Artificial intelligence’s potential offers hope for addressing climate change, but there are still risks to consider
By
The answer to life, the universe and everything was, according to Deep Thought, the supercomputer in Douglas Adams’ Hitch Hiker’s Guide to the Galaxy, 42. The answers to mitigating and reversing climate change are proving more complex but artificial intelligence (AI) is now being promoted as a game changer by scientists and conservationists. The UN, too, is won over and believes that AI can not only meaningfully address climate issues but also enable many of the Sustainable Development Goals (SDGs) to be achieved.
AI tools, such as the algorithms produced by machine learning (ML), can more precisely predict weather, warn of droughts, track icebergs, recycle more waste, find plastic in the ocean, optimise the energy efficiency of heating and cooling systems in buildings, and monitor the distribution of endangered species. All of this is far from the stuff of fantasy; in many instances, they’re already being deployed. ‘The transformative potential of AI for good is difficult even to grasp,’ says Antonio Guterres, the UN secretary general. ‘It could supercharge climate action and efforts to achieve the SDGs by 2030.’ According to Google DeepMind, a team of scientists, engineers and – they’re keen to point out – ethicists, AI ‘can accelerate breakthrough science of climate and its effects.’
This is already happening in real-time. According to the European Space Agency, AI has been trained to measure changes in icebergs 10,000 times faster than a human could do it and can map large Antarctic icebergs in just 1/100th of a second. This will help scientists understand how much meltwater icebergs release into the ocean.
AI is already improving scientists’ understanding of longer climate timescales by, for example, producing vastly improved predictions of the South Asian monsoon. ‘ML technologies are increasingly transformative of current practices and assumptions,’ says Massimo Bonavita of the European Centre for Medium-Range Weather Forecasts. ‘The pace of progress is relentless.’ Traditional weather- and climate- prediction models are becoming less blurry as they are superseded by ML.
‘This is leading to more precise monitoring of environmental conditions and related climate impacts, such as biomass, permafrost extent, polar sea ice and volcanic activity,’ he adds.
According to Roberta Pierfederici, a policy analyst and research advisor at the Grantham Research Institute on Climate Change and the Environment, AI could be effective in accelerating the low-carbon transition. ‘AI can help to redesign complex structural systems, including cities, land use, transport, industry and energy, and run them effectively and efficiently,’ she says. ‘In the energy sector, AI can help to manage predictability in supply– demand balance and improve system productivity.’
The UN believes AI can help deliver quick benefits, such as its Early Warnings for All initiative, which aims to ensure that everyone on Earth benefits from early-warning systems for hazardous weather, water or climate events by 2027. The need for better understanding of the accelerating impacts of climate change and related weather forecasting is pressing because, according to the World Meteorological Organisation (WMO), climate-, weather- and water-related extremes are 15 times more deadly for people in Africa, South Asia, South and Central America, and small island states. Over the past 50 years, nearly 70 per cent of all deaths from climate-related disasters have occurred in the 46 poorest countries.
‘Although the accuracy of traditional weather-prediction models has improved dramatically, the physical processes are not 100 per cent understood,’ says Yuki Honda, chief of the Earth System Prediction Division at the WMO. ‘If future weather conditions are known accurately and in sufficient time, it is possible to prepare for disasters.’ Honda adds that ‘it was a huge surprise’ when ML models surpassed the most accurate traditional models ‘in a short time. There is no doubt that it is a method with great potential.’
Practical deployment of AI is already making a difference in this area, says Pierfederici: Google’s Flood Hub uses AI to help provide more accurate information on riverine floods up to seven days in advance; it’s used to forecast floods in 80 countries. In Kenya, the AI-driven MyAnga app provides pastoralists with more reliable data about droughts and can save them days of scouting for green pastures. In Burundi, an AI project led by UNEP is identifying long-term displacement hotspots to better inform adaptation for pastoralists and other farmers.
The uses of AI in rainforests
Conservation groups are using AI to improve monitoring of illegal logging in rainforests. Companies such as Space Intelligence are mapping more than a million hectares of land from space using satellite data and disseminating the results to NGOs and governments, thereby accelerating understanding of the health of wildlife, forests and other ecosystems. ‘The potential of AI is particularly promising in the Amazon rainforest,’ says Jorn Dallinga, Forest Foresight programme manager at WWF Netherlands, where it’s already being used to predict where illegal deforestation is likely to take place (see box, right). ‘AI is a powerful tool that can rapidly analyse and process large amounts of complex environmental data. This information can be used to inform decision-making around government policies and conservation.’
Existing methods for monitoring deforestation have limitations. ‘We are currently only able to detect deforestation that has already occurred,’ says WWF’s Jorn Dallinga. WWF-Netherlands and partners have developed Forest Foresight, an innovative AI-driven technology for preventing illegal deforestation before it begins. AI reviews the physical features of an area, the population density and past forest cover. Once trained, it reads real-time satellite images, detecting early deforestation predictors, such as expanding roads, and alerts local authorities to threats.
