An award is made to the University of Pittsburgh to develop an automated acoustic platform for locating and counting terrestrial wildlife. Accurate estimates of wildlife populations are central to research on the effects of global change on biodiversity. Historically, population-level data has been time-consuming and difficult to collect, limiting the number of species and habitats that can be evaluated. Acoustic recorders have the potential to inexpensively gather large-scale data on sound-producing species, including birds, bats, amphibians, and insects. However, most current acoustic surveys are only able to detect the presence or absence of a species and cannot count individual organisms in order to estimate population sizes. This award will support the initial development of an automated, open source acoustic platform to survey terrestrial wildlife populations, with the goal of gathering data at larger scales and with better accuracy than human observers. It will also support the training of at least two graduate students and an educational project that targets undergraduate STEM educator professional development at the interface of biology and data science. The creation of this platform and its release under open source licenses will enable professional ecologists, citizen scientists, and large biodiversity monitoring programs to better detect declines in species populations over time, track shifts in species distributions due to climate change, understand the interacting drivers of biodiversity changes, predict future extinction risks, and develop conservation strategies to protect threatened species.
The acoustic survey platform will use GPS time-synchronized recorders to localize sounds in coordinate space. These sound locations will then be used to distinguish and count individual organisms within species. Although applicable to a wide variety of species, the platform will be initially designed for breeding songbirds of the eastern United States. The specific objectives of the award, involving both hardware and software development, include (1) designing an open source, inexpensive field recorder that can collect time-synchronized recordings, (2) developing and training an object-detecting convolutional neural network to identify the boundaries of distinct bird songs in time and frequency, (3) creating a detection-informed time difference of arrival algorithm that localizes each identified bird song in coordinate space and uses this data to count individual birds within species, and (4) completing performance and user testing to evaluate the integrated hardware and software platform. The platform will be evaluated specifically on its ability to estimate breeding bird populations more accurately and precisely than human observers.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
