Acoustic data provides scientific and engineering insights in fields ranging from biology and communications to ocean and earth science. At Pivotal, acoustic data is one of the tools we use to understand the state of nature. Acoustic data is a rich source of information on a wide range of animals, from birds, to bats, frogs and insects, and it is simple to collect. But acoustic datasets can be very large and challenging to analyse, especially the step that determines what sounds were recorded and what species they belong to.
In order to analyse the data at scale, we build and deploy acoustic machine learning (ML) models and collaborate with our global network of ecological experts, to provide crucial quality controls and eliminate bias in the data. We recently sat down with two of our esteemed ornithologists, Magnus Robb and Pedro Fernandes to hear about their journeys in ornithology.
“AI can do amazing things, but it depends on what we feed it.”
Magnus Robb is a musician and composer whose pieces have featured on BBC radio. Since his late teens he has turned his musical ear to the study of bird song and is today one of the world’s foremost experts in ornithological sound. A founding member of The Sound Approach, Magnus has spent three decades making bird sound recordings and continues to travel the world to tune his ear to the beautiful diversity of bird song. He has been a member of our Expert Ecology network since its earliest days, and his deep expertise in identifying birds from their calls has been invaluable to Pivotal ever since.
Tell us a bit about your background. What led to your interest in ornithology?
I have always been interested in wildlife. I holidayed in the Orkney Islands as a child and thought it was totally normal to grow up watching puffins and other wildlife.
In my teens, I drifted away from that passion for a while and instead fell in love with music. I trained as a musician and discovered I was good at composing.
My interest in birds and wildlife came back strongly when I was around 18 or so. I became interested in whale song and fascinated with how this other-worldly sound could be incorporated into music. Then one day I listened to a radio programme where someone sped up whale song and it sounded like a song thrush. I loved this, and it led me to start analysing bird sounds and all the hidden melodies within them.
For example, if you take a snippet of a skylark’s song and slow it right down, stretching it out to eight times the length, you still won’t have unraveled all the melodies and complexities within it. It’s incredibly musically intricate.
Over time my interest in bird sounds themselves became stronger than my motivation to derive my own music from them. The change of emphasis was helped by the fact that I loved the activity of going out to record the birds – I love being out in nature.
How did you go from this to becoming such a deep expert in the identification of birds by sound?
I got to know the bird watching scene in Holland where I was living at the time and I was amazed to discover there were people who were good at identifying birds from just a few little tweets. I thought my musical ear might mean I could be good at that too, and I started to research the practice of identifying birds from sound to see if I could learn.
Pretty quickly I became more serious about it than any of the other bird watchers I knew at the time. I wanted to know more and more. I found that, for example, if I knew details of one or two thrush species, I could make good guesses about ones I didn’t know. And then suddenly all the sounds I had learnt started falling into groups in my brain, and I started to be able to make so much more sense of all these different sound repertoires. Like when you learn a foreign language and suddenly you cross an invisible threshold where you have enough vocabulary to be able guess words from context and that allows you to start to follow conversations. So, my instinctive ear got significantly better, and I found I really loved what I was then able to do.
At this point I became involved with The Sound Approach, which was supposed to be a two-year project – and we’re still going today. The Sound Approach was the door to doing what I love most, and actually being paid for it, which is a pretty rare and special thing.
What’s the hardest ID you’ve ever done that you are most proud of?
Every autumn in the UK two or three species of pipits turn up. Two (the Richard’s and Blyth’s pipits) are from the Siberia / Mongolia region. Blyth’s is rare. A couple of years ago I was sent a recording of a pipit that people had been arguing fiercely about. They were arguing over which of these two species it was and at first, I also couldn’t make it fit. Then I had a moment when the penny dropped, and I suddenly thought “It’s neither of those – it’s a Paddyfield Pipit”. But that is a south Asian species and is not supposed to occur in the UK – at the time, the closest to the UK it had ever been recorded was the United Arab Emirates. So, this was a crazy idea. It was later proven that I was right. No one knows how it got to the UK. I’m proud of that ID because it was so out of the box.
People send me things to ID all the time, but this was one of the best, most rewarding puzzles I’ve managed to solve.
What is the biggest current challenge in your field?
I think there’s a danger that we’re in the last age when people with my level of bird sound identification expertise exist, because we are delegating to AI. Nowadays increasing numbers of people use apps on their phones to identify tricky sounds instead of puzzling them out and learning from the process. If people rely on this technology to make identifications, there is a risk that fewer and fewer people learn what I’ve learnt, and therefore that we’re less and less able to error-check the AI models, and less and less able to know whether what they’re telling us is the truth.
