[Chelsea Benson] So welcome, everyone, to today’s webinar from the Cornell Lab of Ornithology. We’re going to be discussing the amazing research and conservation work being conducted by the Elephant Listening Project. My name is Chelsea. I’m on the Visitor Center team here at the Cornell Lab, and I’m going to be facilitating today’s conversation. And I want to introduce the people who are with us today who are going to be both on screen and in our background, behind the scenes answering your questions.
So with us today are members of the Elephant Listening Project. We have Daniela Hedwig, welcome Daniela.
[Daniela Hedwig] Hi, everyone.
[Chelsea Benson] We have Ivonne Kienast. Welcome, Ivonne.
[Ivonne Kienast] Hi.
[Chelsea Benson] We also have Colin Swider.
[Colin Swider] Hello.
[Chelsea Benson] And Bobbi Estabrook.
[Bobbi Estabrook] Hi, everyone.
[Chelsea Benson] And then, helping behind the scenes is a long term member of the Elephant Listening Project team– we have Liz Rowland.
[Liz Rowland] Hi.
[Chelsea Benson] And we also have Leo Sack who’s on our Visitor Center team who’s going to be helping with the Q&A in the chat. So welcome, everyone. Thanks for joining us today. All right, so we’re going to be hearing more from our panelists soon and I’m going to make some announcements. So panelists, if you want to turn off your audio and video, that’d be great– and we will see you soon.
All right. So, today’s webinar is hosted from Ithaca, New York. And I want to read a statement acknowledging the indigenous people as the original inhabitants of this area.
Cornell University is located on the traditional homelands of the GayogohÃ³:no, the Cayuga Nation. The GayogohÃ³:no are members of the Haudenosaunee Confederacy, an alliance of six sovereign nations with a historic and contemporary presence on this land. This Confederacy precedes the establishment of Cornell University, New York State, and the United States of America. We acknowledge the painful history of the GayogohÃ³:no dispossession, and honor the ongoing connection of GayogohÃ³:no people, past and present, to these lands and waters.
For those of you who aren’t familiar with the Cornell Lab of Ornithology, we’re home to a community of researchers and supporters from around the world who appreciate birds– and elephants– and the integral roles they play in our ecosystems. Our mission is to advance leading edge research, education, and citizen science that help solve pressing conservation challenges.
A few tech-related notes for our audience today. There is closed captioning available on Zoom. If you would like to turn those captions on or off, please click the captions button at the bottom of your screen. Today’s program, each of our panelists are going to be sharing a short presentation about their work, and the final portion of our program is dedicated to Q&A.
For those of you on Zoom, you can click that Q&A button and type in your question. We’re going to be answering some questions verbally and for others, as you saw, Liz and Leo will be helping answer your questions behind the scenes. You’ll be able to see their responses in the answer column. We’re going to only use the Zoom chat for tech support and to share information, and we’re not going to look for your questions there, so be sure to use that Q&A feature.
We’re also streaming live to Facebook. And if you’re watching on the Lab of Ornithology, or the Elephant Listening Project Facebook pages, you can add your questions into the comments, and we also have colleagues there who are going to be helping you out. Please be aware that we have seen spam attempts. So if you see any links that aren’t posted by us, please don’t click on them. OK, that’s all the announcements. Let’s get started. First off, I want to welcome Daniela Hedwig. Hey, Daniela.
[Daniela Hedwig] Hey again.
[Chelsea Benson] Daniela is the leader of the Elephant Listening Project. And I know many people in our audience are familiar with the Lab of Ornithology and might be scratching their heads and thinking, “really, elephants?” But I would love for you to share a little bit about the history of the Elephant Listening Project.
[Daniela Hedwig] Well of course, when you hear or think of the Lab of Ornithology, you first of all think of birds. But overall, we are really interested in preserving and conserving the biodiversity of the natural world in general.
So at the K. Lisa Yang Center for Conservation Bioacoustics, where the Elephant Listening Project is based, we are interested in recording the sounds of a wide variety of species. So for instance, we record the sounds of humpback whales in Alaska, katydids in Central America, gibbons in Southeast Asia, but also the forest elephants in Central Africa.
[Chelsea Benson] Wonderful. Yeah, I love that the Conservation Bioacoustics is such a wide and varied program here at the Lab, and so valuable to the work that we do– not only on birds, but also, as we’re going to learn today, about elephants. So Daniela is going to start our journey today by setting the scene in Central Africa. And she’s going to share what makes African forest elephants special, the threats that they face, and how eavesdropping on elephants contributes to their conservation. So Daniela, take us away.
[Daniela Hedwig] Thank you, Chelsea. All right. I would like to take you all to a little journey to where we work, which is the Central African rainforest. The Central African rainforest is the largest contiguous block of tropical rainforest outside of the Amazon. And we, the Elephant Listening Project are dedicated to protect this habitat by bringing bioacoustics monitoring tools to Central Africa.
The Central African rainforest is a global hotspot for biodiversity. It is home to 10,000 species of tropical plants and over 400 species of mammals. And besides being a biodiversity hotspot, the African rainforest provides important ecosystem services for us. The rainforest stores huge volumes of carbon in its vegetation and tree trunks, and this serves as an important buffer against climate change.
