According to the analysed EURO-CORDEX ensemble, we will see increasing air temperature in the Baltic Sea area during the present century. According to this ensemble, it is a robust result for all seasons, locations, simulations, and scenarios.

For both seasons analysed (winter and summer), the temperature change shows spatial gradients with the strongest warming in the northeast (Fig. 1). Winter warming is larger than summer warming and larger than the global average warming of about 3.7 % (Table 2); in the northeast, it approaches twice the global average warming. Larger warming than the global average is generally expected for land areas, since land heats more quickly than sea areas where also enhanced evaporation tends to reduce warming (e.g. Sutton et al., 2007); it is most clearly seen in winter in the eastern part of the area. The strong winter increase is also influenced by the feedback mechanisms involving retreating snow and sea ice. There is a general pattern of higher warming in the north than in the south, but there is a spread in the magnitude of the change. This is illustrated in the columns of the figures below. As only eight GCMs have been used for these RCP8.5 RCM experiments, the spread between quartiles could be lower than what would have come from an exhaustive downscaling of all CMIP5 global simulations; Kjellström et al. (2016) compared nine GCMs, including the eight GCMs analysed here, to 25 other CMIP5 GCMs and found the nine-member-ensemble spread over Sweden to be comparable in summer but smaller than that in the larger GCM ensemble in winter.

Earlier studies have shown that the increase in winter temperatures is strongest for the coldest episodes (Kjellström, 2004) as well as for extreme daily maximum and minimum temperatures (Kjellström et al., 2007; Nikulin et al., 2011). There is a significant decrease in the probability of cold temperatures (Benestad, 2011). Warm summer extremes are projected to become more pronounced; for example, Nikulin et al. (2011) used an ensemble of six RCM simulations, all downscaling GCMs under the SRES A1B scenario; the data indicate that warm extremes with a present-day (1961–1990) return period of 20 years will be reached 4 times as often in Scandinavia by 2071–2100, with a frequency around once every 5 years in Scandinavia by 2071–2100.

Summer warming in the Baltic Sea basin is smaller than winter warming, and it is relatively homogeneous across the area. A tendency is seen for larger warming over land areas in the northernmost parts of the Baltic Sea basin. These areas are closest to the northern rim of Scandinavia and the Kola Peninsula where warming in summer is as high as that projected for parts of southernmost Europe (Kjellström et al., 2018). In the northeastern part of the region, a large warming may be related to the larger temperature increases further to the north in the Arctic, potentially connected with the ice–albedo and other feedback mechanisms (IPCC, 2021). The strong warming in the southeastern part of the Baltic Sea basin is related to the large-scale pattern of warming in Europe, where the strongest summer warming is seen in southern Europe. Similar results for other GCM–RCM combinations have been reached in, e.g. Christensen and Christensen (2007), Kjellström et al. (2011), and Vautard et al. (2014). A potential source of difference between GCMs and RCMs is the different treatment of aerosols in these models. Many of the RCMs do not include time-varying anthropogenic aerosols leading to weaker future warming compared to GCMs (Boé et al., 2020). The EURO-CORDEX-based results are consistent with the RCM results for 2021–2050 in Déqué et al. (2012). This study found that there is a significant temperature response, even for the relatively short-term 2021–2050 time frame, even though the total uncertainty related to the choice of model combination (GCM–RCM) and sampling (natural variability) is large. Similarly, Kjellström et al. (2013) showed early emergence already in the first few decades of the 21st century of trends in both winter and summer temperature despite large natural variability as represented in the ENSEMBLES RCM projections used in BACC II.

Corresponding changes in the daily minimum temperature and daily maximum temperature (not shown) have the same patterns as the average temperature change, with the expected larger magnitude of warming for minimum temperature. A range of factors may be responsible for this decrease in difference between minimum and maximum temperatures. This could involve changes in the diurnal temperature range (e.g. Lindvall and Svensson, 2015) or changes in the synoptic weather variability in combination with reduced large-scale temperature gradients between the Atlantic Ocean and the Eurasian continent (IPCC, 2021).

The multi-model EURO-CORDEX ensemble relative precipitation change for winter and summer is shown in Fig. 2. The ensemble is the same as that in Fig. 1.

