The climatology of (severe) thunderstorm days is investigated on a pan-European scale for the period of 1979–2017. For this purpose, sounding measurements, surface observations, lightning data from ZEUS (a European-wide lightning detection system) and European Cooperation for Lightning Detection (EUCLID), ERA-Interim, and severe weather reports are compared and their respective strengths and weaknesses are discussed. The research focuses on the annual cycles in thunderstorm activity and their spatial variability. According to all datasets thunderstorms are the most frequent in the central Mediterranean, the Alps, the Balkan Peninsula, and the Carpathians. Proxies for severe thunderstorm environments show similar patterns, but severe weather reports instead have their highest frequency over central Europe. Annual peak thunderstorm activity is in July and August over northern, eastern, and central Europe, contrasting with peaks in May and June over western and southeastern Europe. The Mediterranean, driven by the warm waters, has predominant activity in the fall (western part) and winter (eastern part) while the nearby Iberian Peninsula and eastern Turkey have peaks in April and May. Trend analysis of the mean annual number of days with thunderstorms since 1979 indicates an increase over the Alps and central, southeastern, and eastern Europe with a decrease over the southwest. Multiannual changes refer also to changes in the pattern of the annual cycle. Comparison of different data sources revealed that although lightning data provide the most objective sampling of thunderstorm activity, short operating periods and areas devoid of sensors limit their utility. In contrast, reanalysis complements these disadvantages to provide a longer climatology, but is prone to errors related to modeling thunderstorm occurrence and the numerical simulation itself.

18The environments of severe and non-severe thunderstorms were analysed using 16 421 19 proximity soundings from December 2007 to December 2013 taken at 32 Central European 20 stations. The soundings were assigned severity categories for the hazards hail, wind, tornado 21 and rain. For each of the soundings, parameters were calculated representing the instability, 22 vertical wind profile and moisture of the environment. The probability of the various hazards 23 as a function of CAPE and 0-6 km bulk shear (DLS) is quite different for each of the hazards.24Large hail is most likely for high CAPE and high DLS, a regime that also supports severe 25 wind events. A second severe wind regime exists for low CAPE and very high DLS. These 26 events are mostly cold season events. Storms with significant tornadoes occur with much 27 higher DLS than storms with weak or no tornadoes, but with similar CAPE. 0-1 km bulk 28 shear (LLS) does not discriminate better than DLS between weak and significant tornadoes. 29Heavy rain events occur across a wide range of DLS, but with CAPE above the median for 30 non-severe thunderstorms and are most likely when both absolute humidity in the boundary 31 layer and relative humidity in the low-to mid-troposphere are high. LCL height does not 32 discriminate well between the intensity categories of tornadoes, but higher LCL heights were 33 associated with higher probability of severe hail. Storm relative helicity shows similar results 34 to DLS, but with more overlap among intensity categories. 35 36 37

The occurrence of environmental conditions favorable for severe convective storms was assessed in an ensemble of 14 regional climate models covering Europe and the Mediterranean with a horizontal grid spacing of 0.448. These conditions included the collocated presence of latent instability and strong deep-layer (surface to 500 hPa) wind shear, which is conducive to the severe and well-organized convective storms. The occurrence of precipitation in the models was used as a proxy for convective initiation. Two climate scenarios (RCP4.5 and RCP8.5) were investigated by comparing two future periods (2021-50 and 2071-2100) to a historical period for each of these scenarios. The ensemble simulates a robust increase (change larger than twice the ensemble sample standard deviation) in the frequency of occurrence of unstable environments (lifted index # 22) across central and south-central Europe in the RCP8.5 scenario in the late twenty-first century. This increase coincides with the increase in lower-tropospheric moisture. Smaller, less robust changes were found until midcentury in the RCP8.5 scenario and in the RCP4.5 scenario. Changes in the frequency of situations with strong ($15 m s 21) deep-layer shear were found to be small and not robust, except across far northern Europe, where a decrease in shear is projected. By the end of the century, the simultaneous occurrence of latent instability, strong deep-layer shear, and model precipitation is simulated to increase by up to 100% across central and eastern Europe in the RCP8.5 and by 30%-50% in the RCP4.5 scenario. Until midcentury, increases in the 10%-25% range are forecast for most regions. A large intermodel variability is present in the ensemble and is primarily due to the uncertainties in the frequency of the occurrence of unstable environments.