Artificial intelligence is advancing at a breathtaking pace, fundamentally transforming how scientists study, predict, and respond to a warming planet. While AI is most widely recognized for consumer-facing large language models such as ChatGPT, Claude, and Gemini, its applications across the scientific community are vast and expanding rapidly. From revolutionizing weather forecasting and scanning massive volumes of satellite imagery to optimizing renewable energy grids and shaping global policy frameworks, researchers are increasingly turning to machine learning to tackle the complexities of climate change.
However, this technological leap is accompanied by significant risks and challenges. Experts warn of a dangerous overreliance on unverified data, intense energy consumption, and the potential to exacerbate existing inequalities within global climate science. To understand the true scope of this transformation, Carbon Brief spoke with 10 leading experts working across various climate-related fields, ranging from short-term weather modeling to energy grid management. Their insights reveal a complex landscape of unprecedented opportunities and critical vulnerabilities.
The Benefits and Challenges of Using AI in Weather and Climate Forecasting
One of the most significant recent breakthroughs in the field is the rapid rise of AI-powered weather-prediction models. These systems learn from massive historical weather datasets to predict conditions hours in advance, then repeatedly feed those predictions back into themselves to generate forecasts up to two weeks ahead.
Dr. Doug McNeall, science lead in data-driven climate modelling at the UK Met Office, notes that these models have demonstrated an ability to deliver highly accurate predictions while requiring a fraction of the computing power demanded by traditional numerical models, which rely heavily on complex mathematical equations to simulate atmospheric physics.
Despite these successes, substantial hurdles remain. AI weather models are typically built and trained for short timescales where vast amounts of observational data—such as temperature, rainfall, and sea ice extent—are readily available and performance can be evaluated quickly. Climate prediction, however, operates on an entirely different scale. To generate useful long-term insights, scientists must project future climates that lie entirely outside the existing observational record.
To bridge this gap, researchers are exploring blended or hybrid approaches. By using AI to emulate or accelerate traditional physical models rather than relying entirely on black-box, data-driven systems, scientists hope to make climate risk information more accessible and affordable while preserving the physical laws and scientific principles that underpin reliable projections.
Prof. William Collins, a professor of AI for climate science at the University of East Anglia, describes the current era as an exciting renaissance for weather and climate science. The impact of machine learning on numerical weather prediction has been profound. For instance, the European Centre for Mid-Range Weather Forecasts has begun deploying an operational AI forecast system alongside its traditional forecasting infrastructure, demonstrating that machine learning can compete directly with the world’s leading numerical models.
Yet, Collins points to a fundamental paradox at the heart of the discipline: much of machine learning relies on learning from past data, yet climate science is fundamentally tasked with forecasting a future for which historical observations do not exist. Developing reliable ways to inform models about unprecedented future conditions remains a central challenge for the global scientific community.
Dr. Amirpasha Mozaffari, a researcher in Earth artificial intelligence at the Barcelona Supercomputing Centre, emphasizes that while AI forecast models excel at predicting typical weather patterns, they frequently underperform compared to physics-based models when confronting record-breaking and unprecedented extreme events—precisely the phenomena that matter most for climate adaptation. Solving this limitation through hybrid architectures, physical constraints, or novel evaluation frameworks is currently the primary focus for researchers in the field.
Expanding Horizons: Current Uses of AI Across Climate Research
Beyond traditional forecasting, the proliferation of Earth-observation satellites has unleashed a massive wave of geospatial data, capturing everything from shifting sea ice to widespread deforestation. Dr. Stefano Materia, a researcher in climate variability and change at the Barcelona Supercomputing Centre, explains that AI is increasingly vital for translating raw pixel data into actionable climate intelligence.
Methane tracking and ecosystem monitoring highlight the practical power of these tools. Methane is a potent greenhouse gas, but leaks from industrial infrastructure are often invisible and exceptionally difficult to monitor manually. Today, specialized AI algorithms automatically scan hyperspectral imagery captured by satellites from across the electromagnetic spectrum, pinpointing industrial methane leaks around the globe and alerting operators in near real-time. Simultaneously, machine-learning models process thousands of daily satellite images to detect illegal logging, track land-use changes, and monitor glacier retreat long before human analysts could manually review the footage.
In polar research, machine learning is transforming how scientists study fragile cryosphere systems. Dr. Clare Eayrs, a polar research scientist at the Environmental Fluid Dynamics Lab at New York University, notes that machine-learning applications now span the entire sea ice modeling workflow. Systems like Ice Net have demonstrated that deep-learning models inspired by neural networks can outperform leading physical models in seasonal Arctic forecasts, predicting critical variables such as sea ice concentration, thickness, and drift.
Global coordination frameworks are also adapting to this technological shift. Prof. Cristiana Stan, vice-chair of the World Climate Research Programme (WCRP), highlights that AI has been integrated into major initiatives like the weather prediction model intercomparison project. By placing physics-based numerical models, hybrid systems, and pure AI forecasts side-by-side, the project enables direct comparisons of state-of-the-art technologies, accelerating the development of next-generation operational tools.
