- Over the course of three weeks in July, six teams showcased technology that could potentially automate biodiversity monitoring in rainforests.
- The event in Manaus, Brazil, was the final round of testing for a $10 million competition organized by the nonprofit XPRIZE Foundation.
- Most of the teams deployed drones and AI models, and also used environmental DNA analysis to identify plants and animals in the forest.
- In addition to gauging how the teams presented scalable and affordable solutions, the competition also judged them on how they collaborated with Indigenous communities in Brazil.
If long-term field research is like a marathon, then this was the 100-meter sprint. Over the course of 72 hours, participants scrambled to gather biodiversity data, analyze it, and generate insights about a patch of the Amazon Rainforest in Brazil.
Over the course of three weeks in July, six teams comprising biologists, robotic engineers and AI experts descended on a remote forest near Manaus in the Brazilian state of Amazonas to showcase cutting-edge technology that could potentially automate how scientists and conservationists survey the world’s rainforests. At the end of the 72-hour period, each team handed over a report on the biodiversity that they’d managed to identify in the forest.
“It was like the Olympics for biodiversity,” said Kristy Deiner, the lead for one of the teams.
The event was the final round of testing in a five-year competition with a $10 million prize, organized by California-based nonprofit XPRIZE Foundation. Twelve teams had competed in the semifinals testing round in Singapore last year, but only six made it to the finals in Manaus. In the period in between, the finalists had the opportunity to improvise their technology while also working with local communities to incorporate their knowledge and expertise into their solutions.
At the finals, the teams had 24 hours to gather data in the field. Following this, they were given 48 hours to process and analyze the data. The criteria for judgment included the number of species identified and the teams’ ability to demonstrate efficient deployment of technology that’s replicable, scalable and affordable. They were also judged on how they collaborated with local communities in accordance with Brazilian laws.
Drones, environmental DNA analysis and artificial intelligence models featured heavily in the solutions put forth by almost all the teams. While some teams collaborated with local scientists to build databases for AI models, others organized training workshops on how local communities could use the technology to better manage and protect their lands. The winner of the competition will be announced later this year.
Mongabay’s Abhishyant Kidangoor spoke with all the team leads to understand more about their biodiversity monitoring innovations and how they collaborated with Indigenous peoples and local communities. They also spoke about their biggest learning experiences over the past five years and how they see the future of conservation technology.
Limelight Rainforest
Method: The team used drones to drop 10 rafts onto the rainforest canopy. The rafts, called Limelights, contained cameras, microphones and light traps and bait to attract and collect insects for DNA sampling. The team then used various machine-learning models to identify the bird calls and insect images gathered by the equipment. Fans with filters also collected air samples for environmental DNA (eDNA) analysis. Additionally, the team used a robotic device deployed by a drone to swab leaves and collect water samples for DNA testing. They also mapped the canopies and tree crowns using drones and used machine-learning algorithms to measure tree diversity, carbon storage and forest structure.
Collaborations with local communities: Over the course of the past year, the team supported the formation of a company owned by Indigenous nature guides from Ecuador. The company worked with the team to annotate more than 60,000 bird calls, with the data later used to build a model that identified more than 265 Amazonian birds.
Biggest learning: “I have gained a lot of faith in the human spirit and realized that humans can make some amazing things when they’re given the right conditions,” said Thomas Walla, team lead and professor of biology at Colorado Mesa University.
Where conservation technology is headed: Walla said there’s immense potential in technology since a lot of young people are getting into the field of machine learning. However, he said, it’s imperative to spark their interest and get them to use their skills in conservation and nature monitoring. “It’s hard to get them to focus on problems like tropical trees which don’t have a lot of glory in most ways of application, and not a lot of financial gain either,” he said.
Last word: “It’s been a wild ride,” Walla said.
