Identifying weeds, checking out the pollen map, or discovering new plant life-forms are among the promising wealth of data available to users of PlantNet – a “Shazam!” for plants. Pierre Bonnet and computer scientist Alexis Joly introduced us to the digitally enhanced plant recognition application they developed.
The Conversation: What can you tell us about PlantNet users?
Pierre Bonnet and Alexis Joly: An impact study carried out a few years ago identified that 12% of users used the app for work, either for research, land management, farming or teaching purposes. The large majority of users use PlantNet out of curiosity or personal interest.
Looking at who and where PlantNet users are based reflects the way technology is used in different parts of the world. In Asia, we have quite a few young users, because they are the most connected. We can also observe that the Chinese flora, which is nevertheless very rich, is poorly represented by the users of the application, and this is explained in particular by the fact that non-state or non-Chinese platforms, which are not carried by Chinese companies or partners, are much less present on this market.
Tropical parts of the world are very biodiversity-rich, is PlantNet particularly big there?
P.B. and A.J.: Brazil is in eighth place with just over 600,000 annual users. Indonesia and India are in the top 20. Currently, the bulk of PlantNet’s activity is still in Europe and North America. Several things explain this. PlantNet has already been launched in France and in Europe, and has therefore been further promoted and attracted media attention in this region of the world. Backed by user demand, the application was also initially adapted to the French and Mediterranean flora before it was gradually extended to include other European flora, then North American, and tropical.
It should also be taken into account that in tropical areas, species richness is certainly far greater, but access capacities are often more limited. The road network is less developed; 3G connectivity can be poor, particularly in forested mountain areas.
In the field, plant biodiversity can also be more complex to photograph, take for example, the many tropical plants that are epiphytic that is to say, that grow on another plant, especially at canopy level. When we talk about trees that are several tens of metres high, it immediately becomes more complex to photograph.
All this makes tropical plants and flowers considerably less known. The application covers almost 100% of European flora, compared with coverage of a few tens of percent for tropical countries. This is something that is not unique to PlantNet, and generally applies to all institutional databases, especially since covering tropical areas costs more.
But despite this, we are working with partners in Costa Rica, Guyana, Brazil, Cameroon, Madagascar and Malaysia, in particular to extend our coverage of the number of species.
In the tropics or elsewhere, what can be noted about the plants that users are looking to identify? Are the most common plants the most in demand or not necessarily?
P.B. and A.J.: There is necessarily a correlation, because very rare plants are necessarily rarely observed. But we also note that certain plants are very common, but of little interest, because they are “discreet”. These can be crop weeds, plants found by the roadside but which do not have noticeable flowers, which are pollinated by the wind with slender, greenish flowers, such as species of fescue (Festuca spp.), or bromes (Bromus spp.). They are less observed because they are actually less attractive. We can also note that tree searches are popular, whereas certain herbaceous or epiphytic plants tend to be extremely rare.
People actually often observe plants that they like, or plants that seem useful to them. Incidentally, we get a lot of requests about fruits, berries, and plums, probably because people want to know if they are edible or not.
But the goal of tracking useful plants for humans was not the main objective of the PlantNet project from the get-go. We had to adapt it, however, to meet expectations in terms of this type of use while remaining fairly moderate regarding the volume of information we provide.
At the same time, we are working more and more with people who study animal health, either in human health and who use the PlantNet service or data for their own work. Like for instance, people who worked for ToxiPlant which identifies plants that are toxic for horses. We also regularly consult doctors on different uses of PlantNet for identifying allergenic plants, especially those that cause skin allergies. We also liaise with the French regional agency for monitoring air quality ATMO Occitanie, which uses shared data on flowering plants listed through PlantNet that it integrates into its air quality estimation models indicating pollen counts.
Are there certain plants that would benefit from being photographed more?
P.B. and A.J.: Yes. There are plants that interest us, or our partners, but for which we have very little data. These include allergenic plants for respiratory allergies, such as male junipers, which release pollen when their cones open.
These cones are very discreet. Junipers are therefore photographed, but rarely with illustrations that show the development stage of the cones. Seeing as they cause allergies, some of our partners would like to collect more data on the subject. We hope to remedy this through animation features, either directly through the platform or through social media, to generate interest in collecting data on this type of plant or on rare, endangered plants or plants that have conservation issues, such as Marsilea strigosa Willd. and Arenaria provincialis Chater & P. Halliday (native to the South-eastern France).
We also have the case of certain plants that interest us for agriculture, weeds, for example, which we have barely identified at the stage of young shoots, such as ‘Imperatacylindrica_ (L.) Raeusch. or that Galium aparine L.
Your application encourages users to take several different pictures. This can be photos of flowers, leaves, fruits, bark, or of the entire plant, for example, to help them identify plants. What types of data do you have the most of?
P.B. and A.J.: Data on flowers, especially in their blooming seasons – during spring and early summer. Flowers attract interest and trigger observation. They are also the most effective visually, and have typically been used a lot by botanists in the past.
