How well do convection-permitting climate models represent sub-daily precipitation upper tail in complex orography?

Convection‐permitting climate models (CMPs) give a much more realistic representation of sub-daily precipitation statistics compared to coarser resolution climate models, thanks to the explicit representation of convection. Their higher spatial and temporal resolution allows to used them directly to study future changes in the frequency, intensity, and spatiotemporal patterns of heavy rainfall over complex terrain. However, the high computational requirements of CPM runs restricts the existing simulations to relatively short time periods (10–20 years), too short for deriving precipitation frequency analyses with conventional approaches. Alternative methods, based on the so called Metastatistical Extreme Value Distribution, were recently proposed (e.g. Marani and Ignaccolo, 2015) for deriving frequency analyses from shorter data records, promising improved applications based on CPMs. These approaches rely on the concept of ordinary events, which are all the independent events that share the statistical properties of extremes: once the upper tail of the distribution of ordinary events is known, it is possible to derive an extreme value distribution by explicitly considering their yearly occurrence frequency.

Here, we investigate the CPM ability to represent the upper tail of sub-daily precipitation in a complex-orography region in the Eastern Italian Alps. In this area, different orographic impacts on sub-daily precipitation upper tail were reported at different durations (Formetta et al., 2021), and significant temporal trends in their intensity were reported during the last few decades (Libertino et al., 2019), making it a challenging and interesting test case for CPM simulations. As CPM we used the COSMO model run at 2.2 km resolution over Europe, driven with ERA Interim for the period 2000-2009. We use 180 rain gauges to benchmark the CPM simulation. CPM time series are extracted for the grid points corresponding to the rain gauges, and hourly time series are created from both stations and CPMs. In each time series, independent storms are separated by 24-hour dry hiatuses, and ordinary events for 9 durations between 1 and 24 hours are defined as the corresponding peak intensity of each storm. Ordinary events upper tails are modeled using a Weibull distribution (two-parameter stretched exponential), which was previously reported to well reproduce the statistics of extremes in the area. The ability of CPMs to reproduce the model parameters and extreme quantiles up to 100-year return period, and their dependence on elevation are evaluated, together with the dependence of the biases with elevation. A general overestimation is found for annual maxima (10-40%), and the estimated quantiles (10-60%), especially for short durations. The bias significantly depends on elevation, with increasing overestimation of the 1-hour quantiles with elevation. It seems that CPMs cannot represented well the “reversed orographic effect” reported by previous studies.