Oceanographic regional climate projections for the Baltic Sea 1 until 2100 2

Recently performed scenario simulations for the Baltic Sea including marine biogeochemistry were 13 analyzed and compared with earlier published projections. The Baltic Sea, located in northern Europe, is a semi14 enclosed, shallow and tide-less sea with seasonal sea ice cover in its northern sub-basins and a long residence 15 time causing oxygen depletion in the bottom water of the southern sub-basins. With the help of dynamical 16 downscaling using a regional coupled atmosphere-ocean climate model, four global Earth System Models were 17 regionalized. As the regional climate model does not include components for the terrestrial and marine 18 biogeochemistry, an additional catchment and coupled physical-biogeochemical model for the Baltic Sea were 19 used. In addition to previous scenario simulations, the impact of various water level scenarios was examined as 20 well. The projections suggest higher water temperatures, a shallower mixed layer with sharper thermocline 21 during summer, reduced sea ice cover and intensified mixing in the northern Baltic Sea during winter compared 22 to present climate. Both frequency and duration of marine heat waves would increase significantly, in particular 23 in the coastal zone of the southern Baltic Sea (except in regions with frequent upwelling). Due to the 24 uncertainties in projections of the regional wind, water cycle and global sea level rise, robust and statistically 25 significant salinity changes cannot be identified. The impact of changing climate on biogeochemical cycling is 26 considerable but in any case smaller than the impact of plausible nutrient input changes. Implementing the 27 proposed Baltic Sea Action Plan, a nutrient input abatement plan for the entire catchment area, would result in a 28 significantly improved ecological status of the Baltic Sea and reduced hypoxic area also in future climate, 29 strengthening the resilience of the Baltic Sea against anticipated future climate change. While our findings about 30 changes in variables of the heat cycle mainly confirm earlier scenario simulations, earlier projections for salinity 31 and biogeochemical cycles differ substantially because of different experimental setups and different 32 bioavailable nutrient input scenarios. 33 34 During the time in which this paper was prepared, shortly before submission, Christian Dieterich passed away 35 (1964-2021). This sad event marked the end of the life of a distinguished oceanographer and climate scientist 36 who made important contributions to the climate modeling of the Baltic Sea, North Sea and North Atlantic 37 regions. 38 39 https://doi.org/10.5194/esd-2021-68 Preprint. Discussion started: 4 August 2021 c © Author(s) 2021. CC BY 4.0 License.

to present climate. Both frequency and duration of marine heat waves would increase significantly, in particular 23 in the coastal zone of the southern Baltic Sea (except in regions with frequent upwelling). Due to the 24 uncertainties in projections of the regional wind, water cycle and global sea level rise, robust and statistically 25 significant salinity changes cannot be identified. The impact of changing climate on biogeochemical cycling is 26 considerable but in any case smaller than the impact of plausible nutrient input changes. Implementing the 27 proposed Baltic Sea Action Plan, a nutrient input abatement plan for the entire catchment area, would result in a 28 significantly improved ecological status of the Baltic Sea and reduced hypoxic area also in future climate, 29 strengthening the resilience of the Baltic Sea against anticipated future climate change. While our findings about 30 changes in variables of the heat cycle mainly confirm earlier scenario simulations, earlier projections for salinity 31 and biogeochemical cycles differ substantially because of different experimental setups and different 32 bioavailable nutrient input scenarios.

Analysis 249
Evaluation of the historical period 250 To evaluate model results of the BalticAPP and CLIMSEA scenario simulations during the historical period, 251 annual and seasonal mean biases during the historical period between RCO-SCOBI simulations and reanalysis 252 data (Liu et al., 2017) were calculated. As the reanalysis data are available for the period 1971-1999, we limit the 253 calculation of biases to  First, the monthly average of SST was computed from model output every 48 hours. Then the linear trend was 268 calculated with the Theil-Sen estimator (Theil, 1950;Sen, 1968). The trend computed with this method is the 269 median of the slopes determined by all pairs of sample points. The advantage of this expensive method is that it 270 is much less sensitive to outliers. The significance of SST trends was evaluated from a Mann-Kendall non-271 parametric test with a threshold of 95%. The SST trends were computed by season and annually. In this last case 272 the annual cycle is removed before computing the linear trend.  (Belkin, 2009) making this region prone to marine 286 heat waves (MHWs). Short periods of abnormally high water temperatures have recently been documented for 287 the Baltic Sea (Suursaar, 2020). MHWs can be defined with reference to the mean climatology (e.g. 90th, 95th, 288 98th percentile temperature) or by exceeding absolute temperature thresholds to be defined with respect to end 289 user applications (Hobday et al., 2018). In most cases, MHWs are defined by the number of periods, the 290 intensity, and duration and for specific purposes (Hobday et al., 2018). We here only focus on the general impact 291 of climate change since an appropriate definition of metrics for MHWs suitable for the Baltic is lacking. In the 292 following, MHWs are defined as periods with an SST >= 20°C lasting for at least 10 days to better reflect the 293 sensitivity of ecosystem dynamics. 294 3 Results 295 3.1 Historical period 296