During winter, the relative increases are quite homogeneous, although there are large differences between the lower and upper quartiles. These differences are largest west of the Baltic Sea catchment (Norway) where the amount of precipitation is particularly sensitive to different changes in the large-scale circulation. For summer, there is a clear pattern of more positive change in the north versus less positive change in the south. As expected, winter increases are projected to be larger than those in summer. Roughly, the winter increase is 25 %–35 % over most of the area in the median, and the summer increase is 15 %–25 % for the northern part of the area. This is consistent with the AR5 Climate Atlas, where median increases of precipitation in the area are 10 %–20 % for the winter half year and 5 %–10 % for summer, as these results correspond to the RCP4.5 scenario with around 2.5 % of warming for the periods mapped, whereas the EURO-CORDEX results correspond to a global warming of 3.8 ∘C.

For summer, there is disagreement on the sign of climate change for most of the southern half of the area, indicated by the masked-out area defined as regions where at least 25 % of the models disagree on the sign with the majority. Since the period mapped here consists of the three summer months of June–August, whereas the AR5 Climate Atlas maps April–September, a comparison of the position of the no-change area is difficult. In an analysis of the older ENSEMBLES simulations (Déqué et al., 2012), almost all land points in the Baltic Sea region showed significantly positive summer precipitation signals.

This general picture of change is not surprising. Climate models generally project the global hydrological cycle to become more intense (e.g. Held and Soden, 2006). For Europe, this corresponds to increasing precipitation in northern Europe and decreasing precipitation in southern Europe, both in winter and summer (Christensen et al., 2007). Between these areas of projected increase and projected decrease, only small changes or changes in different directions are projected (see, e.g. Kjellström et al., 2011). The location of the transition zone depends on the season and is located farther to the south in winter than in summer. In summer, this zone shifts into the Baltic Basin: winter precipitation is projected to increase over the entire Baltic Sea catchment, while summer precipitation is mostly projected to only increase in the northern half of the basin. In the south, precipitation change is small for the ensemble mean, and there is a large spread between different models with both increases and decreases. Basically, both increases and decreases are possible in the future.

In Fig. 3, we show scatter plots where the relative change between 1981–2010 and 2071–2100 of precipitation is plotted against the corresponding change of temperature for each model and each scenario. Ensemble means for the three scenarios are indicated by the three larger symbols. This calculation has been performed for various subsets of the Baltic Sea catchment (see Fig. 1): the entire region; only land points; only sea points; only land points north and south of 60∘ N, respectively.

There is a strong correlation between temperature and precipitation in winter with significant regression slopes of around five percentage points per degree and squared correlation coefficients of 0.5 to 0.6 depending on the sub-area. This is an indication of an approximate common sensitivity of precipitation change to local temperature change. This correspondence breaks down for summer, where the plots contain much more noise, indicating large model-dependent influences on the precipitation signal. The north–south gradient in summer precipitation change is apparent in the model averages (compare the northern and southern land point plots), but the inter-model spread is large.

Due to the roughly 20 % higher average global warming in the current RCP8.5 ensemble than in the GCMs underlying BACC II (see Table 2), we would have expected general climate change to be around 20 % larger for EURO-CORDEX RCP8.5 than those presented in BACC II. It is noteworthy that this difference is not generally seen in Fig. 3, where we have plotted temperature and precipitation change for the BACC II simulations (BACC II Author Team, 2015) along with the three scenarios of the present analysis. The BACC II results correspond to the RCP8.5 results both with respect to temperature and precipitation change apart from land areas in summer where the BACC II change is only about 80 % of the RCP8.5 result (+6.5 % vs. +8.2 %).

In Christensen et al. (2019), a thorough comparison of change patterns of mean temperature and precipitation has been performed for the PRUDENCE simulations behind the first BACC report (BACC Author Team, 2008), the ENSEMBLES simulations behind the second report (BACC II Author Team, 2015), and the EURO-CORDEX data behind the present report. This analysis used patterns of change scaled with global temperature change and is therefore useful for pinpointing differences between the BACC reports extraneous to the variations of general scenario strength, i.e. differences in local sensitivity and/or change patterns apart from those due to differences in emission scenarios. The most important differences between BACC II and the current simulations are a slightly reduced winter warming per unit of global warming in EURO-CORDEX compared to BACC II; a smaller wintertime precipitation increase but a slightly larger increase of summer precipitation over the Baltic Sea. These conclusions do not contradict the results from Fig. 3, since a scaling with global warming would increase both local precipitation and local temperature changes for the BACC II ENSEMBLES results relative to RCP8.5.

