2.1 Multimodel Ensembles of Climate Projections

The regional climate projection ensemble includes 55 scenario simulations for the RCP8.5 and RCP2.6 scenarios at 0.11° resolution over the common EURO-CORDEX domain using 8 driving GCMs and 11 RCMs. The details about model availability for each of the variables and indices for both the RCP8.5 and RCP2.6 scenarios are reported in Table 1.

The reader is also referred to Vautard et al. (2020) for model references. Also note that not all variables are available for all models, especially in the RCP2.6 projections. The second ensemble used for comparison consists of 12 GCM simulations (with 8 different GCMs) from CMIP5 (Taylor et al., 2012). These are limited to those simulations used as boundary conditions for the RCM ensemble assessed here (see Table 2). As a third ensemble, 12 simulations from CMIP6 (Eyring et al., 2016) are analyzed and intercompared over the European area.

The CMIP6 ensemble is reported in Table 3 and is based on the availability of data on the Earth System Grid Federation (ESGF) archive at the time when the analysis was completed. EURO-CORDEX and CMIP5 simulations share the same forcing scenarios, that is, RCP8.5 and RCP2.6 (Representative Concentration Pathway; see Moss et al., 2010) while the CMIP6 simulations use the SSP126 and SSP585 scenarios (Shared Socio-Economic Pathways, Riahi et al., 2017). The time slices used to compute the climate change signal with respect to the historical period 1981–2010 are 2041–2070 for the mid-future and 2071–2100 for the far future (end of century). The EURO-CORDEX ensemble results are presented on a regular 0.11° grid, the CMIP5 GCMs on a 2° grid and the CMIP6 on a 1° grid. These grids were chosen to be representative of the resolution of each ensemble. The robustness of the simulated changes of the three ensembles was evaluated by means of a significance test based on the dependent samples of a two-sided Student's t test.

3.1 Mean Temperature

Figure 1 shows the seasonal mean temperature change for DJF (winter) and JJA (summer), the mid-future and far future time slices in the RCP8.5 projections and the EURO-CORDEX, CMIP5 and CMIP6 ensembles.

In DJF all three ensembles show a well-known temperature change gradient, with the largest warming in the north and northeast and the weakest warming in the southwest and over the Atlantic. This warming pattern is to a large extent a result of the Arctic warming amplification, for which an important contribution is sea ice loss as well as the reduction of continental snow cover over vast northern areas leading to the snow albedo feedback (e.g., Dai et al., 2019; Walsh, 2014). Another contribution has been identified in the increased intrusion of low pressure systems, as explained in AMAP (2017). Figure 1 also shows differences between the ensembles in the amplitude of the change. In general, the CMIP6 ensemble shows a much stronger warming compared to the other two CMIP5-based ensembles, except for some areas over the Atlantic in winter. Another difference relates to the RCM ensemble which, in addition to more fine-scale details, shows weaker warming over large parts of the continent compared to the driving CMIP5-GCMs. This is in line, for example, with results from Sørland et al. (2018).

Figure 2 shows the box plot for the NEU and CEU regions of the area average change for each ensemble with colored bars representing the model spread between the 25th and 75th percentiles, while the black bars indicate the 5th, 50th, and 95th percentiles A projected median warming in winter of 4.5°C (2.5°C) is found for the RCM and CMIP5 ensemble in the far (mid-) future time slices, respectively, reaching values of 6°C (3.5°C) and 5.5°C (3.8°C) for CMIP6. Over the MED region a smaller DJF warming is projected, with slightly less than 3.5°C (2°C), for both EURO-CORDEX and CMIP5, and 4.5°C (2.5°C) and 6°C (4°C) for CMIP6. The projections for the 90% interval (range between the 5th and 95th percentiles) are generally less than 1.5° for the EURO-CORDEX and CMIP5 ensembles in all regions, and reach ~3° in NEU and MED for CMIP6.