We focus on forest elephants to conserve the Central African rainforest, because they constitute a crucial keystone species within this habitat. Elephants have an tremendous influence on the community of tree species in the forest. That’s because forest elephants love to eat fruit, and with their large body size and large home ranges, they contribute to the dispersal of the seeds of many large fruit-bearing tree species. And by doing so, they maintain and shape the structure of the rainforest. This not only makes food resources available for other species, but it importantly helps the rainforest to store carbon. And that’s why forest elephants are often referred to as the gardeners or the architects of the rainforest.
So elephant conservation in Central Africa equals biodiversity conservation. Elephants are crucial to maintain the diversity in the rainforest, as well as the ecosystem services it provides for us and for future generations. But really, forest elephants are absolutely fascinating creatures which actually have much in common with us humans, despite looking so different.
They also have a very complex social system, very strong social bonds. Groups typically consist of females with their offspring– and females only give birth for the first time on average with 23 years, and they show extensive parental care. And while daughters leave their mothers as adults to start their own families, mothers and daughters and siblings manage to maintain very strong social bonds throughout their lives. And they’re able to do so because they communicate over long distances with each other through very low frequency rumble vocalizations.
But the architects of the rainforest are under threat, and that’s particularly due to poaching for their ivory. This has developed into a real crisis– it has been estimated that 30,000 elephants are being killed for their ivory every year, and this comes down to 34 elephants that are being poached within the hour. And really, forest elephants bear the brunt of this poaching crisis. Within the home and within their distribution range, two ivory poaching hotspots have been identified.
As a result, their distribution range has considerably shrunk within the last 50 years, and their population size has been estimated to have declined by 62% between 2002 and 2011. And finally now, the International Union for the Conservation of Nature, the IUCN, has now finally recognized that forest elephants do face extremely high risk of extinction in the wild. And they are now listing them as critically endangered on the Red List of Threatened Species.
But the conservation needs of forest elephants are unfortunately still little understood, and most of our knowledge about elephants still derives from savanna elephants. But forest elephants are considered a separate species and their habitat is, of course, strikingly different. And these differences should lead to differences in their behavior, but forest elephant behavior is very little understood. The reason for that is because they are extremely difficult to observe in the dense vegetation of their rainforest habitat. And that’s why we believe that eavesdropping on their vocalizations using passive acoustic monitoring is key to the conservation of elephants in the tropical rainforest they inhabit.
Acoustic monitoring really is a relatively simple method. We venture out into the rainforest and install acoustic recording units in the trees, and we leave them out there to record the sounds of the rainforest for many months, sometimes even years. We collect the SD cards with the sound recordings on, and then we scan those sound recordings for the sounds of the elephants using automated detector algorithms. And we have thousands of hours of recordings of elephant vocalizations like this. And you may think of elephants– of roaring and trumpeting, but the far more representative call is the rumble. And here in this picture, we see a visual representation, also called a spectrogram, of an exchange of rumbles between two elephants. And in the green boxes, you see the calls of one individual, and the blue boxes show you the response, or the answering calls, of the second elephant. And we can listen to this.
[Daniela Hedwig] So these kind of rumble vocalizations really underpin any sort of social interactions between forest elephants– within their social group, between groups, over long and short distances– and they call relatively frequently, one call every 40 minutes. Because of their low frequency, these calls are able to travel over very large distances in the rainforest. And that allows us to detect elephants over large areas, even when they are at low densities. Counting these rumbles in space and time tells us something about where the elephants are, how many they are, and how this changes with time and depending on the habitat. We can create maps of how elephants are distributed across the landscape as well. For example, here in this map.
But we are not only recording elephants with acoustic monitoring– you record the entire soundscape. And another thing that we are able to detect are gunshots of the poachers that are trying to kill the elephants for their ivory, as depicted in this map here with the red dots. And this is really important information for the conservation organizations that we are working with in Central Africa, because it allows them to evaluate and adapt their conservation strategies and anti-poaching patrol efforts objectively.
We are conducting all of our work in close partnership and collaboration with local conservation organizations. Our goal is really to support local conservation organizations in monitoring their elephant populations and poaching so that they can make informed decisions about their conservation strategy. And capacity building is the foundation of all of our work. Our guiding principle is to make our local partners the leaders of their project and conservation in Central Africa in general.
So far, we had the pleasure to implement passive acoustic monitoring in 14 different sites across four Central African countries. For the remainder of this event, we will continue our journey deeper into the heart of the rainforest. We will visit our two major sites and one of the last remaining strongholds of forest elephants, which is the Sangha Trinational Conservation Complex.
The Sangha Trinational is a UNESCO World Heritage site that spans across three national parks in the Republic of Congo, the Central African Republic, and Cameroon. And it is one of the last remaining large blocks of pristine and entirely protected rainforests, and in places like this, it’s where elephants have a true chance of survival.
First we will take you to the Central African Republic to Dzanga Bai, a huge forest clearing deep in the jungle where elephants venture out into the open to form the largest known aggregation of forest elephants. And afterwards we will take you to Northern Congo, to NouabalÃ©-Ndoki National Park, where we are eavesdropping on elephants roaming about in the deep rainforest. Thank you so much for tuning in.
[Chelsea Benson] Thanks for that, Daniela, that was really great. And there’s a lot of wonderful questions popping into the Q&A. So I’m excited to explore that more at the end of our presentation today. So our next presenter is Ivonne Kienast. Ivonne’s a Cornell graduate student and the manager of the Dzanga Forest Elephant Project. Ivonne is going to share this incredible site and the long term data that she and the local research team have been collecting for decades. So I will let you take us on our next destination. Thanks, Ivonne.
