CV Has Stopped Recognizing Eastern Grey Squirrels In Introduced Range

Platform: Website

Browser: Firefox 124.0.2 (64-bit)

URL of relevant observations:


squirrel issue

Description of problem:

Step 1: Upload a photo of an Eastern Grey Squirrel in southwestern British Columbia (the issue does not occur in the EGS’s native range - for example, the CV identifies it correctly in Ottawa, Canada).

Step 2: Check the CV suggestion.

Step 3: The CV never suggests Sciurus carolinensis, despite it being a clear visual match and by far the most commonly-observed squirrel species in the region.

Discussion: I first noticed this in the past few months. My EGS observations used to always receive the correct CV suggestion, but now the CV seems to want to give me only native squirrel species as suggestions (EGSs are introduced to SW British Columbia), even when they clearly don’t match. Given that the CV correctly identifies the EGS in its native range, I think something wonky is happening with how the CV uses range information to inform its suggestions.

How long has it been? There was a major change to the geographic modelling in September, although the model’s expected range for EGS looks fairly accurate (other than including Bermuda, Cuba, the Bahamas, and parts of Europe where it doesn’t exist).

That’s a great question; I wish I’d taken note of it when it first happened. Looking through my observations, I think I noticed the problem with my observations going back to March 17th. My last observation before that was January 12th, and I don’t think it was affected, though I don’t fully trust my memory. Prior to that, my last observation was August 20th, 2023, and I’m confident that one was not affected.

I’m not sure how precise the geomodel is adhered to, but the map you shared might indicate the problem; the northern part of the Vancouver metropolitan area is excluded from the map, even though the squirrels are plentiful in that area. That’s where several of my observations are. I see the map is composed of hexagons; the hex “missing” from the map also includes the mountains north of the city, where the squirrels are uncommon, so maybe that’s why it was excluded?

1 Like

Still recognises it in the UK, although the native squirrel is completely extirpated from most of the country:

What are the coordinates for the observation?

i don’t think there’s a bug here.

clarifying a bit, it’s not that the CV doesn’t recognize and suggest S. carolinensis. it’s that it won’t suggest it if you include only suggestions expected nearby. if you include suggestions no expected nearby, you can see that S. carolinensis is the top suggestion (i.e. the top visual match):

if you look at the geomodel range for S. carolinensis, you can see that the geomodel thinks the taxon could exist in the area (in the unthresholded view), but in your particular location, it falls below the threshold for being expected nearby.

i think part of the issue is threshold, and part of the issue is the granularity (or lack thereof) of the prediction, but the main thing is probably just that no model is going to have perfect predictions for every taxon.


The latest observation I had this with was this one from today; the coordinates are:

Lat: 49.295806
Lon: -123.145903
Accuracy: 6m

It could well be that the model just can’t be granular enough to expect the species nearby in this particular locale; I’m not sure what the policy is on whether the geomodels should err on the side of too inclusive or too exclusionary. I just felt that at least in my local area, the sudden switch to a worse suggestion regime felt noteworthy.

@pisum’s right, the location is juuust outside the geomodel’s range.

It was definitely a topic of discussion, and we had to draw the line somewhere. We tried to find a good balance there will always be cut-offs that cause these issues.


This topic was automatically closed after 14 hours. New replies are no longer allowed.