Evoviz

Sustainability and climate resilience

Climate change, land cover change, and pollution, are existential threats to earth and life systems. The concepts of ‘sustainability’ and ‘climate resilience’ are key to mitigating and weathering (pun intended) the direct and indirect impacts of the their combined effects.

Sustainability is defined as “meeting the needs of the present without compromising the ability of future generations to meet their own needs” (United Nations Brundtland Commission, 1987) and is the conceptual umbrella for decision-making and behavior embodied in the UN Sustainable Development Goals (SDGs). Environmental science and management are how we measure and mitigate the environmental and social impacts of human activity with the aim of us becoming a sustainable society.

Climate resilience is concerned managing the effects of climate change, specifically the ability to prepare for, recover from, and adapt to climate change impacts (C2ES 2019). How resilient a place or demographic is depends on both the threat and vulnerability of people, infrastructure, and ecosystems. Climate resilience is the conceptual umbrella for decision-making and behavior for how we adapt to climate change given vulnerability.

Environmental management and climate resilience are both are emerging, complex, spatiotemporal, scale-free, cross-disciplinary, multi-stakeholder, combined earth-human system problems, that employ conceptual approaches, data, and practical solutions so have the potential inform each other (Mulvihill and Ali 2017). Both address a wide range of real-world problems with some including, flooding and drought addressed by both. Depending on the problem and context, data may be abundant or limited or not available, may be heterogenous and its uncertainty high or unknown. They are topics to which I contribute teaching.

Artificial intelligence

There is no generally agreed on definition of what ‘AI’ is (Russel and Norvig 2020 in Sheikh et al. 2023), with two common definitions being ‘technology that enables machines to imitate various complex human skills’ and ‘the performance by computers of complex tasks in complex environments’. Evocative, but lacking in precision. What are ‘complex skills’ and ‘complex tasks in complex environments’? Having reviewed definitions, Sheikh et al. (2023) settle on the definition: ‘systems that display intelligent behaviour by analysing their environment and taking actions – with some degree of autonomy – to achieve specific goals’. Whichever definition is preferred, ‘AI approaches’ are generally conceptualised as a hierarchy of method complexity.

There are many introductions to AI/ML technologies that are much better than I could provide, but they share common properties. These are my take home messages:

• They are a different and sometimes better ways of asking questions of data.
• They are (very) good at classification and prediction.
• They can integrate data in different formats, e.g. tabular, spatial, images, sound and video, in real-time.
• They are ‘black boxes’ whose internal workings are difficult to illuminate or translate into concepts and processes.
• Their application may be limited by data availability and bias.
• They have social, educational, and environmental impacts that must be mitigated.

AI/ML is restructuring how we analyse and interact with the planet and as a society. It is modern-day ‘magic’ that gives great boons, but works in hidden ways with hidden costs that will workout over decadal time-scales. We are likely in an AI-bubble in respect of the industry, but this is only the first wave of what is to come.

Sustainability and climate resilience AI

AI is being applied to address a wide range of climate resilience and sustainability problems in research and business with ‘environmental applications’ estimated to boost global GDP by 3-4%, reduce GHG emissions by 4%, and create 38 million jobs by 2030 (Microsoft and PWC 2024). The diversity of application areas is illustrated by the topics of recent review papers.

• Climate change (Huntingford, et al. 2019; Mercier-Laurent 2024, Rane et al. 2025; Thulke  et al. 2024).
• Weather forecasting (Zeng, 2024).
• Pollution monitoring and modelling (Alotaibi et al. 2024; Anifowose and Anifowose 2024; Popescu et al. 2024).
• Climate mitgation and resilience (Andigema et al. 2024; Rane 2024; Richards and Worden 2024; Singh et al. 2023; Wani et al. 2024).
• Decision making and govenance (Mehryar et al. 2024).
• Flooding (Bentivoglio et al, 2022; Jones et al. 2023).
• Ecology (Cipriano et al. 2025; Pichler and Hartig 2023; Wilson 2024; Yu et al. 2025).
• Sustainability (Dhiman et al. 2024; Kar et al. 2022; Mercier-Laurent 2024; Schoormann et al. 2023; Siqueira et al. 2024; Rohde et al. 2024; Vinuesa et al. 2020).
• Agriculture and food security (Hrast Essenfelder et al. 2025; Khan et al. 2022; Rane et al. 2025).
• Net Zero (Olawade et al. 2024a).
• Energy management (Ali et al. 2024; Aslam et al. 2025; Javed et al. 2025, Rane et al. 2025).
• Waste management (Abdallah, M. et al. 2020; Olawade et al. 2024b).
• Smart cities (Rane et al. 2025).
• Life-cycle assessment (Preuss et al. 2024).
• Climate and environmental education and participation (Praet et al. 2025; Santos and Carvalho 2025).

Two topics are addressed in this literature, the application of AI to specific problems and the sustainability and ethics of AI solutions.

