Heat extremes in subseasonal hindcasts: a General Extreme Value perspective

Understanding and characterizing temperature extremes is essential for assessing climate impacts and risks. Robust statistical analysis of such extremes requires large datasets, yet observational records often provide limited samples of rare events. Hindcasts, i.e., retrospective forecast model runs for past dates, are typically used to correct model biases, but their potential for extreme event analysis remains underexplored. Approaches such as UNSEEN (UNprecedented Simulated Extremes using Ensembles) have investigated the potential of seasonal hindcast ensembles to provide large samples of events that are physically plausible, particularly for assessing rare events. However, seasonal hindcasts often focus on monthly means.

In this study, we explore whether a similar approach can be applied to subseasonal hindcasts, evaluating their potential to serve as alternative realizations of extreme events at daily resolution.  We use two complementary methods to compare global temperature extremes in ECMWF subseasonal hindcast with ERA-5 reanalysis: (1) the statistical upper bound of daily 2-meter temperature, and (2) the probability of record-breaking daily 2-meter temperature. By leveraging existing subseasonal hindcast ensembles, we aim to evaluate whether these datasets can be repurposed to study temperature extremes that have not yet been observed but are plausible under current climate conditions.