Mapping extreme hot temperatures in Europe and their evolutions: sensitivity to data choices

Electricity generation and demand is highly dependent on the weather conditions and especially temperature. Ongoing climate change has already modified the very hot extremes in Europe, and this is projected to continue in the future. The anticipation of the necessary adaptations in the electricity sector necessitates information on the possible extreme levels susceptible to occur in the next decades or further future periods. This study aims at comparing different ways of producing maps of extreme temperature levels for different future periods. Extreme temperatures are defined here as an example as 20-year Return Levels, that is temperatures reached or exceeded on average once in 20 years over the considered period. The computation of the Return Levels is based on the methodology described in Parey et al. 2019, which consists in applying the statistical extreme value theory to a standardized variable. It can be proven that the extremes of this variable can be considered as stationary. Then, the changes in mean and variance of the summer temperature projected by different climate models from the CMIP5 archive can be used to derive Return Levels for any selected future period.

Producing maps necessitates the use of a dataset with a large geographical coverage over Europe. Such datasets are typically gridded, either based on spatial interpolations of station records or on reanalysis products. However, both spatial interpolation and model assimilation tend to smooth the local highest values. Thus, in order to analyze the impact of such smoothing, the Return Levels computed in the same way from different datasets: the European Climate Assessment and dataset station data, the gridded EOBS database or the ERA5-Land database are computed and compared for different future periods.