Water temperature 297
The climate of the Baltic Sea region varies considerably due to maritime and continental weather regimes. For 298 the period 1970 to 1999, the annual mean SST amounts to about 7.8°C (Fig. 4). The mean seasonal cycle of the 299 SST is pronounced and the northern Baltic Sea is sea-ice covered every winter (not shown). Due to the large 300 latitudinal extension, the Baltic Sea is characterized during all seasons by a distinct SST difference between 301 colder northern and warmer southern sub-basins (Fig. 4). In the southern Baltic Sea, there is also a pronounced 302 west-east temperature gradient, mainly during summer and autumn, which reflects the large-scale cyclonic 303 https://doi.org/10.5194/esd-2021-68 Preprint.  (Fig. 4). In particular, during spring and summer, the shallow coastal zone of the northern and 308 eastern Baltic Sea is too warm. The spatially averaged biases during winter, spring, summer, autumn and the 309 annual mean amount to 0.8, 0.9, 0.8, 1.0 and 0.9°C. The reason for the warm bias is likely a bias of the RCSM. 310 Driven by the reanalysis data ERA40 (Uppala et al., 2005), RCA4-NEMO systematically overestimates water 311 temperatures and underestimates sea-ice cover in the Baltic Sea during the historical period (Gröger et al., 2019;312 their Suppl. Mat. S1). 313

314
In ECOSUPPORT scenario simulations, there is also a systematic warm bias of RCAO driven by GCMs at the 315 lateral boundaries, particularly resulting in too warm winter water temperatures and too low sea-ice cover (Meier 316 et al., 2011c, d;2012c, d). While these biases are found in all three applied Baltic Sea models (Table 3) forced  317 with the RCSM atmospheric surface fields, simulations driven with regionalized reanalysis data (ERA40) 318 showed smaller mean biases . over the open ocean with pronounced west -east gradients. This is related to the predominant south-westerly 322 wind regime with larger wind fetches and higher significant wave heights in the eastern Gotland Basin causing 323 wave-induced vertical mixing. Furthermore, a positive sea -atmosphere temperature contrast favors higher wind 324 speeds ("positive winter thermal feedback loop"; Gröger et al., 2015;2021b). In spring, a weakening wind 325 regime, lowering heat exchange (thereby turning from heat loss to heat gain) and increased solar irradiance lead 326 to a thinner MLD in the southern Baltic Sea while in the northern part melting sea ice and subsequent thermal 327 convection and wind-induced mixing still maintain MLDs > 50 m. During summer, the atmosphere-ocean 328 dynamics is weakest leading to a pronounced thermocline and shallowest MLDs (the so-called "summer thermal 329 short circuit"; Gröger et al., 2021b). During autumn, the atmosphere cools faster than the earth surface and

Marine heat waves 342
Baltic Sea MHWs are here defined as periods of >10 days duration with 1) SST higher than 20°C and 2) SST 343 exceeding the 95th percentile temperature. Figure 6 compares the CLIMSEA climate model ensemble mean with 344 the reanalysis data set generated by the same model (Liu et al., 2017). 345 346 The first index uses a fixed threshold focusing more on the environmental impact of heat waves. In particular, 347 diazetrophic nitrogen fixation becomes effective at higher temperatures. The spatial pattern of such MHWs is 348 strongly related to the simulated SST. The second index is based on a reference climatology, which is taken here as 1976-1999. The number of MHWs 356 ( Fig. 6c) is negatively correlated to their average duration (Fig. 6d). This is somewhat more pronounced in the 357 reanalysis data set. In general, reanalysis data and RCO show similar patterns but the amplitude of spatial 358 variance is higher in the reanalysis data ( Fig. 6c) which assimilated small-scale regional observations. The Since MHWs are predominantly a summer phenomenon in the Baltic Sea, the stability of the seasonal 363 thermocline is likely a key element in the dynamics of MHWs and processes favoring vertical mixing can be 364 considered a benchmark in the models ability to simulate MHW. Taking into account that mixing is highly 365 parameterized in current ocean models, RCO reproduces the spatial pattern of MHW reasonably well. 366