[Ivonne Kienast] Yes, hi, such a pleasure being here. Just let me grab the presentation to take everyone to Dzanga Bai. My name is Ivonne Kienast, and I am the project manager of the Dzanga Forest Elephant Project in Central African Republic. So let me take you on this trip to this unique place Dzanga clearing in the rainforest, also known as the village of elephants. All started 27 years ago with a researcher called Andrea Turkalo, who you can listen to in this short video.
[Video narration by Andrea Turkalo] The Central African Republic is located in the dead center of Africa. A local name for the country is Be-Afrika, which means heart of Africa, and it has a lot of problems, especially at present. Forest elephants are really getting battered because they happen to live in these places where there’s a lot of civil unrest and guns.
[Ivonne Kienast] Forest elephants are heavily poached for their tusks. And if we want to see them around in the future they need to be protected. Andrea spent 27 years at Dzanga Bai observing the elephants. She would observe them from a platform on the edge of the clearing, and identify individuals, families, social behavior, relationships, feeding behavior, and much more, setting the basis for the study of forest elephants at Dzanga Bai. The Elephant Listening Project arrived to Dzanga Bai with Katy Payne, a renowned Cornell researcher, who discovered that elephants communicate using infrasound.
So there was no better place in Africa to study forest elephants than Dzanga Bai. Thanks to Andrea’s hard work and her collaboration, nowadays ELP facilitates the most comprehensive study of a forest elephant population that exists in Africa. We count with long term data available on over 5,000 elephants from over 23 years, and most of ELP’s behavioral research focuses on the long-term monitoring of the elephant population at this one bai in Central African Republic.
In a long term partnership with WWF and funded by Sabine Plattner African Charities, through the Hasso Plattner Foundation, ELP establish the Dzanga Forest Elephant Project in 2021 after Andrea had left Central African Republic in 2017. We now manage the site and are in charge of the research that will allow to continue protecting the forest elephant population of the region. We build up on the database created by Andrea on the elephant demography at Dzanga Bai, and we are re-identifying the elephants to be able to follow the life history of individuals and of entire families.
Our mission is to contribute to the conservation of forest elephants by doing research on elephant communication and social system, by monitoring the demography, visitation rate, activity patterns, social networks, and health condition, by training local researchers building up capacity of specific skills, by the inclusion of Indigenous people in science, by community engagement to improve conservation, and by educational outreach in local schools.
So why is it important to monitor the elephants for so many years? What does long-term monitoring reveal? Looking at how often elephants that visit Dzanga Bai give birth showed us that forest elephants have a much slower reproductive rate compared to savanna elephants. And this has an immediate effect on their survival as a species, because a slower reproductive rate makes it more difficult for them to recover from the ongoing and increasing poaching pressure, unless immediate measures are taken.
To monitor the forest elephants, we collect many different sorts of data. One major focus is on behavioral research. We study elephant vocal communication using passive acoustic monitoring, and at the same time we collect observations of behavior to try to establish correlations between vocal behavior and context. In collaboration with ElephantVoices, an organization that studies and protects savanna elephants, we compare the data that we have collected on forest elephants with data that they have collected on savanna elephants. This comparison will allow us to better understand the correlation of social and vocal complexity in the genus Loxodonta. The call context is important information for the interpretation of passive acoustic monitoring data. And if we identify the correlations, we can monitor behavior using only acoustics, and interpret it– our acoustic data– on a much deeper level.
As I told you before, we also monitor the demography of the forest elephants at Dzanga Bai. Therefore, we seek to identify the individuals that we are seeing. And thanks to Andrea, we have a database full of names, and faces for those names. So we observe the elephants and using specific features like their ears, their trunks, and tusks, we can identify them comparing them with pictures that we have.
Not only do we learn how to do it with the naked eye, but we also use a software developed to do this automatically using pictures, the Elephant Identification System. So it basically works like this– you take a picture, you feed it into the software, and you draw a box around the head of the elephant. Then the system will compare it to each picture that it has been trained for using machine learning, and give you the results on how likely it is to be a certain individual. This makes identification much easier and accessible to others.
Apart from collecting data during the day, we also can collect data during the night by using thermal imaging. For the moment, we can do counts of how many individuals are in the clearing at a certain time and observe and record behavior. We are working on a system that, if successful, can help us identify the elephants also during the night. So once we get to that point, we can also target feeding this information into a machine learning system and have the process automatized in the future.
So we count the elephants coming into the bai, we identify them, we collect data on their behavior– and we also check their body condition by using a chart of categories to give an elephant a score depending on if the elephant is fatter or skinnier. We can use pictures for this, and here comes handy once again to have more than 20 years of pictures. We can follow the body condition of a specific individual during several years and monitor changes on an individual level, but most importantly, on a population level. This information is important because there can be changes in the environment that can cause, for example, a decrease in food, resulting in skinnier animals– information which can give us time to act in conservation.
So, when elephants are in the clearing they dig for resources in the soil and make these resources available for other species. There is so much going on in the clearing and in the forest apart from elephant activity, and we need to understand the ecosystem of the elephants to better contribute to their protection. We monitor, on a smaller scale, other mammals visiting the clearing, such as forest buffalos, and bongos. Though not all of these species are endangered, some populations have been declining according to the IUCN. And even species categorized are not threatened might be shifting someday to another category.
The sitatungas and colobus monkeys see their habitat destroyed by human activity and might become a special interest to poachers one day, when other species are gone. Additionally, for example, the giant hogs on the left side are rare to see in the rainforest, and the clearing gives an opportunity to observe them and count their numbers. We also recall the interactions between species as, for example, elephants and red river hogs– the latter sometimes being chased away or around by young elephants.
