FERAL foxes are widespread and threaten Australia’s biodiversity by hunting native wildlife and livestock. Already they have contributed to the extinction of several species of small mammals and birds, and they pose a major threat to many more.
The strict guidelines on fox control seek to balance cost-effectiveness with the risk to non-target species and humane treatment. The most cost-effective and common method of fox control available is lethal baiting, however not all experts agree on how much baiting should be used in pest control.
Dr Iadine Chadès, from CSIRO’s Conservation Decisions team, has developed an artificial intelligence (AI) algorithm to provide an adaptive solution. Applied to fox control, it could aid the decision-making process by assessing how different native species respond to the reduced density of foxes, so that it’s possible to monitor the progress of pest control and recommend adaptions as necessary.
The extinction crisis
Approximately one million animal and plant species are currently threatened with extinction. This devastating figure was announced in a report from the UN’s Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services (IPBES) issued in May.
The extinction crisis hits close to home: Australia has the fourth-highest level of animal species extinction in the world, according to the International Union for Conservation of Nature. As we lose our native fauna, some of which we rely on to preserve our land and pollinate our flora, we lose part of our identity and ecosystems.
Many factors have contributed to this loss of biodiversity, including feral animals (predators of native species as well as herbivores that compete for the same food sources), weeds, and changes in fire patterns. But one of the most detrimental factors is us.
Australia has implemented multiple programs to protect threatened animals. But unfortunately, we don’t have enough resources to protect them all. Our conservation practices have not progressed fast enough to meet the enormity of the environmental challenges we are facing, such as urbanisation, predation by invasive species, and climate change. We have to make decisions about how we can best use our resources to protect as many species as possible and make every dollar count.
The intelligent solution
Chadès is using AI to improve species detection methods and to advise on better conservation approaches.
AI is the intelligence demonstrated by machines to learn and problem-solve. Using devices that monitor the environment, for example camera traps and telemetry – the transmission of radio signals to locate transmitters attached to animals of interest, the AI program developed by Chadès and team collects data about animal populations and their environment to recommend the best course of action to maximise their survival.
To date, conservation researchers have used AI to observe animals by tapping into data from traditional monitoring methods, but this is the first time that it has been applied to the decision-making process. AI can identify and monitor animals from photos and categorise the many millions of camera trap photos gathered by field scientists over large areas.
“But if we want to help save our species, we can’t just observe them,” says Chadès. “We need people to make good decisions, and there isn’t a lot of time or money.”
She therefore wanted to exploit the ability of machines to learn and develop solutions based on their own data.
Chadès developed mathematical models to tackle complex optimisation problems during her doctorate and she began to apply them to conservation science in 2008. Combining AI with ecological models, she developed a program that can provide optimal solutions in the face of uncertainty that saves money and allocates resources more efficiently.
CSIRO’s interdisciplinary Conservation Decisions team expanded on her early work to develop AI algorithms for modelling adaptive management problems.
Getting out into the field
The team is currently collaborating with the New South Wales (NSW) Office of Heritage and Environment to improve the monitoring and management of the 1000 threatened species in the state as part of the Saving our Species project. This aims to develop an innovative decision-making tool to help inform priorities and the best management practices to protect threatened species across NSW, especially in terms of managing threats such as pests and weeds.
The AI developed by CSIRO is calibrated to suit individual species, as each have its own threats and responses to conservation management.
“We don’t know exactly what each species needs and therefore the program has to be flexible,” says Chadès.
The program assesses the data collected in the past, and models the predicted outcomes of the future to decide on the best recommendations. It can then adapt its recommendations based on the consequences.
There isn’t a wealth of data on our most endangered species, so the benefit of this technology is that it can solve problems without large data sets. It also works as a positive feedback loop: as recommendations are applied, the program monitors the progress of species so that it can adapt further recommendations and will collect data as they are used.
“AI systems can rapidly learn from the successes and failures of the past as well as accounting for all future scenarios. These sorts of assessments are often far too complex for people,” says Chadès. The environment is constantly changing and there is always uncertainty. “We cannot be sure what actions will be best.”
The Conservation Decisions team is a multi-disciplinary group working to solve global conservation challenges. By connecting ecological data with AI and decision theory, they are able to determine what actions to take and how to produce the best outcomes, while taking into account the past, present, and future. This work has already been used to guide practices to eradicate invasive weeds, control mosquito-borne diseases, and protect threatened species from extinction. It has great potential to guide decisions that affect our land, our animals, and our health.