An increasing number of researchers is turning to artificial intelligence (AI) to monitor biodiversity and bolster efforts to help endangered species. Unlike conventional methods that can disrupt ecosystems or require considerable time, labour and resources, AI has the potential to quickly and effectively analyse vast quantities of real-world data.
“Without AI, we’re never going to achieve the UN’s targets for protecting endangered species,” says Carl Chalmers, who studies machine learning at Conservation AI, a UK-based non-profit organization in Liverpool that uses AI technology for various ecology projects.
Species are vanishing at a rate hundreds to thousands of times faster than that millions of years ago, with up to one million species on the brink of extinction. In response, the United Nations set a goal in 2020 to safeguard at least 30% of Earth’s land and oceans by the end of the decade.
AI is “imperfect” but could accelerate important discoveries, says Nicolas Miailhe, Paris-based founder of The Future Society, an international non-profit organization that aims to better govern AI. “We very much need human practitioners in the loop to design models, as well as collect, label, quality check and interpret data,” he says.
Ecologist Jörg Müller at the University of Würzburg, Germany, and his colleagues have shown that AI tools can help to quantify biodiversity in tropical forests by identifying animal species from audio recordings.
In a study published on 17 October in Nature Communications, the researchers used AI to analyse animal ‘soundscapes’ in the Chocó, a region in Ecuador known for its rich species diversity. They placed recorders in 43 plots of land representing different stages of recovery: forests that were untouched by deforestation, areas that had been cleared but then abandoned and had started to regrow, and deforested land actively used for cacao plantations and pasture. They gave the audio files to experts, who were able to identify 183 bird, 41 amphibian and 3 mammalian species.
The researchers also fed their recordings to a type of AI model called a convolutional neural network (CNN), which had already been developed to identify bird sounds. The CNN was able to pick out 75 of the bird species that the experts had, but the model’s data set was limited and contained only 77 bird species that might occur in the region. “Our results demonstrate that AI is ready for more comprehensive species identification in the tropics from sound,” says Müller. “All that is needed now is more training data collected by humans.”
The team says that using AI to precisely measure the biodiversity of regenerated forests could be crucial for evaluating biodiversity projects that must demonstrate success to secure continued funding.
Researchers at Conservation AI have developed models that can scour through footage and images from drones or camera traps to identify wildlife — including critically endangered species — and track animal movements.
They built a free online platform that uses the technology to automatically analyse images, video or audio files, including data from real-time camera-trap footage and other sensors that approved users can upload. Users have the option to be notified by e-mail when a species of interest has been spotted in the footage they have uploaded.
So far, Conservation AI has processed more than 12.5 million images and detected more than 4 million individual animal appearances across 68 species, including endangered pangolins in Uganda, gorillas in Gabon and orangutans in Malaysia. “The platform can process tens of thousands of images an hour, in contrast to humans who can do a few thousand at best,” says Paul Fergus, one of Conservation AI’s lead researchers. “The speed at which AI processes data could allow conservationists to protect vulnerable species from sudden threats — such as poaching and fires — quickly,” he adds. Conservation AI has already caught a pangolin poacher in the act by analysing footage in real time.
As well as monitoring biodiversity in real time, AI can be used to model the impacts of human activities on an ecosystem and reconstruct historical changes. Researchers have used AI to discover how a century’s worth of environmental degradation in a freshwater ecosystem has led to biodiversity loss.
Although it is well documented that human activities have resulted in biodiversity loss in rivers and lakes, little is known about which environmental factors have the largest impact. “Long-term data is pivotal to link changes in biodiversity to environmental change and to define achievable conservation goals,” says Luisa Orsini, who studies evolutionary biosystems at the University of Birmingham, UK.
Orsini and her colleagues developed a model that links biodiversity to historical environmental changes using AI. In a study published in eLife earlier this year, the team obtained genetic material that had been left behind over the past century by plants, animals and bacteria in the sediment of a lake. The sediment layers were dated and environmental DNA was extracted for sequencing.
The scientists then combined these data with climate information from a weather station and chemical-pollution data from direct measurements and national surveys, using an AI designed to handle diverse types of information. Orsini says the aim was to identify correlations among the ‘mayhem’ of data.
They found that the presence of insecticides and fungicides, together with extreme-temperature events and precipitation, could explain up to 90% of the biodiversity loss in the lake. “Learning from the past, we showcased the value of AI-based approaches for understanding past drivers of biodiversity loss,” says study co-author Jiarui Zhou, who is also at the University of Birmingham.
The main benefit of using AI is that it is hypothesis free and data driven, says Orsini. “AI ‘learns’ from past data and predicts future trends in biodiversity with higher accuracy than ever achieved before.”
Miailhe is hopeful that AI can be routinely applied to real-world conservation efforts in the near future. “That’s clearly the way to go,” he says. But he warns that AI consumes computing power and material resources, which ultimately has adverse effects on ecosystems. “Environmental impact assessments should be at the centre of AI risk management,” he says.
This article is reproduced with permission and was first published on October 27, 2023.