Provide cell corners

[florianziemen]

Moin Nicolas,

First of all, thanks a lot for providing the healpix package!

To add HEALPix support to UXarray, an xarray extension for unstructured grids, we need a function that computes the vertices, and ideally also a mapping of the vertices to the cells.
I've played a bit with the healpix package, and found that the _uv functions will provide the corners if supplied with the correct arguments, as in this gist. From there, one can move on with unique functions as in this gist by @philipc2.

I'm sure this is not the most efficient way of generating the vertices, but I was wondering if you would be interested in supporting this operation in your library. I'd be more than happy to create a merge request if that helps.

Pinging @philipc2 and @erogluorhan, as they are core developers of UXarray.

Cheers

Flo

[ntessore]

Hi Flo!

Thanks for getting in touch. It would be great if we added add a healpy-compatible function such as

def boundaries(nside, ipix, step=1, nest=False):
    ...

to the code. In the first instance, I think that using the pix2ang_uv functions at the Python layer, as in your examples, is perfectly fine. In a second step, it could make sense to add the boundaries() function directly to the C library. That might save a bit of overhead -- but it would anyway be good to have the Python functions in place to make the comparison.

How does that sound?

[philipc2]

Thanks for pinging me @florianziemen!

We are excited to use this package to add support for HEALPix within UXarray. To add onto what Flo already mentioned:

• UXarray is written around the UGRID conventions, which requires at least the vertices (node_lon, node_lat) and cell- vertex indicies(face_node_connectivity) to represent an arbitrary 2D unstructured grid. This would be for loading a HEALPix grid into UXarray.
• Going the other way, converting from UGRID to HEALPix can be achieved using a nearest neighbor remap, however for conservative remapping, knowing the boundaries is important

[florianziemen]

Hi,

sorry for the late reply. I've been at the AGU conference and took a couple days off afterwards. I've looked into the boundaries function. In principle, I know what to do to replicate it. I'm not sure, we want to mimic exactly this function, as it kind of assumes a single point going in, and only goes to xyz (if I remember correctly).

Basic thought would be to provide the behavior of boundaries with a signature that's matching the original call for compatibility, and provide a second function, that produces angles in (theta/phi) and (lon/lat) notation.

I won't manage to do this before the holidays, but it's high on my todo list for early next year.

Cheers

Flo