Personnel: Amanda Navine, Stefan Kahl, Ann Tanimoto-Johnson, Holger Klinck, and Patrick Hart
Many of the endangered bird species in Hawaii live in remote habitats that can sometimes be difficult to access, making it challenging for researchers/conservation biologists to monitor how their populations are changing over time. Automated recorders can be deployed in these habitats for extended periods of time, gathering large amounts of bioacoustic data. Researchers can then go through these recordings and listen/look for bird songs and calls to detect presence/absence of species. However, this process can be incredibly time consuming, going through the hundreds to thousands of hours of recordings collected trying to detect rare or endangered species. To significantly reduce the amount of time it takes to process bioacoustic data, the LOHE bioacoustics lab researchers are working in collaboration with the Cornell Lab of Ornithology to develop a machine learning algorithm called BirdNET. BirdNET will allow researchers to be able to automatically detect birdsong and identify the species without manually going through the bioacoustic data. Once this tool has been developed for our birds in Hawaiˊi, it will allow for near constant monitoring of endangered bird species across Hawaiˊi, which in turn will lead to better informed management decisions.
The Google Perch bird vocalization classifier is a machine learning algorithm that can make accurate species predictions for birdsongs within soundscapes with only a small number of training samples per species, which is necessary for rare and endangered birds with few examples available for training. The Listening Observatory for Hawaiian Ecosystems (LOHE) lab at UH Hilo, in collaboration with the Google Bioacoustics Research Group, has recently developed a method for processing the output of such machine-learning classifiers to make accurate estimates of species call densities, which can be used both as an occupancy indicator and as a measure of changes in species abundance at a site over time. We can use such tools to track population trends and make informed conservation management decisions to save native birds from extinction.
Acknowledgements: Compiling this extensive dataset was a major undertaking, and we are very thankful to the domain experts who helped to collect and manually annotate the data for this collection. Specifically, we want to thank Charlotte Forbes-Perry with the Pacific Cooperative Studies Unit, University of Hawai'i at Hawai‘i Volcanoes National Park, The Kaua'i Forest Bird Recovery Project, The Maui Forest Bird Recovery Project, as well as the following current and past members of the LOHE lab (in alphabetical order): Nikolai Braedt, Keith Burnett, Saxony Charlot, Braxton Igne, Meeya O'Dell, Willow Petersen, Esther Gonzalez-Sebastian, Jennipher Himmelmann, Noah Hunt, Erika Kekiwi, Caleb Kow, Elizabeth Lough, Bret Mossman, and Kristina Paxton.
Published Dataset: https://zenodo.org/record/7078499
