Stochastic noise extensions

[gavinevans]

Related to https://github.com/metoppv/mo-blue-team/issues/1268
improver_test_data PR: metoppv/improver_test_data#147

Description
This PR extends the stochastic noise implementation with two main aims:

1. When recycling ensemble members (e.g. for increasing the number of MOGREPS-UK members from 18 to 24), the current approach would produce duplicate members that could only differ due to differences in the originally dry areas, arising from the use of stochastic noise to break ties. This, however, results in ensemble members that are very similar. In order, for the additional realizations to be useful, this PR adds the ability to add stochastic noise to positive (wet) areas, rather than just adding stochastic noise to dry areas for tie breaking. This allows ensemble members to differ in a more discernible way. The use of stochastic noise to wet areas can be limited to the additional >18 realizations.
2. Some forecast sources e.g. MOGREPS-G or ECMWF IFS are coarser resolution models, so don't represent small scale detail, it may be useful to add some stochastic noise to these models to include some small-scale variability.

The stochastic noise values created by pysteps can sometimes be quite large (up to 5 mm/hr), so a wet-noise-amplitude multiplier is added, so that the stochastic noise can be controlled, so that it adds some small-scale variability, without markedly changing the original forecast.

Testing:

• Ran tests and they passed OK
• Added new tests for the new feature(s)

[maxwhitemet]

Hi Gavin. This looks good to me. I've suggested some small changes.

[maxwhitemet]

Happy with the changes made. Approved 👍

[mo-jbeaver]

I've added some comments below. The unit tests passed successfully but the acceptance tests failed:

FAILED improver_tests/acceptance/test_stochastic_noise.py::test_scale_non_positive_noise - ValueError: Checksum for stochastic_noise/scaled/kgo.nc is 45940447df6131b21f66081fcb105b8d9b2e7ffe79b3db7c97791eafab0e2d88, expected 707e828564189398fb705eeafeff94ba368714697bfe20f22dfe6a8f67cc2616 in /home/users/jasmine.beaver/impr... FAILED improver_tests/acceptance/test_stochastic_noise.py::test_dry_realizations[False] - clize.errors.UnknownOption: improver stochastic-noise: Unknown option '--wet-noise-floor'. Did you mean '--non-positive-noise-floor'? FAILED improver_tests/acceptance/test_stochastic_noise.py::test_dry_realizations[True] - clize.errors.UnknownOption: improver stochastic-noise: Unknown option '--wet-noise-floor'. Did you mean '--non-positive-noise-floor'? FAILED improver_tests/acceptance/test_stochastic_noise.py::test_positive_regions[False] - clize.errors.UnknownOption: improver stochastic-noise: Unknown option '--wet-noise-amplitude'. Did you mean '--positive-region-noise-amplitude'? FAILED improver_tests/acceptance/test_stochastic_noise.py::test_positive_regions[True] - clize.errors.UnknownOption: improver stochastic-noise: Unknown option '--wet-noise-amplitude'. Did you mean '--positive-region-noise-amplitude'?

Accidently approved instead of requesting changes.

[mo-jbeaver]

Happy with the updates made and all tests passed successfully.

[gavinevans]

I've updated the checksums in this commit: bbe9a2c, which fixes a forecast_trajectory_gap_filler acceptance test failure that I think was introduced by #2436.

[mo-jbeaver]

Happy with the changes made and all tests passed successfully.