Add reader kwarg to 'ahi_hrit' to disable exact start_time

[pnuu]

The 'ahi_hrit' by default takes the exact start_time from the data it self. The start_time is down to microsecond level, and is different for every dataset. This will cause for example Sun zenith angle correction to be computed separately for every VIS channel.

This PR introduces a reader keyword argument use_acquisition_time_as_start_time to control whether nominal (parsed from the filename) or exact acquisition start time will be used. The default, due to significantly better performance, is the nominal time parsed from the filename (reader_kwargs={'use_acquisition_time_as_start_time': False}).

The exact times are always available from the channel coordinates:

scene.load(["B03"])
print(scene["B03].coords["acq_time"].data)
array(['2021-12-08T06:00:20.131200000', '2021-12-08T06:00:20.191948000',
       '2021-12-08T06:00:20.252695000', ...,
       '2021-12-08T06:09:39.449390000', '2021-12-08T06:09:39.510295000',
       '2021-12-08T06:09:39.571200000'], dtype='datetime64[ns]')

The first value represents the exact start time, and the last one the exact end time of the data acquisition.

Another option would've been to ensure all the angle computations were using a rounded time (to closest minute or so), but that might not be desirable for some use cases.

• Closes AHI HRIT reader has gotten slower #1384
• Tests added
• Fully documented

[codecov]

Codecov Report

Merging #1986 (e0168d8) into main (cccbe06) will increase coverage by 0.02%.
The diff coverage is 96.15%.

@@            Coverage Diff             @@
##             main    #1986      +/-   ##
==========================================
+ Coverage   93.54%   93.56%   +0.02%     
==========================================
  Files         279      281       +2     
  Lines       41322    41480     +158     
==========================================
+ Hits        38654    38811     +157     
- Misses       2668     2669       +1     

Flag	Coverage Δ
behaviourtests	4.77% <0.00%> (-0.02%)	⬇️
unittests	94.09% <96.15%> (+0.02%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
satpy/readers/hrit_jma.py	97.98% <80.00%> (+0.04%)	⬆️
satpy/tests/reader_tests/test_ahi_hrit.py	100.00% <100.00%> (ø)
satpy/tests/enhancement_tests/test_viirs.py	100.00% <0.00%> (ø)
satpy/tests/enhancement_tests/test_enhancements.py	100.00% <0.00%> (ø)
satpy/readers/msu_gsa_l1b.py	100.00% <0.00%> (ø)
satpy/tests/reader_tests/test_msu_gsa_l1b.py	100.00% <0.00%> (ø)
satpy/tests/writer_tests/test_cf.py	99.69% <0.00%> (+<0.01%)	⬆️
satpy/writers/cf_writer.py	94.35% <0.00%> (+0.05%)	⬆️
satpy/resample.py	79.73% <0.00%> (+0.38%)	⬆️

---

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 cccbe06...e0168d8. Read the comment docs.

[coveralls]

Coverage increased (+0.02%) to 94.025% when pulling e0168d8 on pnuu:bugfix-ahi-start_time-slowdown into cccbe06 on pytroll:main.

[djhoese]

To be clear, this PR does not fix any of the performance issues you ran into right?

I know the exact start time should be used by default, but it is too bad for performance that we can't set this new option to False by default. Otherwise, this looks good to me.

[sfinkens]

LGTM, very useful! Just nitpicking...

[sfinkens]

I'd favor "nominal" over "approximate", what do you think?

[sfinkens]

And I think there are two whitespaces in the beginning of that sentence.

[pnuu]

Thanks, I'll use nominal. And the double whitespace is a leftover reflex from using Emacs, which thinks wrapping works properly only if there are two whitespaces after a comma...

[pnuu]

To be clear, this PR does not fix any of the performance issues you ran into right?

No, but setting use_exact_start_time=False helps.

I know the exact start time should be used by default, but it is too bad for performance that we can't set this new option to False by default. Otherwise, this looks good to me.

So I should make the nominal time the default?

The exact scanline times will any way be available for any scientific users.

[pnuu]

I opted to read Dave's comment so that False should be the default. Updated the docstring and tests to reflect this.

[pnuu]

For comparison, generation of 6 full disk composites that use SZA correction. First with use_exact_start_time=True (the previous default):

Elapsed time: 49.85 seconds. Maximum memory usage approximately 9.2 GB.

And with use_exact_start_time=False (new default):

Elapsed time: 39.99 seconds. Maximum memory usage approximately 7 GB.

[mraspaud]

LGTM, I just have a suggestion:
I think use_exact_start_time is not precise enough. How about use_nominal_time_as_start_time or use_acquisition_time_as_start_time?

[pnuu]

Changed the kwarg to use_acquisition_time_as_start_time.