In addition, we are establishing a study to monitor the gray parrot population, as this bird species is also endangered due to poaching pressure. Ongoing is also the design of a study on gradients of the biodiversity, trying to assess the impact of anthropogenic activity, a study that intends to assess environmental health by using an insect species which could be a vocal species of birds or insects, and the monitoring of the fruiting patterns of elephant key food species, to understand better food availability and changes in fruiting or flowering patterns of the trees.
So maybe now you’re asking yourselves who I mean by “we”, and who all the hard work in the field does. So please meet our team, part of which has worked already with Andrea for many years. We are a team of six Indigenous trackers and research assistants, one local researcher, and me. We count with increasing our team next year by at least two more local members, and I insist on local. One of our main objectives is to train Central Africans to become leaders in conservation.
During the pandemic, we had to adapt to remote training. And as we come with people from different backgrounds and ethnic groups with a variety of levels of education, we identify skills and target specialized capacity building. We work with Indigenous people from the Ba’aka tribe, we work with research assistants, with students, and we do collaborate with other field sites in the trinational area. Last, but not least, we also focus on the community that has lived there since before the arrival of researchers and conservationists. We ask ourselves, how can we approach science to the local community? How can we help with human-wildlife conflicts? How can we enthusiast kids about conservation?
The future of the elephants is in the hands of everyone, and the local communities can lead this endeavor by becoming the voice of the forest elephants in the Central African Republic. Thank you very much for your attention.
[Chelsea Benson] Thanks, Ivonne. That was wonderful. Next up, we have Colin Swider. Colin is a PhD candidate at Syracuse University, and has been researching African forest elephants in collaboration with the Elephant Listening Project since 2017, both in the lab and in the field in Central Africa. And Colin’s going to share the work that goes into monitoring elephants at landscape level, and the different sounds they detect, elephants and more. All right, thanks Colin.
[Colin Swider] So thank you, Chelsea, and hello, everyone. Thanks for tuning in. As Chelsea mentioned, I’m going to switch gears here and talk a little bit about how we use acoustics to study forest elephants at the landscape scale. And by landscape scale, what I mean is that we’re basically going to zoom out from talking about our studies at specific individual sites– like the forest clearings on the bai that Ivonne just described– and now we’re going to consider a much larger region, or spatial extent, as well as a more continuous time frame.
So outside of those forest clearings like Dzanga Bai, forest elephants are really hard to study visually for a few reasons. They are very elusive and they’re hard to find. Their habitat, the rainforests of the Congo basin which is pictured here, for the most part, is very hard to access for us humans. And if you were to fly over this and try to make observations from above, you would not be able to see through the canopy in order to observe any animals. So what we do to get around this is we use a suite of techniques known as passive acoustic monitoring, where we set up grids of autonomous acoustic sensors that cover relatively large tracts of rainforest, and that can potentially record for a years at a time.
So here’s a little footage of how we set up the acoustic sensors and install them in the treetops. These sensors that you can see here are called the Swift recorders, and they’re made in-house by our engineers at the K. Lisa Yang Center for Conservation Bioacoustics, which is where we are based at Cornell Lab of Ornithology in Ithaca, New York. And one of the main advantages of using these sensors and these methods in general is that we can install them in the forest, and then we can leave and they continuously record audio data without anyone having to be present. So essentially, it’s a form of remote sensing.
However, because we have these sensors recording nonstop for years at a time, they do have to be visited intermittently in order to change the batteries and retrieve the audio data from the memory cards. So you can imagine that if you have many sensors spread out across a huge tract of not-so-easily traversed rainforest, this can be a tremendous challenge to keep that grid up and running. And the grid that is currently our main focus, and our main source of data is shown here on the map on the right.
This is our study area in NouabalÃ©-Ndoki National Park, Republic of Congo. It’s right near the border with the Central African Republic and Cameroon. And each of these black dots that you see here is an acoustic sensor that has been deployed in the canopy of the rainforest. This grid contains 50 of those Swift recorders that you saw in the video, and they’re spread out across about 1,250 square kilometers. That covers both the National Park itself as well as these adjacent logging concessions to the South, and they’ve been recording data continuously for four years. That fact, along with the sheer expanse of this terrain that it covers, made this the first passive acoustic grid of its kind in Africa.
So no one had really implemented data collection at this scale prior to this in Africa, and it posed some tremendous challenges for us. Most of which we have been able to resolve and overcome by working in collaboration with our partners over in Congo. So we’ve been working very closely with the Wildlife Conservation Society from the beginning. WCS plays a huge role in our logistics for this project and they maintain a permanent presence in the National Park.
We also have an amazing field team of three Congolese researchers, Phael, Frelcia, and Onesi, who have been able to more or less keep this grid up and running consistently for the past four years. And so every three to four months they do a massive circuit on foot through this study area, visiting each and every one of these 50 sensors. And they swap out the batteries, and they collect the acoustic data from the memory cards so that the sensors can continue to function, and so that we get a new round of data every four months. So these guys are just phenomenal. They put in an astronomical amount of work into this project– this is definitely not easy terrain to traverse for weeks at a time on foot, and it’s really great to have these guys on our team.
