In today’s rapidly changing world, not only are our cultural society and technology evolving, but our environment is undergoing significant transformations. You may have noticed this firsthand: recent extreme heatwaves have shattered our summertime dreams, and the beaches you cherished as a child are slowly disappearing. Around the globe, astonishing and “statistically rare events” are occurring with alarming frequency.
Consider the severe droughts in the western United States, where water scarcity impacts over 40 million people. This crisis has devastated agricultural production and caused billions of dollars in economic losses (NIDIS, 2024). At the same time, unprecedented heatwaves across Europe in the summer of 2020 pushed energy infrastructures to their limits, resulting in widespread power outages (EEA, 2024). Even more dramatic are the locust swarms in East Africa, driven by unusual cyclonic activity and weather whiplash, which have ravaged crops and left over 2.5 million people facing food insecurity (FAO, 2024; Liu et al., 2024). Growing evidence shows that these catastrophic events, reminiscent of disaster movie scenes, are likely driven by anthropogenic climate change. These climatic shocks pose substantial threats to our water, food, and energy supplies which in turn jeopardize human health, economic stability, and global security.
Effective disaster response in Mozambique
During Cyclone Freddy in March 2023, Mozambique's early warning system proved remarkably effective, leading to fewer deaths and displacements compared to previous similar cyclones. Moreover, AI-powered drones played a vital role in risk mapping, evacuation route planning, and search and rescue operations, underscoring their invaluable contribution to disaster management.
Source: INGD
Combating climate change is one of humanity’s greatest challenges. In this context, artificial intelligence (AI) has emerged as a prominent ally. To fortify climate defence, we need a dual strategy: mitigation and adaptation. Climate mitigation aims to cut down greenhouse gas emissions to slow down global warming, while climate adaptation enhances our resilience against its ongoing and future impacts. In this era of big data and unprecedented computational power, AI steps in as a masterful guide, offering revolutionary tools to understand climate dynamics, predict extreme weather, and develop sustainable strategies for both mitigation and adaptation.
CLIMATE MITIGATION
One of AI’s most revolutionary contributions to climate mitigation lies in carbon capture and removal technologies. Recently developed advanced AI algorithms can enhance the efficiency and feasibility of these processes by optimising capture methods and improving the materials used in CO2 absorbers. For instance, AI-driven models can predict the performance of novel absorbent materials, enabling quicker and more costeffective deployment (Ibrahim et al., 2024). This optimization is crucial for reducing atmospheric greenhouse gases towards net-zero carbon.
AI’s benefits also extend into energy management, significantly improving energy efficiencies within buildings and industrial operations. In building sectors, smart AI-based systems can monitor and use real-time data to regulate indoor heating, ventilation, and lighting systems. This not only achieves substantial energy savings but also ensures that comfort is not compromised. Similarly, in industrial settings, AI can optimise processes through predictive maintenance and operational adjustments, significantly reducing energy use and associated emissions. Moreover, a recent study by Oxford University develops an advanced machine learning model using historical and planned power plant data in Africa to predict the success or failure of power-generation projects (Alova et al., 2021). The insights from this model are critical for avoiding the lock-in of carbon-intensive infrastructures and promoting the adoption of renewable energy sources, essential for long-term cost-effective climate mitigation.
Machine learning for energy projections
Researchers analysed nearly 3,000 African power plant projects, using their machine learning model to predict success or failure based on project-specific and country-level factors. They found that project-specific factors like plant size and generation type are more critical to successful commissioning than country-level factors such as country population and politics.
Photo: Chris Kruger/ Shutterstock
CLIMATE ADAPTATION
Adapting to climate change is meant to be a long run and involves more than just reactive measures. Latest advances in AI have significantly improved the accuracy of weather-related early warning systems. In regions where data scarcity often hampers the accuracy of weather forecasts, AI models fill critical gaps. These systems utilise advanced algorithms to forecast severe weather events with impressive accuracy and increasingly long lead time (Wang et al., 2023). From hurricanes to flash floods, such forecast capabilities are not just technological achievements, they are lifesaving tools. They provide the crucial minutes or hours needed to evacuate or prepare for the vulnerable communities, supporting them to better adapt to a warming climate.
For example, Google’s DeepMind has developed an advanced AI model, GraphCast, that utilise machine learning and Graph Neural Networks (GNNs) to deliver accurate 10-day weather forecasts in less than a minute. This capability is crucial for early warning systems, allowing for preemptive actions against potential flood risks and extreme temperatures that could devastate communities (Lam et al., 2023). Moreover, Google has expanded its Flood Hub platform to provide real-time flood forecasting and visualisation across 80 countries, benefiting over 460 million people by offering essential data that can lead to timely evacuations and minimise the flood impacts (Lam et al., 2023).
