I am working through the Donaciinae leaf beetles of North America, as they were my doctoral critters and I’ve done the taxonomic revisions of the genera. Many observations were ID’d as Palaearctic species and all are corrected to the extent possible. At least, all those so ID’d have been changed back to genus if I couldn’t put a correct species name on them.
As I’m filling in the ID field, Palaearctic taxon suggestions keep coming up, though some Nearctic taxa are now appearing as well.
Why are taxa not known to be in North America even appearing in suggestions at all? This can lead only to ID errors.
My guess would be that there were RG observations of Palearctic species in North America at the time the CV model was trained. The computer vision should improve as the incorrectly IDed species are corrected.
There’s a way to get the CV to suggest only (or primarily) things that are “seen nearby,” isn’t there? If so, how does one make the CV do that? Lately I’m getting a lot of suggestions that are FAR out of range, and I’d like to change that.
Unfortunately, after a wildly out-of-range CV suggestion is seconded by someone who assumes that iNat CV ought to be right, it does create the potential for a “seen nearby” error that is often then propagated through the system by other users who are also lacking knowledge about species ranges. It’s a common thing, in my experience, that frequently creates pages and pages of erroneous IDs.
In my experience, this has become worse with the update from “seen nearby” to “expected nearby” for some reason. The system has somehow adopted a broader definition of “nearby”.
And yes, there are big influences due to single observations - which looks like something that could be easily fixed.
When I see an obviously wrong seen nearby - I go to the distribution map - find that one (sigh which is sometimes multiple) and hammer them back. If we don’t correct the mistake, it gets worse and worse.
But now CV is updated about once a month, it can get tidied up reasonably quickly.
My copypasta - You have chosen a South African plant
On the one hand, it also works in the opposite direction. Yes, I too am sometimes a bit annoyed by the need to correct obviously incorrect identifications of European observations as belonging to Nearctic species. On the other hand, entries from the Palaearctic to the Nearctic and back again are so not uncommon that I created a project for such observations:https://www.inaturalist.org/projects/invisible-travelers-hitchhiking-animals. Whether we like it or not, the fauna and flora of the Holarctic are becoming increasingly similar ¯_(ツ)_/¯.
The problem with the last one is the CV refuses to suggest Phytocoris! 10,000 observations of the genus on iNat (~1,000 are RG) and it always shows me Parthenicus, even for a centered, cropped photo on a white or neutral background. For example: https://www.inaturalist.org/observations/182479995 (CV = Parthenicus) and https://www.inaturalist.org/observations/136766993 (CV = Miridae, 2nd option Parthenicus).
So it’s been 5 weeks since I posed this question. All Nearctic observations have been reviewed if ID as Palaearctic taxa, and have been correct to species if possible, or back to genus for later review. Most of the ID to species I’ve done were then agreed to by the observer, and the right Nearctic taxa are appearing in suggestions but still Palarctic taxa appear as well. But I’m following the subfamily and immediately correct any that come in wrong. So the CV training seems to have taken place at least once, in order for this to be happening. While it’s going the right direction, it’s looking like it’ll take many months, till I do IDs using my reference collection and keys, and work through the 1000s of “specimens”
How long does it take for the geomodel to update the “expected nearby” suggestions?
For quite some time the CV has been suggesting the North American species complex Lasioglossum gemmatum as “expected nearby” for bee observations in Europe. I believe the initial cause was a mis-ID’d observation in Europe for the taxon, but I have been regularly checking and correcting these observations, and I think there should have been at least one CV update in a period in which there were no European observations that had this as an ID.
Is it continuing to count non-withdrawn IDs even though they have been disagreed with? Or is the flora and fauna of continental Europe otherwise similar enough to North America that it thinks Europe is plausible?
OK, so what needs to happen so that it stops suggesting a North American species complex as “expected nearby” in Europe?
There were one or two European research grade observations of a species in this complex (Lasioglossum tegulare; misclick by an expert subsequently agreed to by the observer) which I pushed back to genus/subgenus months ago – long enough that there has been more than one CV update since then.
In any case, it is suggesting the complex, not the species – but if this regular trickle of unconfirmed and quickly corrected CV suggestions is enough that the CV continues to think that it is “expected nearby”, something doesn’t seem to be working as it should be.
(For many of these observations, the subgenus is probably correct, just not the species complex within it. This is a difficult group to ID, so subgenus is as far as we can go most of the time for European observations. So part of the problem here is the issue with the CV not being trained on higher taxa if a more specific taxon is included. The irony here is that this geographically incorrect suggestion means it is actually suggesting something within Lasioglossum rather than a completely different genus, which is what it would probably be doing otherwise.)