FANTASTICALLY ELUSIVE BIRDS AND HOW TO FIND THEM
JOHN BIRMINGHAM discovers how the fledgling science of ecoacoustics is transforming conservation.
Unless you’re a fugitive, an ecologist or a crocodile, swamps are terrible. Yes, yes, yes – very important ecological super niches and all that. But the mud, the quicksand, the insects, the predators; the deep, abiding discomforts of brute creation; the steam-press heat and humidity; the particular stink of marsh gas bubbles and the generalised stench of rot and genesis – it all pretty much sucks. Unless you’re Liz Znidersic.
“For me it’s bliss,” she says. “It’s muddy. It bites. Every mozzie there is as big as a small bird. You drive through a cloud of them [and] it’s like bullets hitting the windscreen.”
But it’s bliss.
Specifically, the sort of bliss an audiophile supernerd feels when slipping on a new pair of Sennheiser Orpheus cans – they’ll set you back a hundred thousand dollarydoos full retail – to listen to a virgin wax pressing of their favourite band at 33rpm. Dr Znidersic likes to listen.
Specifically, she likes to listen for cryptic bird species, the sneaky ones, who hide out in the world’s gnarliest wetlands, staying silent for weeks at a time, almost as though they know she’s out there, listening for them, and they will be damned if they’ll give her the satisfaction of a single tweet.
Until recently, hunting for fugitive species, especially birds, in remote and punishing wilderness was expensive, difficult and more often than not futile. A researcher might embed themselves in the big muddy for a couple of weeks, listening and recording, but the scope was finite, the returns contingent, and the points of failure many.
Advances in audio technology, batteries, and increasingly in solar-cell power began to fundamentally and rapidly rewrite the equation. Small recording units with weeks or even months of power could be placed throughout an area of interest and left to record the soundscape of the local environment 24/7.
In some ways sound is an even richer resource for studying ecologies than direct or recorded visual observations. Like the reader of a first-person novel, a camera sees only what is in the light cone ahead of it. Sound travels – sometimes over great distances. Where many species are furtive and even deliberately clandestine in their movements, the needs of the genetic line still demand they reach out to potential mates. They do so through calling.
There are other advantages. Any number of researchers can listen to the same recordings any number of times, to improve interpretation. The data can be revisited by future researchers with better analytic tools, and of course the passive arrays of recording devices create much less disturbance in the local ecology than the continued presence of even one human observer.
Is there a downside? But of course.
Even a single day of audio from a single digital recorder generates a vast and almost impenetrably dense trove of data that simply cannot be analysed minute for minute. At least not by human beings.
This was where Znidersic found herself as a postgrad with an interest in natural resource management. A traditionalist who had thrilled to the idea of sitting for a week in a swamp, increasingly dizzy from blood loss to mosquitoes (“It’s bliss!”), she was told that acoustics were the future and she’d better get with it.
“It was just thrust on me that one of the tools I had to use was acoustics, and I was like, ‘Aw, gawd’. I started to collect all this data. I really didn’t have much faith in it.”
Not just data. Lots of data. Terabytes, to be specific. Her supervisor, Professor David Watson at Charles Sturt University, Albury-wodonga, was already in contact with Professor Paul Roe at Queensland University of Technology’s School of Computer Science, binding up the early threads of a crossdisciplinary approach that would eventually create the data science of ecoacoustics.
Recalls Roe: “We started thinking about it over 13 years ago. I came from an escience background and I knew how data was revolutionising many sciences, and the value of preserving data. I’d created an escience centre and we had several projects, including some involving sensors and some in bioinformatics. David Watson had been working for me on some bioinformatics projects and we brought some of the ideas together.”
Watson suggested Znidersic reach out to the Queenslanders and soon she found herself in the laboratory of a bearded gentleman called Towsey, who wanted nothing at all to do with swampworlds of sucking ooze and venomous snakes, even if there were mysterious and cryptic little feathered fellows to be found in there.
Michael Towsey had his own problems.
“I was being given these 24-hour recordings,” he says. “That’s a huge block of sound. I couldn’t even open them with the software I had at the time. I was under a lot of pressure. My job was on the line. Twentyfour-hour recordings, and I can’t even open the file. I haven’t got a clue what’s going on!”
Years later he still sounds stressed.
But necessity and some desperation being the mother of invention, he came up with the idea of breaking the recordings into one-minute segments. He could open a one-minute audio file, no problemo. Analyse the heck out of it, too. The 1,440 minutes he could then stitch back together for a whole day’s output.
But only for Towsey.
The output was a spectrogram: a fuzzy monochrome blur of data visualisation that represented 24 hours of audio recording – thousands, maybe tens of thousands of instances of insect chirrups, bird calls, frog croaks, rainfall, aircraft and the strangely terrifying growl of the koala – presented as the sort of impenetrable greyscale scan an oncologist might hand to a life-long smoker with a sad shake of the head. The eureka moment was colour.