With the assistance of AI, urban designers can plan the placement of green spaces, optimize traffic flows to reduce emissions, and manage water resources in a more sustainable way.
These AI-driven initiatives are particularly vital in developing countries where traditional monitoring systems are sparse. For example, AI models employing long short-term memory networks have demonstrated their ability to predict extreme flooding events in ungauged watersheds with remarkable accuracy, significantly enhancing the lead time and reliability of flood warnings (Nearing et al., 2024).
Beyond forecasting, AI’s role in climate risk quantification cannot be overstated. Nowadays, AI systems provide crucial insights into how climate risks impact different sectors and communities. For example, McKinsey has developed an AI-powered system to quantify how climate-related risks affect property values, allowing insurers, investors, and policymakers to prepare more effectively for the financial impacts of climate change (McElhaney et al., 2023).
AI is also widely used in urban planning. This is a critical component of urban climate adaptation that will directly affect our daily lives. Latest deep reinforcement learning models can help design cities to counteract future climate scenarios while minimizing environmental impacts. With the assistance of AI, urban designers can plan the placement of green spaces, optimise traffic flows to reduce emissions, and manage water resources in a more sustainable way (Lam et al., 2023). Additionally, AI plays a significant role in predicting how land-use changes affect various species and ecosystems. This helps to conserve biodiversity in cities and ensures that urban expansion does not come at the expense of ecological imbalance (Kerakos, 2024).
Empowering resilience
Empowering resiliencePopulations in low- and middle-income countries comprise nearly 90% of the 1.8 billion people vulnerable to flood risks. The World Bank estimates that upgrading flood early warning systems in these regions to the standards of developed countries could save an average of 23,000 lives per year.
Photo: REUTERS / gdagys / iStock
CLIMATE ADAPTATION POTENTIAL, PITFALLS, AND PATHWAYS
The potential of AI in combating climate change is monumental. The explosion of big data and enhanced computational resources provides rich information that AI can utilise for improved decision analytics. From satellite remote sensing and in-situ measurements to social media and model-based simulations, these data resources, coupled with advanced physics-based and AIdriven models, allow for detailed and scalable environmental assessments.
Recent breakthroughs in large language models (LLMs), such as ChatGPT, have transformed how we can handle unconventional data types and automate complex tasks. The generative abilities of LLMs enable them to predict and create scenarios of unseen climate extremes, and their human-like reasoning and interaction facilitate better integration of human feedback into climate models. There is also huge untapped potential of using LLMs for operational purposes, such as real-time control of water systems, power grids, and optimised irrigation in farming practices. Tech giants, including IBM, Microsoft, Nvidia, and Google, are racing to create Digital Twins to replicate Earth’s complex climate processes. This could revolutionise our ability to predict climate change impacts and optimize resource management.
This image was generated using Midjourney based on the prompt: “A 3D illustration of how AI can be used for climate mitigation and adaptation with a focus on extreme weather.”
Image: Dr Xiaogang He
Regions most vulnerable to climate change often lack the technological resources and capacities to leverage advanced AI solutions, even though they have contributed least to global emissions.
However, deploying AI for climate-related tasks remains challenging. One key obstacle is the “black box” nature of many AI systems, making their decision-making processes untransparent. This opacity complicates the interpretation and reliability of AI outputs. Although approaches like Explainable AI and physics-informed AI models can address these issues to a certain extent, they are not yet fully resolved at this stage. Additionally, the environmental footprint of AI is questioned for its sustainability due to its substantial energy and water consumption. Moreover, as data collection scales up, the focus must shift towards enhancing data quality rather than just quantity. Poor-quality data can reinforce biases and lead to unequal benefits or exacerbate vulnerabilities in different regions (Ravuri et al., 2021).
Addressing these challenges requires a pathway that prioritises equity and accessibility. Regions most vulnerable to climate change often lack the technological resources and capacities to leverage advanced AI solutions, even though they have contributed least to global emissions. To balance the global response to climate challenges, ensuring that these regions can access and deploy AI technology is crucial. This implementation involves embedding human knowledge and ethical considerations into AI models to ensure they are equitable and universal. For instance, integrating deep generative models with augmented and virtual reality can enhance public engagement and awareness of climate impacts, providing a more comprehensive approach to both mitigation and adaptation strategies (Synced, 2020).
