Region boundaries may be too coarse for robust automation

[ezwelty]

The GTN-G / RGI region boundaries used to assign glaciers to regions are coarse approximations of the drainage divides they follow. This results in unexpected (?) situations:

• RGI outlines that straddle a region boundary (but not a drainage divide). This means that region assignment (e.g. based on largest overlap) for a given glacier will not be robust to changes in the outline, which has implications for the stability of future revisions, RGI versions with different target years, and the split vs dissolved products. In many cases, the glacier's CenLat, CenLon is not in the region to which the glacier is assigned, meaning point-wise analysis will not yield the same region assignment.
• RGI outlines that are in the same drainage basins but in different regions. Which feels counterintuitive (and awkward for certain kinds of analysis?) and could be a problem for consistency across versions (e.g. if these glaciers were joined together in the recent past).

Probably unrelated: Some RGI v6 glaciers are not assigned to the region that contains them, but I presume this has been addressed by assigning regions in an automated way.

See examples of these situations, with maps, at https://gitlab.com/wgms/fog/-/issues/29. I'll note that in all these cases, geopolitical boundaries available in OpenStreetMap follow the drainage divide and would be good candidates for region boundaries.

[fmaussion]

Thanks Ethan! Yes these issues are well known. There are several other problems with the subregion outlines, for example in RGI5 (the problem there is that there are no subregions).

We decided the following:

• we implemented only very necessary changes to RGI12 and RGI19 documented here: https://github.com/GLIMS-RGI/rgi7_scripts/blob/master/workflow/preprocessing/01_rgi7_reg_files.ipynb
• we will tackle the problem of the WGMS-RGI regions at a later stage, in one batch and in consensus with the community
• RGI7 glaciers are attributed to a region when their representative point is - which "solves" the problem of the overlaps

Actually I think the WGMS would be well placed to run the discussion about the regions together with the RGI. Something to discuss.