An event-ensemble approach to investigate hourly extreme precipitation over Italy using convection-permitting reanalyses

Localized convective precipitation, characterized by short duration and high intensity, remains difficult to investigate using observations alone. Convection-permitting regional reanalyses provide valuable information on such events and their temporal evolution, particularly when adopting event-based approaches. Several convection-permitting reanalyses are available over Italy (e.g. SPHERA, MERIDA HRES, MORE, CHAPTER and VHR-REA_IT), obtained by dynamical downscaling of ERA5. They provide the opportunity to analyse extreme precipitation at hourly resolution and using an ensemble approach. Although a multi-model ensemble can improve precipitation average statistics, simple averaging of model fields tends to smooth intensity peaks, thereby degrading the representation of extreme events. To address this limitation, this study proposes an event-ensemble framework that combines information from multiple convection-permitting reanalyses without directly averaging precipitation fields.

First, extreme precipitation events are independently extracted from each reanalysis dataset. Event identification is based on the Extreme Rain Multiplier (ERM), defined as the ratio between hourly precipitation at a given grid point and the climatological mean of the RX1hour index (that is, the annual maximum 1-hour precipitation) at that location. This reference threshold is also compared with observational estimates derived from a high-resolution gridded hourly dataset (GRIPHO). The ERM provides a physically interpretable measure of event rarity and enables the classification of extreme events across datasets. In a second step, events identified in the different reanalyses are combined probabilistically within an event-ensemble framework. This approach allows uncertainty to be quantified through the spread in precipitation structures extracted from different reanalyses in terms of event timing and location, peak values, spatial extent.

The proposed methodology opens several perspectives for investigating extreme precipitation over Italy and is transferable to other regions where multiple convection-permitting reanalyses are available. First, it allows detailed investigation of individual exceptional events and their robustness across different reanalysis datasets. Moreover, it supports analyses of long-term variability and emerging changes in hourly extremes, including the detectability of such changes against background variability. Overall, this work demonstrates how the growing availability of convection-permitting reanalyses can be leveraged to study extreme precipitation at short timescales using a probabilistic, event-based and multi-dataset approach that explicitly accounts for model uncertainty while preserving the intensity of extreme rainfall events, offering a new framework for the assessment of regional extreme-precipitation variability and change in Italy.