Sounds of recovery: AI helps monitor wildlife during forest restoration

Benjamin Thompson

Welcome back to the Nature Podcast. This week, the ecologists gathering data by listening...

Shamini Bundell

And how people can follow simple instructions while fast asleep. I'm Shamini Bundell...

Benjamin Thompson

...and I'm Benjamin Thompson.

<Jungle sounds>

Noah Baker

This is the sound of a tropical forest in Ecuador. And ecosystems like this are vital.

<Music>

Jörg Müller

They matter for all humans living on this globe. Even if you're totally not interested in tropical forest, the tropical forests drive our climate.

Noah Baker

That's Jörg Müller, a conservation ecologist from the University of Würzburg in Germany.

Jörg Müller

But as a conservationist, you're interested in the second important topic, they host the highest terrestrial diversity on Earth.

Noah Baker

But forests around the world are facing a lot of challenges.

<Chainsaw sounds>

Noah Baker

Timber extraction, agriculture and other industries along with climate change are threatening tropical forests. And those challenges don't only change the landscape, they also change the sounds.

Noah Baker

As forests are converted to open farmland, new sounds appear. The balance of noise changes even the acoustics of the environment itself morph. And scientists are now asking if these changing soundscapes could provide them with an opportunity. In recent years, various efforts to reclaim agricultural land and restore forests have been launched with a multitude of goals from carbon capture to biodiversity increases, but monitoring the progress of efforts like this is not always easy. After all, regrowth doesn't happen overnight, and it's impacted by many factors. This creates what scientists like Jörg refer to as a 'restoration gradient’. Now tools like satellite imagery can be used to quantify the regrowth of trees and plants, and by extension, estimate carbon capture, but wildlife regeneration, well, let's hand back over to Jörg.

Jörg Müller

This is much more difficult — or it's impossible from the satellites. Because it's very cryptic, you can imagine a small hummingbird you cannot easily track them from the satellites. What we need is our standardised methods where we can investigate a whole graded in time and space.

Noah Baker

And that's where the sounds come in. Jörg and his team wondered if soundscapes could provide another part of a solution.

Jörg Müller

No method captures everything. Every method has some biases towards some species groups. And in this approach, we used all vocalising vertebrates, which is predominantly by birds and amphibians and a few mammals, and then check how they describe the gradient of forests recover over time.

Noah Baker

Whilst imperfect, Jörg still believed that analysing soundscapes could provide a useful measure. But they needed somewhere to test their theories. And they found it in the form of a reserve in Ecuador created by an NGO.

Jörg Müller

Inside the reserve or at the edges, you have forests which are still under agricultural use. And some areas are abandoned now for two years, five years, 10 years, 34 years, something like that. And they're also old-growth patches, so where you have primary forests.

Noah Baker

And so they set out to record soundscapes, at a series of plots throughout the reserve.

Jörg Müller

We did it everywhere. We set the recorder in all these plots, along the gradient from active agriculture over recovering forests to old-growth forests.

Noah Baker

And then they set about analysing the data.

Jörg Müller

So our starting point was that we selected from each plot the same minutes from two days within two weeks, and hand over these to experts for birds and experts for amphibians, and they were able to identify a lot of animals on the sound excellent. And this was our, let's say, gold standard, the starting point.

Noah Baker

But they didn't stop there. You see, these kinds of expert analyses can be very time consuming, and Jörg and his team had big ambitions.

Jörg Müller

We use these plots to investigate the role of sound and AI to identify these communities.

Noah Baker

They wanted to automate the process using AI. And so on the same data, they used an acoustic index analysis. Now that doesn't pick out individual species, but rather assesses sounds broadly based on fundamental sonic properties like frequency or pitch. But they also employed an AI assisted programme, specifically a neural network, which had been trained to identify 75 species of bird. These birds were from the region, but not from this specific plot. And they're only a relatively small subset of the animals which could be heard in the recordings, but Jörg hoped that it would still be enough to get a reasonable proxy. And indeed, the AI software was only able to pick out about half the species the experts did, owing to its limited training. But what stood out to Jörg his team is that all their assessment methods tracked onto one another, and onto models of regeneration in the reserve, reliably predicting where on the regeneration gradient a forest sits. And that's something which Jörg says is not always easy to tell, even for experts in the field.

