Platform: iOS
App version number: 1.0.10 (179)
URLs (aka web addresses) of any relevant observations or pages: https://www.inaturalist.org/observations/320468730
Screenshots of what you are seeing:
Suggestions in iNat Next when using location:
Suggestions for the same observation in iNat Next when ignoring location:
Suggestions for the same observation on the web when ignoring location:
Suggestions for the same observation on the web with “only view nearby suggestions”:
Description of problem:
Step 1: Create observation in iNat Next, with location (in southern Brazil) added from photo metadata, and use “ID with AI” button to display a list of suggested IDs. For the sake of this example, all suggestions are based on the first photo.
Step 2: Toggle between “ignore location” and “use location”. Regardless of what is selected, the top suggestion is of Furina diadema. The problem is that this is a species that is not found in South America, and the geomodel does not predict it outside of Australia. The third suggested species, Ninia diademata is only expected in Mexico and Central America. The fourth, Pliocercus euryzonus, is also not expected anywhere nearby, and so on. This is not, therefore, an issue with a single species or with the geomodel, but with how the computer vision suggestions are being filtered for display.
Step 3: Go to the same observation on the web, and observe that when viewing only suggestions filtered for “expected nearby”, the Australian species is no longer offered as a suggestion. For what it’s worth, the iNat Classic app on iOS offers the same suggestions in the same order as the web interface, with the only difference being that it gives an extra two suggestions when ignoring location (ten instead of eight).
There appears to be an issue, which I have suspected for some time but hadn’t taken the time to document such a clear example, that the “expected nearby” list includes a bunch of species which are not predicted as expected nearby by the geomodel. This appears to affect only the suggestions in iNat Next and not in the web interface.
The one species that is in fact “expected nearby”, according to the web interface (Oxyrhopus guibei, which is legitimately expected locally) appears in second place when filtering by location on iNat Next, but does not appear in the top ten at all when location is ignored. It appears in third place on the web interface. Does this indicate that the location filtering, rather than being “in” or “out” is applying some sort of weighting system that is going wrong in the app?
My issue is not with the specifics of which species are being presented as suggestions (I suspect that none of the suggestions are in fact correct in any of the interfaces, in this specific example) but that species not found or expected on the same continent are being presented as “expected nearby” in iNat Next.