Andrea Zampetti spent three months in the Brazilian forest with a camera trap pointed upward, watching for flickers of movement in the canopy above. The Ph.D. candidate from Sapienza University of Rome was chasing something most artificial intelligence models had ignored: the hidden world of tree-dwelling animals that live beyond the reach of conventional wildlife monitoring.
The result is TropiCam-AI, an artificial intelligence model designed to identify arboreal species—primates, birds, small mammals, and others that make their homes high in the forest canopy. It represents a meaningful shift in how scientists can study ecosystems that depend on creatures we rarely see.
Until now, AI models for camera-trap analysis have focused almost exclusively on ground-dwelling animals. But arboreal species play an outsized role in tropical rainforests. Primates, small mammals, and birds together consume up to 90% of plant species in these forests, serving as essential seed dispersers that keep ecosystems functioning. Yet these tree-dependent creatures face unique threats—especially from deforestation—making it critical to track and monitor them. The technology simply hadn't caught up with the need.
Working in collaboration with the TROPECOLNET project at Spain's National Museum of Natural Sciences, Zampetti built TropiCam-AI by gathering training data from across the tropical Americas. His expedition to Brazil provided the foundation, but he expanded the dataset by collecting camera-trap images from researchers working in Peru, Costa Rica, and French Guiana. He also drew from iNaturalist, a massive citizen scientist platform. Each image required manual annotation—researchers reviewing them to identify the species—before the algorithm could learn what to recognize.
The model now recognizes 84 taxa, including 63 distinct species, with an overall accuracy of 95%. More impressively, 50 of those 84 taxa achieve better than 90% precision and recall, meaning the model confidently identifies what it sees. The system is designed with a practical humility: when an image is too unclear for reliable species identification, rather than forcing a wrong guess, the AI steps back and tells users the species belongs to a particular genus. That honesty matters when decisions about conservation depend on accurate data.
What makes TropiCam-AI transformative is its potential to accelerate the work of field ecologists and conservation practitioners. Processing camera-trap data can mean analyzing millions of images—a task that once required months of human labor. "It's critical we speed up the pace at which we can gather and analyze data and transform it into useful information," Zampetti explained. The tool allows scientists to feed their footage into the system and receive automated species identifications, freeing researchers to focus on the ecological insights hiding within the data.
Yet the team sees this as a beginning, not an endpoint. Zampetti and his collaborators continue refining TropiCam-AI with new training data from researchers eager to contribute. The model can only recognize what it has learned, so expanding the dataset means improving accuracy and expanding coverage to more regions and species. As tropical forests face mounting pressure from human activity, having tools that let us understand and protect the creatures living in their canopies has never mattered more.