Sustainable AI

The main focus of sustainability of AI research is energy and water use. Every use of a computer uses energy and hence water, but AI requires considerably more of both than conventional systems. For example, Google (2019) estimated a single search consumes approximately 0.3 Wh of energy, compared to AI processes that may consume from 23 to 1500 times more energy (Chen, 2025). This is a huge increase that may not provide equivalent information gain.

AI usage is contributing to the estimated 3-4 fold increase in global data centre energy demand by 2030 (Chen, 2025). While energy use is not insignificant and is rising, figures must be considered in the context of other consumption, with data centers and transmission consuming only 2-3% of global energy production and AI applications just 1/10000th (International Energy Agency in Luers, 2024).

Data centres produce heat that is often cooled with water with a Google data centre in the near-desert town of Dalles, Oregon consuming 25% of the local water supply leading to local shortages, with plans to build other data centres in the area (Pascual/EL PAÍS 2023). This example is just one example how effects are localised where data centres are located for economic reasons.

AI for sustainability and climate resilience – a review of reviews

Some notes on a few of the review papers listed above.

AI applications have been evaluated in respect of whether they address the environmental, social, or govenance dimensions of sustainability and their alignment to SDG gaols. Dhiman et al. (2024) report applications relating to the natural environment, such as carbon emission measurement, energy consumption optimization, and the energy costs of running large ML models are the most common, followed closely is the social studies that explore ethical considerations, educational implications, and issues of equality related to AI. The economic dimension of AI such as economic growth, labor market dynamics, and novel business models has the fewest publications. The third-biggest category includes papers that integrate all three dimensions and examine sustainability through a more comprehensive lens that addresses complexity by encompassing a broader perspective. In contrast, Schoorman et al (2023) report 78% of 95 studies relate to social SDG goals (primarily health) compared to just 24% relating to the economic dimension and 18% to the environmental dimension.

Kar et al (2022) identified transportation/communication/electric/gas/sanitary services as the most common applications of AI for sustainability. The second most frequent class was ‘non-classifiable sectors’ that included ‘society’, IT, environmental issues and the food sector. Kar et al (2022) report the most common methods are regression (54%), decision support system (36%), support vector machine (32%) random forest (26%), and artificial neural network (23%). Schoorman et al (2023) also report ML classification and regression as the most common analytical approaches applied to AI for sustainability problems.

‘AI for Energy’ (US Department of Energy2024) identifies nearly 50 oppotunities for AI applications across a wide range of sectors (planning, permitting and siting, operations, resilience, transportation, buildings, industry and manufacturing, agriculture, and cross-cutting) in the quest from clean energy.

Olawade et al (2024a) reviewed ‘AI and Net Zero’ reporting most useage to optimize electricity grid trasmission, forcast energy production, smart building efficiency, intelligent transportation, sustainable agriculture, waste management and the circular economy, and emissions tracking and monitoring. Optimisation is achieved thought ML, while deep-learning improves climate prediction. AI and robotics are increasingly used in agricultural production and waste management. Block chains provide a means of carbon-market transparancy. Natural language processing is used for communication and sentiment analysis. AI applications are often underpinned by the ‘Internet of Things’ (IoT). Aslam et al. (2025) give six examples of how energy companies are using AI to improve productivity and reduce emissions including General Electric who report a 20% increase in wind farm yield worth $100 million. A review of AI and building energy management report 37% energy saving in offices and 21-23% in education and residential buildings (Ali et al. 2024).

Mehryar et al. (2024) reviewed ‘AI and climate resilience’ reporting the most frequent applications being in disaster management (28%), flood or drought management (16.0%), agriculture and food production (15%), transportation and critical infrastructure (12%), and water governance and quality (7%). Machine learning is by far the most used approach (73%) followed by uncertainty modelling and analysis (12%). The most common application is risk assessment, with fewer studies appraising policy.

‘AI for Energy’ (US Department of Energy2024) identifies nearly 50 oppotunities for AI applications across a wide range of sectors (planning, permitting and siting, operations, resilience, transportation, buildings, industry and manufacturing, agriculture, and cross-cutting) in the quest from clean energy.

Preuss et al. (2024) review LLMs for life-cycle analysis. They conclude, “LLMs offer greater time efficiency for completing LCA tasks, can provide expert-level LCA advice, increase the accessibility of LCAs, and aid in LCI data retrieval and processing tasks. These benefits are limited by the challenges of using LLMs for LCA tasks, including a lack of metrics and benchmarks to quantitatively measure the quality of the LCA, LLM data privacy issues, the difficulties experienced by LLMs when performing calculations, and large variations in quality of the response. Moreover, there are risks in using LLMs for LCA tasks that must be acknowledged, including accountability for generated output, the inclusion of hallucination and bias in generated output, and a lack of transparency in how the LLM generates content.”

Conclusions

A summary of ‘strengths and opportunities’, and ‘challenges and risks’ extracted from the reviews.