Climate modelling of precipitation extremes in NE Italy: biases and uncertainty

Accurate modeling of extreme precipitation in regions with complex topography, such as Northeastern Italy, remains a major challenge for hydrological risk assessment due to persistent model biases and uncertainties. This thesis investigates the biases and uncertainties associated with sub-daily precipitation extremes derived from an ensemble of high-resolution climate models over Friuli Venezia Giulia. The analysis focuses on mean annual maxima (AM) and 20-year return levels (RL20) for precipitation durations ranging from 1 to 24 hours. Climate model simulations are evaluated against observations from 79 regional rain gauges using statistical, spatial, and elevation-dependent approaches. The results reveal a systematic underestimation of precipitation extremes across all durations and elevation classes. The strongest negative biases and highest uncertainties occur for short-duration precipitation events (1 h and 3 h), particularly in mountainous regions. In contrast, model performance improves progressively with increasing precipitation duration, with the best agreement observed for 12 h and 24 h events, indicating a stronger ability of the models to reproduce large-scale synoptic precipitation compared to localized convective storms. The analysis further demonstrates that 20-year return levels exhibit substantially larger negative biases and greater inter-model uncertainty than mean annual maxima, highlighting the increasing difficulty of reproducing rare and high-intensity precipitation extremes. Although biases generally decrease with increasing duration, significant inter-model spread persists across all temporal scales and elevation classes, indicating considerable structural uncertainty within the climate model ensemble. Overall, the findings highlight the limitations of current high-resolution climate models in representing sub-daily convective precipitation extremes in complex Alpine terrain. The study emphasizes the importance of multi-model ensemble approaches, explicit uncertainty quantification, and bias correction procedures when applying climate model outputs to hydrological design, flood risk assessment, and climate impact studies.