On PlantNet, after flowers, we notice a decreasing degree of performance for fruit, leaves, and then twigs and bark, which are sometimes more difficult to take pictures of, even if users are always encouraged to combine several criteria (flower and fruit, flower and leaf…) each time to maximise their chance of correctly identifying a plant.
What happens when PlantNet can’t match a photograph with an existing species?
P.B. and A.J.: Troubleshooting “no results found” is still a tricky business for all forms of AI, and PlantNet is no exception. Some species are very poorly represented, and it is very difficult to teach the model behind the app to differentiate between something that is very rare in the learning database and something that we do not have at all. We therefore seek to quantify the uncertainty, rather than decide when the model is uncertain or not. We are working with our team on creating confidence intervals. That is why, in some cases, the application will provide several species.
One thing that makes this work difficult is that new species often resemble existing species. A new species is thought to be very picturesque, but it isn’t always the case, hence the confusion.
There is also the issue of hybrid, ornamental plants. You will always come across new plant creations. We cover them, but not as well as other plants, even though we are trying to overcome this difficulty.
We also set out to simply identify more and more plants. Today PlantNet covers 85,000 species out of an estimated 400,000, with 2,000 new species being discovered each year. These discoveries are made by taxonomists worldwide and via the World Flora Online network, which brings together several dozen universities, herbaria and botanical gardens, which PlantNet joined in 2025.
Working closely with these networks will enable us to improve cases where PlantNet performs poorly, particularly when new species emerge thanks to the research carried out by the taxonomists, who actually further knowledge by dividing what was previously thought as a single species, but which represents, in fact, several of them or, conversely, by grouping together what were thought to be different species, but which turns out in reality to be only one plant.
Let’s come back to the 10% of users who used PlantNet for work. Who are they? What line of work are they in?
P.B. and A.J.: PlantNet’s data is very useful for building species distribution models predicting where a given species is commonly found. There are certainly biases in our data, depending on where people do and don’t go, but they can help us to gain a better understanding of the factors that influence these distributions, including climate change. This provides answers to questions like, “will species change habitats or stay put?” or “what is the human impact on species distribution?” Many ecologists download the PlantNet data and couple it with other data for species distribution modelling.
PlantNet data can also help in the detection of invasive species that are often notable for their size, density, visual appearance, or novelty, such as Carpobrotus edulis (L.) N.E.Br or Mirabilis jalapa L. We are working on this subject with the French Biodiversity Office, which is interested in using plant monitoring cameras to detect the presence of invasive plants, in order to contain their expansion within the areas where they are found.
We have an ongoing project that focuses on improving the classification of species that are present in farming environments and recognition of plant pathogens whether they are viruses, bacteria, fungi…
Have you discovered any unusual uses of the app that have surprised you?
P.B. and A.J.: Perhaps the most surprising was a Dutch museum which used PlantNet to identify plants featured in the paintings it had on display.
Other surprises have included people using PlantNet to identify a tattoo of a plant, or a plant-inspired pattern of an oilcloth… along with other more fun uses of the app like on the Explorama or Geo Quest apps which use our automated identification service.
PlantNet supports the diversity of possible uses, not by trying to integrate them, but rather by sharing its plant-identifying services. We have more than 20,000 accounts that have incorporated PlantNet’s service into their own mobile or web application, or in their data processing workflow.
And what other uses for PlantNet could be developed?
P.B. and A.J.: We have processed just over 1.3 billion plant ID requests. Among this data, there is a lot of material for characterising environments and species… However, the data is still difficult to use due to its sheer volume. These photos most probably contain new species and data on species that do not exist elsewhere. Photos users have posted provide potentially interesting information about the associated communities (not just in the foreground, i.e.; the plant photographed, but in the background) and about plants’ surrounding environments. While it’s not what the photo primarily sets out to do, it does offer potentially very interesting data on plant associations, that for the moment, remain untapped.
Little use has been made of our data to study the impact of current, fairly fast-moving climate change. PlantNet has, in this respect, collected extensive data over the last five to ten years that could allow us to gain greater understanding on how plants react to such swift environmental changes. From one year to the next, there may be a lot of variability, but for now, it is difficult to measure this impact.
What can PlantNet users do to help with further research going forward?
P.B. and A.J.: Creating a user profile significantly increases the benefits for research. By creating an account, people agree to terms of use and facilitate re-exploitation for research. The more qualitative the data is, the more relevant it is to research activities. Geolocation is, for example, very valuable to us. It is also beneficial for users with a potentially higher level of determination.
We encourage researchers around the world to feel free to explore the full potential of PlantNet, whether it be via our shared data or the services we offer.
Interview by Gabrielle Maréchaux, Environment Journalist at The Conversation France.
A weekly e-mail in English featuring expertise from scholars and researchers. It provides an introduction to the diversity of research coming out of the continent and considers some of the key issues facing European countries. Get the newsletter!
