search

pySTEPS / pysteps

Python framework for short-term ensemble prediction systems.
https://pysteps.github.io/
BSD 3-Clause "New" or "Revised" License
441 stars 160 forks source link

Hackathon July 2023 #351

Closed mpvginde closed 3 months ago

mpvginde commented 4 months ago

This pull request brings all developments made during the July 2023 pySTEPS-hackathon at the RMIB.
Many thanks to all contributors: @ladc, @RubenImhoff, Simon De Kock and Ricardo Reinoso-Rondinel.

The work was focussed on the STEPS blending routine.

NEW FEATURES

  1. Scientificaly correct handling of following cases when using STEPS blending (linked with #309)
    • No rain in radar fields | No rain in NWP fields
    • No rain in rader fiels | Rain in NWP fields

REFACTORING (all linked with #332)

  1. Initialisation of the noise cascade with random noise (with correct spatial correlation) to have correct power
  2. Renormalize radar extrapolation cascade after AR(2)-evolution to not lose power
  3. STEPS Blending extrapolation component:
    • Precipiation and noise cascade are recomposed before advection
    • Both fields are advected
    • They are decomposed again for blending with NWP cascade This removes the loss of power in the smallest scales due to the interpolation needed for the semilagrangian advection.
      In addition computational cost is reduced because the number of extrapolation is greatly reduced.

BUGFIXES

dnerini commented 4 months ago

some of the tests were failing for the same reason as in #352 , @mpvginde can you please merge master into this branch to see if that solves the issue?

there are however many other failing tests, are you looking into these?

RubenImhoff commented 3 months ago

I made some changes which should (hopefully) fix the issues that came from the tests. @ladc, which tests did you still want to add? We might be good to go after that.

RubenImhoff commented 3 months ago

The tests on Ubuntu fail due to a problem with uploading the Codecov report - not sure what is causing it, but it seems unrelated to the PR.

dnerini commented 3 months ago

yes I agree, we can ignore the failing test in the context of this PR. If @ladc has no further comments or changes to add, I think we can merge it!

mpvginde commented 3 months ago

Thanks everyone for helping with this PR!