Welcome to the Jungle
Method: Drones deployed 11 lightweight sensor packages fitted on a partly biodegradable platform. The package was equipped with audio recorders, DNA traps and sensors to measure temperature, altitude, humidity and geolocation. The platform also had air filters to collect eDNA samples and contained cameras that could be lowered into the canopy to capture images. Machine-learning models were later deployed to analyze the images and sounds and identify species in the forest. A second, smaller part of the team flew drones equipped with multispectral and lidar cameras to map the forest cover.
Collaborations with local communities: Members of the Baniwa tribe in Brazil were part of the team and used their knowledge to corroborate the findings, especially in confirming tree species identified using the technology. The team also worked with experts in Asia and Africa on ways to make their models applicable to rainforests across the world.
Biggest learning: “It is just how much fun engineering and field work can be,” said Matthew Spenko, team lead and professor of mechanical and aerospace engineering at the Illinois Institute of Technology. “Getting out there and designing something for extreme conditions, and then testing them in those environments can be anxiety-ridden, but it’s also a lot of fun.”
Where is conservation technology headed: According to Spenko, there’s nothing “too new or crazy” that will magically solve the hurdles in environmental monitoring and protection. “We have cameras, we have drones, we have audio sensors, we have lidar,” he said. “But it’s not trivial to put these together and get them launched in a remote location in a harsh environment. I think that’s what’s really important here — to combine those technologies into one system or platform.”
Last word: “I think this is a major kick-start for conservation technology,” Spenko said.
Brazilian Team
Method: The team first split the test plot into 24 clusters, then deployed drones to map the area and identify trees with the help of Pl@ntNet, an AI-based tree identification platform. The findings were corroborated by local scientists and experts. They also deployed drones fitted with thermal cameras at night to detect wildlife. Additionally, the team deployed seven platforms on tree canopies fitted with various sensors as well as acoustic recorders and light traps for insects. To study water bodies in the plot, they deployed drones with pumps as well as plankton traps to collect samples.
Collaborations with local communities: Being based in Brazil, members of the team have long worked with local communities, training them to scientifically name plants and animals in a bid to link local expertise with academic knowledge. A few of these trained botanists also joined the team for the finals to help build a biodiversity database and survey the forest.
Biggest learning: “The best learning was the integration of different areas, such as robotics and zoologists working together,” said Vinicius Castro Souza, team lead and associate professor of biology at the University of São Paulo.
Where conservation technology is headed: Juliana de Paula-Souza, professor of botany at the Federal University of Santa Catarina, said that while the technology continues to evolve, the need of the hour is something else. “The technology is now getting similar and converges to the same thing,” she said. “If we want an autonomous solution for surveying biodiversity, we still have to work on libraries so we can leave it all to the machines to do the work.”
Last word: “The technologists are the inventors and we have problems,” de Paula-Souza said. “Hopefully they will fill the gap.”
ETH BiodivX
Method: Canopy rafts equipped with light traps, microphones and sticky tapes were deployed on tree tops using drones. Separately, another drone collected eDNA samples from inside trees, water bodies and the ground. The team used a new type of machine that used aluminum instead of plastic to analyze the eDNA samples, from which they even found traces of malaria, likely from monkey feces. Machine-learning models aided the team in classifying sound recordings and identifying nearly 100 bird species. The team also used AI models to automatically identify insects caught in the traps and trees that were mapped using a drone fitted with an RGB camera.
Collaborations with local communities: David Dao, one of the three team leads and co-founder of the nonprofit GainForest, said feedback from local communities was critical for improving their insect traps as well as their algorithms and models. Indigenous scientists were also part of the team. The team also set up a mechanism where local communities were paid for helping train the algorithm.
Biggest learning: “I don’t think I ever would have imagined that we could bring the time frame down so much, skip so many steps and still get quality data,” said Kristy Deiner, co-lead and assistant professor of environmental DNA at ETH Zurich. “That will change the future of the frequency with which we can gather and analyze data.”
