Projected changes in sub-daily extreme precipitation: comparing temperature-scaling approaches and convection-permitting models across an Alpine gradient

• Optimal scaling 5.5–8.4% K −1 varies with event duration and return period. • High-elevation basins show strong CPM–scaling agreement; lowlands weak links. • Seasonal shifts degrade scaling reliability, especially for longer-duration storms. • Event-based temperature outperforms mean annual temperature in T-P scaling. Quantifying future changes in sub-daily extreme precipitation in mountainous basins is essential for adapting to increasingly intense flash floods and debris flows. Often, extreme precipitation-temperature scaling rates are used to anticipate future extremes. In this study, we evaluate the reliability of temperature-scaling approaches by comparing their projections of extreme precipitation to outputs from convection-permitting models (CPMs) across a complex mountainous region in northeastern Italy. We use hourly data from an ensemble of five CPMs at a 3 km resolution, comparing historical (1996–2005) and future (2090–2099, RCP8.5) periods and using average temperatures in the 24 h prior to the precipitation event to compute the scaling rates. Our analysis employs a robust methodological framework to identify optimal scaling factors that replicate the extreme precipitation estimates produced by CPM simulations. We find that the optimal scaling rates vary with duration and return period. For moderate extremes (2- and 5-year return period), the optimal scaling rates are typically between 5.5 and 6% K −1 , while more severe events scale with higher rates, up to 8.4% K −1 for 3‐hour, 100‐year events. Future shift in precipitation seasonality also plays a role. In high‐elevation areas, where extremes are prevalent in the summer in both present and future conditions, estimates based on temperature-scaling rates align closely with CPM outputs. Conversely, poorer performances are found in areas where large shifts in the seasonality of extremes are expected. Overall, careful consideration of local variability, seasonality, duration, and return period is necessary to avoid potential inaccuracies when deriving projections from temperature-scaling relationships.