Temporal insights into ecological community: Advancing waterbird monitoring with dome camera and deep learning.

Biodiversity monitoring is critical for conservation and management. However, efficient species monitoring is often hindered by the complexities of ecological dynamics and the constraints of conventional techniques. This study presents an automated observation system by integrating dome camera with cascade neural networks (CNNs) to map the dynamics of waterbird communities across a semi-enclosed wetland in Dianchi Lake, southwestern China, a well-known important bird habitat. The trained model achieved performance with a Top-1 accuracy of 96.83 %, a Top-5 accuracy of 99.55 %, an F1 score of 93.54 %, a recall rate of 93.38 % and precision of 93.44 %, demonstrating its reliability for precise and well-balanced classification performance. Automatic and manual monitoring performed simultaneously showed highly significant correlations for community abundance (R = 0.89, n = 68, p < 0.0001), underscoring the value of this system as a tool for waterbird communty monitoring. Analysis of the monitoring results showed significant differences in species richness (n = 595, p < 0.0001) and community abundance (n = 595, p < 0.05) between morning and afternoon sessions, suggesting the need to consider ecological community changes across different time scales when conducting biodiversity surveys. In 2023, the system identified 17 species of birds in 5 orders, 6 families. By analyzing the data obtained from this system, the community composition, diversity changes, the arrival and departure times of waterbirds were revealed. Enabling high-frequency, continuous, and long-term monitoring at a lower cost, this system provides a reliable, alternative tool for developing species conservation and habitat management strategies.