Predict invasive plant distribution with Pl@ntNet to manage them better

You may forget it by using it, but Pl@ntNet is a citizen science project. The data collected in the application is used to make the research go forward. Today, it’s Christophe Botella, PhD student in statistics applied to ecology at INRA and Inria, who uses the information from locations of 5 invasive plants in France for his thesis. Its objective is to develop a model for predicting the distribution of these species.

Invasive plants
From left to right: Erigeron karvinskianus DC., Acer negundo L., Reynoutria japonica Houtt., Carpobrotus edulis (L.) N.E.Br., Opuntia ficus-indica (L.) Mill. Licence CC-BY-NC-SA: Jean-Jacques HOUDRÉ, Steven J. Baskauf, Hervé Goëau, Daniel Kieffer, Michel GAUBERT

“Invasive species represent a major economic cost to our society (estimated at nearly €12 billion a year in Europe) and are one of the main threats to biodiversity conservation”

To establish his model, Christophe Botella used the observations of species automatically recognized by Pl@ntNet from your photos. This geolocalised data made it possible to model and predict the presence of species thanks to environmental variables, including bioclimatic, topological or hydrological variables. Nevertheless, the distribution of Pl@ntNet observations is strongly correlated with human presence. Its model therefore includes urban areas, proximity to roads and the distance to the coasts to cancel this sampling effect, also called “pressure of observation”.

There are three types of specimens that form from his model: first, cultivated plants, i.e. those present in gardens or parks, which are grown or maintained by humans, second, “casual invasive”, due to their closeness in relation to human activity, and lastly, newly inventoried invasive plants (which are the most interesting to successfully manage in their proliferation).

In the end, the Pl@ntNet data does not exactly correspond to the distribution of plants according to the reference data (INPN), but this is explained by these three typologies and by the still imperfect methods of corrections of the biases due to the pressure of observation. However, this model makes it possible to detect new ornamental species that become invasive, or to clarify ranges of already invasive species.

A still limited choice

This approach could therefore be used as a decision support tool for the management of invasive species, but it has its limits. It is effective on easily recognizable species with Pl@ntNet, i.e. with visibly large patterns (flowers, leaves, fruits, etc) (those distinguishable from other closely related species). The phenology of the species is also important: the organ of interest of the plant must be present for a long time, particularly at the moment when users use the application (for example, especially in spring and summer in France). Finally, being totally subjective, this species must interest the public from its visual appearance to its smell and so on, to be pushed to be identified with Pl@ntNet, and thus to have its observation shared. One of the other limitations of this model is the presence of garden specimens, or indoor specimens. Being maintained by humans, environmental criteria, such as soil composition or water intake, is completely skewed with respect to the natural environmental conditions of the same area. Finally, some invasive species develop very easily in urban areas, which happens to be the favorite place to observe users of the application, causing a magnifying effect that will have to be countered.

This model, while imperfect, provides a new and promising vision for the future management of invasive species and thus, inevitably, the protection of biodiversity.

Complete article: Species distribution modeling based on the automated identification of citizen observations