Informal Seminar: Exploring the predictability of Eurasian extreme events: a comparative study of dynamic and AI models

Relaxation experiment, which nudges model forecasts toward observed conditions, is an effective method for identifying sources of predictability. Using this technique, the predictability of extreme events in Eurasia with both dynamic and AI models was investigated. Results indicate that that forecast skill for extreme events over Eurasia can be improved through Arctic relaxation, although the improvements are less significant in AI models compared to dynamical model. This difference may be due to AI models’ weaker representation of the Arctic-Eurasian linkage. Overall, the linkages between different latitudes are weaker in AI models compared to dynamic models, particularly the link between the tropics and Eurasia. This might be attributed to the lower variability in the AI models in the tropics. Finally, the impact of AI model architecture and potential directions for future predictability research with AI models will be presented.