“By drilling down into the frequencies, represented by the falsecolour spectrograms, certain species revealed themselves.”
And realising that you could apply filters to an audio spectrogram the same way you could to a photograph.
There’s a lot happening in any given minute of audio recording in a wilderness – and at particular times of the day and night, that complexity explodes. Filtering out the complexity wasn’t possible for ecologists like Liz Znidersic sitting in the mud or, to be honest, sitting back in the sound lab listening through headphones.
But it was possible with enough processing power. Originally, this power was harnessed to create software to identify the calls of individual species in the chaotic, unconstrained soundscape of the wilderness. Think of it as Shazam for bird calls. Or frog croaks, or whatever. But in the same way the little smartphone app that tells you the name of the absolute banger
playing in the bar can be overwhelmed by background noise, even the best, most expensive species audio “recognisers” fail. And of course, they are silent on the content of recordings in which the uncooperative or absent species remain silent.
Treating long-duration recordings as soundscapes, rich with different categories of audio sources, opened the path to sorting those categories into biophony (produced by living animals), geophony (the sounds of wind, rain or crashing waves we’ve all been listening to through pandemic lockdowns) and anthrophony (for any human contributions).
Once you know what to ignore, 24 hours (or 1,440 minutes) of recording shakes out into a simpler (but still complex) audio map of biophonic sound sources. And those sources can be filtered by different indices to create different patterns of acoustic structure.
“It’s like photography,” Towsey explains. “You can put filters on a camera lens and you’ve got the same scene but you’ve got three different views of it because the filter removes this and that. We’re not filtering a photograph, however – we’re filtering sound. If I apply three different filters, or indices, I get three entirely different views into the sound world.”
Towsey’s breakthrough moment arrived when he realised that if he took these different indices, or views, into the sound world, and assigned them to the red, green and blue channels of the visual image, he suddenly had a false-colour spectrogram, instead of a grey haze.
“It’s exactly the same thing except that instead of visual filters, I’m applying sound filters. You put them all together and bingo! The first one I saw I was
The breakthrough moment was colour. And realising you could apply filters to an audio spectrogram the same way you could to a photo.
just amazed at the amount of detail I could see in a 24-hour recording. Compared to what had previously been possible it was… it was… well, let’s just say it was a lovely feeling.”
The spectral acoustic indices about which Towsey was growing so excited on our Zoom call describe certain precise features of the distribution of acoustic energy in each of the frequency bins of the one-minute recording segments he’d been forced to create through the constraints imposed upon him by the limited software with which he was then working.
Liz Znidersic was more poetic.
“Look at the false-colour spectrogram the way an ecologist looks at the landscape,” she says. “We learn
WIRED FOR SOUND: THE A2O
The Australian Acoustic Observatory (A2O) is a continental-scale acoustic sensor network, designed to collect data over five years from 90 sites across seven different Australian ecoregions.
Funded by an
Australian Research Council (ARC) Linkage Infrastructure, Equipment and Facilities grant of $1.8 million, A2O is futuristic and… well, hard to explain.
“It’s not actually the traditional sort of scientific project where you say, ‘hey, we’ve got a question – let’s try and answer that question’,” says Professor David Watson, of Charles Sturt University, who’s one of the A2O’S five chief investigator managers.
“We call it an observatory … because it’s borrowing the astronomers’ use of the word and that is: ‘Hey, we’ve all got questions. So let’s all pool our money. Let’s get a big grant. Let’s buy some big kit. And let’s all address our questions with that kit.’”
The “kit” at each site is four solar-powered acoustic sensors (which retail for about $1,300) purpose built by Brisbanebased Frontier Labs.
The number of sensors, Watson says, reflect some built-in redundancy “if something happens – if a cow leans on a machine, if a bushfire comes through”. Two sensors are placed near wet habitat and two in a relatively dry zone. That’s about 360 sensors across the entire network.
“Each machine collects about a gigabyte of data a day,” says Watson. “And so when you start doing some sums, [the data set] gets really big, really quick.”
So who in science is accessing the data, and what are they doing with it? Watson says the current project of Liz Znidersic (see main story) is a great example.
“Liz is working with a whole bunch of sneaky birds,” he says. Watson says Znidersic is using A2O kit to “eavesdrop” on those calls, and then, with her collaborator Michael Towsey, using falsecolour spectrograms (“a really fancy approach”), to reduce the complex soundscape into something that humans can analyse and query.
But the sensors are collecting information about “way more than just birds”, Watson says. He cites insects, raindrops, mammals (“goats make a lot of weird sounds”) and anthrophony – human sounds: “In desert areas that are very still, you can [record] planes from a very long way away.”
