Predicting Climate Change with AI: "Providing Policy Analysis and Response Solutions"

An artificial intelligence (AI)-powered “climate research model” has emerged, enabling the prediction of climate change and the simultaneous analysis of its social and economic impacts to inform policy responses. The development is significant, as it suggests that AI could become a core technology providing solutions to climate change—a threat to humanity—beyond its role as a commercial tool.

(From left) Professor Jeon Haewon, Professor Kartik Mukavilli, Professor Oh Hyeyeon. KAIST

On May 13, KAIST announced that a joint international research team, led by Professor Jeon Haewon of the Graduate School of Green Growth and Sustainability, Professor Karthik Mukkavilli (joint appointment), and Professor Oh Hyeyeon of the School of Computing, together with Peking University in China, Imperial College London in the UK, Politecnico di Milano in Italy, University of Maryland in the US, and the International Institute for Applied Systems Analysis (IIASA) in Austria, has presented an AI-based “integrated climate research framework.”

Climate change is a complex issue that affects all aspects of humanity’s economy, energy, and industry, and cannot be limited to natural phenomena such as global warming. However, climate change research has thus far mostly been conducted separately in the areas of physical climate prediction, social and economic impact analysis, and energy policy assessment. Each area has used different data sets and analytical systems. For the same reason, integrating these approaches to inform policy decision-making has required considerable time.

To address these challenges, the joint research team proposed the “AI-Based Climate Research Foundation Model.” This model enables AI to understand and analyze various large-scale datasets—such as earth observation data, energy and economic scenarios, and policy indicators—in a unified manner within a virtual analytical environment.

This analytical approach allows for rapid and precise forecasts by considering both the physical phenomena of climate change and its economic and social impacts at the same time.

In particular, the team implemented a “Mixture of Experts (MoE)” structure within the model, enabling AI models with different roles to cooperate across domains. By integrating physics-based computational modules with statistical learning-based AI modules, the model enhances predictive accuracy and reliability. It also allows for swift analysis of the effects of greenhouse gas reduction targets or renewable energy expansion policies on industries and the economy.

This research is meaningful in that it proposes an “AI-based climate research direction” that goes beyond simply predicting climate change to simultaneously analyzing the social and economic impacts of policy changes.

Multimodal Composite Expert Foundation Model. KAIST

Furthermore, the KAIST research team has unveiled an AI-based predictive model that actually implements this framework. The model has been developed as a pilot “Energy–Greenhouse Gas Forecasting High-Speed Emulator.” The key feature is that AI now handles the complex calculations of the traditional “Integrated Assessment Model (IAM)”—which combines energy and carbon emissions—rapidly and efficiently.

Whereas conventional integrated assessment models required significant time and computational resources to analyze a single policy scenario, the AI model developed by the research team can analyze thousands of policy scenarios in a short period. This makes it easier to quickly predict the effects of carbon neutrality and energy transition policies and utilize these predictions for policy decision-making, according to the team.

Professor Jeon Haewon stated, “The Climate-AI model will serve as a bridge between climate scientists and policymakers. Most importantly, the ‘high-speed AI emulator’ released by the KAIST research team can be used as a core technology to provide climate response solutions through near real-time policy analysis.”

Professor Oh Hyeyeon emphasized, “In the future, AI technology must contribute to solving the climate crisis that threatens humanity’s survival, going beyond its role as a commercial tool. This international joint research demonstrates that AI can serve as a global public good in addressing complex social challenges.”

Meanwhile, master’s student Yen Shin was the first author of this research, with Professors Oh Hyeyeon and Jeon Haewon as co-corresponding authors. The results were recently published as a preprint in the journal Geoscientific Model Development, a leading scholarly publication in earth science model development.

This content was produced with the assistance of AI translation services.