In the next few slides, I’ll talk a little bit more about the types of sounds that we’re listening for, which are mostly the elephant rumbles and gunshots from poaching. Because one, this tells us where the elephants are spending time across the landscape, what types of habitats are important, and things like that. And two, it allows us to identify potential hotspots for poaching. However in addition to those sounds, there’s also a wealth of other information contained in the recordings. So the soundscape that we record in this rainforest contains signals from a huge diversity of species, including birds, frogs, insects, primates, et cetera. So the opportunities to study different facets of biodiversity are enormous.
And in terms of human sounds, in addition to the gunshots we can also pick up noises from logging activity. And this presents the opportunity to study the impacts of resource extraction on different animal populations. But again, our main target sound that we’re trying to find in the recordings are those vocalizations from forest elephants, particularly those rumble calls that Daniela introduced earlier, which you can see an example of here in the spectrogram on the left. So because we have, for the first time, this grid of sensors set up across such a large expanse and recording for so long, this is the first time that we’ve been able to track elephant populations across such a large study area like this.
So as I play this animation of the map, you’re going to see the shifting distribution of the elephants that we detected on a month-by-month basis. We have the white and the pink colors indicating more detected rumbles, and the yellow and green indicating fewer rumbles. And so, as you can see, this sort of allows us to visualize how the distribution of the elephant shifts around over time, and it also allows us to make statistical models that include different habitat variables. So for example, the density of rivers or the specific type of forest at the different sites. And then we see how the detections of the elephants might correlate with those variables. And some of our preliminary modeling results suggest that elephants prefer certain types of forest, like the mixed forest with a high diversity of tree species, over other forest types, such as flooded forests and swamps.
Because the grid extends from the National Park into both active and inactive logging concessions, we’re also able to gain some insight into how logging activity affects forest elephants. So for example, they seem to prefer the old inactive concessions. Perhaps they’re attracted to the secondary regrowth of vegetation that happens following logging in the old concessions and that’s why we find them there, because they use it as a food source. Conversely, we found that they avoid the active concessions, which makes sense that they would be deterred by the human activity that is associated with the ongoing logging. And one of our major ongoing objectives at ELP is to keep adding our incoming data into these models, and then consequently to make more and more refined models to kind of see which of these landscape variables are the most influential, in terms of forest elephants and their distribution.
So as I mentioned before, the other major target they will listen out for are gunshots from poaching. And this is a spectrogram showing a series of gunshots that I’ll play for you now. You probably have to listen carefully because the gun was quite far from the microphone, so it’s not particularly loud.
[Colin Swider] It’s an AK-47.
[Colin Swider] So unfortunately, this is very likely an instance of an elephant being poached. And just like with our detections of elephant vocalizations, we can use detections of these gunshots to map out where the poaching is taking place across our study grid. And then we end up with maps like this, which show the relative density of gunshots that we detected over an approximately two-year period with the brighter orange sites indicating more detected gunshots.
So the aim here is to understand when and where poaching is most likely to take place, because if we can predict when where it’s likely to occur, we can improve the chances of preventing elephants from being killed– or at least we can improve the chances of apprehending the people that are responsible for it. And we do that by working in conjunction with our partners at WCS, who manage the eco-guard anti-poaching patrols.
So these guys are just incredibly brave humans. They risk their lives every single day for conservation and for forest elephants. And the work that we are doing in collaboration with WCS and the anti-poaching effort is a great example of applied science, or data collection for conservation action. And bringing the acoustics into it, we believe that the acoustic method of collecting poaching data is a major improvement over how it has been done up until now.
So prior to this, the detection of poaching has mostly been left up to chance, with the eco-guard patrol routes determine somewhat randomly, or perhaps sometimes based on information, such as tips from informants on poachers whereabouts. And that’s kind of an inconsistent and subjective type of information, but now that we have this established acoustic grid, we systematically collect gunshot data in an objective, unbiased, and much more reliable fashion. And this will allow us to understand patterns of poaching pressure in a scientific framework. And perhaps more importantly, the park management receives our gunshot data every four months, so the eco-guards then have the opportunity to adjust their patrol routes accordingly. So our results are directly translated into conservation outcomes.
Another analysis that we can do is look at how elephants respond to particular instances of gunfire. So obviously, at least one elephant is typically targeted in a poaching event, But there are likely a bunch of other individual elephants in the vicinity that hear or witness the gunfire, but that aren’t necessarily the direct targets. So this map shows one of our poaching event detections, like the gunshots that I played for you, which is indicated by this red explosion symbol. And then the distribution of elephant rumbles that we detected at sites all around the poaching site are represented by the purple color.
So we can see how the elephant population responds to these poaching events by interpreting this figure, which shows the number of elephant detections per sensor. So we compare our elephant rumble detection rate during the 24 hours before a poaching event, which is here in the middle, to the 24 hours after the poaching event, which is here on the right. Then we compare those two rates to control periods where there is no poaching event at all. So this control period without any gunshots whatsoever gives us a baseline rumble detection rate with which to compare the rates from the before and after periods.
And when we made those comparisons, we noticed the really interesting pattern. There’s a significant decrease in the call rate during the 24 hours before a poaching event, compared to these baseline rates that we see in the control period. And this decrease is statistically significant. So this is really intriguing, because we did expect there to be a change in call rate following a poaching event, but the change actually takes place before the poaching incident even occurs. That suggests that forest elephants are responding to the presence of poachers in the forest before shots are even fired.
So as the poachers are moving through the forest and preparation to shoot an animal, it seems like the elephants might actually be reducing their vocal activity– perhaps becoming more vigilant, potentially as a mechanism to avoid being detected by the poachers. Or, alternatively, they could be totally abandoning the area where the poachers are and then relocating away from the danger, and that could be why we have fewer rumble detections during these times surrounding the gunfire events.
