Follow-up: bottom topography interaction for AABW particle tracking

[imariaguiar]

Question

Question

Hi,
I recently posted a discussion here about particles getting stuck on bottom topography, following up now with two approaches I'm considering, and would really value your take on which makes more sense (or if there's a better option I'm missing).

Context: I'm tracking Antarctic Bottom Water (AABW) as it exits through the Vema Channel and continues into the deep South Atlantic. Since I need particles to stay in deep water, deleting on contact with topography loses most of the signal, so I'm looking for a way to deflect particles horizontally around obstacles instead.

Option 1 — Displacement field. Same concept as your "Preventing stuck particles" tutorial, but built from a bathymetry mask (invalid below a reference depth) instead of a coastal landmask. I'd compute the Laplacian/gradient of that mask to get a vector field pointing away from topography, normalize it, and apply a fixed-magnitude push (analogous to your 1 m/s) whenever a particle gets close.

Option 2 — Radius search with line-of-sight check. When a particle would hit the bottom, the kernel searches outward in expanding rings (0.1°, 0.25° and 0.5°), testing 8 directions per ring. For each candidate, it checks that the straight-line path to it (not just the endpoint) has valid bathymetry, then picks the candidate with the strongest real current among valid ones. If nothing is found at any radius, the particle is deleted.

My concern with Option 1 is that it only uses the local gradient, in a narrow, V-shaped channel like Vema, I've seen similar local-gradient approaches fail to find a way out and end up pushing particles all the way to the surface. Option 2 is more expensive computationally but "sees" further and checks the path, which seems more robust for this kind of geometry.
Does either of these seem reasonable to you? And do you have a sense of which would make more sense for this specific case, narrow, irregular abyssal topography rather than a coastline?

Thanks again for your time!

[erikvansebille]

Hi @imariaguiar, I'll give a first quick reply. First of all, this is a follow-up from #2900 (good to link to that post)

What yup of grid is your model on? A-grid or C-Grid? In #2900 you mentioned OFES, is that the output you're using? How do you load in the data into a FieldSet, with what interpolation method?

In any case, perhaps you can have a look at what @michaeldenes wrote for the plastic.parcels-code.org project. For example, he has a kernel that seems somewhat similar to what you propose as option 2 above:
https://github.com/Parcels-code/plasticparcels/blob/9b59910ab7a3db285451c5eb16145165575fde4d/plasticparcels/kernels.py#L600-L695

He also wrote a reflect-at-bottom kernel:
https://github.com/Parcels-code/plasticparcels/blob/9b59910ab7a3db285451c5eb16145165575fde4d/plasticparcels/kernels.py#L732-L759

Note that these are still old v3-kernels. But are these useful?