Jörg Müller

As an ornithologist, sometimes it's hard to see if this is a recovery forest, or already an old growth. So you are in the forest, there are a lot of trees around you, after 30 years of tree growth, some of them grown one metre or more per year, and that it's hard to identify what's going on. But if you ask the birds, they show you exactly if this is an old growth, or this is still a recovery forest. So ask the birds and they tell you something about the progress is the story in a nutshell.

Noah Baker

And their data showed something else, the best indicator they found of the status of regeneration wasn't the number of species recorded, but rather the composition of the species in an area.

Jörg Müller

The number of species is a very weak indicator. And what is the reason I would like to explain it in a very simple example. So if you go there to an agriculture farmland, and then you will find a bunch of birds. And you can record let's say you find 10 birds there in one morning, and then you go to the old growth, and you'll find also 10 birds. So when you compare, there is no difference. But in fact, these are totally different species. And what we learned is the species in the agricultural land are species which are common, more southern parts of South America, where the habitats are naturally more open, and therefore it turned out that the community composition, so the similarity in species, is a much better indicator to describe the pattern of recovering biodiversity.

Noah Baker

Now, this automated system isn't perfect. The one thing there are many, many animals that don't make prominent sounds and so weren't detected by the system. But Jörg still think that the measurements are useful. In fact, he tested it by comparing their analyses with a totally independent dataset.

Jörg Müller

So of course, this is a crucial question and at first we have to say, soundscape are about vocalising animals. That's it. But in our study, we combine the data preliminary with another data set, which was based just on metabar coding, so sequencing bulks of insects collected with light traps, so they have nothing to do with our sound. And there are almost no vocalising insects in this dataset. And what we saw that it's quite well correlated, even with our sound indices, which indicates that the birds are very integrative. And if the birds are shifting the species composition, even other communities are shifting. And so maybe birds can be used as a major surrogate for this recovery forest. But this is too early to say that overall, because we have not correlated these to soil diversity, for example, or plant diversity, this is just an assumption which has to be tested further.

Noah Baker

More work needs to be done. But Jörg see systems like this opening up new opportunities, for example in the emerging biodiversity credit market. In these models, landowners are paid by companies, individuals, even governments to focus their efforts on regeneration of biodiversity. The idea works in a similar way to carbon credits and offsets. But those kinds of transactions require reliable reporting mechanisms.

Jörg Müller

And there are no tools available at the moment. And we have no solution. So we come in, in our area, we can easily say, okay, this is the status of your area now, where you collect sound data and five years or 10 years later you do it again. And we can say exactly, you are more close now to the primaeval forest by 20%. And this is paid. And this is really serious, well recorded baseline data and can make your baseline for this new upcoming market.

Noah Baker

So what's next for the system? And as AI has become more prominent, does this spell the end for the expert ecologist? Well, Jörg thinks not.

Jörg Müller

Yeah, I think it needs definitely both sides of the coin. So we need experts in the long run, we will further develop these AI models with the help of expert identification that they label specific species and provide the snips and say this sound snip is exactly the species and then you can feed your AI models with that. So the models are hungry, data hungry, and without the expert we cannot feed them, then the expert will never be able to run millions of sound files. But AI can do that. And this is the big advantage. So I think we need both sides. And in principle, it works. So scientifically the story is done. Now we need better labels from the still missing species in the AI models to have better models. And then we can run it on a large scale. So often people fear that now the experts are run out of jobs, but in fact, it's vice versa. So people are more and more interested in their expertise than ever before.

