WOW - a World model of Our World

Global climate change is leading to long-term changes in global and regional weather patterns, with profound local impacts on societies and ecosystems. Modern AI methods offer the potential to revolutionize modelling at every step of this process chain. WOW is exploring a so-called "AI world model" - an approach to couple these individual AI models along the entire process chain and across different spatial and temporal scales.

Initially, separate AI models are developed: AI versions of global climate models, AI weather models and local AI impact models (e.g. for forest fires). Each of these models learns a so-called internal "latent representation" into which it maps any input data. In world models, these latent representations are linked together in an innovative approach that allows the complex interaction of environmental data to be learned very efficiently.

The research questions include the transfer of the world model concept to climate and environmental sciences, the understanding of non-linear interactions between the atmosphere, water cycle and land surface as well as the improvement of AI methods with regard to their energy efficiency. These developments could motivate similar approaches in many other areas of the natural sciences.

The aim is to use the world model to provide climate and environmental information in unprecedented quality, speed and locality to enable optimal planning of climate adaptation measures and assessment of climate risks.