UCLA

In order to investigate how birds are adapting to a changing environment, it is imperative that we have a better understanding of avian nesting phenology. The research outlined in this paper is intended to be a first step toward developing procedures and data processing pipelines that can be adapted to monitor nesting phenology at a continental scale. Here, we investigate the effectiveness of bioacoustic monitoring approaches to study the relationship between singing activity and nesting phenology. Using a mix of xeno-canto and field data recordings, we developed and applied species-specific, image-based song classification models for three bird species: Hooded Warblers, Grasshopper Sparrows, and Gray Vireos. We compared our proposed classifiers to BirdNET, a multi-species classifier developed by the Cornell Lab of Ornithology. Our proposed classifiers received an average AUC score for the ROC curve of 0.89, 0.89, and 0.81, and an average AUC score for the precision-recall curve of 0.85, 0.79, and 0.85, respectively. Our proposed Hooded Warbler and Grasshopper Sparrow classifier performed better or the same as the BirdNET classifiers on at least 4/5 accuracy metrics, whereas the BirdNET classifier outperformed our Gray Vireo classifier on all accuracy metrics. Using our proposed Hooded Warbler and Grasshopper Sparrow classifiers, along with the BirdNET Gray Vireo classifier, we explored the relationship between singing activity and nesting phenology using meta-data collected at each study site. We observed little to no significant difference in song occurrence and singing rate within the standard nesting cycle for all three species. For Gray Vireos, there was a significant decrease in song occurrence after failure, fledgling, and at a nest without eggs compared to the song occurrence and singing rate during the nesting cycle. This distinct difference was not observed for Hooded Warblers and Grasshopper Sparrows. Overall, the methods proposed in this paper can be broadly adapted and applied to any vocal avian species. Given that this research is the first step in a potentially much larger project, there are many opportunities to fine-tune the approaches and extend the analysis.