So overall, I hope that gave you some indication of how we use passive acoustics to study elephants– their distribution, their behavior, and poaching pressure at the landscape scale. Thanks for tuning in and expressing interest in what we’re doing. And I’ll turn it over now to my colleague, Bobbi.
[Chelsea Benson] Thanks so much, Colin. It’s definitely really heavy information that you just shared, but it is encouraging that the work that you’re doing is really having direct impact on elephant populations, and the work that people on the ground are doing. So thanks for sharing that.
[Colin Swider] Thanks for tuning in.
[Chelsea Benson] All right. Bobbi is our final presenter and then we’re going to move to our Q&A after Bobbi’s presentation. So this is Bobbi Estabrook, everyone. Bobbi’s a research analyst with the Elephant Listening Project. Bobbi takes hundreds of thousands, or perhaps millions of hours of recordings– I’m not sure how much you have there– but, the Listening Project– and uses a special software to analyze the data and detect all those things that Colin was just talking about. So I’m really looking forward to seeing Bobbi’s work here in action. So, thanks Bobbi.
[Bobbi Estabrook] Thank you, Chelsea. Yes, so I’m going to talk about– just give you an overview of the work that we do, and how we actually go through all of the data that we collect. And you’re right, Chelsea, it’s a lot of data. And I’ll also walk through a little bit of how we do the sound analysis for this work. And I’ll be using the NouabalÃ©-Ndoki data set as an example here, which is what Colin had just discussed with you.
This data set– we started collecting it in 2017 and it’s ongoing. It’s continuous and simultaneous data that’s collected across 50 sites. So far, this amounts to about 1.3 million sound hours, which is about 148 years, so it’s a lot of data. And it’s a lot to go through, so we need specialized software to be able to do this.
So far, the project has collected about 66 terabytes of storage. We store the sound files on servers at Cornell, but we’ve also been storing our sound data up on the Amazon Web Services, which is actually open to the public. So if you were interested in looking at our sound files, you could access that. There’s more information on ELP’s website if you’re interested.
So how do we convert all of these sound files into usable data for elephant conservation? Approximately every four months or so, the field team trek through the really dense forest and they retrieve and deploy the Swift recording units. So this forest habitat is extremely challenging to move through, as you can see from the photo on the left, and it can take the team about a month to make their way to all 50 sites for this study. After they collect the memory cards from the recording devices, they load them onto hard drives, and then they send those hard drives to us in Ithaca.
From there, we upload the sounds onto our servers and onto the cloud storage, and then we check the sound files for any problems. We then run them through automated detector algorithms, which help make our work possible. Automated detector algorithms are detectors that have been programmed and designed specifically to automatically detect certain sounds or a set of sounds in the audio files. The two detectors that we primarily use for this project are the rumble detector and the gunshot detector. But as we’ve already mentioned, there are plenty of other animals that produce sounds, or other sounds of the force that are of interest. So we can run other detectors to detect things like bird signals, primates, or manmade activities like chainsaws.
And this is an example of one of our detectors. It’s the interface for our gunshot detector. What we do is we direct it to where our sound files are, and then we specify certain parameters in the detector, and then we click “run”, and we let it go. And after some time, it saves a text file like this that you see on the screen here. And this text file contains all of the detections for the detector output files, and so every row in this table is a separate detector event.
Now, our detectors are not perfect. It’s really difficult to have a perfect detector. So it’s really important for the work that we do that we review all of the detections and mark them as either a true detection or false detection. We don’t want to accidentally claim more elephant presence than what we actually found. So what I can show you here in this figure is that here’s an example where our rumble detector correctly detected an elephant rumble, but it also found a vehicle noise that has similar properties to an elephant rumble. And similarly, our gunshot detector here correctly detected several gunshot events, but it also detected tree branches breaking, and those are false detections in this case. So by reviewing all of the detections, we can evaluate how well our detector is doing, and we can ensure that we are only including the sounds that we want in our data set. For example, we only include elephant rumbles and we exclude all vehicle sounds.
So now I’m going to just give you a really brief demonstration of how we review our detector events using Raven, which is an in-house sound analysis software program that was designed by our team. I’m going to load in a selection table, the same that you just saw, into Raven. And I’m just going to clean up this spectrogram really quick and make it look a little bit nicer. OK. So now what you see is a spectrogram you’ve seen earlier in the presentation, where we have time along the x-axis and frequency along the y-axis, and our selection table is right here. And if I click on the first event, Raven takes us to that detection event So we can see what the detection is. And we have some begin time, end time, low frequency, and high frequency information about that event, which creates our event boundary box.
We can also look at which site this detection came from. And we can see which date it occurred on, as well as a detector score. So in Raven, we can move through all of the detector output and start annotating sounds and find elephant rumbles, like here. That’s a nice rumble right there. But you can see we can have quite a few detections– lots of detections– especially in this big data set. So we use this program, or a tool within Raven, which is called selection review, and it helps us to review the large volume of detector events much more efficiently than looking at it one by one as I just showed you.
What this will do is convert each selection event or each detector event into a small spectrogram. Here, I’m loading up 48 spectrograms, which are 48 detector events that we could just look at really quickly. And we get this small view of each of them. This is a selection table that I’ve already annotated before, so I’ve already marked the events as elephant or as something else. And let me just collapse the table real quick and make this look a little bit nicer.
