Abstract:

Aims: Traditional avian diversity surveys have largely relied on manual observations. However, in recent years, voiceprint monitoring technology has been gradually applied, providing a new approach for studying avian diversity. Avian diversity is a key indicator for assessing the quality of wetland ecosystems. This study aims to compare AI-based voiceprint monitoring with line transect surveys, offering a case reference for the application of bird voiceprint monitoring devices in wetland parks nationwide.
Methods: In January (winter), April (spring), August (summer), and October (autumn) of 2024, we conducted comprehensive and systematic avian diversity surveys at five sites within Xixi National Wetland Park in Hangzhou, Zhejiang, where line transect surveys and voiceprint monitoring devices overlapped. We applied a confidence threshold of 77.5% for output data. Based on Simpson dominance index (C), Shannon-Wiener diversity index (H′), Pielou evenness index (J), and Margalef richness index (M), the applicability and limitations of two methods were evaluated.
Results: (1) Voiceprint monitoring detected 105 bird species across four seasons, outnumbering the 89 species recorded by line transect surveys and demonstrating superior species richness (S) performance. (2) In terms of residency status, voiceprint monitoring showed higher detection efficiency for migratory birds, whereas line transect surveys performed better for resident birds; moreover, voiceprint monitoring also contributed new regional records. (3) The seasonal variations of the indices obtained by the two methods were not entirely consistent. The values of C and M were higher in most seasons under voiceprint monitoring, which performed better in capturing dominant species signals and reflecting S. By contrast, H′ and J were higher in most seasons in line transect surveys, better representing overall community diversity and the evenness of species distribution. (4) By region, the highest H′, J, and M values under voiceprint monitoring were observed at Lüdi/Shuixiachanglang area, whereas under line transect surveys, the highest values were recorded at Lianhuatan area.
Conclusion: Overall, voiceprint monitoring is suitable for long-term and wide time-scale dynamic monitoring, with broad application prospects, and can serve as a complement to line transect surveys. An evaluation system incorporating recognition confidence and quality control thresholds is recommended to enhance the accuracy and comparability of the method.

Key words: avian diversity, voiceprint monitoring, line transect surveys, Xixi Wetland, AI recognition