Acoustic biodiversity monitoring with hopping robots

Understanding how animals are distributed cross our planet, how they move and how they behave is essential if we are to protect biodiversity in the face of increasing human pressures. However, in practice, collecting insightful monitoring data at meaningful scales and resolutions remains extremely challenging.

By Sarab S. Sethi, lecturer in ecosystem sensing, Department of Life Sciences / I-X, Imperial College London

Traditional approaches to collecting biodiversity data rely on manual surveys conducted by trained experts. For example, in a bird point count an ornithologist will stand at a site of interest and record every single bird that they see or hear over a fixed period. Whilst data quality can be exceptional (e.g. identifying sex, age, or breeding stage is possible), the slow speed and high cost of surveys very quickly becomes a major barrier to scalability.

Acoustic monitoring has recently seen a surge in popularity as a scalable alternative to surveying animals in the wild. Inexpensive audio recorders are deployed to capture natural soundscapes from sites of interest over weeks and months. Audio is then analysed with machine learning algorithms to identify species from their vocalisations (e.g. birds by their calls and songs) or other unique acoustic cues (e.g. mosquitos by their buzzing frequencies). Still, whilst recorders themselves can be cheap, deploying, maintaining and retrieving them can become prohibitively expensive and time-consuming when species need to be monitored on landscape scales or over extended time periods.

Above: Drones carrying acoustic sensors might be able to transform the scale at which we can monitor biodiversity

Drones on the hop
In a pilot project funded by the UK Acoustics Network (UKAN+), Dr Peggy Bevan led explorations into whether autonomous robots (e.g. drones) carrying acoustic sensors might be able to transform the scale at which we can monitor biodiversity.

Using a vast dataset of audio recorded from ~300 sites across tropical rainforests and agricultural lands on the Osa Peninsula, Costa Rica, we simulated autonomous drones hopping between sampling sites recording an hour of audio at a time. Whilst the original dataset had complete temporal data coverage at each site, our simulated drones could only record short snippets of audio while they briefly landed at a sampling site.

We found that even when simulating very lightweight sampling networks (e.g. one drone per five-10 sites), we could reconstruct previously published patterns in bird biodiversity and spider monkey occupancy across the region. We also found that adaptive sampling – using real-time data to inform which site the drone visited next – improved the reliability of our downstream biodiversity data.

Reducing costs and increasing scalability
In other research from our group at Imperial (the Ecosystem Sensing Group) and collaborators across Europe, we are developing technologies that will make autonomous biodiversity sensing systems of this type a reality. For example, Mili Ostojic is developing an autonomous drone that can navigate through complex natural environments and Dr Clementine Boutry and Javad Bathaei at TU Delft are developing fully biodegradable sensors that could be deployed from robotic platforms of this kind.

More engineering research and development work certainly remains before robotic sampling systems reach full maturity and are available at reasonable price points. Nevertheless, our research has paved a clear pathway for autonomous robotic platforms to deliver reliable and impactful data whilst reducing costs and increasing scalability of biodiversity surveys in the future.

Above: Collaborators throughout Europe are developing technologies that will make autonomous biodiversity sensing systems a reality