Mili Ostojic: the PhD student monitoring biodiversity with drones

(By Emilia Griffin on 6th October 2024)

Sound can paint a vibrant picture, especially in a habitat full of bird life. But some areas have varied environmental conditions, affecting biodiversity and our ability to measure biodiversity across a location. Until now. Imperial PhD candidate, Mili Ostojic, is deploying autonomous drones to capture the sounds of areas with variations in environmental conditions such as where deforestation is occurring.

Using her background in engineering, Mili is building an autonomous drone, to identify bird species using sound recordings in hard-to-reach locations. She flies the drone to a chosen area, lands it safely and begins recording using acoustic sensors on the drone. Further down the line in Mili’s research, the audio data will be transmitted to a remote computer system which uses AI to analyse it and predict the species present. The sound alone tells us a lot about the diversity of species that live there. A combination of surveying techniques gives a full picture of the environment, with sound recordings picking up even the most inconspicuous species.

Birds of different species sing unique songs, recognisable to the trained ear, like a music fan recognises a track from their favourite artist. In this case, the song is recognisable to a fine-tuned AI system, BirdNET. Mili’s drone transmits the sound recordings to a remote computer where BirdNet can predict which bird species are singing in that recording. BirdNET was created by the Cornell Lab of Ornithology and works by predicting which bird is singed based on the comparison of a spectrogram to archival data. A spectrogram is like a signature or a thumbprint for each birdsong, visually representing the changing notes in a song unique to each species. Of course, the system might get it wrong if species’ sounds overlap or sound too similar. To deal with this, Mili does some manual validation of the results in the preliminary studies to ensure the system is getting the species correct.

As global biodiversity declines, the songs of the forest gradually become whispers. In areas with gradients in environmental conditions, we might expect the sounds we can hear to also be on a gradient. Mili is aiming to deploy the drone in areas with different environmental conditions, such as a forest with parts that have been deforested. Whilst a drone can sample a greater number of locations in a given area than is even feasible for a human to survey, the drone records at each location for less time than a human. Mili is interested in whether recording with the drone in more locations with a shorter recording time is better than the manual surveys for studying the difference in birdsong across areas with a difference in environmental conditions. This could tell us more about how biodiversity varies in deforested areas, for example.

There are other possible applications for drone surveys because drones can reach areas that are inaccessible to people. A new legislation, Biodiversity Net Gain, has been introduced to ensure that wildlife is increased in new property developments. Mili says that ecologists could use drones “for developments to collect the information they need for these biodiversity surveys, especially if it’s a development that would be a lot easier for a drone to survey instead of a human, which would save labour costs”. Some of these locations are difficult to access for ecologists so they might not have data on biodiversity gain without a drone. Another possible location for Mili to deploy the drone is in decommissioned mines that are carved out of hillsides and dangerous for people to access. Mili says “if they’re planning on rewilding the mine, you can monitor the rewilding efforts in comparison to what they’re trying to achieve. It’s a lot easier for a drone to be able to monitor that than it is for a human because of the way the landscape is”. The hope is that by monitoring locations with a variety of environmental conditions, or that are inaccessible to humans, we might hear a crescendo back to diverse life.

With the wave of AI and drone usage across the world has come an influx of fear around technology and the potential loss of jobs. However, in most cases, and indeed in this case, drones and AI technologies are additional tools that fill some gaps in the current market. For an ecologist, an autonomous drone with bioacoustics sensors can be used alongside other surveying methods, allowing them to move into inaccessible areas or record many locations over a shorter time. It can take years of practice to train the ear to individual bird songs; Mili says this can fill a training gap that currently prevents some from accessing the job market.

AI systems are not unheard of in conservation science, with many keen birders turning to the Merlin app to identify species. This is an excellent app that uses similar machine learning to BirdNET, also made at Cornell University. The benefit of using a drone is that the sound recorder can be moved around remotely to new locations, widening its surveying ability compared with a static recorder. Of course, this system still requires someone to deploy the drone, so we needn’t worry that technology is taking our jobs anytime soon.

Similar technology is being adopted elsewhere across the world, improving our understanding of global diversity through sound. For example, The Sound of Norway is a project that uses static recorders across different sites to monitor biodiversity across Norway. Mili is still in the initial stages of her research but with her excellent public engagement at Imperial Lates, and a passion for engineering and ecology, it’s clear to see she will go far.