Instant Genius - Why ecoacoustics is the future of conservation

Episode Date: May 24, 2026

Traditionally, ecologists wanting to take stock of the diversity, health and population levels of animal species within a given environment have needed to tie on their boots, get out in the field and painstakingly record what they are able to observe manually. But the emerging field of ecoacoustics – the use of sound recording to survey the biodiversity within ecosystems – is promising to be a game-changer in the way researchers are able to approach this work. As part of our Science of Sound miniseries, we’re joined by Dr Sarab Sethi, the head of the ecosystem sensing research group at Imperial College London. To talk about his work in this exciting new field. He tells us how advances in technology are providing new methods of monitoring wildlife in greater detail than ever before and why cross-collaboration between engineers and conservationists is vital in the fight against biodiversity loss. Learn more about your ad choices. Visit podcastchoices.com/adchoices

This episode is brought to you by Activia. You might already be eating yogurt, but not all yogurts are created equal. Activia contains over 1 billion probiotics per serving to survive and reach the gut alive. When it comes to gut health, Activia is the number one family doctor-recommended probiotic yogurt brand. Choose Activia. Feel good from the inside out. Visitactivia.ca for more details. This spring, Denham gets a softer, lighter update.

Introducing Old Navy's drapey denim wide leg, a new fit that moves with you. It's everything you want denim to feel like for summer. Easy, breathable and effortlessly cool. With a fit that creates natural movement and a wide leg that feels modern, not overwhelming. Plus, that signature, wait, for this price, moment. Old Navy's drapey denim wide leg.

Hello, and welcome to Instant Genius, a biotized master class in podcast form. Every Monday and Friday you'll hit. are world-leading scientists and experts talking about the most fascinating ideas in science and technology today. I'm Jason Goodyear, commissioning editor, a BBC science focus. Traditionally, ecologists wanting to take stock of the diversity, health, and population levels of animal species within a given environment have needed to tie on their boots, get out in the field, and painstakingly record what they're able to observe manually. But the emerging field of eco-acoustics, the use of sound

recording to survey the biodiversity within ecosystems is promising to be a game changer in the way researchers are able to approach this work. In this episode, we're joined by Dr. Sarab Sethi, the head of the ecosystem sensing research group at Imperial College London, to talk about his work in this exciting new field. He tells us how advances in technology are providing new methods of monitoring wildlife in greater detail than ever before. And why cross-collaboration between engineers and conservationists is vital in the fight against biodiversity loss. So welcome to the podcast. Thanks so much for joining us. Thanks for inviting me. So today we're talking all about your work on eco-acoustics. So I'd venture this may be

the first time that some of our listeners have heard about this field of research. So let's start off with, can you explain what exactly that means? Yeah. The first I'd heard of this field was when I entered it as well, so I wouldn't blame your listeners at all. Essentially, the problem that we're trying to solve is that many people are interested in monitoring animals and how they're distributed, how they behave, how they move, how they're affected by human pressures, et cetera, et cetera, for many different reasons for farming, for biosecurity. But actually getting that information is incredibly difficult. typically used to rely on quite manual ways of getting this biodiversity data, which is what we call it. Typically, you know, you'd go into a forest, you'd start counting the insects you found under relief, you'd say stand at a point, you'd count the birds in the trees, but you can only do that

for so long or for so many different places. So eco-acoustics is basically a field that's starting to use acoustic monitoring data, so microphones that are left unattended in natural spaces to monitor the animals around by their vocalizations or sounds they make passively and getting all of this biodiversity data at sort of real scale. Yeah, so as you said that, this is quite a new field and one that's not widely known. So I'd just like to ask you, how did you get involved in this in the first place? So my undergraduate and master's degree were in engineering. So I'd done general engineering, a bit of civil, but a mechanical, and I'd sort of started to lean towards electronics and computing towards the end. So I had no ecology background and then came to the end of my master's.

I started looking for jobs. And there was just one really interesting PhD advert that I saw that was, use your engineering skills to build sensors. But you can do it in Borneo in the middle of a tropical rainforest along with a bunch of scientists. lots of travel included, lots of fun problem, but, you know, seemed solvable. And I was like, well, let's see what that is. I'd never really considered academia even as a field. So it was just a, this seems like an interesting thing to do for a few years.