<Jungle noises>

Benjamin Thompson

That was Jörg Müller from the University of Würzburg in Germany. To read Jörg's paper head over to the show notes for a link. This piece was produced by Noah Baker and Nick Petrić Howe.

Shamini Bundell

Coming up, the research showing that being asleep doesn't necessarily stop someone from following simple commands. Right now though, he's back again. It's Noah with this week's Research Highlights.

<Music>

Noah Baker

A team of scientists have created a living composite material, which glows when it's squeezed. The material is made with bioluminescent algae called dinoflagellates. In the wild, these microscopic plankton emit flashes of light when disturbed to ward off predators. And this presented researchers with an opportunity, the team embedded the plankton in a soft, yet durable polymer called alginate that's extracted from seaweed. Then they used a 3Dprinter to sculpt the mixture into various shapes ranging from balls and pyramids to spiderwebs and spirals. Finally, they solidified the structures in a process known as curing. The resulting materials glowed when squeezed, stretched or twisted, and were sensitive enough to glow even when a lightweight ball was rolled across their surface. When coated with a rubber-like polymer, the structures retained their form and light-emitting properties for up to five months with minimal maintenance. The team say that this ability to convert mechanical force into radiation could have applications in soft robotics and biomedical devices, perhaps even acting as mechanical sensors. You can read more on that in Science Advances.

<Music>

Noah Baker

A 50-million-year-old bat skull is hinting that echolocation is an ancient tactic. Today, most of the world's bats emit high-pitched calls and listen for the reflected sound to navigate and catch their prey, echolocation, but it's a strategy which some scientists suspect only evolved in modern bats. And pinpointing the origins of the behaviour is tricky, in no small part because it requires palaeontologists to study delicate fossilised ear bones, and they tend to get lost or damaged over millennia. In fact, the oldest bat fossils ever found are too flattened to let researchers determined conclusively whether the creatures could echolocate, like their modern cousins. That is, until a team from Australia recovered a nearly complete unsquashed bat skull from a cave in France. The skulls ear bones demonstrated that the bat, a previously unknown species, could probably echolocate in much the same way as modern bats even though it wasn't their direct ancestor. The skull was also found alongside the fossilised remains of around 20 more bats, hinting that the animals roosted together in caves. You can locate that research in Current Biology.

<Music>

Benjamin Thompson

Finally on the show, it's time for the Briefing Chat where we discuss a couple of stories we've read in the Nature Briefing. And I think I might go first this week Shamini and it's a story that I read about in Nature. And it's based on a paper in the Journal of Neuroscience. Now, this is going to be a bit of an aside, but trust me, this is going to make sense. So I had ice cream for dessert when I had my dinner last night. Okay, now I am a big fan of ice cream. And I think a lot of our listeners probably are as well okay. And these rich, high-fat foods like ice cream are loved, obviously, not just for their taste, but also for their mouthfeel, the physical sensation they produce while they're being eaten.

Shamini Bundell

That's definitely true for ice cream because as it's gotten colder in our northern hemisphere winter, I've really fancied like ice cream but hot, and there's nothing that's quite the same texture. And yeah, mouthfeel as you say.

Benjamin Thompson

Yeah and so this mouthfeel is obviously important. But what isn't necessarily understood is how food texture, how mouthfeel like this influences, maybe, eating habits and how high-fat foods exert such a pull on us why our brains seem to enjoy them so much.

Shamini Bundell

Oh, so this is sort of the gastronomical science of what is it about particular foods that really appeal to us?

Benjamin Thompson

Yeah. And so a way that a team of researchers did this is they made a variety of different milkshakes—

Shamini Bundell

—ah—

Benjamin Thompson

—and these milkshakes varied with fat and sugar content. But this is where things take a slight turn. So what they did was they use two pigs tongues that they've got from a local butcher—

Shamini Bundell

—what?—

Benjamin Thompson

—and kind of made a sandwich with these two pigs’ tongues and put some of the milkshake in between. Okay, so stick with me here. And then by sliding the tongues across each other, they could work out the amount of friction that each milkshake was creating. And so its smoothness because higher fat gives lower friction. And this is really, really important to how something feels in your mouth.

