsky to have less arguments.

For the parallax correction, update "version added" to 0.37.