As we continue to explore and leverage the evolving capabilities of AI, we are more confident in moving closer to a future where the goals of the Paris Agreement are not only met but surpassed.
AI holds the key to unlocking unprecedented advancements in our fight against climate change and meeting Sustainable Development Goals. The continuous emergence of breakthroughs motivates us to further integrate AI into our climate strategies, paving the way for a resilient and thriving world for generations to come. ∞
Dr He acknowledges Xuejie Zhang and Jack Shi Wei Lun for their assistance with the literature review.
REFERENCES
- Alova, Galina, et al. “A Machine-Learning Approach to Predicting Africa’s Electricity Mix Based on Planned Power Plants and Their Chances of Success.” Nature Energy, vol 6, no 2, 11 Jan 2021, pp 158–166. https:// doi.org/10.1038/s41560-020-00755-9.
- Al-Sakkari, Eslam G, et al. “Carbon Capture, Utilization and Sequestration Systems Design and Operation Optimization: Assessment and Perspectives of Artificial Intelligence Opportunities.” Science of The Total Environment, vol 917, 20 Mar 2024, 170085. https://doi.org/10.1016/j.scitotenv.2024.170085.
“Climate Change Impacts, Risks and Adaptation.” European Environment Agency (EEA), 3 May 2024. https://www.eea.europa.eu/en/topics/in-depth/climate-change-impacts-risks-and-adaptation.
- “Desert Locust Crisis: Urgent Actions Needed.” Food and Agriculture Organization of the United Nations, 2024. https://www.fao.org/ emergencies/crisis/desertlocust/en/.
- “Impact of Drought in California.” National Integrated Drought Information System (NIDIS), 2024. https://www.drought.gov/drought/ states/california.
- Kerakos, Emil. Managing Climate Overshoot Risk with Reinforcement Learning: Carbon Dioxide Removal, Tipping Points and Risk-Constrained RL. KTH Royal Institute of Technology, 2024. https://www.diva-portal. org/smash/get/diva2:1851716/FULLTEXT01.pdf.
- Lam, Remi, et al. “Learning Skillful Medium-Range Global Weather Forecasting.” Science, vol 382, no 6677, 14 Nov 2023, pp 1416–1421. https://doi.org/10.1126/science.adi2336.
- Liu, Xinyue, et al. “Unveiling The Role of Climate in Spatially Synchronized Locust Outbreak Risks.” Science Advances, vol 10, no 7, 14 Feb 2024. https://doi.org/10.1126/sciadv.adj1164.
- McElhaney, Doug, et al. “Reducing Climate-Related Property Risk with AI”. McKinsey & Company. 1 May 2023. https://www.mckinsey.com/ industries/financial-services/our-insights/reducing-climate-relatedproperty- risk-with-ai.
- Nearing, Grey, et al. “Global Prediction of Extreme Floods in Ungauged Watersheds”. Nature, vol 627, 20 Mar 2024, pp 559–563. https://doi. org/10.1038/s41586-024-07145-1.
- Ravuri, Suman, et al. “Skilful Precipitation Nowcasting Using Deep Generative Models of Radar.” Nature, vol 597, 29 Sept 2021, pp 672–677. https://doi.org/10.1038/s41586-021-03854-z.
- Synced. “Bengio and Mila Researchers Use GAN Images to Illustrate Impact of Climate Change.” Medium, 25 Feb 2020. https://medium. com/syncedreview/bengio-and-mila-researchers-use-gan-images-toillustrate- impact-of-climate-change-4cac3a78ef47.
- Wang, Huimin, et al. “Forecasting Fierce Floods with Transferable AI in Data-Scarce Regions.” The Innovation, 2024. https://doi.org/10.1016/j. xinn.2024.100652.
DR XIAOGANG HE
Dr Xiaogang He is an Assistant Professor in Civil and Environmental Engineering at the National University of Singapore. Dr He is a Princeton-trained PhD hydrologist with experience in economics, machine learning, and environmental policy. His research interests focus on the fundamental understanding of how climate change, variability, and human interventions affect drought and flood risk across scales, and how to implement an integrative framework to reduce their societal impacts on the interlinked water-food-energy sectors. Dr He has published more than 50 papers in peer-reviewed journals, including PNAS, Nature Water, Nature Communications, and Science Advances as the lead/corresponding author. His research has gained lots of media attention, including the New York Times, Associated Press, Bloomberg, USA Today, The Straits Times, KQED, and many others.
JULY 2024 | ISSUE 12
NAVIGATING THE AI TERRAIN