The JJA warming has its maximum signal over the Mediterranean land regions with values above 4°C (2.5°C), 4.5°C (2.5°C), and 6.5°C (3.5°C) for RCM, CMIP5 and CMIP6, respectively, in the far (mid-) future (Figure 2). This result, which is in line with previous analyses (e.g., Giorgi & Coppola, 2010; Giorgi & Lionello, 2008) is likely connected with a marked decrease in spring and summer precipitation (as shown on Figure 3), increased soil drying and decreased cloud cover.

The NEU and CEU regions warm less in summer, with values in the far (mid-) future of 3.5°C (2°C), 3.6°C (2.3°C), 5.8°C (3°C), and 3.5°C (2°C), 4°C (2.3°C), and 6°C (3.4°C), respectively, for the EURO-CORDEX, CMIP5 and CMIP6 ensemble. The spread connected to the projections is lower for the first two ensembles compared to the CMIP6. The RCP2.6 scenario has a general warming between 1°C and 2.5°C for both seasons and for the three ensembles, with CMIP6 being at the upper end of the distribution. In summary, the warming projected by the EURO-CORDEX ensemble is comparable with that in the CMIP5 ensemble in DJF, while it is about 0.5° lower on average in JJA.

These findings confirm previous results based on a more sophisticated ANOVA analysis (Christensen et al., 2019; Déqué et al., 2007; Evin et al., 2019) from which most of the temperature change signal is largely influenced by the choice of GCM, while the RCM mostly dominates the change pattern in regions of complex topography, sea ice, soil moisture or snow. The difference between the RCM and GCM warming during summer also confirms the finding of Sørland et al. (2018) and Boé et al. (2020), that is, that CMIP5 GCMs warm notably more than RCMs even on a pair by pair basis (not shown). Possible explanations for this result are multiple. Some studies have at least partially attributed it to the lack of aerosol forcing in the RCMs, since aerosol concentrations over Europe are expected to decrease in the future as a result of stricter emission control measures (Boé et al., 2020; Gutiérrez et al., 2019; Nabat et al., 2020). Other reasons can be related to clouds (Bartók et al., 2017) or differences in the representation of plant physiological effects (Schwingshackl et al., 2019). The consistently higher warming values of the CMIP6 ensemble, along with the much higher uncertainty spread, is in line with the higher equilibrium climate sensitivity (ECS) of some of the members of CMIP6 (Forster et al., 2020; Zelinka et al., 2020), whose origin remains unclear and is under investigation.

Maximum (TXx) and minimum (TNn) temperature seasonal projections are shown in Figures 3 and 4, while the median and percentile values are reported as box plots in Figure 2. Both variables show a similar behavior as the mean temperature. Maximum winter warming occurs in the NEU and CEU regions, with a change of the TXx median value of about 4°C (2.5°C) (EURO-CORDEX and CMIP5) and 5.3°C (3–3.5°C) (CMIP6) at the far (mid-) century. In winter, TNn shows greater warming in NEU and CEU compared to TNx about 4.5°C (2.5°C) for EURO-CORDEX and CMIP5 and 7°C (4°C) and 5.5°C (4.5°C) for CMIP6 far (mid-) century, probably due to the snow albedo feedback. For the MED region, the TXx and TNx warming in DJF has values of 3.5°C (2°C) for the RCM and CMIP5 ensembles and 4.5°C (2.5°C) for the CMIP6 ensemble far (mid-) century time slice, with a very limited model spread (about 1°) for all the three ensembles probably due to the lower temperature variability over the ocean compared to the land. In summer, the changes in maximum and minimum temperature in the three European sub-regions have similar values (Figure 2). For NEU the median change value for both variables by the far (mid-) 2100 are around 3.5°C (2°C) for EURO-CORDEX and CMIP5 and 6°C (3°C) for CMIP6; TNn for CEU has a 3.5°C (2°C) median increase for the regional ensemble, while the GCM ensembles show stronger warming by 1° (0.5°) for CMIP5 and 2° for CMIP6 for the far (mid-) century time slice. TXx shows the same behavior. In the MED region, TXx warms by about 0.5° more than TNn, with median values of 4.5°C (2.5°C), 5°C (2.7°C), and 6.5°C (3.7°C) for CORDEX, CMIP5 and CMIP6 by the far (mid-) future, respectively.

3.2 Mean Precipitation

Figure 5 shows the seasonal mean precipitation change in percent of reference values for DJF and JJA. The well-known north-south dipole is evident, with positive changes in the northern part of the domain and negative changes in the southern part. The zero-change line is positioned in the southernmost location during winter over the Pyrenees, Alpine and Balkan Mountains, and shifts northward during the spring and summer season when it reaches the northernmost position placed over central eastern Europe and southern Scandinavia (Giorgi & Coppola, 2007).

For NEU and CEU in DJF, the EURO-CORDEX ensemble shows a median precipitation increase for the 2071–2100 (2041–2070) time slice of 15% (5%) and 20% (10%), respectively, with the smallest inter-model spread for the NEU and the largest over the CEU in all ensembles (see Figure 2). Change patterns are similar in the CMIP5 ensemble, with an amplitude of the spread about doubled over NEU and halved over CEU compared to the EURO-CORDEX one. The CMIP6 ensemble exhibits a 25% (15%) and 22% (12%) median value of increase over the two regions, respectively, and a spread similar to the CMIP5 one in NEU but greater in CEU, and lower than the RCM spread. In the Mediterranean region the three ensembles project a median negative DJF precipitation change slightly above 10% (5%), with a wider spread for the RCMs.

For the summer, over the NEU the three ensembles project an increase of about 5% for both the mid-future and far future time slices and for both the RCP8.5 and RCP2.6 scenarios, with a larger spread for the far future time slice compared to the mid-century one and a range of 20% between the 5th and 95th percentiles. Again, the RCM ensemble shows the widest spread. In CEU and MED all ensembles agree on negative changes of precipitation for far (mid-) century, spanning the range of −7% (4%) for EURO-CORDEX to −20% (−10%) for CMIP6 over CEU and from −22% (−10%) to −32% (−20%) in the Mediterranean. In all cases the CMIP5 ensemble has intermediate values between the EURO-CORDEX and CMIP6 ones. For the MED region far future projections, the range between the 5th and 95th percentiles is roughly 40% in the RCM ensemble and slightly lower (35%) for both CMIP ensembles.

3.3 Temperature Indices

3.3.1 Hot Extremes

Extremely hot days, as characterized by a daily maximum temperature larger than 35°C or 40°C, have a well-marked increasing frequency for RCP8.5/SSP585, particularly in regions where their occurrence is already significant in the reference period (see Vautard et al., 2020). In Southern Spain and low-lying areas of Italy and the Balkans, the number of days (35°C/40°C) is robustly projected to increase by more than 30 days/10 days in the mid-century and 50 days/20 days for the end of the century (Figure 6). This may have important impacts on agriculture and health, as these are typical critical thresholds above which these sectors are impacted.

3.4 Wet and Dry Extremes

3.4.1 Precipitation 99th Percentile, SDII, RX1day, and RX5day

Figure 7 shows the extreme precipitation indices P99 (often connected with the pluvial flooding hazard), RX1day, RX5day and the precipitation intensity index SDII for the mid-future and far future time slices in the RCP8.5 scenario. All three indices for all the RCM and GCM ensembles show an increase in extreme precipitation events, with a maximum by the far (mid-) century above the 20% (10%) median increase over the European continental region and Scandinavia (Figure 9 box plot). The most intense changes are shown by the EURO-CORDEX ensemble over the CEU and MED regions, while the CMIP6 ensemble shows the largest increases in the northeast uppermost corner of the NEU region.

