Weather and climate extremes, such as intense precipitation, tropical and extratropical storms, drought, heatwaves, and compound events, pose substantial risks to society, infrastructure, and ecosystems across the globe. Yet directly and explicitly modelling such extremes and their future behaviour remains challenging, owing to the high computational cost of resolving relevant processes, incomplete understanding of many driving mechanisms, and the chaotic nature of atmospheric dynamics, which further limits predictability at longer lead times. Consequently, model-based projections often carry large uncertainties, especially in how complex and non-linear interactions among atmospheric variables shape these extremes in a future warmer world.

Observational constraints and statistical approaches offer a complementary pathway. Observations from ground-based networks, radars, and satellites, together with reanalysis simulations, can provide information on the characteristics and physical drivers that can be used to evaluate and constrain models. Statistical and data-driven models, when constrained by associated temperature, moisture, storm characteristics, and regional and global circulation teleconnections, can provide viable alternatives for problems that remain too computationally expensive, while incorporation of constraints in explicit physical simulations can help narrow uncertainties and reduce model biases.

This session aims to bring together research on the use of observational constraints, physical theory, and statistical approaches for improved understanding, prediction and projection of past and future extremes. We welcome contributions including, but not limited to, assessing the drivers of extremes, scaling relationships based on physical theory, storyline approaches, and statistical and hybrid approaches, as well as studies critically evaluating the physical basis and robustness of the proposed constraints. Contributions spanning various types of extremes, spatial and temporal scales, and observational and modelling frameworks are encouraged.