Shamini Bundell

You were talking about milkshakes. And I was ready to volunteer for this study. I was like, yes, I'll help and now you've got into pig tongues, which is not what I was expecting.

Benjamin Thompson

Well, actually, it turns out that there is more to it than this. So, so obviously, the pig tongue is to get a sense of the smoothness. Okay, but then we move into humans. Okay, so this is where you could have volunteered because there actually were 22 participants in the study. And they were given a taste of some milkshakes with the same compositions as those that were tested to work out how smooth they were okay. And after the participants had tasted this milkshake, they were asked to say how much they'd pay for a full go on this milkshake. Okay, so like give their value of each one—

Shamini Bundell

—oh, yeah—

Benjamin Thompson

—and it turns out that samples with a lower sliding friction, so higher fat typically elicited higher value.

Shamini Bundell

So fattier equals smoother equals basically more delicious via the medium of I'd pay more for this.

Benjamin Thompson

Essentially, as I understand it, but wait, there's more. So accompanying this was some fMRI scans, some brain scans. And these showed a few things one than an area of the brain called the orbitofrontal cortex, which is involved in reward processing, detects the smooth texture of fatty food on the tongue, and this activity reflected the milkshakes’ different textures. And the researchers also showed that brain activity patterns reflected the value that the participants put on each milkshake. Okay, so higher orbitofrontal cortex activity led to a higher valuation. And this suggests that this brain region links mouthfeel to the value placed on foods, so it seems like higher fatty food has a higher value placed upon it. And in some real-world experiments, they offered the volunteers some different sorts of curry, vegetable korma, as I understand it, with different levels of fat in and showed that the results kind of mirrored what they'd seen in the MRI studies with some of the volunteers eating more of the fattiest curry compared to the other ones, even though they looked identical.

Shamini Bundell

So this has like real-world implications in how we behave what we eat. So can this particular finding, could this prove useful in some way?

Benjamin Thompson

So the researchers said that it might. So these results show that the reward system senses you know, the fat content of food, and that this part of the brain has an important role in evaluating you know food textures and, and preferences for high-fat foods. And I think understanding this neuronal mechanism could be important to understand overeating, for example—

Shamini Bundell

—mmm—

Benjamin Thompson

—and also, obesity is clearly you know, a serious health issue across large parts of the world. So the researcher said that perhaps this could be used to formulate lower-calorie foods that mimic the mouthfeel of these high fat foods—

Shamini Bundell

—oh—

Benjamin Thompson

—as it passes over the tongue or what have you.

Shamini Bundell

I will definitely be volunteering for the experiment involving the low-calorie delicious milkshake, not the bit with the pigs tongues, just the taste testing.

Benjamin Thompson

Well, let's move on rapidly then Shamini to your story in this week's Briefing Chat, what have you got?

Shamini Bundell

So I've got a Nature article here based on a Nature Neuroscience paper. We're diving into the brain science of sleep. And this is actually kind of a follow up to a previous Briefing Chat of mine. We often say like, oh, will be interesting to see what happens next with this research. Well, here it is. This was early 2021, there was some research on lucid dreaming—

Benjamin Thompson

—mmm—

Shamini Bundell

—and what it showed was that the researchers could almost have sort of conversations with people while they were lucid dreaming. So this is sort of people who have this experience of almost being able to control their dreams. I don't know if you've ever tried to do lucid dreaming, apparently, is something you can learn.

Benjamin Thompson

I've only ever had one lucid dream. And I was walking around the Shinjuku area of Tokyo following a visit there, which is amazing. But yeah, all right. So it's kind of you know that you're dreaming. But you're saying then that previously, it was shown that while these folk were asleep, they could communicate with someone in the real world as it were.