The Rx1day extreme index exhibits the strongest positive change among the three indices, with only less than 5% non-significant negative changes limited to the Gibraltar Strait, Pyrenees, Morocco coasts and the Peloponnesus in Greece. Due to the higher resolution, the EURO-CORDEX ensemble indicates a pronounced increase in extreme 1 day precipitation over the Padania plain in Italy and in Tuscany. The land-sea contrast is quite evident between the Tyrrhenian and Adriatic seas and the Italian peninsula, with more extreme values over the ocean and in particular a maximum over the northernmost part of the Adriatic Sea where the Venice lagoon is located.

A minimum of the daily maximum precipitation increase (RX1day) is evident over all the European mountains, such as the Pyrenees, the Alps, the Apennines, the Carpathians and the Norwegian mountains, and a maximum along the U.K. and Ireland western coasts, Northern France and Germany. The same characteristics are also evident in the P99, RX5day, and SDII indices, although for the later the signal is somehow more damped. All these small-scale features are not resolved in the GCM change plots due to the coarser resolution which does not allow the GCMs to adequately resolve orographically forced circulation dynamical and thermodynamical effects and land surface feedback (Chan et al., 2014; Giorgi et al., 2016). The EURO-CORDEX projections have in general a wider spread for both the mid-future and far future time slices followed by CMIP5 and CMIP6, except for the NEU region where the CMIP6 have the highest projection uncertainty.

3.4.2 Drought Frequency Change per Decade and Consecutive Dry Day

The mean annual CDD and the drought frequency change per decade (DF) change are shown in Figure 8 for two time slices and for all ensembles. For both indices a significant maximum is observed over the Mediterranean region with the CDD being positive everywhere and the DF showing the well-known north-south dipole structure (Spinoni et al., 2014), that is, an increase of number of droughts in the south and a decrease in the uppermost northeast European areas and Scandinavia. The CDD spatial structure is quite similar among the three ensembles, with the CMIP6 having longer dry spells for a broader region covering the whole Mediterranean Sea up to the southwestern French coasts and the Alps. An increase in the number of droughts per decade in the Mediterranean is projected by all model ensembles, with a median value of 3 (1.8) for the RCM ensemble, 3.5 (2) for CMIP5 and 4 (2.5) for the CMIP6 ensemble by the far (mid-) century (Figure 9 box plot). The projection uncertainties are quite small (between one and two events per decade) as shown by the CMIP6 and EURO-CORDEX ensembles, respectively. Also, in the NEU region the three ensembles show a consistent (low uncertainty) decrease of the number of droughts between one and two events per decade by the end of century or around one event for the mid-century. The most uncertain region is the CEU, where the zero line is positioned, with a pronounced latitudinal difference between the RCM and CMIP6 ensembles. The EURO-CORDEX ensemble has the zero line placed in the southernmost location, going from the United Kingdom through Germany, Austria and the Balkans; the CMIP6 has the line just over southern Sweden and the Baltic Sea, while the CMIP5 zero line lies in between. This uncertainty is reflected in the box plot of Figure 9, where median values spanning from negative to positive are reported when moving from the regional model ensemble to the CMIP6 ensemble (CMIP5 is positioned around the zero-change values) for all the time slices.

3.5 Simulated Changes for Downward Surface Shortwave Radiation (W/m²)

Figure 11 shows the periods for the surface downwelling shortwave radiation (rsds) in both scenarios and ensembles.

The common behavior to all ensembles is a tripole pattern, with an increase in rsds over the Mediterranean area and Central Europe and a decrease in Northern Europe and the extreme South of the domain over the Sahara regions. This tripole is likely due to a tripole-like pattern in cloud cover change. Local values reach up to −10 W/m² over Northeastern Europe and more than +15 W/m² over the Balkans and Southeastern Europe. Note, however, that this tripole pattern is less visible in the CMIP6 ensemble for which the northern pole nearly disappears and the median change over the NEU zone is positive (see Figure 12). For the EURO-CORDEX RCMs and CMIP5 GCMs, the values and spatial patterns obtained in Figures 11 and 12 are in line with previous studies (Bartók et al., 2017, 2019; Boé et al., 2020; Gutiérrez et al., 2019).