And then I'll figure out how to get a real job afterwards. Yeah, sure. So like from the engineering background, so that's quite interesting. Let's sort of perceive that a little bit. Do you use specialist equipment to do the recordings? Yeah, so really, I've now got my own research lab at Imperial, which is called the ecosystem sensing group, and a large theme of our research is in developing the equipment that we use to collect this data. So it comes from, we, for example, develop sensors that are lower cost and deliver real-time data from really remote regions. They can weather tropical storms and such. We're also developing robotic approaches to deploying sensors in forests or in natural spaces where they're otherwise really hard to reach. They could even be robots with sensors that hop around the forest autonomously, come back and

recharge themselves. So yeah, there's a lot of specialist equipment and there's a lot of research developing the next generation of specialist equipment to collect this kind of data. So how does the project work then? So, you know, is it like a whole team of researchers from different backgrounds all coming together? There's every scale of project. really. There are PhD students who kind of do their own project self-contained where they work on small but really important problems and solve them a little bit at the time. But then there are also large European projects that we're involved in where they've deployed, for example, a network of ecocoustic sensors through Norway, Netherlands, France and Spain down a massive

bird migratory highway. And there you have experts such as from our team who are working on developing the sense of themselves. We also have AI experts in the Netherlands who are developing the classifiers. You have policy experts else in Spain where they're linking this into biodiversity reporting. So we're involved in lots of different projects. Some are very small, some are very big and it isn't necessarily correlated with interestingness, I'd say. So you mentioned there that you did your PhD out in Borneo. It would be nice to have a chat about that. So, because you might think, someone listening might think, well, you know, if you're designing the centres and the equipment, you might just be doing that, you know, in London and then sending

them off. But you actually went over there and, you know, how did that work? And, you know, what did you do? What sort of experience did you have out there? Yeah. So I was based actually at Imperial for my PhD. And the reason it was in Borneo is because I joined a lab group of tropical forest ecologists where they were working in this large. experiment where they had virgin tropical rainforest but also nearby log forest and in between it all oil palm plantations and this this was quite a big like international research effort and teams of ecologists who are studying freshwater fish to insects to mammals to birds and everything were all

working the same place and one thing they all was was emerging was this field of acoustics and they were wondering like one can acoustics give us any useful data here, but too, can, you know, this is a very challenging 100% humidity, tropical rainforest, dorms every day. Like, the kind of, back then, the field was much younger. And so, whatever equipment there was, it wasn't really necessarily trialed in the harshest environments. So my supervisor at the time was like, this is a problem that we all are facing that we can't collect enough biodiversity data to really answer the questions that we have. And I want to start exploring what methods there are around the corner. And so, yeah, Borneo was, it was there to

answer that question for that team. But you're absolutely right. You know, we could have, and we did do lots of prototyping and stuff in London and then take them out. But what you learn from prototyping to device in a very temperate urban environment is very different from what you learn from deploying it in a very sort of challenging, humid complex, insect-ridden tropical environment. Yeah. So once you've deployed one of these senses, you know, What's the kind of idea? Is it to single out the health of certain populations, the species, or to just get a broad overview of the whole ecosystem,

or, you know, maybe a bit of both? Yeah, absolutely, it's a bit of both. So throughout my career and really more widely in the field of eco-acoustics, people tend to take two approaches. One is, I know what my blackbird in my garden sounds like. So every time I hear that song, I know a blackbird. there and when it's not there, I infer that it's not there. That of course opens up a whole can of worms, which is maybe it's just there but silent,

but let's just skip over that for now. But you're getting presence or absence data or maybe even abundance data, which tells you how many of the blackbirds are there. If you can correlate somehow vocal activity to a number of blackbirds, then you can do that, right? And you go through and you train your classifiers for as many species as you can with all the training that you have, and you can say, for my observable species community,

I know the numbers and how this varies by different sites across this landscape or different hours of the day. There's a really important nuance to this, which is even today with how advanced AI models are and incredible performing this in the news every day, you really can only get the most common species identifiable in this kind of data. And the limitation really is the core natural history understanding we have of most species about what sounds they make. and also the data sets that we have available to train on. So actually doing this species level classification, picking out and identifying species from their unique vocalizations or acoustic signatures,

is only possible for common species and only a small handful of them. So there is this other part of the field of eco-acoustics where people try to use aggregated metrics of, say you listen, I think the perfect analogy or metaphor for this, similarly, I don't know, one or the other, is when you, if you're outside a football stadium and you hear a whole roar of the crowd as something's happened, you know something's happened, you don't necessarily know exactly what's happened, but you know that perhaps a goal's been scored. And there's some collective response of the soundscape that tells you about what's going

on. And so there is a similar sort of feature-based way of analyzing soundscapes, which says, within a minute of this tropical forest recording I have, what's the average volume or what's the average activity within frequency bands that I know birds to be active within or what's all these different types of features that you can define. And then you say perhaps that serves as a proxy for biodiversity, a proxy for the overall ecosystem health, and you can go from there. And then, yeah, the nuances of whether that works, whether that scales, whether that generalises, how do you do that is exactly the kind of work that our lab is also involved in. Spotify, it's Jen.

