Skip to content

Rock glacier landforms in RGI7? #64

Open

Description

@fmaussion

RGI version
RGI v7.0

Describe the issue
Reported by Owen King:

"RGI7 was used to measure glacier surface elevation change from 2000 onwards, alongside the elevation change estimates of Hugonnet et al. (2021). A subset of RGI7 was extracted where RGI7 and RGI6 polygons intersected. This was done to remove rock glaciers, perennial snowfields and other ice-debris landforms included in RGI7 in GTN-G 17. These features display little change in the elevation change grids of Hugonnet et al. (2021) and are so abundant (several thousand) that they bias catchment level elevation change signals."

Detective work that followed

Here are the outlines submission IDs available in GLIMS when RGI reg17 was generated:

  • 700 is the Argentina glacier inventory. 708 is an update of 75 of its outlines, for unclear reasons (the outlines are still in 700)
  • 730 is the Chilean glacier inventory

Both 700/708 and 730 have the same issues (rock glaciers) as for RGI16. Furthermore, the Argentina / Chile border makes both inventories pretty much useless as is (both inventories stop at the border). This situation is why submission 764 was created, a merge work done by Frank Paul and Philip Rastner. It's 764 that was used for the RGI, as single submission for all of RGI reg17.

Image

I noted back in the day (1 Jul 2022) that all glacier classes were lost in the merger (primeclass attribute, containing categories for glacieret, rock glacier, etc.), to which Frank replied:

"Yes, unfortunately the topologic changes resulted in a more or less complete loss of all attribute information. With a spatial join it was possible to recover the satellite image IDs, but here we often had one to many and many to one issues so it was not 100% automated. It might also be possible to reassign the Primeclass attribute (or others) with a spatial join, but it might only be valid then for the unchanged glaciers."

This was not pursued further. If someone wants to have a look to investigate, here are the files of the state of GLIMS at the time RGI reg17 was created: https://cluster.klima.uni-bremen.de/~fmaussion/misc/rgi7_data/l2_sel_reg_tars/RGI17.tar.gz. A workflow should have a look at submissions 700/708, 730 and 764 and try to trace back what happened.

Activity

  1. fmaussion commented on Sep 3, 2026

    @fmaussion
    ContributorAuthor

    More context from Frank:

    Thank you for checking and setting this up!  I now had a closer look at the RGI7.0 outlines and the two original national inventories from Chile and Argentina. As it seems, everything is correct :)  We removed what could be removed and you used the correct submission. What I can see is:

    For Chile we have removed the type ‘Glacier Rocoso’ (rock glacier), but kept the type ‘glacierette' as it also contained some larger glaciers. However, most of them are very small and show likely very little elevation change. We have 13,000 of these in Chile.
    For Argentina we removed the types GCGE, GEA and GE but kept the type MN (similar to glacierette). As for Chile, this type contained a huge amount of very small glaciers, but also a few larger ones. We have about 6,000 of these.

    Collectively, about 19,000 out of 30,600 glaciers in Region 17 might be useless or bias the results for GlaMBIE. Similar to Fabien's suggestion I would also suggest to either use the primeclass (or add a new item ’Type’) that contains a specific number when it is not a glacier and/or something that specific users would like to de-select. Theoretically, this could be assigned automatically to the attribute table as the polygons will not change. It might thus be implemented for RGI 7.1 and we can check the best way to adjust it for region 17.

    As a note, there are some other regions (e.g. 2, 4, 9, 10, 16) that contain large numbers of ‘glacierettes’, perennial snow fields, coastal ice, snow-filled depressions, etc. that might also bias mean elevation changes over a region. But assigning these in the attribute table would be manual work and take some time. As a quick fix, one might consider increasing the lower size threshold, e.g. to 0.1 km2. That would also remove a large part (but not all) of these features with limited change._

  2. fmaussion commented on Sep 3, 2026

    @fmaussion
    ContributorAuthor

    Action: add the primeclass attribute from the two national inventories to 7.1

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    RGI 7.0refers to RGI 7.0

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions