Object- and event-level representation of extreme events for machine learning

Develop a proper object- or event-level representation of individual extreme episodes that captures onset, duration, spatial propagation, and morphology in machine-learning-compatible label structures.

Background

The EastAsiaClimateExtremes dataset provides grid-point-level weekly extreme labels and event-based metrics for anomalously high temperature, heavy rainfall, and marine heatwaves over East Asia. Although these products summarize extreme characteristics within regular temporal windows and include a temporally regular extremeness index, they do not fully represent individual episodes as spatially coherent objects or events.

The unresolved issue is to create labels that encode the onset, duration, spatial propagation, and morphology of individual extreme events in a form suitable for artificial-intelligence applications. Such a representation would extend the current grid-point framework toward object-based analysis and improve the ability of machine-learning models to learn the structure and evolution of extreme episodes.