AI Generalizability for Non-stationary Environmental Regimes: The Case of Hydro-climatic Extremes

A robust model generalisability is considered a fundamental challenge in artificial intelligence. A model is deemed robust if it continues to perform reliably even when confronted with previously unseen data. Environmental conditions are constantly changing and therefore require equally robust models to ensure dependable predictions.

The GENAI-X project focuses on climatic extremes such as floods, landslides, droughts, and late frost, as well as their impacts on ecosystems. Its goal is to improve the understanding and prediction of these and similar environmental phenomena. To achieve this, AI methods are being developed that can adapt to changing data patterns and uncertainties.

Methods such as hybrid and causal modelling, equation discovery, dimensionality reduction, and uncertainty quantification are employed. By combining the strengths of different modelling approaches, investigating cause–effect relationships, and identifying mathematical equations that describe complex environmental systems, these approaches enable a comprehensive analysis and more nuanced predictions of environmental phenomena. Through simplified visualisation and the assessment of uncertainties, complex model structures are made interpretable, facilitating balanced decision-making.

The project is based at the ELLIS Unit (European Laboratory for Learning and Intelligent Systems) in Jena and represents a collaboration between Friedrich Schiller University Jena, Jena University Hospital, the Max Planck Institute for Biogeochemistry, and the Senckenberg Institute for Plant Diversity.