So if I click through each of these events, we can then look at it in the spectrogram above and see what’s around it. We’ve got some really great rumbles here, but it looks like our rumble detector– yep, it found some really good ones. But it found vehicle noise as well, which can be very tricky for the detector. It’s got these horizontal lines from vehicle noise, and let’s see– we’ve got an insect here as well. So we can quickly look at the events and mark them as elephant or something else. And now I can just jump ahead to the next 48 events and review them and mark them as what they are.
And in this way, we get to exclude things that are not rumbles, and we also get a better understanding of what our detector is finding. Like this buffalo here, which looks an awful lot like an elephant rumble, if you notice over here, but it’s not a true detection. And again, we get vehicle noise sometimes in this. But what we’ve just looked at now is 100 events pretty quickly. And so the Raven software and these features in it help us to go through a lot of data, along with the detectors, a lot of detector output pretty efficiently.
So then those reviewed selection tables will eventually become data in this format that you see here, where we can generate heat maps of detections across the survey area over time. We can learn about the ecology and the distribution of forest elephants, and we can observe gun hunting trends. Also with data like this, we can provide park managers with illegal gun hunting activity every four months. And this, so far, has allowed them to evaluate the effectiveness of their patrol efforts on gun hunting. And you can see in this figure here, there’s an inverse relationship in the number of patrols and gun events. And then, we also can share our results with the broader community through scientific papers, reports, and presentations. But ultimately, all of these data and all of our collective efforts aim to contribute to the conservation of African forest elephants. Thank you for your attention.
[Chelsea Benson] Thanks so much, Bobbi. I would love it if the rest of our panel could rejoin us and we’re going to spend the remainder of our time together taking your questions. I’m really encouraged to see so many great questions coming through. And I thought we should start with some of the basics about the elephant calls. So maybe Daniela, with the first spectrogram that you showed, people are wondering how far away can the recording devices detect those calls that are being made?
[Daniela Hedwig] This particular recording was made at Dzanga Bai, that forest clearing that Ivonne talked about at approximately 20 meters difference. So we have this observation platform and that particular interaction took place right in front of the observation platform. As for the detector– as for the passive acoustic monitoring devices, we have estimated that, on average, a rumble can be detected at 500 meters. But it can go all the way up to 2 kilometers, and depending on the amplitude at which the elephants are producing the rumble, that can also be much, much less.
[Chelsea Benson] And so as you’re listening to all these calls, are you detecting anything interesting? Put one way, somebody who has like a “words” or surprising patterns that emerge from these recordings that you’re all listening to, and the data that Bobbi is analyzing?
[Daniela Hedwig] Well one thing that we are very passionate about working on is what we call the Elephant Dictionary. So we’re trying to better understand what’s the information or content– what are the elephants are trying to say with their rumbles? And we’re hoping that this will allow us to use such a dictionary to better interpret the sounds that we record from the elephants deep in the forest, where we can’t observe them directly.
So, we know from savanna elephants that the rumbles show specific acoustic structures that are particular to what the elephants are doing. There are rumbles that indicate that they’re very agitated and stressed, or that indicate what the elephants are doing. For instance, the very, very short rumbles are usually rumbles that are produced by calves that are trying to nurse. And the very long, flat-lined rumbles, they are used by elephants in contexts where they have to coordinate their foot movements. And we are seeing very similar patterns in forest elephants too, now.
[Chelsea Benson] Yeah, that was one of the other questions people asked is how different is the African forest elephants’ vocalization from the savanna elephants’ vocalization?
[Daniela Hedwig] That’s actually an active line of research that we are pursuing right now. We’re working together with ElephantVoices’ Joyce Poole, who’s got this huge database of savanna elephant rumbles with a very detailed annotations of context. And we are combining our data from Dzanga Bai now with hers, to do this really huge comparative analysis. We have already shown that these patterns that are the very short rumbles are nursing rumbles, and the very long rumbles are those group movement rumbles– that’s very similar in the two species. But overall, because forest elephants are smaller than savanna elephants, their rumbles are higher-pitched in general compared to the savanna elephants.
[Chelsea Benson] That’s great. So Ivonne, this question’s for you. As we were looking at the bai and all the activity that was happening there, one of our audience members asks if it’s possible to– maybe it’s already there, we just don’t know it– is there like a real-time video? Can people tune in to watch that activity?
[Ivonne Kienast] Well, that would be wonderful. I think the problem we have in all these remote sites is connectivity. So this is something that all research sites who are remote like us try to solve, and I really hope that this could be something in the future. But right now it is very difficult, because you don’t get the right satellites or the understory of the rainforest is very close, so in general you don’t get any type of connectivity in a place like that. But for sure, we have been working on that for many years and we will continue to try to find a solution for it. And hopefully come up with that, too.
[Chelsea Benson] This series of questions is about the field work that goes into it, and with your local research team. And people are curious– how dangerous is the field work for you and for the team that lives there? And not just confrontations that might happen, but we’re still in the midst of a pandemic. So how is all of that impact the work that goes on with the Elephant Listening Project? Not sure who that’s a good question for.
[Colin Swider] I can touch on that a little bit. It is pretty hairy sometimes, the field work, like living in the jungle. We basically live at a very primitive research camp for months at a time. No electricity, we get our water from a river, from a stream– it’s super off-the-grid. And so we’re right in the middle of the jungle, and there’s gorillas around, there’s elephants around, and you know, I think the elephants are probably the biggest danger.
