Kaskawulsh glacier - erroneously connected glacier

[drounce]

RGI version
RGI v7.0

Describe the issue
The links dataframe links two glaciers to Kaskawulsh Glacier (RGI2000-v7.0-G-01-16268). The two glaciers are ['RGI60-01.14443', 'RGI60-01.16201']. 'RGI60-01.16201' is correct, while 'RGI60-01.14443' appears to be an error. I'm not sure how many other glaciers are like this.

Expected outcome
Update the attribute table. Preferred solution: check for glaciers that are double-counted and quality control them.

Suggested solution
Do a quick check for glaciers that have two links and quality control them. In this example it appears that it's a small overlap with the glacier on the opposite side of the mountain range. A simple threshold of a minimum overlap area would resolve this issue. Here the overlap error of the erroneous connection is 0.000523 km2.

[fmaussion]

Thanks for the report @drounce ! I assume you used the data from #41 , correct?

Note that the 200m2 threshold was chosen with quite some care: ezwelty/rgi_links#6 - it might be slightly too small for this case. The connectivity files should give you the overlap info, right? i.e. you can use your own logic if you want to connect things differently.

[drounce]

That's correct. The simple solution was to just find all the linked RGIv6 glaciers and then select the one with the most overlap. In this case, it was incredibly obvious. That said, it seems like a fairly straightforward quality control issue to put in some logic to generating an improved links dataset that would avoid this issue altogether. I'm not sure how many glaciers is happens to and if it just so happened that it happened to the second glacier I was testing on!

Again, not a big issue, but flagged here because one of the two links is clearly the wrong glacier that is connected.

[fmaussion]

Thanks @drounce - as I said, the overlap threshold of 200m2 was chosen by @ezwelty and I based on some reasoning that is explained here: ezwelty/rgi_links#6 . The "wrong" overlap you report is 523 m2, i.e. according to us this is not a bug (yet), because some outline updates might be the result of a genuine update on the outline, while we found that overlaps < 200m2 are probably due to rounding/geometry errors and are not to be trusted. In this particular case, it seems that 200m2 is too small because its obvious that the outlines did not change between RGI6 and 7, but you can find smaller glaciers where an overlap of 500m2 might be a genuine change of outline.

The idea was that we list all overlaps, and let the users decide on the minimal threshold they want to apply to their use case - does that make sense?

We can revisit the choice of 200m2, but it would be good to be sure that we are not missing anything else.

[drounce]

Right. I don't think the threshold of 200m2 needs to change, but rather the quality control. For this case, for example, it seems to be overlap based on the delineation associated with the top of the mountain range. Hence, the two glaciers are clearly not meant to be linked.

Another potential solution would be to check if the multiple "linked" glaciers greatly exceed the area of the initial glacier, which would indicate that it's not a problem of 1 glacier splitting into multiple glaciers (which I wouldn't be surprised if that happened) but could be one way to quality control this.

Again, not a big deal, but was just flagging in case there are future improvements. Users who don't pay attention might be quite surprised otherwise!

[ezwelty]

@drounce Indeed, the links table concerns itself only with the outlines themselves, not whether they are intended to represent the same glacier. (Perhaps they would be better named "overlaps" or "intersections" table). There was discussion of correcting for georeferencing shifts in some RGI 6 regions before calculating the links so that even two outlines that do not overlap at all could be strongly linked, but we opted to stick to the outlines themselves. After all, a geodetic mass balance calculated from the shifted outline and correct DEMs has no relationship with a corrected outline that does not overlap, even if intended to represent the same glacier.

Regarding your suggestion, the development version of the RGI7 user guide has a better description of the links table columns and their meaning:

https://www.glims.org/rgi_user_guide/products/glacier_product.html#product-files
(see RGI2000-v7.0-G-01_alaska-rgi6_links.csv)

Several statistics are precomputed to help the user apply their own filter criteria and identify different scenarios. If both rgi6_area_fraction and rgi7_area_fraction ~= 1, it is a strong 1:1 match. If rgi6_area_fraction ~= 1 and rgi7_area_fraction << 1, the RGI 7 outline contains one or more (see n_rgi6) RGI 6 outlines (e.g. removal of ice divide, or in the opposite case, addition of ice divide). Most of the trouble is that there are innumerable cases in between, with different ice divides, where their is no clearly-best answer.

That said, I'm open to producing a more opinionated link table if we can identify a single set of criteria expected to serve the needs of most users.