Shetty, are you one of those media strategy people? Scrolling through spreadsheets, searching for an audience that pays twice as much attention to your ads than they do on social? Let me introduce you to fans. And they're here with me on Spotify. Trust me, I know fans. They don't skip, they stay for hours. They don't move on, they manifest. They're not a demographic group, they're fans.

Spotify Advertising. You're among fans. Sticking with data collection then, are the sensors just continuously recording? Yeah, so different sets-ups are different questions as always. But for sure, for example, that European project that I mentioned earlier, the sensors are 24-7 recording, compressing data to quite high-quality MP3 and sending it over mobile networks. So we're getting 24-7 data. But one paradigm, as I mentioned, that we're exploring with robotic sensor deployments,

is this idea that rather than having a sensor at 10 different sites, could you just have a robot that or a drone that hops between each of these sites, recording little snippets of data? It's obviously much more resource-unintensive, resource-efficient. You don't have to have 10 sensors. You have one sensor that's mounted on a robot. But then how do you understand the ecological data you're getting because you're not getting simultaneous data from all of those 10 different sites then?

And then increasingly actually we're talking about batteryless intermittent sensing where your senses themselves are harvesting energy from the environment. Like maybe it's sunlight, maybe it's a bit of wind. There's even some incredibly science fiction looking work that we've seen recently where energy is harvested from sounds themselves. So a microphone can record powered by the sound that it's recording. So there you're not continuously recording, but the benefit is you're not continuously recording. but the benefit is you haven't got a finite battery that's going to run out. You can essentially deploy these things indefinitely. So I guess like that's the cases I've described are very niche and very forward-looking,

but I'd say, yeah, in most deployments these days, people either record 24-7 or they record during the hours that the animals they're interested in are likely to be vocalising. Yeah, that's fascinating. So you mentioned a couple of times this idea of robots there. So I'd like to talk about that. Because obviously the first thing that's come into my mind, if I say obviously, but as the kind of job that I do in the person that I am in my taste, etc.

I just think of like some sort of little Star Wars robot walking around like the, with the Ewks in the joke or something. So what exactly would that look like? So a PhD student in my group is developing an autonomous drone. Like we think this is a bit more feasible than necessarily like a ground robot. robot because anyone who's been into these kind of like a complex, even if they're managed, right, like managed woodland still has a very complex understory and very challenging environment. So if you use a drone, all you need to be able to do is get up above the canopy flow

wherever you want to go and then find a way to land. And in fact, that we're even working with some robotics labs where they're developing mechanisms that robots can land on branches and have grippers that grip onto a branch and then they don't even have to make that perilous journey down through thick foliage to the ground. So at the first instance, we're exploring autonomous drones as our robotic platforms, but I don't by any means think it's the only way to do it, right? And as groundroving robots and these robotic dogs and everything become more capable, then why not? But I guess another thing that's good about drones is due to other commercial uses that are becoming really increasingly cheap and accessible, right, which is not true

necessarily for robotic dogs or humanoids or any of these other formats. Obviously, as we sort of touched on there, you're collecting huge amounts of data. So once we've got all of this data, you know, how do we start combing through it? Yeah, AI, as I said, right, really is the way. And as I said, there's two different ways to analyse this data. One is to pick out and identify species, and the other is to define features and see how those features are shifting as a whole. But you can really use, you know, AI for both and we regularly do use AI for both. And in fact, non-AI, AI methods to detect or identify species or to characterize soundscapes are becoming increasingly rare. It doesn't

mean that these are solved problems at all. It just means that the AI models work better than the last generation of algorithms that were available to the community. Yeah, so as you've mentioned, you've worked on, you know, several projects all around the world now at this point. So what are some of the highlights you can pick out? and like some of the findings, you know, some of your sort of proudest achievements, I guess I'd say. That's always a very imposter syndrome-inducing question, you know. Most of our work is methodological. I'd say like the really, the moments that make you proud are when we managed to do something

that the experts in the field that we started working with told us wasn't possible at all. So I remember, for example, when I started working in Borneo, really well-meaning colleagues of mine were like, it's a really cool project you've got, but there's absolutely no way that you'll possibly get a sensor to collect data for a year from this. It's just too wild and too difficult. And then actually it is possible, right? And it's a bit tricky and you've got to figure out solutions to a thousand problems in between.

