ABSTRACT Despite the scarcity of comprehensive studies at a global scale, many regional analyses report increases in extreme hourly precipitation values. The growing interest in assessing trends in extreme hourly precipitation has outpaced the development of new statistical tools tailored to their features. Typical analyses employ Extreme Value Theory (EVT) in a regression framework, where the parameters of EVT distributions are modeled as functions of exogenous covariates. While this approach is common in daily precipitation analyses, it may be less suitable for the higher‐frequency hourly data. We propose a dynamic EVT approach, where the distribution parameters evolve according to an autoregressive‐type dynamic. Our model accommodates the high‐frequency nature of the data without requiring arbitrary choices regarding the covariates as in the regression approach. Applied to a group of Midwest US cities experiencing increases in hourly extreme precipitation, our method reveals dynamics in extreme high quantiles and outperforms a reference model in the EVT regression approach.
Dupuis et al. (Thu,) studied this question.