When we’re hiking through the forest from one point to another, we’re always accompanied by the Indigenous tribe that we work with, the Ba’aka, and they just know everything there is to know about the forest. They’re just amazing. And so they keep us safe for the most part. I mean, we do have encounters with wild animals, especially elephants, and we get chased and we have to run really fast. And it can get pretty close sometimes, but we’re all still here and alive and well. It’s exciting to say the least.
[Chelsea Benson] Colin, this one might be for you, too. How often do people– or poachers– locate these listening devices and perhaps try to damage them, knowing that you’re collecting pretty valuable research that could impact their illegal efforts?
[Colin Swider] Yeah, that’s one of the reasons why we try to put them up pretty high in the canopy, and so that they’re harder to find. But occasionally, they do find them. Actually, Daniela and I were working at Dzanga Bai a few years ago, and we had recorders all around the bai. And one of them at the end of our three month field season had been stolen. So it does happen. And we just try to hide them as best we can. I don’t think necessarily people know what they are when they find them, but poachers obviously don’t want– it looks like it could be something that could get them in trouble. So they take them and I don’t know what they do with them, but–
[Chelsea Benson] So thinking about that four month lag that happens between the recording and taking it back and analyzing the data and reporting it out– our audience is curious, is there any way to– as you just said, these recorders are hundreds of miles apart. They’re very difficult to get to. Is there any hope for some type of real-time detection of these sounds that you’re picking up?
[Daniela Hedwig] Well I can answer that. That’s a very good question. It’s kind of the Holy Grail that we really want to achieve. I mean, getting the poaching information to the park managers at four months intervals is already a huge improvement, but we need to be faster as well. And it still takes a lot of effort to get the data out of the forest. In northern Congo, we have two teams out, each for a month, to collect all the data. It’s very useful, but it’s still a lot of work, and real-time monitoring is what everybody wants to achieve at the moment. There are many groups working on it, but it’s difficult in the Central African rainforest to implement.
For instance, you need to power a real-time unit and it’s very difficult to do that in a sustainable way with solar power, for instance, under a closed canopy. And then we don’t have cell phone coverage there, so how do you get the data out? So you rely on often-sketchy satellite connections. So it’s definitely a challenge, but I’m really convinced that over the next five years there will be a major breakthrough, and that will be a game changer.
[Chelsea Benson] Bobbi, I have a question for you, and then I want to end with a question for all of you. So Bobbi, they’re curious about what’s the error rate for your detection when you’re using that software?
[Bobbi Estabrook] Oh, that’s a good question. It can vary by the area. So sometimes, if the recording site is really close to a road, for example, the false detections are going to be a lot higher at that site than it would for sites that are a lot further away from the road. So it’s kind of variable, depending on what signals are also happening around the unit that could trick it. Right now, I would say I think it’s about 0.65 of a true positive rate detection, I think in some cases. But again, it’s better in other cases and worse in others.
So that’s part of what we try to do is when we use these detectors in specific projects, we have to evaluate the performance pretty regularly and for each project separately so that we can really get a better understanding of what information we are getting from it. We also care a lot about missed detections– and I didn’t show that here but it’s very important. So that’s a good question. Thanks.
[Chelsea Benson] Our final question before we end our presentation today is our audience wants to know, after you shared all this great information, what can they do to help support the Elephant Listening Project and the work of your team, and also just protecting African forest elephants in general?
[Daniela Hedwig] Thank you so much for that great question. I think the most important thing that people can do is not to buy ivory when they see it. I know it’s very pretty, but it’s killing elephants that are individual personalities. So please don’t buy ivory, and tell your friends about how amazing elephants are and how important they are to preserve biodiversity in Central Africa.
And the other thing that’s very important is of course, there is a lot of deforestation going on in Africa. And that’s threatening elephants as well besides the poaching. So if you want to buy wood, stay away from tropical hardwood. Those are the two major things that people can really do that have an effect on the ground.
[Chelsea Benson] And I see that Leo and Catherine are behind the scenes and they’re adding links to the Elephant Listening Project website into the comments in Facebook and the chat here in Zoom, so that you can learn more about the Elephant Listening Project through the website. So Daniela, Ivonne, Colin, Bobbi, Liz– I just want to thank you so much for your time today and sharing the incredible work of the Elephant Listening Project. So really, hats off, you’re doing incredible work. So thank you for sharing it with us.
[Colin Swider] Thank you for having us.
[Daniela Hedwig] Thank you. Thanks, everyone.
[Chelsea Benson] I’ll be emailing our Zoom attendees tomorrow with this webinar and the resources we discussed. If you’re watching on Facebook, you can check the comments for those links. This webinar is part of a series. We spotlight programs and research from around the Cornell Lab. If you enjoyed today’s webinar, it’s funded primarily by people like you who become a member. So if you enjoyed it, please visit birds.cornell.edu and consider becoming a member of the Lab of Ornithology. So thanks again. Thanks to our panelists and to our audience, and everyone have a great afternoon. Bye.
[Daniela Hedwig] Thank you, bye everyone.
[Ivonne Kienast] Thanks. Bye.
[INTERPOSING VOICES]End of transcript
For more than two decades the Cornell Lab’s Elephant Listening Project has helped understand and conserve the critically endangered African forest elephant, a little-known species that’s been hit hard by forest loss and poaching for the ivory trade. Join our experts and learn how studying the rumbles and calls of this elusive giant is helping to keep them safe in the rainforests of Central Africa.