Tracking habitat use of migratory birds in a human‐dominated stopover site using deep learning and acoustic indices

Abstract Bird migration is a fascinating behavioral phenomenon on earth, with annual movements along migratory routes forming complex migration networks. Stopovers, which serve as fuel stations for migratory birds, are critical to the success of long‐distance migrations. However, there is growing concern that stopover habitat has been converted and degraded due to intense human disturbances, which severely threaten migratory populations. New remote automated approaches for collecting data, such as passive acoustic monitoring (PAM) technology, provide a promising avenue for the continuous measurement of vocally active species. In this study, we applied PAM to monitor migrating birds in the stopovers of the Jingxin wetland in China, aiming to explore the activity and habitat use of migratory species through soundscape and deep learning approaches. We collected acoustic data from October 16, 2022, to December 15, 2022 (autumn migration season) and from February 19, 2023, to April 28, 2023 (spring migration season) across three habitats: degraded wetland, farmland, and forest. We applied multilabel classification via the ResNet50 convolutional neural network (CNN) to identify a total of 2.45 million 10‐s audio clips collected. Our results revealed that the 1–2‐kHz vocal signals of Anatidae dominated the soundscapes of the two migratory periods. Two automated measures—compound acoustic indices and a CNN‐derived migratory bird activity—reflected avian habitat use gradients and diel patterns in two migratory periods, with the compound indices model explaining 52% and 47% of the variation in migratory intensity, respectively. Furthermore, farmland is the most intensively utilized habitat by migratory species because of the food resources available. This novel use of combining reproducible acoustic data with deep learning can be used to track the temporal changes and spatial distribution of avian migrants effectively and highlights the importance of agricultural ecosystem management at dominated‐human stopover sites. Managers should consider using cost‐effective acoustic sensors for long‐term monitoring of avian movements and for refining conservation practices in a rapidly changing world.