So what are the wider possibilities for research from the A2O? Watson says that only now, three years in, are they starting to be revealed. He thinks that, just as GIS (geographical information systems) took many years to develop
“as a whole way of doing spatial science”, that it’s going to take “decades, decades” for all manner of scientists to realise the possibilities of sound data. “That’s where we are now,” he says. “I get like three emails a day from people who say, ‘hey, I want to do this’.” To which he responds: “If you really want to do that, you’re going to need to work out how to do it because no one has.”
All the data collected is open access, so there’s plenty of time to work out how to do things.
to read and understand parts of that landscape. We see eucalyptus, reeds, whatever, and we associate certain species with it.
“A false-colour spectrogram is a representation of vocalisations during a 24-hour period. It’s a soundscape, so you’ve got all the different species vocalising in different frequencies. Some species vocalise all in the one area creating what we call the ‘cocktail party effect’. Everyone is trying to talk above each other so they can be heard.”
By drilling down into the frequencies, however, represented in the false-colour spectrograms by flares of red, blue or green, certain species revealed themselves.
“I’d been using camera traps and call playback methods to detect the cryptic wetland birds,” says Znidersic. “David Watson suggested I apply acoustic monitoring to see if it was useful. I collected all these terabytes of data and was at a loss what to do with it. Michael then analysed the terabytes into the false-colour spectrograms. Visually they were beautiful, intellectually they were mind-bending. Once I started to read the spectrograms as a visual landscape, seeing species and groups of taxa clearly, I was literally taken on an amazing trip into sound and colour.”
She remembers walking into Towsey’s lab at QUT, listening to him talk about false-colour spectrograms, and staring at them for a very long time, trying to interpret them. “And I remember saying to Michael, ‘I think I can see Lewin’s rail calls in there.’ That was new, because we hadn’t gone down to species-specific level. And I remember he turned to me and said: ‘Well, you better prove it. You get the spreadsheet together and present that to me.’”
Towsey smiles at the memory.
“We weren’t married at that stage,” he tells me. Reader, they are now.
In a 2018 paper in the Journal of Ecoacoustics called “Long-duration, false-colour spectrograms for detecting species in large audio data-sets”, the two scientists describe Lewin’s rail as “a furtive wetlanddependent bird which inhabits thick vegetation, calls rarely and is seldom seen”. A chubby little redhelmeted introvert, Lewin’s rail likes foraging for invertebrates at the edge of shallow water and minding its own damn business.
There are eight subspecies of Lewin’s rail, although one of them, Lewinia pectoralis clelandi, was last seen in the southern reaches of Western Australia in 1932, so its chances of a comeback aren’t looking good – the last known live Tasmanian tiger was still walking around four years later.
Rail species worldwide are among the most threatened bird species in the world, due to invasive species. They’re also shy.
Until passive recording and long-duration false-colour (LDFC) spectrograms, one of the main ways of finding Lewin’s rail was what’s known as a play callback, where an ecologist “plays” the sound of a bird and waits for the real thing to respond. Unfortunately the response of the Lewin’s rail to such an entreaty might just be to abandon its nest and get the hell away from an unexpected competitor.
As Towsey and Znidersic wrote with their coauthors: “Repeated use of call playback (which simulates a territorial intrusion) may negatively affect resident pairs, resulting in territory abandonment or nest failure. Additionally, this methodology also requires a costly extended survey effort to enable high confidence levels inferring absence. Lewin’s rail vocalisation repertoire changes temporally from an acoustically simple contact call to a complex call
repertoire with harmonic elements that is thought to be associated with breeding. Vocalisations are also sporadic, and of either short or long duration. To establish their current distribution and evaluate their population status, a monitoring approach is needed that can reliably detect small numbers of individuals unobtrusively.”
In other words, the species doesn’t like to be shouted at and it talks rarely, often using funny voices. It’s a perfect candidate for the new approach.
The researchers chose Tasman Island for their study, a forbidding, sort of oval-shaped plateau rising nearly 300m above sea level off the stormy south-east coast of Tasmania. Populated from the early 20th century until 1977 by lighthouse keepers who grazed sheep and left behind a murderous litter of feral cats, Tasman has been largely denuded of its original forest cover. But its steep cliffs and table top heights still play home to tens of thousands of fairy prions, little penguins, swamp harriers and shearwaters – their numbers enhanced over the past decade by a cat-eradication program declared successfully completed in 2011.
When Tasman’s avian choristers get rolling they generate quite the cocktail party effect.
Znidersic placed a Wildlife Acoustics SM3 sensor on the island for 10 days in November 2015, resulting in 240 hours of recording. Recognition software
“Ecoacoustics is transforming ecology into a big-data science, a data-driven science, rather like bioinformatics.”
eventually delivered up to 49 positive identifications, or “instances”, including 18 easy hits, 12 more difficult examples and 19 “very difficult instances”. To compare: 70 confirmed observations of the Tasmanian subspecies of Lewin’s rail in the preceding 20 years. Then 49 instances in 10 days.