Forest Foresight has already successfully predicted and averted illegal deforestation in Kalimantan and Gabon. Monitoring a combined area of more than eight million hectares, 33 interventions against deforestation have now been carried out. Informed by Forest Foresight, rangers in Gabon uncovered illegal gold mining activities in time to save an estimated 30 hectares from illegal deforestation, thereby preserving rainforest that would otherwise have been lost.
Following these successful trials, WWF has begun work to understand more about Forest Foresight’s potential in the Amazon rainforest. WWF plans to grow Forest Foresight’s operation across 12 countries by 2027. ‘We’ve worked hard to develop a system that predicts where forests will be cut down by connecting big data and human activity,’ says Dallinga. ‘This allows us to work together with governments and local communities.’
Can AI help wildlife?
The prospect of being able to confidently analyse an entire ecosystem, monitor its checks and balances, and identify where best to reintroduce or relocate species or remove invasives is the stuff of conservationist dreams. If the claims are to be believed, tools such as those deployed by Google DeepMind will make a profound difference. In the Serengeti National Park, DeepMind is collaborating with ecologists and conservationists to develop ML methods to help study behavioural dynamics across the park and its 70 large mammal and 500 bird species.
Until now, the millions of photos captured by motion-sensing cameras in the park have had to be counted and analysed by hand, creating a bottleneck of up to a year before information can be deduced and actions planned. ML can dramatically speed up this process, ensuring conservation measures can take place in real-time rather than retrospectively. Another DeepMind ML project, called Turtle Recall, is helping to better identify individual turtles, animals that are notoriously difficult to distinguish even when they’re tagged, in coastal waters around Ghana and Benin.
A step forward in green farming?
AI is also seen as potentially beneficial by those seeking less-intensive use of farmland chemicals. Pierfederici points to companies such as John Deere, which has created an AI visual-recognition tool, See & Spray, that distinguishes between cultivated plants and weeds, greatly reducing herbicide and fertiliser use.
In the UK, software provider ABACO, in partnership with crop science organisation NIAB, has trialled a digital application, known as the integrated soil health scorecard, that will improve the targeted use of nutrients, increase yield resilience, reduce fertiliser use and improve biodiversity. The process is hands-on: the farmer takes a soil sample, posts it to a lab and gets results inside a week on whether, and where, they need to apply chemicals. ‘The system does the colouring in,’ says Elizabeth Stockdale, head of farming systems research at NIAB.
‘It will tell you whether everything is fine or where there are issues. The more data a farmer has, the more dots they can join, the better. The focus is on pulling together physics, chemistry and biology.’
AI and ocean software
According to Robert de Vries, who works on remote sensing of coastal environments and rivers for the Ocean Clean Up, AI has added ‘a new realm of possibilities’ with its ability to analyse large amounts of camera data for plastic debris and create detailed maps of ocean litter in remote locations. Previously, mapping large plastic items in the oceans could only be done by trained visual observers and with surface trawl nets, ‘methods that are limited in scalability and involve the large costs of operating open-ocean research expeditions,’ he says. AI has already enabled the Ocean Clean Up to collect and analyse more than ten million images and study thousands of large floating plastic objects.
The data also help us understand how, when and where particular items end up in the ocean; AI data have shown that plastic from specific rivers accounts for 80 per cent of ocean plastic pollution. ‘This was nearly impossible to record through visual observations or physical sampling,’ says de Vries. ‘AI has helped us create additional eyes on the ground to get accurate counts of the plastic concentrations in remote locations and better target our clean-up operations.’
Soil intelligence
Soil health is crucial to enabling agriculture to reduce its contribution to climate change, and several companies are developing innovative soil-analysis mechanisms
that seek to embrace AI and machine learning (ML).
‘We currently have a renaissance in looking at our soils and measuring their condition,’ says Russell Lawley, product development team leader for the UK Soils Observatory. NIAB is also leading for the UK Agriculture and Horticulture Development Board on a data-use project, utilising yield-map data more effectively to target sampling and inform management decisions. The focus is on soil variation, impacts of management on soils, and other in-field variations, such as shading. The project is being trialled in the UK at Strategic Cereal Farm East.
‘Yield maps are still generally pretty maps,’ says Elizabeth Stockdale from NIAB. ‘Until now, they haven’t really guided practice or been applied to strategic decisions.’ Trials at the farm using AI and ML have indicated potentially huge savings through informed, targeted spraying of nitrogen.
Climate change is making such technology and its insights more crucial.
‘We used to have a lovely temperate climate; we’re not going to have that in the future,’ says Stockdale. ‘Soil degradation – compaction, erosion, carbon loss – costs at least £1.2 billion annually across UK agricultural land.’ AI and ML, she argues, can deal with the complexity and diversity of links between soil function on a specific farm and industry practices, and avoid general prescriptions in a way not previously possible. ‘You can put in data from 100 farms and an average will come out – but, by definition, no farm is average. Improving soil health is essential if the triple challenge of food security, net zero and improved biodiversity is to be met.’
What impacts does AI have upon humans?