The water-holding capacity of the atmosphere increases with increasing temperature. Therefore, precipitation extremes are projected to increase with climate warming (e.g. Lenderink and van Meijgaard, 2010). Several studies, some of which are described in the following, indicate that extreme precipitation is likely to increase in the future, even in areas and seasons, where the average precipitation does not increase. One example is the IPCC Special Report on extreme events (Seneviratne et al., 2012), where it was shown that higher extremes of precipitation consistently show larger increases than lower extremes, and higher increases than averages.

Already simulations from the PRUDENCE project (Christensen and Christensen, 2003) showing a considerable decrease in average summer precipitation in large parts of southern Europe at the same time showed an increased probability of very extreme precipitation in that area as well as in the north, where average precipitation was not projected to decrease. Quite generally, more intense precipitation can be expected on all timescales, from single rain showers to synoptic-scale precipitation.

Nikulin et al. (2011) investigated an ensemble of RCM simulations following the SRES A1B scenario with the RCA model; they showed that the 20-year return value of precipitation extremes in Scandinavia in the period 1961–1990 was projected to decrease to 6–10 years in 2071–2100 for summer over northern Europe and to 2–4 years in winter. Similarly, Larsen et al. (2009) analysed a high-resolution RCM integration and reported that the return period for 20-year rainfall events at hourly duration decreased to about 4 years for Sweden.

Collected results from 90 of the models from the EURO-CORDEX project are illustrated in Fig. 4, along with results from the coupled models discussed below. For data availability reasons at the time of writing, not all simulations have been analysed for extreme precipitation. We will here use the 10-year return value as representative of extreme precipitation. This is defined as the daily precipitation amount, which is only exceeded once every 10 years on average. The model-median signal has a consistently positive sign across the domain for the areas where more than 75 % of the model results have the same sign. The temperature dependence of the increases in the Baltic Sea basin (slopes in Fig. 4) are generally larger in summer than in winter with the southern land points as an exception, the same area where the average precipitation (Figs. 2–3) decreases. The inter-model spread is considerably larger in summer than in winter, illustrating the greater influence of local processes in this season; it should be noted that the increase from 19 to 90 in the number of models analysed, compared to Christensen and Kjellström (2018), results in a considerably more robust positive signal in the summer 10-year return value.

The relative changes of extreme precipitation in winter (Fig. 4 upper panels) are quite similar to the relative change in average precipitation (Fig. 2), indicating no change in the shape of the intensity distribution function. For summer, however, the projected change in extreme precipitation is consistently more positive than the change in average precipitation. While the temperature sensitivities (slopes in Figs. 3 and 4) for winter average precipitation and winter extreme precipitation are almost identical, the sensitivity of extremes in summer is larger than that for winter, while it is insignificant for the average precipitation in summer. This feature is, however, less apparent in the EURO-CORDEX results than in the PRUDENCE results of BACC (BACC Author Team, 2008) and the ENSEMBLES results described in BACC II (BACC II Author Team, 2015). It is not clear if this difference is due to the fact that the RCMs are run at different horizontal resolutions in the three projects (i.e. 50, 25, and 12.5 km, respectively), or if it is a consequence of different model formulations in the projects or of the large-scale climate change signal as imposed by the underlying GCMs that also differs between the experiments.

