10-year return levels of sub-daily precipitation over Europe and their internal variability

Abstract:

Sub-daily extreme precipitation events can cause high societal and economic impact inducing flash floods, urban flooding, riverine flooding and landslides. Hence, public authorities, civil security departments and engineers need information about the frequency and intensity of these events. However, observational coverage of sub-daily rainfall measurements is sparse, and data is often not publicly available. Therefore, we provide an impact-orientated data set of 10-year rainfall return levels based on climate model simulations over Europe. We apply a high-resolution single-model large ensemble (SMILE) of the Canadian Regional Climate Model version 5 (CRCM5) with 50 members for the time period 1980–2009 in order to calculate 10-year rainfall return levels for hourly to 24-hourly durations. The quality is evaluated by comparison to a large, newly compiled, observation-based data set of 16 European countries from 32 different sources. For the area covered with observations, rainfall return levels of the CRCM5 reproduce the general spatial pattern of extreme precipitation with centred Pearson product-moment coefficients of linear correlation > 0.7 for all sub-daily durations. The biases of the rainfall intensity amount to -19.3 % (hourly) to +8.0 % (24-hourly) averaged over the study area. Especially for durations between 3 and 24 hours, the return levels based on the CRCM5 show good agreement with observations. However, for hourly duration and topographically complex regions, we argue that higher-resolution convective-permitting climate model simulations could improve the results.

As the 50 CRCM5 members only differ due to slightly perturbed initial conditions, their spread can be interpreted as representation of the internal variability of the climate system. For the return levels of the applied 50 members, this internal variability results in a range of -15 % to +18 % around the median.

The 10-year return level data are publicly available (https://doi.org/10.5281/zenodo.3878887) at the spatial resolution of the climate model, which equals 12.5 × 12.5 km².