Shamini Bundell

Yes. So generally, while you're asleep, you're not moving your muscles. Otherwise, we'd all be sleepwalking all the time. But there are some muscular movements, obviously, REM sleep, rapid eye movement sleep, your eye muscles are moving, that's why you got that kind of twitching. So they'd done things where they were asking the lucid dreamers questions and looking at their eye movements. And as I said, this was a couple years ago, this new study sort of follow up to that what they wanted to know was, what about if you're not a lucid dreamer, you're a normal dreamer, as it were, can you still react to the outside world while you're asleep? Is there still some level of awareness of what's going on outside? And could you actually even potentially respond to that?

Benjamin Thompson

So rather than just being when you're out, you're out kind of thing? Like, is there still this extra sense, I suppose to an extent of the world around you?

Shamini Bundell

Yeah. Because traditionally, one of the sort of defining features of sleep is that you are not aware of external stimuli, you know, you're unconscious, you're asleep. You're not kind of responding. But they wanted to know, how kind of hard-line that was like, is it that black and white, you're either asleep, or you're not. And you know, obviously, their previous research suggests not. So what they did this time is they got two groups of people, roughly half of whom had narcolepsy. So they had both a lot of daytime sleepiness and they had a lot of lucid dreams. And then they had another group of people without narcolepsy, and got them all to sort of come into the lab and have some naps during the daytime, basically. So they do point out, they don't know about night-time sleep, maybe that's different. But they were using, like EEG, electroencephalography, sort of electrodes on your scalp, to actually confirm that these people were asleep. And also look at the different stages of sleep, including REM sleep, slow-wave sleep, different types of sleep. And then they basically asked them to either smile or frown. So this was a really simple experiment—

Benjamin Thompson

—mmm—

Shamini Bundell

—and across both groups, when they were repeatedly asked, the participants responded accurately to at least 70% of the prompts.

Benjamin Thompson

So not awake, then, but aware, to an extent.

Shamini Bundell

Not that they remembered afterwards. Or that can also depend on whether you're a lucid dreamer yet, but at the time, yeah, it seems like they're both hearing and processing and responding to this external stimuli, i.e. the researcher saying, you know, frown or smile. And, you know, I mentioned that they were tracking the brain activity as well. And they basically were kind of correlating it with, in states with more awake like brain activity, they were more likely to respond. And in some deep sleep, slow-wave sleep, they were much, much less likely to respond. So it isn't all the same kind of thing. But what it does show is that sleep certainly isn't an on off type thing, and actually, might be much more of a sort of gradual spectrum of consciousness to unconsciousness than we thought.

Benjamin Thompson

So we now have these two sets of results, then, what next, I suppose where do we go from here?

Shamini Bundell

I think these researchers are actually quite interested in how this could actually be used as a tool to help study sleep. So obviously, this is interesting for our understanding of sleep in itself. But you could also then use this kind of method of like, well, let's give people tasks while they're asleep, let’s give them stimuli while they're asleep, and then use that to further explore sleep. And again, you know, we've mentioned a lot of these participants had narcolepsy, or you know, maybe people might have issues with sleepwalking. This is a very relevant area for understanding those conditions better.

Benjamin Thompson

Well, that's a neat story, and I'm sure some of our listeners have been using the podcast to fall asleep to and I hope they're smiling rather than frowning after hearing that one. But let's leave it there for this week's Briefing Chat. For more on those stories, head over to the show notes where you'll find links to both of them. And you'll also find a link on where you can sign up for the Nature Briefing to get even more stories like this delivered directly to your inbox.

Shamini Bundell

That's all for the show. But keep an eye on the podcast feed later this week. We've got a Podcast Extra looking at how seismic signals measured by NASA's InSight lander on Mars have helped give researchers a better idea of what's going on beneath the surface of the Red Planet. Find that wherever you get your podcasts.

Benjamin Thompson

And as always, you can keep in touch with us on X, we're @naturepodcast, or send us an email to podcast@nature.com. I'm Benjamin Thompson...

Shamini Bundell

And I'm Shamini Bundell. Thanks for listening.