Epsilon Archive for Student Projects

Abstract

Monitoring of forest restoration efforts is essential to ensure healthy, self-sustaining tropical rainforests. Passive acoustic monitoring is used to monitor vocal activity of birds, which play a key role in forest ecosystems as seed dispersers. Communication between birds seems most profitable during a peak of bird singing in the morning, known as the dawn chorus. Anthropogenic disturbances leading to increased light levels affect the timing of this chorus in individual species. This research sheds a light on the effect of forest restoration on the dawn chorus using automatic detection methods to identify bird sounds from acoustic data. Machine learning methods like clustering and pattern matching were used alongside a manual analysis to describe the dawn chorus in protected forests as well as restoration sites around Ranomafana National Park, Madagascar.
Restoration sites were found to have lower species richness and increased interference from insect sounds. No difference was found between timing of the dawn chorus in both forest habitats. This can possibly be assigned to changes in community composition and decreased detectability of species in insect-dominated landscapes. Future research could further disentangle these effects, by filtering of acoustic data, development of workflow pathways and the use of stronger machine learning methods that allow for more reliable species-specific detection. In the current state of automatic acoustic methods, close cooperation with local experts is recommended to achieve effective monitoring in tropical rainforests.