367
In the literature, MHWs in ECOSUPPORT scenario simulations have not been analyzed. 368

Salinity 369
The annual mean sea surface salinity (SSS) distribution shows a large north-south gradient mirroring the input 370 of freshwater from rivers, mostly located in the northern catchment area, and saltwater inflows from the North 371 Sea (Fig. 7). The SSS drops from about 20 g kg -1 in Kattegat to < 2 g kg -1 in the northern Bothnian Bay and 372 eastern Gulf of Finland. For the period 1970 to 1999, the annual mean SSS of the Baltic Sea including Kattegat 373 amounts to about 7.3 g kg -1 . Occasionally big inflows of heavy saltwater from Kattegat ventilate the bottom 374 water of the Baltic Sea, filling its deeper regions (Fig. 7). Due to almost absent tides, mixing is limited and the 375 water column is characterized by a pronounced vertical gradient in salinity, and consequently also in density, 376 between the sea surface and the bottom. reanalysis data (Fig. 7). The spatially averaged, annual mean bias amounts to -0.4 g kg -1 . In the climate models, 381 bottom salinities in the Belt Sea, Great Belt area and the Gotland Basin (most pronounced in the northwestern 382 part) are considerably higher and in the Bornholm Basin considerably lower than in the reanalysis data (Fig. 7). 383 The spatially averaged, annual mean bias amounts to +0.3 g kg -1 . Hence, the vertical stratification in the Belt 384 Sea, Great Belt area and the Gotland Basin is larger in climate models than in the reanalysis data.

Sea level 391
Due to the seasonal cycle in wind speed, with wind directions predominantly from southwest, the sea level in the 392 Baltic Sea varies considerably throughout the year, with highest sea levels of about 40 cm relative to Kattegat 393 during winter, at the northern coasts in the Bothnian Bay and at the eastern coasts in the Gulf of Finland (Fig. 7). 394 For the period 1976 to 1999, the annual mean sea level amounts to about 16 cm, with a horizontal north-south 395 difference of about 35 cm (not shown). This sea level slope is explained by the lighter brackish water in the 396 northeastern Baltic Sea compared to the Kattegat and by the mean wind from southwesterly directions which 397 pushes the water to the north and to the east (Meier et al., 2004a). 398 399 Differences in mean sea level between CLIMSEA climate models and reanalysis data are small (Fig. 7) and the 400 spatially averaged, winter mean bias amounts to +0.6 cm only. Sea levels in some parts of the coastal zone such 401 as the western Bothnian Sea are higher in climate models compared to the reanalysis data probably due to lower 402 salinities. The negative sea level bias in the eastern Gotland Basin suggests an intensified, basin-wide cyclonic 403 gyre. The seasonal cycle of the ensemble mean sea level is relatively well simulated, with overestimated sea 404 level in early spring and underestimated sea level in summer at all investigated tide gauge locations compared to 405 both observations and a hindcast simulation driven by regionalized ERA40 data (Fig. 8). 406

407
In ECOSUPPORT scenario simulations, sea levels were not systematically analyzed. In one of the three models 408 (RCO-SCOBI), seasonal mean biases comparable to the biases in CLIMSEA scenario simulations were found 409 (Meier et al., 2011d).  observations. However, simulated monthly mean nitrogen fixation during 1999-2008 showed a prolonged peak 482 period in July and August while the observations showed a peak more confined to July. However, it should be 483 noted that the RCO-SCOBI version that has been used for scenario simulations discussed here (e.g., Saraiva et 484 al., 2019a) did not contain a CLC model. In the vertical, temperature trends are largest in the surface layer compared to the Baltic Sea winter water above 551 the halocline causing a more intense seasonal thermocline (see Section 3.2.2) with largest trends in spring and 552 summer (not shown). Elevated trends are also found in the deep water due to the influence of saltwater inflows 553 that will be warmer in future climate because the inflows originate from the shallow entrance area mainly in 554 winter. Hence, the deep water below the halocline in those sub-basins that are sporadically ventilated by lateral 555 saltwater inflows such as the Bornholm Basin and the Gotland Basin warm more than the overlaying 556 intermediate layer water. 557

