Renaissance herbarium sheds light on floral change

‘Inestimable importance’: 500-year-old cache of pressed flowers reveals new secrets (

1 Like

These old collections if preserved well can also be used to compare DNA and other chemicals e.g. The Rothamsted Sample Archive | Rothamsted Research one of the collegues here at OU did DNA work with these old plant samples.

In the article was mentioned Silver Cranesbill; this seems to be Geranium argenteum. As it isn’t ticked in my iSpy Alpine list (meaning I haven’t knowingly seen it) I looked it up in Flora Europaea. The description of the leaves is: densely silvery-sericeous on both sides, divided 95% of the radius. lobes deeply divided , each with usually 3 iner-oblong segments The habitat is given as :Calcareous rocks and screes.

I picked a GBIF record at random - this one.

The leaves - there are only leaves in the photo - are nothing like the description.

Then I clicked on the source of the observation. The source is given as PlantNet - is that wherein lies the problem?..

So what is the algorithm for accepting these erroneous offerings?

I then looked at some other, randomly selected, gbif records of this species and they look at least a good match.

I used to think gbif was a good site - now I am not so sure.

As I understand, PlantNet was originally trained on large samples of curated images of a variety of Western European Plants. But it’s subsequently used user-supplied data for further training, which involves smaller and less reliable data sets.

I looked through the images for several species of Malvaceae, and there’s a (finger in the air) error rate of 10% - for example Abutilon x milleri, Abutilon striatum, Abutilon x hybridum and even Crinodendron hookerianum misidentified as Abutilon megapotamicum (which you’d think is an easy plant).

PlantNet seems to have a higher error rate for Geranium argenteum - there are quite a few plants in there that look like Geranium molle and Geranium pyrenaicum, and some that look like Geranium argenteum except that they’re glossy green rather than silvery. Of the 4 observatons that PlantNet admits are questioned one was Geranium robertianum, one was Geranium pyrenaicum, one may be the pale form of Geranium sanguineum; the last has Geranium cinereum as the other offered identity, but Geranium cinereum is beyond my competence.

One of the drawbacks of PlantNet is their doesn’t seem to be any feedback (notifications) when you offer alternative identifications of make comments. It tells me that I’ve made 104 determinations, and 10 have been accepted as the correct name, but I don’t even see a way to relocate the relevant observations.

GBIF has a whole page on its site devoted to the way in which they ensure that data meet their standards. Perhaps someone should write and give some examples of where PlantNet fails to meet those standards?

Thanks for the Data Standards page which I will read leter today.


From Data Standards I moved to the GBIF page Pl@ntNet observations

Where I read:


Quality control

The validation is based on three main criteria:

  • the identification confidence score greater than a given threshold, the score being inferred from (i) the output of the automated identification algorithm (categorical probability) and (ii) the species names proposed by the members
  • the contributor has a reputation score higher than a given threshold
  • the species name matches the checklist considered as the most trusted one for the country where the observation was done.

I don’t know about points 1&2 but G. argenteum is not in Flora Helvetica, so the photo which I cited does not pass GBIFs own criteria.

they have a contact page: Contacts
but how well it is monitored, I have no idea.

Thanks re contact page; I saw it and wondered… If any of us gets beyond wondering and contacts them, we could share their response.
Later: I notice they give directions to their office …
Visitors arriving from Copenhagen Airport can simply take the Metro M2 from the airport, change line to Metro M3 (direction via Østerport) at Kongens Nytorv St. and get off at Vibenhus Runddel St.

So no excuse really…

Even later… So I have emailed them.

It needs to be someone who has enough knowledge about flora to be taken seriously - in other words, not me.

I have had an immediate reply from GBIF. It is informative and, if there is interest on this forum, I am happy to post it for others to read.

1 Like

Yes please. I am interested.

OK here is the GBIF response. I thought it informative. The specific query relates to Geranium argenteum (you can scroll up to see it)
… …
From GBIF helpdesk

helpdesk [email protected]

Dear Jo,

Thank you for contacting us and thank you for the feedback.

As you noted, the data were provided by Pl@ntNet to GBIF. Each GBIF data provider is responsible for the quality of the data they share online. The criteria mentioned on the Pl@ntNet dataset page on GBIF are their own.

With that in mind, we always forward the feedback we receive to the original data providers. They are often keen on correcting the mistakes reported. I have sent your feedback to Pl@ntNet.

In general, the quality of the data shared on GBIF is heterogenous. Some users might choose to exclude data from citizen science initiatives (like Pl@ntNet) entirely, or only trust records of preserved physical specimens. We have a few resources which can help users navigate the data on GBIF:

Unfortunately, there is no way to automatically detect misidentification. We rely the expertise of the data providers and on the feedback we receive. If you notice more issues, please let us know, it really helps. Thanks!

I hope this helps. Please let us know if you have any question or if you encounter more mistakes and issues. Thanks!
All the best,


Thanks for sharing. So like NBN, GBIF relies on data suppliers to ensure data quality. That is fair enough. It would be a massive job with much duplication of effort for these data amalgamators to try to do verification. But then you have iNaturalist taking the view that it doesn’t matter if they send low quality data to GBIF because it is up to the end user to ensure the data they use are of good enough quality. That is reasonable if you are preparing a paper and using a small number of records, but it doesn’t work if you just want to dip in to the dataset to look up a distribution.

Exactly. It seems we cannot use gbif records with certainty. Marie says she’s informed PlantNet of my feedback. I wonder if they’ll take any action.

Lavateraguy said there are Malvaceae problems.
Marie says « Please let us know if you have any question or if you encounter more mistakes and issues ». So I guess we could send her more if we want to.

Interesting to find out this stuff, which just came to my notice as a plant that I might have seen; but it’s clearly In rather limited habitats. People do seem to like to post rarities, without really checking. Oh well.

I was wondering if you, Lavateraguy, had sent your comments on Malavaceae & Geranium argenteum to Marie at GBIF as she has asked for issues.

I just checked to see if anyone at GBIF had removed the G. argenteum Swiss error; it’s still there.

Following up, I see 3 Swiss records for G. argenteum on GBIF. The other two seem likely to be valid. One is listed as a Specimen in Bern museum. The other was as a Human Observation by Rita Gerber who is (or was, in 2009) the Secretary and Membership Support for the Bern Botanical Gardens.
Rita probably had a list of the plants growing in their (quite famous) Alpine Gardens.

The Wonders of the World Wide Web eh?

Update on G. argenteum. GBIF still has the wrongly identified Geraneum argenteum. L. So I have now emailed PlantNet to ask them how they deal with reports of erroneus data, which have been trawled and displayed by GBIF database.
I’m not holding my breath, but maybe they’ll respond.

Have you heard anything yet?

Thank you for asking. I got this:

Hi Jo,

We try to update our data on gbif regulary. In this case, contacting us through this is the correct way to report this issue.

Vanessa will double-check the observation and the next GBIF update will have revised data for this observation.

Pl@ntNet is a citizen science platform, any users can help change or improve the determination of observations.

Best regards from Montpellier, France,

1 Like

An update: From Vanessa, my contact at PlantNet,

Hi Jo,

Thank you very much for your report!
I checked the observation mentioned which is in fact misidentified… Unfortunately I realized that many other observations are also misidentified… It seems that our AI is getting a little lost with all these sterile Geraniums. This is not just one observation, we have work to do in cleaning this group.

Thanks again for warning, we definitely need users’s help to track errors in the app!

Happy new year!