Recently, several research institutes have started employing convection-permitting regional models (CPMs). Such models are able to run in much higher resolution, where traditional hydrostatic RCMs with fully parameterized convective precipitation release may produce convective precipitation explicitly as well as parameterized, CPMs avoid this possible double counting at high resolution. With CPMs grid distances below the “grey zone” of 3–5 km are possible. In Lind et al. (2020), results are presented with the CPM HIRLAM ALADIN Regional Mesoscale Operational NWP in Europe -Climate version (HARMONIE-Climate; HCLIM), produced in a common Nordic model collaboration (NorCP) with participation from Sweden, Norway, Denmark, and Finland. Comparing a CPM version of HCLIM in 3 km resolution with a non-CPM version in 12 km, it was concluded that the high-resolution model showed better results for precipitation intensity distribution, including extreme precipitation at subdaily timescales, for the summer precipitation diurnal cycle, and for snow in mountains. Such better agreement now shown for the Nordic region has previously been shown for other regions in Europe and elsewhere (e.g. Kendon et al., 2014; Lind et al., 2016; Gao et al., 2020).

Based on convection-permitting models, it has been argued that changes in precipitation extremes of a shorter duration may be larger than those for longer timescales (e.g. Kendon et al., 2014; Lenderink and van Meijgaard, 2010). However, other results indicate (Ban et al., 2015) that convection-permitting models may give roughly the same increase also for shorter durations, consistent with the Clausius–Clapeyron scaling of around 6 %–7 % per degree of warming. In a study of idealized warming experiments repeating present-day observed weather under warmer and moister conditions with the HCLIM model, Lenderink et al. (2019) showed that the increase in precipitation extremes is strongly dependent on moisture availability.

Changes in the climatology of 10 m wind speed are even more uncertain than is the case for the precipitation climate, both for seasonal mean conditions and for extremes on shorter timescales (e.g. Kjellström et al., 2011, 2018; Nikulin et al., 2011).

In a study by Donat et al. (2011), annual 98th percentile daily maximum wind speed changes in RCM simulations from the ENSEMBLES project were analysed, for the middle of the century as well as the end of the century. The ensemble average, like the driving GCMs, increased in a region from the British Isles to the Baltic Sea and decreased in the Mediterranean area. Nikulin et al. (2011) found increasing wind speed extremes (20-year return periods of annual maximum 10 m wind speed) over the Baltic Sea in five out of six simulations, based on an ensemble of one RCM downscaling six different GCMs under the A1B scenario.

In BACC II (BACC II Author Team, 2015), an analysis of 13 ENSEMBLES simulations showed a very small insignificant median increase in the southern part of the Baltic Sea area; the signal is consistent with the findings by Donat et al. (2011) but with a large spread between models.

Figure 5 shows average changes over the Baltic Sea for the 72 EURO-CORDEX RCP8.5 simulations, the 22 RCP4.5 simulations, and the 30 RCP2.6 simulations, which were used (Table 1). In Figs. S13–S18, we show median and quartile maps for summer and winter for each of the three RCP scenarios. There is very little agreement between the models about even the direction of change for winter in the Baltic Sea area unlike the tendency for reduced average wind speed outside of the study area over the North Atlantic (not shown). Over the northernmost part of the Baltic Sea basin, the Bothnian Bay, there is an indication of larger wind speed increase (or less decrease) over the sea than over surrounding land areas. This feature has previously been pointed out by Kjellström et al. (2011), Meier et al. (2011), and Tobin et al. (2016), and has been related to decreases in sea ice in the future warmer climate, leading to consequent changes in stability conditions of the lower atmosphere. See also the comparison between regional coupled and uncoupled simulations in Fig. 12, where the probably more consistent treatment of ice–albedo feedback leads to a slightly larger increase in winter. As seen in Fig. 5b, the slight increase in mean wind over the Baltic Sea in BACC II is not projected in the current simulations.

Summer results show consistent but small reductions of wind over land of about 2 %–6 %. Again, in summer, there are differences between land and ocean areas with generally larger increases, or smaller decreases, over the Baltic Sea than its surrounding land areas.

The relative change in extreme wind speed is shown in Fig. 6 as the relative change of the 10-year return value of daily maximum wind speed for the EURO-CORDEX RCP-based and the BACC II SRES-based simulations considered, as well as for the coupled RCA4-NEMO simulations. The correlation between temperature and extreme wind is quite small, which indicates that there is no significant signal.