558
In the literature, trends in ECOSUPPORT scenario simulations were not analyzed. 559

Mixed layer depth 560
In Figure 17 Another way to analyze MHWs is to calculate them with respect to the 95th percentile temperature of the 590 historical reference climate (Fig. 19). For the historical climate, such periods are in most regions less than 20-30 591 days. In the southern Baltic Sea, especially west of the Baltic proper they are more frequent. The climate change 592 signal is characterized by more frequent MHWs of longer duration. Already in RCP4.5 MHWs occur at least 593 every year. The strongest increase in frequency is near the coasts whereas their average duration increases less 594 compared to the open sea (Fig. 19). This is probably related to repeated cold water entrainments from the open 595 sea that interrupt warm periods because of the larger variability of the coastal zone compared to the open sea. In 596 addition, shallow areas are, due to their lower heat storage, more sensitive to cold weather events and the 597 associated oceanic heat loss. 598

Salinity 599
In the CLIMSEA ensemble, salinity changes are not robust, i.e. the ensemble spread is larger than the signal 600 . The ensemble mean signal is small compared to the ensemble spread because the impact of 601 the projected increase in total river runoff from the entire catchment (Fig. 3)  In BalticAPP and CLIMSEA scenario simulations, sea level changes are small (Fig. 12). On the other hand, sea 619 level changes in ECOSUPPORT scenario simulations are larger, particularly in spring, because one member of 620 the multi-model ensemble considered Archimedes' principle (not shown). Note that in Figure 12

Phytoplankton concentrations 685
Annual mean changes in surface phytoplankton concentration (expressed as chlorophyll concentration) follow 686 the changes in nutrient concentrations and are confined to the productive zone along the coasts (Fig. 23). In 687 ECOSUPPORT projections, annual mean Secchi depths are decreasing in all scenario simulations (see Fig. 24  688 and Table 7). On the other hand, in BalticAPP projections, area averaged Secchi depths generally increase, 689 except in the combination of RCP8.5 and BAU scenarios (Table 7) (illustrated by the differences between BalticAPP and CLIMSEA ensembles and between CLIMSEA ensemble 693 mean and high SLR scenarios) have only a minor impact on the water transparency response ( Table 7). The 694 overwhelming driver of Secchi depth changes are nutrient input scenarios (illustrated by the differences between 695 ECOSUPPORT and BalticAPP/CLIMSEA ensembles highlighted by even contradictory signs in the changes). 696

Biogeochemical fluxes 697
In CLIMSEA under the BSAP, primary production and nitrogen fixation were projected to considerably decrease 698 in future climate (Fig. 22) scenario, it is found that the NAO shows high interannual variability. By applying a wavelet analysis, it is found 714 that the calculated NAO index contains some decadal variability, which differs for every model (not shown). By 715 comparing RCP4.5 and the high emission scenario RCP8.5, it can be seen that the spread of the ensemble 716 increases with enlarged greenhouse gas concentrations. Furthermore, Figure 25 shows the running correlation 717 between the NAO index and the area averaged SST. Indeed, the correlation remains positive but it is not constant 718 in time. By comparing RCP4.5 and RCP8.5 it is found that there are no systematic changes between both 719 emission scenarios. However, for RCP8.5 a slightly larger ensemble spread is found. 720 https://doi.org/10.5194/esd-2021-68 Preprint. Discussion started: 4 August 2021 c Author(s) 2021. CC BY 4.0 License.

Knowledge gaps 721
As in this study only four ESMs were regionalized using one RCSM, the CLIMSEA ensemble is still too small 722 to estimate uncertainties caused by ESM and RCSM differences. It should be noted that recently even nine 723 ESMs with the same RCSM were regionalized but without running modules for the terrestrial and marine 724 biogeochemistry (Gröger et al., 2021b). Therefore, we have not considered these simulations in our assessment. The available ensembles of scenario simulations are now larger than in previous studies. It was shown that the 825 uncertainty caused by ESM differences became now also larger. However, the ensemble size might still be too 826 small and the model uncertainty is very likely underestimated. Further, natural variability might be a more 827 important source of uncertainty than previously estimated. 828

829
In present climate, the climate variability of the Baltic Sea region during winter is dominated by the impact of 830 the NAO. However, during past climate the correlation between NAO and regional variables such as water 831 temperature or sea ice varied in time. These low-frequency changes in correlation were projected to continue and 832 systematic changes in the influence of the large-scale atmospheric circulation on regional climate and in the 833 NAO itself could not be detected, although a northward shift in the mean summer position of the westerlies at 834 the end of the twenty-first century compared to the twentieth century was reported earlier (Gröger et al., 2019).