Our ClimateAI Team focuses on the exploration and application of deep learning and Artificial Intelligence (AI) tools on climate data. Our primary research topic is foundation models – deep learning models that are trained on large datasets, are multi-modal, and task-agnostic. Over the past few years, these models have revolutionised weather forecasting and have outperformed classical, physically-based numerical weather prediction systems across various tasks. However, extending their application to climatic timescales and various atmospheric feedback processes has not been sufficiently explored. By leveraging a wide range of petabyte-scale data that supports multiple grids and resolutions, our ClimateAI Team aims to build a comprehensive foundation model for climate science capable of achieving atmosphere-ocean-sea-ice coupling across seasonal to decadal timescales. We aspire for this model to accurately and efficiently facilitate future climate predictions, revealing the causes of climate change and the effects of global warming.
Through our excellence and close collaboration with climate researchers and AI experts from various institutions within the Helmholtz Association, we aim to ensure a strong interdisciplinary approach, advance climate research, offer new insights into the Earth’s climate system, and support decision-making in climate policy. Building on the success of the AtmoRep initiative, our contributions are part of the HClimRep Project and the Helmholtz Association Initiative (HFMI).
Team Goals
- Develop a reliable AI foundation model as a baseline for novel climate simulations.
- Achieve atmosphere-ocean-sea-ice coupling by using a wide range of available data.
- Employ petabyte-scale data on multiple grids and resolutions.
- Introduce reliable seasonal-to-decadal AI-based predictions using novel rollout techniques.
- Simulate future climates by learning forcing factors, such as greenhouse gas concentrations.
- Investigate downstream tasks related to counterfactual scenarios and extreme events.
Projects
Members
Last Modified: 15.04.2025