Extreme events cause severe losses to both society and the environment. Although their simulation and prediction have long relied on physics-based models, progress remains constrained by high computational cost and limitations in process parameterization. Recent advances in machine learning offer promising solutions, but challenges remain in physical consistency, representational fidelity, and interpretability. Hybrid physical–machine learning models provide a path forward for improving the simulation, prediction, and projection of extreme events and their impacts. Realizing this potential, however, requires advances in identifiability, interpretability, and uncertainty quantification to ensure robust and trustworthy predictions. We welcome contributions on hybrid modeling methods, explainable and uncertainty-aware approaches, causal and attribution analyses, and applications to diverse extreme events and their impacts. This session aims to bring together researchers from physical modeling and machine learning communities to advance the next generation of extreme event prediction and understanding.

• A - Atmospheric Sciences
• B - Biogeosciences
• GC - Global Environmental Change
• NH - Natural Hazards