From citizen to scientist: evaluating ant observation accuracy on a Mediterranean archipelago

Citizen science platforms, such as iNaturalist, have become invaluable tools for biodiversity monitoring, allowing non-experts to contribute photo-based observations that are then identified by the user community. However, one of the major challenges of these platforms concerns the accuracy of these identifications, especially for taxa requiring specialized knowledge or more sophisticated means of identification. In this study, I assess the reliability of ant identifications in the Balearic Islands by comparing community-generated identifications with expert validations across multiple taxonomic levels. Based on 300 iNaturalist observations, I analyze the influence of user experience, AI-generated suggestions, image quality, and community participation on identification accuracy. The results indicate that species-level accuracy was 69.91%, increasing to 90.91% at the genus level and 99% at the family level. Experienced users significantly improved accuracy, while early-stage identifications at finer taxonomic resolutions increased the likelihood of correct consensus. AI-assisted identifications performed well for frequently recorded species but struggled with underrepresented taxa. Surprisingly, photo quality had minimal impact, as common species were often identifiable even from low-resolution images. Additionally, the dataset documented exotic species, including the first record of the Formica rufibarbis complex in Mallorca. These findings highlight both the strengths and limitations of citizen science in taxonomic research and emphasize the need for strategies to enhance data reliability for conservation and biodiversity monitoring.