Pl@ntNet

Pl@ntNet previously relied on a heterogeneous taxonomic backbone with uneven quality and incomplete coverage across countries. To address this, the platform transitioned to a globally standardized system based on Plants of the World Online (POWO), maintained by Kew Gardens and integrated with the World Geographical Scheme for Recording Plant Distributions (TDWG). This migration required significant updates to the database, AI workflows, and interface, enabling users to select region-specific floras through a map or GPS. 

Aligning Pl@ntNet with POWO enhances interoperability with other biodiversity platforms, facilitating the integration of resources like IUCN species status and native ranges. It also ensures consistent taxonomy, improving data quality and enabling easier data sharing with platforms like GBIF. Regionalized floras now support case studies tailored to specific areas, including southwestern Europe for Spain and France, Eastern Europe for Greece, East Asia for Cyprus, and the western Indian Ocean for Madagascar.

The major benefit for the users is that they can now select the flora best suited to their location. This can be done either by selecting it on a map (see here for the web version), or automatically with the GPS if the user has authorized the use of his position in the application settings. Selecting the appropriate flora in such a way rather than working at the world flora’s level enables significant gains in the identification performance. The other benefit is that all other screens and features of the Pl@ntNet app are contextualized with this regional flora. The IUCN status of species, for instance, has been integrated as part of GUARDEN’s requirements. The native range of species provided by POWO is also used to determine whether a species is native, introduced, or probably erroneous in a particular regional flora. 

Secondly, the core AI technology enabling species identification in vegetation survey images was developed as part of the third work package, while the integration of this functionality into the Pl@ntNet application was carried out under work package 4. It required extensive work on the database built with ArangoDB and the data access layer developed via JavaScript. Previously, the platform did not support HD images, and the existing data schema was unsuitable, as classic Pl@ntNet observations typically included one individual per species rather than multiple individuals of various species. A new data schema was therefore created to accommodate this type of data, allowing surveys to include multiple HD images of the same plot and link them to individual plant observations within the plot. Database query mechanisms, such as index optimization and data verification procedures, were also developed to ensure consistency, including validating that plot-associated observations align with nearby GPS coordinates. Additionally, new routes were implemented to enable front-end access to plot data and interaction with the AI engine for species identification.

The front end, developed in Vue.js, comprises three main views. The home page provides a summary of all plots and allows users to create new ones. The user page displays plots associated with a specific user, while the plot page focuses on an individual plot, offering a range of interactive features. Users can edit metadata (title, description, date, EUNIS habitat label, flora, and geolocation), add HD images and classical Pl@ntNet observations, and leverage automated tools for species identification, coverage percentage calculation, and phenological stage determination. Additionally, users can manually edit, complete, or correct the species list, coverage, individual counts, and phenological stages to ensure data accuracy and completeness.

The third new feature developed in Pl@ntNet as part of GUARDEN is to enable the import of observations by batch within the database. We previously noted that many institutions and private individuals have large batches of plant observations and data that they would like to share on the platform. However, importing observations at present can only be done by uploading them one by one, which takes a considerable amount of time. This constitutes a major obstacle to integrating existing data, particularly for the rarest species and the least surveyed regions not observed by general Pl@ntNet users. 

The interface provides a comprehensive toolset for managing and identifying plant images efficiently. Users can select and import batches of images through the file explorer, along with any available EXIF data or metadata supplied in CSV or Excel files. The system automatically identifies the most likely species for each image, while also allowing users to manually edit or revise species names and associated metadata, such as date, location, and comments. Multiple images can be consolidated into a single Pl@ntNet observation, ensuring streamlined data organization.

To maintain data quality, users can flag potential issues such as non-plant images, observations missing dates, or unidentified images. Finally, the interface offers the option to select the relevant flora where the observations will be uploaded, ensuring accurate contextualization within the Pl@ntNet platform.