“That’s the novel and amazing thing about this,” Znidersic says. “I can sit in a marsh for two weeks and not hear this bird call. You leave the recorder out there for a month and – what do you know? – as soon as I leave, it calls.”
It was, she thinks, the true crossing over of the disciplines.
“Something really tangible for an on-ground outcome. That’s where I come from. How can we find the birds, minimise impact on species and the environment, competently infer absence? If they’re not there, we’re really confident they’re not there.”
But they were there.
T “here’s a huge future coming out of this one tool,” Znidersic promises. “We’re looking at how can we answer different ecological questions using false-colour spectrograms. We’ve been working on a project for a couple of years looking at pre- and post-feral cat eradication on an island. We’re also looking at pre- and post-fire.”
But it’s not close to being a mature technology, her partner warns.
“From an acoustics point of view, there’s still a lot to be done,” says Towsey. “There isn’t really any effective software for automated bird recognition. Environmental recording is extremely difficult. You can and you do get everything. We get gunshots, we get human speech, we get wind, rain; we’ve had bat wings knocking the microphone. Anything that can happen in the acoustic world does happen in environmental recording.”
Towsey sees the future in what he calls content description: a simple explanation, minute by minute, of the content.
“You know, this minute contains bird calls, this minute has frogs and insects,” he says. “The advantage of that is, it can be put into a text database and text databases are very efficient for search. So an ecologist would be able to search a database and pull out all the minutes that contain frogs, or frogs in a particular bandwidth. You’d be searching these terabytes of acoustic recording simply by searching the text database. That’s what I started on, and it’s probably a bigger job than I’m going to be able to finish.”
Bigger still perhaps is the idea that started to emerge from discussion between Paul Roe and David Watson over 13 years ago. As Roe’s IT specialists and data scientists cooperated with ecologists such as Znidersic on increasing numbers of projects, and as the cost of computer storage and hardware came down, their original idea of an acoustic observatory moved from speculation to execution.
The Australian Acoustic Observatory (A2O) now exists as a continent-spanning network of sensors continually recording across multiple ecosystems (see “Wired For Sound”, page 35).
To understand the importance of such data, imagine the same technology had been available to Sir Joseph Banks when he first arrived in Australia aboard Cook’s Endeavour. We would not just have his detailed notes and paintings of the undisturbed wilderness, but an exquisitely detailed, super-dense audioscape of the existing ecologies. That is what A2O is gifting to future generations.
“It will provide badly needed baseline data, which for much of the world we don’t have,” Roe explains. “In many regions we simply don’t know what species are where. Without such information we have no idea how our world is changing in response to bushfires, invasive species and climate change, nor can we effectively audit the environment or monitor remediation strategies to know if they are working.”
Roe sees a future in which acoustics will increasingly be used with other sensors, often remotely, to scale environmental monitoring.
“And this will in turn support environmental accounting and green banking,” he says. “I expect there to be progress in integrating different data streams from different devices to yield comprehensive information on how our world is changing. Ecoacoustics is transforming ecology into a big-data science, a data driven science, rather like bioinformatics.”
“I can sit in a marsh for two weeks and not hear this bird call. You leave the recorder out there for a month and – what do you know? – as soon as I leave, it calls.”
The accelerating slide into climate chaos also drives Liz Znidersic. “Eighty per cent of the world’s wetlands are disappearing or have gone,” she says. “And we need water in this world. Wetlands are critical, so to monitor a wetland with very basic ecological indicators is essential to us. It will play a vital part in our world existing into the future.”
Her husband leans forward, his brow furrowed like a man who has just been handed a massive data dump he can’t even begin to open, let alone analyse.
“The interface between ecology and computer science is still not easy,” Michael Towsey says. “There’s very few people who can step across those boundaries. But increasingly it is happening and ecoacoustics is being recognised as a discipline in itself. That’s only happened in the last 10 years. There’s even a journal.” Validation, at last.
Towsey still doesn’t find bliss in brute creation. The gators and crocs growling at night, the insects hitting the windscreen like bullets – none of it is as agreeable to him as a nice air-conditioned lab, with a decent coffee machine and a really interesting data set.
“But,” Towsey finishes, “the point that Liz made about wetlands being crucial to ecology of the continent is what drives this work. It’s the raison d’être for our careers.”
JOHN BIRMINGHAM is a Brisbane-based bestselling writer of non-fiction and science- and fantasy fiction. His most recent book is The Shattered Skies. This story is part of our New Ways of Seeing series, enabled by a grant from the CAL Cultural Fund.