A team at Cornell had been developing Birdnet, which is an AI classifier of all sorts of different birds all around the world. And increasingly, the consensus was like, yeah, it's a cool toy, but it doesn't really work, and it won't tell us anything ecologically meaningful. We still absolutely need expert surveys to do anything. Now, I think that what we did was we applied that classifier to a bunch of different data sets across the world, and then we got experts to validate the data and the detections, and then we actually went into each of those data sets and picked up new and interesting biodiversity trends. And I think, like, the really satisfying moment there was showing that this tool that many people thought was a bit of a toy,

actually already today is effective for delivering some new biodiversity insight. Now, that's not to say it's perfect, not by a million miles, right? And there are many caveats to how you should interpret the data and validate it, etc., to get useful insights. But I think the really satisfying moments of my career have been taking things that perhaps from an ecologist or biologist's point of view might have been seen as a bit of a novelty toy and showing actually, no, these can deliver really new and exciting and insightful data already.

So you mentioned the PhD student that you're supervising working on at the drone work. So what are the sort of current projects are you working on or people in your team working on or that are coming up in the pipeline? You know, what exciting stuff can you share with us? Yeah, it's all over the place and it's kind of what keeps the job fun, right? So there are three PhD students in the group. Another one is working on insect pollinator activity detection in farmland specifically, but kind of generally using acoustic.

So can you detect a bee from its buzzing as it comes near a microphone, and can you even classify the type of bee from its buzzing? And then with that data distributed around the farmland, can you understand whether there are pollination hotspots or even areas which are not being pollinated and therefore your crop suffering? and could that link into adaptive land management, right? Could the farmer then be planting wildflowers specifically in this back corner of the field where they know the pollinators aren't being encouraged to go enough to pollinate and such?

Another student is working on how do soundscapes, so again, this is going away from the species classification and more towards soundscapes, how does soundscapes respond to typhoons in Taiwan across a sort of decadal time scan? And every time there's a typhoon, there's obviously an enormous distinctions. disturbance to the ecosystem, but then eventually we assume and we see in time that the soundscape returns back to its baseline. Now, does that return to baseline tell us something about the stability or the redundancy within this ecosystem? And is that a metric kind of for free that we get of ecosystem resilience, which is a topic that's incredibly important and hot in the world of

ecology, but incredibly difficult to measure? And then I guess finally an exciting project that should be kicking off later this year is in Indonesia, the government are building a new capital city, and it's called Nusantara, and it's in the, like, on the island of Borneo, because Jakarta is sinking in some places that's sort of 10 centimetres a year. So this enormous infrastructural project, right, 10 to billions of pounds, at least, where they're building airports, schools, a whole new city in what used to be pretty just small villages in a forest. And here we want to understand how. How is this rapid urbanization and rapid like influx of people into this very biodiverse area

that used to just be tropical rainforest? How is that going to interact with disease spillover from the bats, the rodents, the macaques in the area? So it's a really cool project where there's four different PIs. Our role is to monitor biodiversity across this area and how that's changing. Another person's role is to monitor the pathogens found within the animal community. and other is to sort of sequence human diseases that are emerging and we start to understand how this interaction between people,

nature and disease outbreaks is modulated and whether that could have implications for sort of safeguarding as human societies, I guess. Yeah, so it sounds like we're only just sort of scratching the surface of the potential of this field. Yeah, yeah, for sure. I think basically, you know, ecology, forgetting even like agricultural land management and other such fields that are sort of to the side.

Ecology itself is an enormous field, not one that I pretend to know in depth across any sort of way. But almost every ecologist could do with better, more scalable biodiversity data. And the kinds of questions that they're asking could be accelerated and they can expand the scope of what they can even ask and look at. And I think that just starting to find these applications and develop methods that work to solve specific problems, is, it's kind of what our lab is. It's also looking at. Thank you for listening to this episode of Incent Genius, brought to you from the team behind BBC Science Focus.

That was Dr. Sarah SEPFes. If you liked what you just heard, then please do consider subscribing to InsentGenius on your preferred podcast platform. If you'd like to see our guests and hosts in person, then why not check out our YouTube channel at Science Focus. The current issue of BBC Science Focus magazine is out now. Pick up a copy wherever you buy your favorite magazines or download us on your app store of choice.

You can also find us on Apple News or online at sciencefocus.com. Hey y'all, it's Kelly Clarkson with Wayfair. Ever order furniture online and wonder what if? Like, what if it doesn't hold up? That sofa was four days old. You should have ordered from Wayfair. With Wayfair, there's no what if. Just style you love and quality you can trust.

Visit Wayfair.ca.