Protecting our ecosystems with smart surveillance technologies

Dr Katerina Taškova from Waipapa Taumata Rau, University of Auckland’s School of Computer Science is creating impact-driven artificial intelligence solutions to globally relevant ecological problems. Her current research focuses on reliable machine learning, spatial temporal data mining, and computational sustainability applications.

There’s a reason New Zealand takes its biosecurity very seriously – we are one of the only countries in the world with no large mammalian predators. However, for our native birds, frogs, lizards and plants, this is far from the truth.

Having evolved in isolation for millions of years, our native animals lack vital defence behaviours needed for survival against introduced predators including cats, possums, rats and stoats, and many are now threatened with extinction.

Dr Katerina Taškova from the University of Auckland’s Machine Learning Group is working towards Predator Free 2050, a government-funded initiative headed by the Department of Conservation, aiming to eliminate these harmful predators from Aotearoa in the next 26 years. More than 30 organisations across Aotearoa are working towards this common goal, with Dr Taškova co-leading the Biosecurity Technology Spearhead Research Project tasked with developing new, impact-driven biosecurity technology by leveraging cutting-edge science and mātauranga Māori.

Funded by the Science for Technological Innovation National Science Challenge (SfTI), with co-funding from Biological Heritage National Science Challenge and Predator Free 2050 Ltd, the project’s multidisciplinary team has developed a swarm of smart networked sensors that use artificial intelligence (AI) to detect the last predators in vast and complex landscapes across Aotearoa.

Katerina explains how the mission-led design, including workshops and subsequent consultation with scientists, industry experts and key stakeholders, shaped the project significantly. “You cannot mitigate threats with high confidence if you are not able to reliably detect them first. The workshops identified a critical need for scalable predator surveillance technology that will be functional in complex and remote environments, which are otherwise difficult and expensive to manage with conventional tools.

“Conventional surveillance methods use stationary devices, such as trail cameras and traps, and rely on target predators moving and encountering the devices. To maximise encounter rates, devices need to be placed at relatively high densities – this is expensive for large and complex landscapes, or ineffective in case of a small number of surviving predators after large eradication campaigns. For some species, even having just two left can repopulate an entire area.

“A swarm of mobile networked sensing devices could self-navigate through large spaces and automatically detect and identify invasive predators. Advances in AI, robotics, wireless communication, sensors and battery (energy) technology coupled with well-established predator control expertise could make this sophisticated technology a commercially viable production in a matter of several years.

Katerina explains that conventional active surveillance from helicopters or hunters with dogs is expensive and requires intensive human effort. However, AI poweredsurveillance using swarms of low-cost wirelessly connected sensing devices moved and serviced by drones might be the most cost-effective solution long-term.

“We want a technology that could scale and adapt to different environments, and function autonomously long-term; as such, AI is going to play a key role in the proposed technology. However, the success of the new technology will depend as much on science and clever engineering as it will co-design and social licence.

“We were fortunate to have worked with an inspiring team of Māori researchers; they created a framework for mātauranga Māori co-design of ngahere technology, which was used to inform the drone design including shape, colours and materials, with inspiration taken from traditional practices like weaving and carving. The same approach can be used to inform the design of the sensing devices, including the algorithms for predator detection and predator search strategies.”

Although the SfTI project formally concluded in June of this year, further testing will continue later this year, thanks to the co-funding from Predator Free 2050 Ltd. Katerina explains, “Our current efforts focused on proof-of-concept prototypes,tested in small-scale field trials and using humans to move sensing devices, collect data and replace batteries. Our last field trials will test long-time deployment of the swarm in a forest, and a pilot study that demonstrates a no-human-needed system with drones. The goal is to have a robust operation and full autonomy in the long term. The key to that will be reliable machine learning models for predator search and navigation of rugged native forest reliably, a considerable challenge due to the abundance of branches and vines.”

Artificial Intelligence is a multifaceted vehicle not only driving solutions to existing ecological problems, but also steering our scientific efforts towards the prevention of future climate emergencies. Katerina’s most current project focuses on assessing the impact of the longspined sea urchin population expansion on kelp forests in and around New Zealand.

Scientists are currently witnessing a potential boom in sea urchins in Northern and Eastern New Zealand waters, which directly impacts our kelp forests – a pivotal life supply for many marine species.

Katerina explains, “Climate change driven sea urchin proliferations and overgrazing pressure can turn productive kelp forest habitats into persistent underwater deserts, called ‘urchin barrens’. With drastically reduced productivity and biodiversity, urchin barrens provide fewer ecosystem services than kelp, while significantly reducing support for important cultural practices, fisheries and tourism.

“Longspined sea urchin expansion, due to climate change, is the most urgent threat to kelp-dominated reefs in south-eastern Australia, as well as our kelp forests. However, the extent of this habitat loss is unknown, due to data deficiency. Robust ecological monitoring at spatially relevant scales is therefore needed to gain an understanding of the current impact, and to prioritise areas for protection and active management.”

With funding from Climate Change AI, a global non-profit organisation, Katerina (Co-PI) is leading the development of a machine learning (ML) toolbox that will enable rapid surveillance of kelp forest changes and the spread of sea urchin barrens. In a joint effort with researchers from across Australia and New Zealand, including Dr Arie Spyksma (PI) from the University of Auckland’s Leigh Marine Laboratory, they have “developed open source ML algorithms and models for sea urchin detection and habitat classification, leveraging historic benthic imagery sourced from across south-eastern Australia and northern New Zealand. Although this project is yet to finish, our models are already proving effective in improving the accuracy and speed of image annotation, with ongoing development to improve model generalisability across new locations as new data is becoming available, including detection of new urchin species.”

Arie adds, “A well-designed AI tool, like we have developed, can significantly reduce the time required to analyse large imagery datasets. Across datasets containing 1000s of images, what would have taken days to analyse is now taking hours and still producing highly accurate results.”

Katerina further highlights, “Models and data generated in this project have already been integrated within Squidle+, an open-source platform for centralised marine image data management and annotation, used widely by marine scientists in Australia. This will increase the utility of the models and data and allow us to perform post-deployment evaluations of the models. The latter will inform further refinements and development of models and algorithms so they can be applied to kelp forests and urchin species globally."

Although the team currently focuses on streamlining the automated data annotation, the goal is to use the data to inform mitigation strategies. Katerina adds, “Automated analysis of timely collected data could unlock the full potential of AI-driven solutions for proactive management of underwater ecosystems, freeing up valuable resources for conservation activities and restoration projects.”

The modelling framework underpinning the sea urchin barren detection can be repurposed to address other ecological questions with relative ease, provided there is data available to train it with. Katerina says, “The team has data from the Hauraki Gulf during the 2021/2022 summer marine heatwave; leveraging them to detect sponge species showing signs of necrosis or bleaching as a result of marine heatwaves, which will allow us to better monitor sponge health prior to, during and after future marine heatwave events.”