Abstract: To investigate the effectiveness of AI-assisted passive acoustic monitoring in avian diversity surveys, we installed 14 acoustic monitoring devices in the Xiaolongmen and Baihua Mountain areas of Beijing Baihua Mountain National Nature Reserve. Continuous monitoring was conducted using a triggered recording mode for one year from June 2023 to June 2024. Audio files were transmitted via 4G networks and analyzed by an AI recognition model for bird species identification. Audio clips with species recognition confidence thresholds above 50% underwent manual verification, yielding 116584 valid recordings documenting 90 bird species. Notably, Tundra Swan (Cygnus columbianus) and Bean Goose (Anser fabalis) were first-time regional records. The detection capabilities for nocturnal species like nightjars and owls were stronger. Under a higher confidence threshold, it reduced the number of identified bird species to a certain extent and did not significantly affect the results regarding aspects such as the spatial differences, seasonal variations, and activity rhythms of bird communities. Passive acoustic monitoring of birds can play a crucial role in the long-term monitoring, management, and scientific research of nature reserves. However, standards for data collection, identification, and analysis should be standardized. For bird groups that are not vocal, it is necessary to integrate methods such as line transect surveys and infrared camera monitoring to build a comprehensive, multi-level, and multi-angle monitoring system, thereby enabling more accurate assessment of local bird diversity.

Key words: passive acoustic monitoring, acoustic recognition, bird diversity, nature reserve