Where conservation technology is headed: “We could see that progress is faster when you integrate knowledge and bring people from different backgrounds, expertise and cultures together,” said Stefani Mintchev, a co-lead and professor of environmental robotics at ETH BiodivX. The team leaders also emphasized how all the teams in the competition were testament to the fact that, if need be, technology can be made low-cost, scalable and equitable.
Last word: “Our biggest competitor is not any of the other teams, but the climate and biodiversity emergency,” Dao said. “That’s a very tough one to compete with and we need to do all we can.”
Providence Plus
Method: The team deployed what it called a DROP — Deep-Rainforest Operation Platform — an umbrella-shaped solar-powered solution that enabled the autonomous collection and processing of sound, images and eDNA on canopies and water bodies. A motion sensor on DROP enabled photos to be captured automatically whenever movement was detected. Audio data and images were processed, analyzed and transmitted in real time using built-in AI models. The team also collected soil and water samples for eDNA analysis by deploying custom-made solutions carried by a drone.
Collaborations with local communities: The team members have worked with local communities and conservation organizations for over a decade to train them on the use of different tech tools. The team has said that its priority objective is to provide local communities with the processed data they gathered to help them manage their resources and protect their lands.
Biggest learning: “It has been to recognize the critical role of incorporating technology with the traditional knowledge of Indigenous people and local communities,” said Soraya Hidalgo, team lead and head of strategic partnerships and institutional fundraising at the Polytechnic University of Catalonia. Hidalgo emphasized the importance of ensuring that the technology and data derived from it be shared with the communities so they can work to protect their lands.
Where conservation technology is headed: Michel Andre, team lead and director of the laboratory of applied bioacoustics at the Polytechnic University of Catalonia, said the competition demonstrated that there’s a lot of work being done to address biodiversity loss. “If any of the teams prove to have a tech-ready solution for rainforest conservation, it should be immediately and globally implemented,” he said.
Last word: “We don’t have time,” Andre said. “Time is really tight to address the huge challenge of assessing the health of the Amazon and other rainforests.”
Map of Life Rapid Assessments
Method: Low-cost and, mostly, autonomously operated drones collected images, audio and eDNA data. The team’s MVP, however, was a suite of software tools for survey execution and species assessment. The tools leveraged the larger Map of Life platform, built by a team at Yale and a network of partners, and its AI-supported biodiversity database. The tools execute autonomous drone survey missions, predict species; list, identify and classify the collected data; and map the tree canopies in 3D. Remote experts corroborated the identification made by the tools. An online dashboard also displays historical and global data on the identified species, including protection status, habitats, evolutionary distinctiveness, and the site’s significance on a regional, national and global scale.
Collaborations with local communities: The team has developed tools and resources to support local communities, including a photographic field guide available on the dashboard and a website that documents the biodiversity of the area. The team also developed a module that allows local experts to contribute their knowledge about large vertebrate populations as well as provide insights about the identified species.
Biggest learning: “It offered something complementary to the usual academic grind,” said Walter Jetz, team lead and professor of ecology at Yale University. “We need to think of ways in which we can really put the technology into the hands of people that are not directly tethered to a university setup or a commercial entity somewhere in the Northern Hemisphere.”
Where conservation technology is headed: Jetz said there’s great urgency to come together and democratize access to the rapidly growing technology. “We are scaling up, but in a way that startups are doing their own thing and keeping to their closed ecosystems,” he said. “We need to engage with one another in order to capture as much biodiversity as robustly as possible.”
Last word: “Let’s build on the competition in a collaborative way, and not so much in a competitive way to deliver to the larger community in the most synergistic manner,” Jetz said.
Banner Image: The site of the final round of testing of the XPRIZE Foundation’s rainforest competition. Image courtesy of ETH BiodivX.
Abhishyant Kidangoor is a staff writer at Mongabay. Find him on 𝕏 @AbhishyantPK.
Credits
