fix internal wave refraction

[raphaeldussin]

current code uses halo (or land) points in the computation of the gradient of Internal Wave group velocity, which results in spurious gradients and unexpected refraction close to boundaries. This PR computes IW group velocities on U and V points, discarding contribution from halo or land, before computing the gradient.

[codecov]

Codecov Report

Merging #1458 (bcbf43c) into dev/gfdl (e88cdaa) will decrease coverage by 0.00%.
The diff coverage is 0.00%.

@@             Coverage Diff              @@
##           dev/gfdl    #1458      +/-   ##
============================================
- Coverage     29.13%   29.12%   -0.01%     
============================================
  Files           235      235              
  Lines         71038    71045       +7     
============================================
  Hits          20694    20694              
- Misses        50344    50351       +7     

Impacted Files	Coverage Δ
...c/parameterizations/lateral/MOM_internal_tides.F90	0.00% <0.00%> (ø)

---

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 e88cdaa...bcbf43c. Read the comment docs.

[marshallward]

After a chat with @raphaeldussin we came up with the following recommended changes:

• Loops at L799 and L811 should swap the i and j loops, so that i is innermost

• The c / (wgt1 + wgt2) should probably replaced with a precomputed mask containing 0., 1., or 0.5.
  Two benefits:

  • Eliminate the floating point division (since it's unlikely a compiler will
  • Remove the if-blocks with some sort of mask

  But at the moment I don't have a good suggestion on how to efficiently construct this mask.

We will sit on this one for a bit while Raf looks into some other issues, and will settle on a solution when that other work is finished.

[marshallward]

The method by @raphaeldussin for computing cn_[uv] looks good to me 👍

[marshallward]

Gaea regression: https://gitlab.gfdl.noaa.gov/ogrp/MOM6/-/pipelines/13373 ✔️