The political and societal implications of AI are real and serious, says Pierfederici, and acting upon the knowledge AI gives us may not be straightforward,
as the potentially enormous change brings risks that require further thought and attention. ‘The AI revolution, combined with the net-zero transition, will generate mass displacements affecting workers and communities – the phasing out of fossil-fuels industries, workforce automation and replacement induced by AI. This will need to be actively managed by governments.’
AI also risks exacerbating the North-South divide, as today’s AI research and application is predominantly led by research institutions and corporations in
the Global North. ‘AI is far from a silver bullet for addressing the complex challenges faced by the Global South,’ says Dallinga, who believes the ‘digital divide’ is exacerbating existing problems. As of 2021, 37 per cent of the global population, roughly 2.9 billion people, still lacked internet access. This largely affects the Global South, where infrastructure and affordability affects connectivity. In the context of conservation, while the Global North may have huge datasets, the Global South, home to far more diverse ecosystems such as the Amazon rainforest, often lacks thorough documentation. This makes existing inequalities in technology and scientific knowledge even more severe.
The potential of AI isn’t only being picked up on by scientists and NGOs. If AI is portrayed as offering copper-bottomed data, Dallinga is conscious that it could be used to make greenwashing harder to expose (a 2021 European Commission report found that 42 per cent of ‘green’ claims from consumer websites were either exaggerated, deceptive or false). ‘It’s important to be aware that AI is trained on data, which could potentially include misleading content, such as greenwashing or exaggerated sustainability claims,’ says Dallinga. ‘If someone asks AI about a company’s potential involvement in deforestation, it may repeat its sustainability claims.’
The limitations of AI
AI isn’t infallible – like any software, it’s only as good as its training dataset. At the Ocean Clean Up, De Vries has found that the presence of waves and light effects can still trick the AI. Another risk is that AI technologies base their future predictions on patterns of past data.
‘This means that we must be careful when using them as guides through structural changes that are likely to be profoundly different from the past,’ says Pierfederici, pointing out how current models (without AI) have systematically underestimated the power of innovation and adoption of low-carbon technologies at scale.
‘The question is how we can make sure that AI will be able to predict aspects of the future that will necessarily be very different from the past.’
AI raises the issues of capacity and competence. It’s one thing being more confident than ever about when extreme events may strike, or a species’ population will fall off a cliff; it’s another to be able to do something about it. To address this, governments must play a crucial role in enabling effective and judicious scaling and deployment of AI through financial incentives and support for research and development in AI technologies, suggests Pierfederici, ‘to provide clarity on what is the vision of the future and how to get there. There are still very few studies out there on this, with very limited scope.’
The problem of the ‘rebound effect’, where efficiency gains can actually lead to greater energy consumption is also real, says Pierfederici. However, she feels the overall trend is positive.
‘Rebound effects have indeed been measured in some sectors, like transport or home heating and cooling, but if we look at the whole economy, energy efficiency has been critical in decoupling economic growth from rising energy consumption and greenhouse gas emissions.’
But the carbon footprint of AI applications could still be significant, as the computational needs of AI are highly energy intensive.
‘There are still a lot of uncertainties concerning what contribution AI can bring in reducing global GHG emissions,’ Pierfederici adds.
So what’s the verdict?
De Vries recognises there’s a need for ‘human-based checkpoints’; Stockdale cautions that sound judgement will always be required. Looking at her own specialism in agriculture, she notes: ‘The most important technology, the heart of good understanding, is still, and will be for a long time, a spade, and conversations with other farmers and advisors.’ Dallinga adds: ‘AI cannot take care of everything. It offers valuable support for decision making, identifying patterns and predicting outcomes, but it cannot enact change on its own, it cannot replace the role of communities and Indigenous peoples in safeguarding our planet. The root causes of environmental problems are often deeply embedded in socioeconomics and political systems. Solutions to environmental challenges such as biodiversity loss, deforestation and climate change are complex and need to be addressed by combining AI with other strategies.’
While AI may be transformational, it seems it will have its limits in our complex, interconnected world. As Honda at the WMO says: ‘In principle, the weather is considered to behave chaotically. Small errors will always cause uncertainty in the future.’
One thing at least is clear: AI isn’t going to clean up our mess by itself. Expecting AI to enable us to continue with business as normal will not cut it, says de Vries. ‘Pollution continues primarily because of the lack of consistent and widespread recycling infrastructure, a lack of government intervention and learned behaviours. Intelligence in a vacuum is not going to clean the oceans. Behaviour change would be a good start in helping heal our planet.’
Pierfederici agrees that AI won’t enable humans to wriggle out of their responsibilities. ‘This is a long- standing problem that goes well beyond the capacity to use information from AI,’ she says. ‘Scientists have been warning the world for decades now about the effects of climate change, but progress on commitment has been slow, so this is really about governments and societies being able to act on warnings coming from science.’
The time to act and utilise AI and ML is now, adds Pierfederici.
‘The next ten years will be crucial to bring clean technologies to market. AI and ML can accelerate the process of scientific discovery in ways that were unimaginable only years ago. AI is not a magic wand that can solve all the risks posed by climate change alone. For this to be translated into meaningful actions, we still require an active role from governments, the private sector and civil society.’
