Acoustic phenology of tropical resident birds differs between native forest species and parkland colonizer species

Most birds are characterized by a seasonal phenology closely adapted to local climatic conditions, even in tropical habitats where climatic seasonality is slight. In order to better understand the phenologies of resident tropical birds, and how phenology may differ among species at the same site, we used ~70,000 hours of audio recordings collected continuously for two years at four recording stations in Singapore and nine custom-made machine learning classifiers to determine the vocal phenology of a panel of nine resident bird species. We detected distinct seasonality in vocal activity in some species but not others. Native forest species sang seasonally. In contrast, species which have had breeding populations in Singapore only for the last few decades exhibited seemingly aseasonal or unpredictable song activity throughout the year. Urbanization and habitat modification over the last 100 years have altered the composition of species in Singapore, which appears to have influenced phenological dynamics in the avian community. It is unclear what is driving the differences in phenology between these two groups of species, but it may be due to either differences in seasonal availability of preferred foods, or newly established populations may require decades to adjust to local environmental conditions. Our results highlight the ways that anthropogenic habitat modification may disrupt phenological cycles in tropical regions in addition to altering the species community.

Methods
This is an acoustic phenology dataset. Soundscape recordings were collected 24/7 in Singapore over the course of 2 years. The machine learning software Kaleidoscope Pro was used to make species classifiers for 9 species of birds. Species classifiers are able to automatically detect all occurrences of the target species' song within the 2-year-long dataset. Automatic outputs were manually verified to ensure accuracy.

In addition to the cleaned species detection dataset, we also provide the Kaleidoscope species classifiers. These classifiers can be used to detect the 9 focal species in your own audio data.