Climate Knowledge in Large Language Models

About this episode

This episode discusses the limitations of large language models in accurately predicting climate data and the importance of geographic context.

Climate Knowledge in Large Language Models Kuznetsov, I., Grassi, J., Pantiukhin, D., Shapkin, B., Jung, T., & Koldunov, N. (2025). Alfred Wegener Institute, Helmholtz Centre for Polar and Marine Research. LLMs have an internal "map" of the climate, but it is fuzzy: Without access to external tools, Large Language Models (LLMs) can recall the general structure of Earth’s climate—correctly identifying that the tropics are warm and high latitudes are cold. However, their specific numeric predictions are often inaccurate, with average errors ranging from 3°C to 6°C compared to historical weather data. Location names matter more than coordinates: The study found that providing geographic context—such as the country, region, or city name—alongside coordinates reduced prediction errors by an average of 27%. This suggests models rely heavily on text associations with place names rather than possessing a precise spatial understanding of latitude and longitude. Performance struggles with altitude and local trends: Models perform significantly worse in mountainous regions, with errors spiking sharply at elevations above 1500 meters. Furthermore, while LLMs can estimate the…

People in this episode

Host: Amirpasha

Topics covered

• climate knowledge
• large language models
• prediction accuracy
• geographic context
• temperature change
• scientific caution

Keywords

• climate
• large language models
• prediction errors
• geographic context
• temperature dynamics

Mentioned in this episode

Organizations: Alfred Wegener Institute, Helmholtz Centre for Polar and Marine Research

More episodes of Earthly Machine Learning

• Aligning artificial intelligence with climate change mitigation · May 9, 2026 · 19 min
• Machine learning for the physics of climate · May 3, 2026 · 20 min
• Atmospheric Transport Modeling of CO2 With Neural Networks · April 27, 2026 · 21 min
• On the foundations of Earth foundation models · April 20, 2026 · 18 min
• Whose weather is it? A fairness framework for data-driven weather forecasting · April 14, 2026 · 22 min
• Learning predictable and informative dynamical drivers of extreme precipitation using variational autoencoders · March 7, 2026 · 18 min