Modeling the Earth system with an AI world model

Forest fires, floods or droughts: artificial intelligence (AI) from the Karlsruhe Institute of Technology (KIT) is set to help predict such events worldwide more precisely, faster and more energy-efficiently than conventional physics-based simulations on supercomputers. In the "WOW - a World model of Our World" project, researchers are developing an AI world model that efficiently links various AI models to simulate climate, weather and the environment. The Carl Zeiss Foundation is funding the project with almost six million euros.

Global climate change is changing the weather and the environment in the long term - both worldwide and locally. The consequences range from more frequent extreme events to far-reaching effects such as the collapse of entire ecosystems. "Numerical climate, weather and environmental models can now be used to model many aspects of these developments relatively consistently on different time and space scales," says tenure-track Professor Peer Nowack from the Institute for Theoretical Computer Science at KIT and coordinator of the project. "Modern AI methods, in turn, offer the potential to either emulate such models cost-effectively or even learn correlations directly from observational data. As a result, they often achieve better results in weather forecasting, for example. This offers great opportunities for environmental modeling as a whole."

In the "WOW - a World model of Our World" project, Nowack and seven other KIT researchers are now going one step further: they are investigating how several such AI models for different processes in the Earth system can be coupled into a single AI system across spatial and temporal scales. To do this, they want to pursue an approach from computer science that makes it possible to link the various models - to create an AI world model.

How the AI world model is created

With WOW, the scientists are developing methods that can link different AI models together. These include emulators for global climate models, AI-based weather models and models that depict local phenomena such as forest fires or flooding. The aim is to link these previously separate systems into a continuous process chain in order to depict the local effects of global changes even more consistently. In order to make this progress possible, new developments in AI methodology are needed in particular.

By linking the different models, the researchers also want to better understand the interplay between the atmosphere, water and land surface. "We are interested in how changes in one part of the Earth system affect other areas - for example, how droughts or changes in cloud formation trigger feedbacks in the climate," says Professor Almut Arneth from the Institute of Meteorology and Climate Research - Atmospheric Environmental Research, KIT's Campus Alpin in Garmisch-Partenkirchen, who is also involved in the research project. "It is possible that we can make previously hidden connections of climate change visible."

Also relevant for other fields of knowledge

In the medium term, the new AI world model could already help to better assess risks - and make well-founded decisions on climate protection and adaptation. "In the future, our methods could also be used in other natural sciences in which complex systems are modeled," explains Dr. Markus Götz from the Scientific Computing Center at KIT, another scientist involved in the project. "If we learn to couple AI models efficiently, we can understand complex relationships faster and more precisely. We can also integrate observational data much more directly into the modeling process. Overall, this offers great opportunities for science." The Carl Zeiss Foundation is funding the WOW project for five years with a total of six million euros.

Further information on the project

Details on the KIT Center Climate and Environment

Details on the KIT Center Information ˑ Systems ˑ Technologies