In Fig. 7, we study the change in incoming solar radiation in the ensemble. In winter, most of the area shows a considerable relative reduction of the order of 10 %. This has been proposed to be linked to the more extensive cloud cover in northern Europe in most EURO-CORDEX RCMs for the future (Coppola et al., 2021). It should be noted (Bartók et al., 2017) that global and regional models frequently disagree considerably about the change in incoming radiation in a changing climate, with global models having a more positive trend; this discrepancy is connected to different projections of cloud cover, with GCMs frequently projecting a decrease, while RCMs frequently show no significant change. We repeat here that the different treatment of aerosols in GCMs and RCMs plays a role as many of the RCMs do not include time-varying anthropogenic aerosols as in GCMs (Boé et al., 2020). It has also been suggested that reduced snow cover (see Sect. 3.6 below) could contribute to attenuate gross downward solar radiation flux as the reduced surface albedo reduces multiple reflection between the surface and the clouds (Ruosteenoja and Räisänen, 2013).

Future snow cover is expected to decrease with climate warming, both because more precipitation is projected to fall as rain, and because snowmelt accelerates. As an indicator of less cold conditions, Coppola et al. (2021) show that the number of frost days decrease by more than 2 months in large parts of the Baltic Sea basin comparing a set of EURO-CORDEX RCMs under RCP8.5 for 2071–2100 with 1981–2010. Simultaneously, there is an increase in winter precipitation in Scandinavia, which may partly compensate for these effects.

Räisänen and Eklund (2011) analysed data from RCM simulations from the ENSEMBLES project. The study found a decrease of snow volume across all of Europe in the future with the only exception that the Scandinavian mountain areas may experience a slight and statistically insignificant increase. Räisänen (2021) found a widespread future decrease in northern Europe for snow water equivalents also for a set of EURO-CORDEX RCMs. It was shown that a smaller snowfall fraction together with larger reduction of snow on the ground more than compensated for increasing precipitation, as seen in several of the RCMs. In relative numbers, the decrease was found to be larger in southern warmer parts of Scandinavia, while changes in absolute numbers are larger in the north. Similarly, the results were ambiguous for the most high-altitude parts of the Scandinavian mountains where some models indicate increasing snow water and others a decrease. A potential increase in the latter region was also proposed by Schuler et al. (2006) in a detailed study for Norway based on two RCM simulations with different GCM drivers. The study concluded that the maximum amount of snow in extreme years could be greater than in extreme years of the recent past in spite of decreasing average snow amount.

Winter snow cover is one of the most drastically changed climatological quantities (Fig. 8). There is agreement between models about a reduction of average wintertime snow amount of around 50 % on average for land grid points north of 60∘ N for the RCP8.5 scenario, and almost 80 % reduction for land grid points south of this latitude. Northern grid points probably have a lower reduction due to the generally colder climate and smaller amount of solar radiation. In addition, there is a significant amount of mountain grid points, where the warming temperature does not reach the freezing point as frequently as in lower-lying regions even if the frequency is increasing in a warmer climate (Nilsen et al., 2021). The reduction in snow amount is slightly larger than in BACC II (BACC II Author Team, 2015). This is consistent both with the fact that the RCP8.5 scenario on average projects larger warming than the SRES A1B scenario used in BACC II and that the precipitation increase is smaller in the RCP8.5 scenario than in SRES A1B, at least north of 60∘ N (see Fig. 3c).

It is only in high-altitude parts of central and northern Scandinavia that changes are limited with relatively large amounts of snow also in the future. At high altitude, the increase of winter precipitation may be compensating for the increase in melting with higher temperature. Also the fact that increasing temperatures may not reach the melting point is significant; see, e.g. Gröger et al. (2021a) Fig. 12b. However, also in these high-altitude regions, the warmer future climate results in a shorter snow season with accumulation starting later and spring melt starting earlier, which acts to reduce the total amount of snow (Räisänen, 2021).

Sea-ice cover is not a product of the RCM but rather an input originating from the driving GCM. We will show the changes in interpolated sea-ice field for the RCP8.5 scenario in Fig. 9, as these changes are large and are decisive for the change in climate between the periods. In order to compare to a more consistent description of sea ice, we also show in Fig. 10 the corresponding figures for the eight-member RCA4-NEMO coupled regional simulations. The main difference is that the present-day simulations with the coupled model have some extent of coastal sea ice in the southern Baltic Sea, which is disappearing in the future.