Portal de Periódicos da CAPES

Can citizen science and social media images support the detection of new invasion sites? A deep learning test case with Cortaderia selloana

2024; Elsevier BV; Volume: 81; Linguagem: Inglês

10.1016/j.ecoinf.2024.102602

1878-0512

Ana Sofia Cardoso, Eva Malta-Pinto, Siham Tabik, Tom August, Helen E. Roy, Ricardo A. Correia, Joana R. Vicente, Ana Sofia Vaz,

Environmental DNA in Biodiversity Studies

Deep learning has advanced the content analysis of digital data, unlocking opportunities for detecting, mapping, and monitoring invasive species. Here, we tested the ability of open source classification and object detection models (i.e., convolutional neural networks: CNNs) to identify and map the invasive plant Cortaderia selloana (pampas grass) in mainland Portugal. CNNs were trained over citizen science images and then applied to social media content (from Flickr, Twitter, Instagram, and Facebook), allowing to classify or detect the species in over 77% of situations. Images where the species was identified were mapped, using their georeferenced coordinates and time stamp, showing previously unreported occurrences of C. selloana, and a tendency for the species expansion from 2019 to 2021. Our study shows great potential from deep learning, citizen science and social media data for the detection, mapping, and monitoring of invasive plants, and, by extension, for supporting follow-up management options.