If you ever have time to do this I would be very interested! I have just started a traineeship on edaphic ecology at the Pyrenean Institute of Ecology and I would love to help.
5 Likes
Glad to hear it! I have some drafts for identifier guides and should hopefully have those released soon.
9 Likes
Thank you for taking on this task, fixing mine and explaining. I love to learn.
2 Likes
Here’s the new numbers since the last CV update, to compare with my original tally that spanned a few different updates:
Out of 1624 worms identified using the CV, 47 were correctly identified to species, 480 correctly to a taxon above species, 710 incorrectly to species (of those, 440 to Lumbricus terrestris), 369 incorrectly to a non-species taxon, and 18 observations were not actually of earthworms.
All IDs: 32.5% correct, 67.5% incorrect.
Species IDs: 6% correct, 94% incorrect.
Nonspecies IDs: 56.5% correct, 43.5% incorrect.
It seems like when observers use the CV cautiously, it’s a little more right than wrong (seeing as how there is only one widespread order of earthworms, that’s still pretty bad! most family and genus IDs are wrong) but it’s even worse than the CV of some months ago when trying to ID species. Not very encouraging, but I’ll still hope the next update changes something.
My thanks to the few people who are uploading identifiable photos and/or IDing cautiously!
18 Likes
Please enjoy this great Instagram reel that I heard was inspired by this forum post. 
9 Likes
In the UK and Ireland, I manage (as a volunteer) the National Earthworm Recording Scheme. This involves curating a database of confirmed or trusted earthworm species occurrence records from research projects and biological recorders. Research Grade earthworm records flow into a UK-based recording platform called iRecord, where they are checked by me (as the National Recorder for Earthworms) and assigned a verification status base don our verification protocol: https://www.earthwormsoc.org.uk/verification
I presented the issues with iNaturalist earthworm records, including the algorithm problems, to the iNaturalist UK User Group in September 2022.
I’ve set up a small unfunded project to try and fix the algorithm, by encouraging a small group of volunteers to collect live earthworms and bring them to monthly sessions that I host at the Natural History Museum in London. We photograph the earthworms live and then preserve the specimens so that we can identify them following best practice (using a microscope and features such as the head type, setae spacing and location/shape of the tubercula pubertatis). Each specimen is identified by at least the volunteer recorder and myself, and then we add our identifications to iNaturalist. I don’t expect to make the algorithm perfect, but I hope that we can at least improve the accuracy of the confidence level which it assigns to the suggestions it makes.
More info can be found here: https://biologicalrecording.co.uk/invertebrate-study-days/
19 Likes
Welcome to the Forum and sounds like a great project! Just remember to correctly mark the observations as captive/wild based on whether observations are made for time/place of capture (wild) or the workshop (captive).
1 Like
Sounds like a large-scale equivalent to quite a few other invertebrate groups, many of which have cryptic species, and genitalia-only IDs possible. I actually did get some training in earthworms, but it was over 30 years ago. Still have my Soil Biology Guide, though no idea how up-to-date the taxonomy is now.
The suggestion to host training sessions is an excellent one! Since being invited to those by Zachary, I’ve ended up a devoted fly identifier (along with others in the group). I was always going to become an identifier, and had already started, but because he likes flies, that seemed to set my area, too. A similarly subversive effort to make others want to do worms might well work.
5 Likes
Soil Bio Guide is still the most recent key to the endemic US genus Diplocardia, and I recently got myself access to a copy. It is a horrible key to use (intestinal origin, my least favorite character).
2 Likes
Would a task list of taxa that identifiers can help work with (without requiring training) be a useful thing to provide? For example: CV identified genus-only Metaphire are almost universally in need of moving back to family or are misidentified members of other families; Octolasion shouldn’t be identified to species unless segments of the head can be counted, etc.
1 Like
Write a journal post - then it is easy for interested identifiers to use. Or is there an earthworm project ? with journal posts - which would be the easiest to find. iNat staff have promised an ID-a-Thon for December, perhaps you can use that momentum too.
3 Likes
Here’s how things have gone after the October CV update, I’m starting a new tally for November’s now:
Out of 1622 worms identified using the CV, 99 were correctly identified to species, 511 correctly to a taxon above species, 649 incorrectly to species (of those, 279 to Lumbricus terrestris), 347 incorrectly to a non-species taxon, and 16 observations were not actually of earthworms (there were some fishes).
All IDs: 37.6% correct, 62.4% incorrect (32.5%, 67.5% previous CV)
Species IDs: 13.2% correct, 86.8% incorrect (6%, 94% previous CV)
Nonspecies IDs: 59.6% correct, 40.4% incorrect. (56.5%, 43.5% previous CV)
Incorrect L. terrestris: 17.2% of total earthworm IDs (26.7% previous CV)
Slight improvements since the RG Lumbricus terrestris purge, but still pretty terrible overall. Instead of everything being misidentified to Lumbricus terrestris, now they’re just CV misidentified to Aporrectodea caliginosa and terrestris! The geomodel for terrestris is still the entire planet despite the corrections made–will it take correcting all the Needs ID misidentifications for that species to stop showing up “expected nearby” in equatorial locations where it has never been present?
