Add .sel and .isel equivalent that returns all vars on all grids related to the indexers?

[willirath]

I'm working on a project that aims at providing observation-like sampling for Ocean model data living on a C grid. What we aim at is keeping the C grid logic intact as long as possible. For this, something like

ds.xgcm.isel(X=3, Y=5)

returning a full dataset restricted to x_c=3, x_l=3, x_r=3, y_c=5, y_l=5, y_r=5and a correspondingds.xgcm.sel(...)` would be ideal.

A related question is: Which grid points are "related" to a given position?

If we consider the tracer grid of a finite volume model the basis of our selection, then for a complete description of the dynamics of the selection, we'd not only need the vars defined on the faces for one side (per dim), but for both sides of the box.

(Thanks @jbusecke for the chat yesterday. Definitely helped me getting a clearer idea on what I'm looking for.)

[jbusecke]

I think this would be a really nice addition to xgcm. I have been looking for this functionality for a long time to subset large datasets. This could also be a partial solution to e.g. the issue @miniufo mentioned in #193.

I would propose that a call could look like this

grid_sub, ds_sub = grid.isel(ds, x_c=3, y_c=slice(0,5), to=‘outer’, boundary=‘fill’)

By using the actual dimension (x_c) and a specifier (not sure if to is the most intuitive), the user has maximum flexibility.
In your case you want to select based on tracer points but get both neighboring velocity points. ‘outer’ will do the trick and also keep existing xgcm lingo for clarity.
If for some reason, another user wants to select based on velocity points, and the select only the tracer point to the left, that is also possible.

xgcm should set some reasonable defaults for to and boundary based on the original grid object to keep a simple sub setting command short.

Internally I’d like to wrap all of the isel/sel logic from xarray (profiting from possible future improvements) and just getting a logical index from there. Xgcm then would have to deal with constructing matching indicies for other dimensions along the same axis.

[rabernat]

Good idea. this would work well in combination with #197 (xarray accessor). Then we could do

ds.grid.isel(X=slice(5, 10))

I slightly prefer @willirath's API over @jbusecke's. The whole point of this is to not have to worry about the different dimension names and be able to select on an axis.

[jbusecke]

Fair point. Could we maybe add some more keyword to allow the flexibility if wanted? e.g. from could default to center but also be left/right?

This would be hidden for the casual use, but still make the use case, I described possible?

Would love to see this working with the accessor. What would the output of that call be? Just a dataset or both a new grid and ds?

[dcherian]

This lines up with the cf xarray discussion: pangeo-data/pangeo#771.

[dcherian]

If you have the axis attribute set, you can now do this with cf-xarray: https://cf-xarray.readthedocs.io/en/latest/examples/introduction.html#Slicing

EDIT: that's not exactly right. It won't take into account the grid layout. cf_xarray implements the grid-unaware version of this feature

[jbusecke]

Oh this is super cool. I will try to have a closer look soon. Do you think this will overlap in functionality with xgcm? I will have to dive deeper into cf-xarray, but want to make sure that we are not duplicating efforts here.

[jbusecke]

I am still very keen on having this feature. Just wanted to drop a usecase that I encountered the other day:
It would be awesome if xgcm.sel() could retroactively figure out how to select fields on other grid locations. Say I have a tracer field and selected a subset, I think it would be amazing if there would be a way to select the velocity fields in a coherent way. This might however be a bit more complicated, depending on how the dimension labels are given. Just wanted to drop this here, so it doesnt get lost.

[dcherian]

See pydata/xarray#4979 (comment)

[github-actions]

This issue has been marked 'stale' due to lack of recent activity. If there is no further activity, the issue will be closed in another 30 days. Thank you for your contribution!