Large testing of AI-based camera traps on a forest defoliating pest in Europe

This is the authors' abstract. We don't add key points for this paper.

Background Recent advances in AI-based surveillance technologies have shifted agricultural pest monitoring from reactive to proactive, yet their application to forest pests remains limited despite the need for efficient monitoring across large, remote and difficult-to-access forest landscapes. The pine processionary moth, Thaumetopoea pityocampa , is one of the most damaging forest pests in the Mediterranean region and a species undergoing climate-driven distributional and phenological shifts. Methods We evaluated the performance of the AI-based camera system for monitoring the insect under field conditions. Automated traps were compared with conventional traps across four European countries (France, Italy, Portugal, and Slovenia), encompassing contrasting habitats and population densities. We assessed trapping efficiency, automated image recognition accuracy, probability of early detection, relationships between adult trap catch and local nest density, environmental drivers of flight activity, and non-target captures. Results Automated traps consistently outperformed conventional traps by catching significantly more moths in France, Italy and Portugal, while demonstrating the same positive trend under the low-density conditions found in Slovenia. The smart trap also detected adult activity earlier and significantly increased the probability of detection during the first weeks of the flight season. Manual validation confirmed excellent performance of the AI recognition algorithm, with near-perfect precision, recall, F1-score, and overall accuracy throughout the flight season. For both trap types, adult trap catches showed positive and statistically significant relationships with subsequent nest density, indicating that higher catches were associated with greater larval infestation. The meteorological data provided synchronously by automated traps showed that temperature generally promoted moth flight activity, whereas high relative humidity hindered it. In addition, automated traps significantly reduced non-target catch compared to conventional traps. Conclusions Overall, this study provides the first large-scale evaluation of an AI-based camera trap for monitoring the pine processionary moth and demonstrates its considerable potential to improve forest pest surveillance.