An important feature of Figures 11 and 12 is the strong difference between the change signals of the three model ensembles over Central Europe, where the EURO-CORDEX ensemble median values show a decrease in the shortwave radiation reaching the surface for CEU and both periods, whereas both CMIP ensembles show a large increase. This is one of the rare cases of GCM-RCM inconsistency in the climate change response at large scale and confirms the results obtained with CMIP5 and EURO-CORDEX smaller ensembles by Bartók et al. (2017), Boé et al. (2020), Gutiérrez et al. (2019) and Nabat et al. (2020). It has been recently attributed to the different way aerosols are represented currently in GCMs and RCMs (Gutiérrez et al., 2019). In addition, Bartók et al. (2017) showed that cloud cover trends behave very differently in CMIP5 GCMs and EURO-CORDEX RCMs, with a decreasing cloud cover trend in GCMs and an absence of trends in RCMs.

To our knowledge, what differs most between RCMs and GCMs is the temporal evolution of the aerosol forcing (a general decrease of the aerosol load over continental Europe during the scenario period), which is taken into account in CMIP5 and CMIP6 despite large uncertainties, whereas it is not taken into account in most of the EURO-CORDEX RCMs. More specifically, only three RCMs (RACMO22E, ALADIN53 and ALADIN63) apply evolving aerosol forcing in the future, resulting in only 11 runs out of 55 for RCP85 and 5 out of 22 for RCP26.

Looking at individual RCM future responses, for example, for the RCP8.5 and the near-future, we note a strong increase (maximum value of +20 W/m²) over Central Europe and the Mediterranean for the ALADIN and RACMO22E runs regardless of the versions used and the driving GCM, along with a negative or weakly positive signal for all other RCMs, except for one of the RegCM4–6 runs (not shown). Plotting multimodel mean values for the RCM ensemble in Figure 11 therefore gives a large weight to RCMs without evolving aerosols. Note that the GCM-RCM discrepancy in rsds change also have implications when computing potential photovoltaic production in Europe (Gutiérrez et al., 2019).

It is also worth noting that the change response is largely stronger in CMIP6 than in CMIP5 models. Moreover, the choice of the socio-economic scenario and of the temporal horizon have weak impacts on the results. For example, for the end of the 21st century and the RCP2.6 and SSP126 scenarios, CMIP5 GCMs project an increase by +4 W/m² (median value) for the CEU region whereas CMIP6 GCMs project an increase by +7.5 W/m² and CORDEX RCMs a decrease by −2 W/m². The uncertainty in the future change of surface shortwave radiation is thus dominated by the choice of the model ensemble. Concerning the surface radiation change signal over the European seas, on average the EURO-CORDEX ensemble projects a decrease in surface shortwave radiation over the Baltic and North Seas and a close-to-zero change for the Mediterranean and Black Seas, whatever the scenario and the temporal horizon. Conversely, the CMIP5 ensemble projects nearly no change for the Baltic and North Seas and an increase for the Mediterranean and Black Seas, whereas the CMIP6 ensemble projects a clear and strong increase for all European Seas. These differences in the radiation change signal over the European Seas could have strong impacts on the regional sea physics and marine biogeochemistry scenario simulations, often driven by or coupled to Atmosphere-RCMs (Darmaraki et al., 2019; Gröger et al., 2019; Soto-Navarro et al., 2020). An interesting small-scale feature in the radiation change pattern is the specific signal obtained over the Alps with the CORDEX ensemble for the RCP8.5. Contrary to the GCMs, the RCMs simulate a contrast of response between the Alps (strong decrease in surface shortwave radiation) and the surrounding areas (weak signal or increase). This interesting contrast could be the signature of a potential local added value of the RCMs in future climate response as already shown for summer precipitation by Giorgi et al. (2016). It remains unexplained so far but could be linked to a stronger response of cloud cover over the Alps or to a locally lower influence of the aerosol load decrease.