In the latest update several species were added to the CV, some of them very useful, like Octolasion lacteum and Metaphire hilgendorfi. Amynthas agrestis is lined up for training, and with Amynthas tokioensis and hilgendorfi in the CV, hopefully it will start suggesting family rather than species for these nearly identical pheretimoids. Won’t be until next summer that we can see if it works, though, as frosts have done in most of those in the US for this year. As a bonus there are some nice colorful endemics like the beautiful Fletcherodrilus fasciatus in the queue that the CV ought to perform well with eventually.
Still dragging my feet on making identifier guides, but I really hope that I can get them out before spring. This year’s CNC was the worst yet for earthworms and I still haven’t gotten through it yet, let alone any previous years’. Thinking about a 2026 CNC with even more worms and even worse photos is highly discouraging.
11 Likes
It’s complicated, I would start out systemically wiping out observations by region. All of Africa could be a start, then the Middle East, South America, etc. Starting with managable areas that likely have lower observation counts.
It seems it may use any verifiable observations. When I did my cleaning of Diamesa some time ago, really I focused on all observations which was about 3.4k. The genus was actually unlearned and only has about 70 observations now.
https://www.inaturalist.org/geo_model/345991/explain
6 Likes
@thirty_legs Hi, I have noticed your IDs on many of my worms and am always very grateful since no one else seems to ID them.
Since I’m fascinated by polychaetes and my annelid literature includes some stuff about Clitellata as well, anyway, I‘m willing to help. I can prioritise that group for now over the bristle worms.
I‘d also be interested in helping with a photo guide, if you want. In Winter, there‘s not that much else to photograph, anyway. :)
3 Likes
How are you determining what the CV suggested vs. what the user chose?
I mean, I’m sure the CV is responsible for a lot of wrong and completely implausible earthworm IDs; however, I frequently see wrongly ID’d bee observations where the correct ID was among the CV suggestions (sometimes even the top species suggestion) but the user chose a different CV suggestion. So it’s a bit more complex than the CV having bad or limited training data and these mistakes automatically being perpetuated in every new observation. Improving the CV will no doubt help, but it won’t fix user error or overconfidence in choosing an option from very similar-looking taxon photos.
1 Like
Yeah, I realized that a while back: some fraction of CV misidentifications are on the user, who had a chance to pick the right option but didn’t, and some identifications using CV are correct because the user actually knows the worm and just used the CV as a shortcut instead of typing the whole name out. Still, from what I’ve seen, simply offering species IDs in the CV for people to choose will lead to near-universal misidentifications/unverifiable IDs because the majority of earthworms identified to species aren’t even identifiable to family from the photos. If you’ve taken blurry photos of some worm in India identifiable to order only but all the CV gives you are three temperate European species and a North American genus to pick from, all of those are going to be wrong.
My goal for this tally, though it is probably a waste of time doing so, is to prove that the CV cannot improve past a certain point. I do not expect we will ever reach a point where even half of the uploaded observations of earthworms are identified correctly by the CV. This is because there are so few earthworm species that can be identified from photos, and these species are scattered throughout the phylogeny of earthworms in such a way that the CV’s method of forgetting higher taxa it once knew in favor of only leaf taxa inevitably causes misidentifications. For example, Pontoscolex corethrurus is the only Rhinodrilidae that can be reliably identified from photos because it has been introduced worldwide. Because it is pinkish (unpigmented) but the Rhinodrilidae include some 380 species of which most look nothing like Pontoscolex (dark gray-blue pigment, large size), the CV now and probably forever will assume they are members of some other family, having “forgotten” what Rhinodrilidae look like. Instead, some change needs to happen in the CV. I do not want to have to be correcting hundreds of earthworm observations a day for as long as iNaturalist exists.
6 Likes
I already did that in the spring and the CNC + summer observations while I was away reversed all my work… ugh. Makes sense though. Hopefully northern hemisphere winter will slow down worm observations and I can focus on trying to fix geomodels now, it’s hard just getting through the daily uploads.
2 Likes
How many people know worm species? You have to have heard of something before to even type its name. Unlike bees, worms and Chironomids alike are much much more niche. The odds of somebody knowing a scientific name of either is extremely low. Hence almost all lower IDs like species come from the CVs suggestions. Or if a user has observed a species so much they remember it.
For example, I only know one worm I could assign a species ID off the top of my head only because it is such a problem. Every other worm I would need the CV, or to do my own research to know what their name is to even type.
Do you want help? I’m assuming one doesn’t need to know anything about Lumbricus terrestris to help in pushing them to a higher rank because they don’t live there. Anybody can push observations if told where they certainly aren’t living.
1 Like
Sure. For anyone interested, the map of Research Grade Lumbricus terrestris is roughly where this species should occur: https://www.inaturalist.org/observations?quality_grade=research&subview=map&taxon_id=81545
Canada and USA except for southeastern states, most of Europe and Russia, New Zealand, and some of the southernmost South America. It’s apparently found in Tasmania but nowhere else in Australia; no observations of it there yet.
Lumbricus terrestris is always brownish red with a clitellum on segments 32-37. If it isn’t brownish red, at least in the head, it isn’t L. terrestris. Observations of unpigmented (pink) worms are easy to rule out. Identifying to Crassiclitellata or Oligochaeta is best if you are not familiar with family characters in earthworms. There are similar-looking species in Europe and the US so when in doubt there is no need to add agreeing terrestris IDs. Feel free to @ me in any worm observation for any reason.
4 Likes