Yeah that doesn’t make any sense though. The CV is clearly recognizing the genus. I just tested this though. There are only two valid species in the genus Aiolochroia. One is essentially endemic to California, the other is only found in the western Atlantic. Their ranges don’t even come close to overlapping. It no longer recommends A. Crassa, only the genus for what is very obviously A. crassa in the Caribbean now. It’s doing the same thing for Amphimedon nodosa and Desmapsamma anchorata. In each of these cases there is either only one or two local species and some of them are some of the most common sponge species recorded on the site and the CV never had any issue identifying them before. I also just ID’d a Halisarca caerulea which the CV never used to have any issues identifying. Its only suggestion now is an octopus. And this was a clear picture. This is literally the worst iteration of the CV I have seen since using the site.
I just checked the geomodels for several species and that wouldn’t explain it either because there’s still only one species in the genus described from their ranges even in iNaturalist’s internal systems for some of these.
I found an example of a [Desmapsamma anchorata](https://www.inaturalist.org/observations/135181942) that exhibits the behavior you’re describing. If you select “include suggestions not expected nearby,” the top CV suggestion is Desmapsamma anchorata; if you select “only view nearby suggestions,” Desmapsamma anchorata does not appear in the list of suggestions. This is because Veracruz is not included in the geomodel for Desmapsamma anchorata, despite several nearby RG observations of the taxon.
There’s a similar situation with this observation of Aiolochroia crassa. The observation falls outside the geomodel, so the CV does not list Aiolochroia crassa when suggesting nearby suggestions, although it does suggest the genus.
I think you are assuming that the geomodel uses more intuition than it actually does. It does not think “this looks like Aiolochroia, and among Aiolochroia species there is only one in the Carribean, so I will suggest that taxon, namely Aiolochroia crassa.” It goes “this looks like Aiolochroia, but the location does not lie in any of the blue hexagons in any of the geomodels for any of the Aiolochroia species, so I will not suggest a species-level identification.” It doesn’t matter whether the location is one inch outside the hexagon, as far as I can tell.
As mentioned above, there are many examples discussed in this thread where the geomodels are wrong and, for whatever reason, exclude locations with many RG observations. It seems like this issue is most pronounced in areas with significant elevation differences (e.g. a mountain range) and near/in oceans.
In summary: the issue you are seeing seems to stem from bugs in the geomodels and perhaps the rigidity with which the CV uses the geomodels to filter out suggestions. The issues disappear when you toggle off the restriction to nearby suggestions.
It looks like the CV is not trained on Halisarca caerula, so that taxon cannot be suggested by the CV. This is because there aren’t enough research grade observations of that taxon to train the CV on reliably. Perhaps there used to be more RG Halisarca caerula observations on the site, so maybe the CV was previously trained on this taxon.
I’m pretty sure H. caerulea is a species that the CV used to be able to ID. And I’m pretty sure the sponges I was selecting were not in Veracruz. It kind of doesn’t make any sense as to why some of these species are not appearing in the geomodel when there’s multiple instances of them from that region. But then it is appearing in places where there’s no observations of them apparently, like the Pacific coast of Central America, so I don’t know what that’s about. And again, this doesn’t really explain why it’s suggesting to the species level animals where there are zero observations in the entire ocean they’re being suggested from by the CV. Especially when there’s a similar one that’s common in the area like the Xestospongia example I mentioned. Maybe the latest training data they used for the most recent iteration had one or two examples? But again that makes no sense to pick the one that’s 100x less common. I just checked the geomodel for X. muta. It’s not listed anywhere outside of the Atlantic. Well aside from that thin strip on the Pacific coast of Central America again for some reason.
I agree this is frustrating but this is a known problem. Some of why it happens is discussed here:
https://www.inaturalist.org/blog/115962
I really don’t get why they don’t add in some hard rule that, e.g., any location within 10km of an existing obs counts as “expected nearby” no matter what and any location >1000km away from the existing obs cannot count as “expected nearby”. I’m sure this would be far from perfect, but feels like it would be a substantial improvement over what we have now.
Another one. This is the only observation for this species outside the entire Indo-Pacific region. CV recommended ID. C. apion/alloclada is a common species in the area.
The web CV isn’t suggesting that species at least. That observation and ID are coming from the iNat Next app where the CV is more eager to recommend species and isn’t kept as up to date.
No, I just checked. The CV doesn’t recommend any Cinachyrella, until I include suggestions not expected nearby. Then it only recommends C. australiensis.
CV does not recommend C. apion/alloclada because they don’t have enough observations (exactly 47 each, coincidentally). The iNaturalist CV only includes species with at least 60 observations and 100 photos. The only way to get the CV to start recognizing and recommending these species is to observe or ID more of them.
Edit: I suspect the observer may have clicked “include species not expected nearby” to get their ID, but I’m not certain.
This is one of those issues that casual users unfortunately won’t understand. I eventually learned to go higher up in the taxonomic tree, to genus or even family. But that takes a level of knowledge many don’t have (or forgot from science class).
I can only agree wholeheartedly, I have spent so so much time correction automatic suggestion where a genus level ID would have been correct. Taxons difficult (edit: or impossible) to identify on habitus alone (aka the vast majority of described biodiversity) should see something changed with the CV.