AI can do amazing things, but it depends on what we feed it. If we don’t quality control it, we risk that it learns inaccurately and becomes biased. And we can’t quality control it unless there are people with the expertise to do this. So, we’re at a junction in this field where we have choices to make about the skills we value and the accuracy we care about. Of course, it could be that help from AI will inspire more people to get involved by helping them to get started and learn. I hope that’s the case. But if we don’t involve experts in the AI pipelines and quality controls, as Pivotal does but many others do not, there’s a risk that my kind of expertise disappears, and then we won’t be able to further improve what the AI is doing or know whether it is learning accurately.
What do you enjoy about working with Pivotal?
Working with Pivotal brings me great new challenges and puzzles to solve, which I really love. It forces me to sharpen up my ID skills for some of the common species. Identification of common species isn’t a challenge that comes up often in the ornithological world (precisely because they’re common), but some of them can actually be quite difficult to identify.
For example, distinguishing common and spotless starlings can be really difficult, especially if the recording contains only a small fragment of song. Working with Pivotal forces me to work harder on these species, because Pivotal includes even common species in their quality control processes. It is important to error check AI models even on the common species, which they should in theory be better at, because AI is not yet advanced enough for us to know where it is making important mistakes.
“I get to do what I love – labeling sounds – and it is nice to listen to familiar bird voices in familiar and unfamiliar places alike.”
Pedro Fernandes is an ornithologist and science illustrator from Portugal with a passion for natural history. Pedro was an early founding editor for eBird, where he has been managing the Portuguese team of bird watchers for over a decade, and contributes to the Merlin Bird ID app catalogue of bird sounds. In addition to publishing numerous studies on ResearchGate, Pedro has also illustrated for several publishing houses and non-governmental organisations, including the Cornell Lab of Ornithology, Princeton University Press, Storey Publishing, Éditions Fleurus, Schwager & Steinlein Verlag among others. Since joining our Expert Ecology network, his expertise in identifying birds from their calls has been invaluable in accurately identifying even the rarest of species in the Iberian Peninsula and Mediterranean coast.
Tell us a bit about your background. What led you to become interested in ornithology?
I was born and raised in Lisbon and became an avid nature book reader early on. I’ve always enjoyed birds, but it wasn’t until my university years that I started birdwatching. At that time, I mostly visited the Tagus Estuary Nature Reserve, which is the largest wetland in Portugal and one of the most important in Europe, as a sanctuary for fish, molluscs, crustaceans, and especially to birds that stop over on their migration between northern Europe and Africa.
I later studied earth sciences in Lisbon, with a focus on paleontology, and then obtained my certification in science illustration from the University of California Santa Cruz, with a one-year internship at the Cornell Lab of Ornithology in Ithaca, New York.
Throughout my career, I have alternated between natural history illustration, field ornithology and sound labeling. In addition to working with Pivotal, I have been part of the eBird regional editor team since eBird went global, managing the Portuguese team of volunteers for over a decade, and contributing to labeling birds on the Merlin Bird ID app for the Iberian Peninsula. I’ve also had a chance to travel quite a bit – in addition to Portugal and the US, I’ve lived in Ireland, Turkey, Morocco and am currently in Croatia.
In all your travels, what’s the hardest ID you’ve ever done that you are most proud of?
A Hume’s Warbler that I heard in Rabat, Morocco. I couldn’t see the bird, so I recorded it to listen back. The recording sounded promising but there were no previous IDs of this species of Warbler in Morocco (or Africa at all!), so I reached out to an esteemed colleague for additional review. It was a huge relief when Magnus Robb listened to the recording and confirmed it (thank you once again, Magnus!). Eventually I did see the bird, but the views are never as satisfying as the calls.
What made the ID challenging?
Hearing bird calls requires you to be able to isolate their sounds accurately and consistently from other environmental sounds, like the wind. At times, when the winds are strong, the recordings can be very painful to listen to! I would love to see further advances in microphone windshield technology.
What do you enjoy about working with Pivotal?
Working with Pivotal has been a very positive experience. I get to do what I love – labeling sounds – and it is nice to listen to familiar bird voices in familiar and unfamiliar places alike. The instructions for annotations are clear, and the communication with the team is straightforward. I also really appreciate the flexibility Pivotal provides, along with reliable support from the Operations team. It has been great so far and I look forward to continuing working for Pivotal!
If you’re interested in joining Pivotal’s network of world-class ecological experts, please contact us on annotator-support@pivotal.earth.
To discuss how we can help you secure real, auditable evidence of on-the-ground changes in nature, so you can make solid decisions and trustworthy claims, contact us on info@pivotal.earth.