Meanwhile, major international bodies are formally recognizing the shift. The Intergovernmental Panel on Climate Change (IPCC), currently undertaking its seventh assessment report (AR7), has integrated model intercomparison projects specifically dedicated to AI-driven weather and climate forecasting. Dr. William Collins notes that this milestone represents an exciting proof of concept that will shape the future of climate assessments. Furthermore, Dr. Amirpasha Mozaffari points out that advancements like diffusion models, foundation models, and generative ensembles now migrate from computer science into climate research within months rather than years, dramatically lowering the computational cost of running global forecasts.
Driving Climate Policy, Energy Transition, and Action
The utility of artificial intelligence extends deeply into climate mitigation, energy management, and policy design. Prof. Priya Donti, an assistant professor at the Massachusetts Institute of Technology and co-founder of Climate Change AI, emphasizes that the integration of massive amounts of wind, solar, and other renewable sources into power grids requires entirely new management strategies.
AI plays a crucial role here by generating vastly improved forecasts for solar and wind power generation, effectively reducing uncertainty and allowing grid operators to plan with greater foresight. Additionally, machine learning is deployed to optimize and control diverse power grid components, including power generators, battery storage systems, industrial loads, and flexible energy demands from electric vehicles and smart heating and cooling systems.
Dr. Carlos Rodriguez-Pardo, a postdoctoral researcher at Politecnico di Milano and an affiliated scientist at the Euro-Mediterranean Center on Climate Change, notes that machine learning is now useful across every single stage of the climate policy cycle. In the measurement phase, AI processes satellite data to estimate emissions, enhances air quality monitoring, organizes fragmented policy databases, and maps expansive scientific literature.
During the scenario-modeling phase, AI-based emulators replicate the behavior of computationally expensive economic and climate models at a fraction of the cost, allowing researchers to explore thousands of scenarios and quantify uncertainties that were once entirely out of reach. Finally, in the evaluation stage, causal machine learning extends the toolkit researchers use to rigorously determine whether a specific policy actually achieved its intended outcome.
Future Horizons and Emerging Misconceptions
Looking ahead, the role of AI in climate research is poised to deepen even further. The WCRP has established a dedicated task team to examine modeling, projections, data infrastructure, and the scientific trustworthiness of machine-learning methods. Furthermore, experts predict that upcoming international assessments, such as future IPCC reports, may heavily utilize large language models to assist in synthesizing vast bodies of scientific literature, helping researchers identify "known unknowns" and direct future scientific inquiry.
However, experts are eager to correct widespread misconceptions regarding AI’s role in the field. Dr. David Rolnick, an associate professor at McGill University and co-founder of Climate Change AI, stresses that the general public and funding bodies often conflate specialized climate-focused AI tools with massive, energy-intensive generative models developed by major technology firms. While the massive energy and water consumption associated with large-scale generative and agentic AI systems represent valid societal concerns, they bear little resemblance to the targeted, efficient machine-learning algorithms actually deployed to fight climate change.
Prof. Priya Donti adds that these public and institutional misunderstandings can distort funding priorities, potentially diverting critical resources away from the specific scientific applications that yield the greatest climate benefits.
Addressing Regional Inequalities and Emerging Risks
Despite its tremendous potential, AI also threatens to exacerbate existing inequalities within the global scientific community. Dr. Amirpasha Mozaffari warns that frontier models are built almost exclusively in the Global North, trained predominantly on data collected from regions with dense observation networks, and validated against benchmarks reflecting those same wealthy areas.
Simply providing access to cheap inference—allowing researchers in developing nations to run pre-existing models—does not solve the underlying imbalance. Without deliberate efforts to share data infrastructure, develop inclusive evaluation metrics, and ensure genuine international co-production of knowledge, AI risks automating and amplifying existing disparities in the global climate information system.
Dr. Carlos Rodriguez-Pardo suggests that if developed responsibly, AI-based Earth system models could eventually lower barriers to entry, enabling small research groups or agencies in data-poor regions to adapt pre-trained models on modest hardware. This is particularly crucial because the greatest vulnerability to climate change often coincides with the sparsest observational data and the widest resolution gaps.
Other experts emphasize the irreplaceable value of fundamental scientific rigor. Dr. Clare Eayrs cautions that the apparent success of AI systems must never be used as a justification to reduce investments in physical observations. Artificial intelligence is not a substitute for empirical data collection, physical understanding, or rigorous independent evaluation, particularly in data-scarce regions like Antarctic sea ice research.
Dr. William Collins echoes these concerns, noting that computer models will always provide an answer when asked, but researchers must continually verify whether that answer is physically valid rather than merely plausible. Ultimately, experts agree that artificial intelligence must serve to augment human judgment and democratic climate action, rather than replace the fundamental research, physical understanding, and equitable international cooperation required to address the global climate crisis.
