Nordic Seas Acidiﬁcation

. With prevailing low temperatures, deep winter mixing, and cold-water coral reefs, the Nordic Seas is vulnerable to ocean acidiﬁcation. Here we present a detailed investigation of changes in pH and aragonite saturation, and its impact on cold-water corals, in the Nordic Seas, from pre-industrial times to 2100 by using in situ observations, gridded climatological data, and Earth System Model (ESM) projections. From pre-industrial to present, the Nordic Seas surface pH has dropped by 0.06 on 5 average, and the aragonite saturation horizon has moved from a depth of 2500 meter to 2000m, which is well below the cold-water coral habitats. Between 1981 and 2019 pH decreased by, on average, 0.10 in the Nordic Seas surface waters. The pH drop, mainly driven by an uptake of anthropogenic CO 2 , is signiﬁcant all over the Nordic Seas, except in the Barents Sea Opening. We also ﬁnd that the acidiﬁcation has penetrated relatively deep, in some regions down to 2000 m. This has resulted in a signiﬁcant decrease in the aragonite saturation state, which are close to undersaturation in the depth layer of 1000-2000 10 m in the modern ocean. Model projections indicate an additional surface ocean pH decrease of 0.1-0.4


Introduction
Since 1850, human activities have released 650 ± 65 Gt of carbon to the atmosphere, of which about 25% have been taken up 20 by the oceans (Friedlingstein et al., 2020) where it has added to the pool of dissolved inorganic carbon (C T ). The increasing C T has resulted in surface seawater pH decline of approximately 0.1, which corresponds to an approximately 30% increase in hydrogen ion (H + ) concentration (e.g., Doney et al., 2009;Gattuso and Hansson, 2011;Jiang et al., 2019). This ocean acidification is a serious threat to many marine organisms, in particular those having shells and skeletons consisting of calcium carbonate (CaCO 3 ), such as pteropods and corals (Guinotte et al., 2006;Turley et al., 2007;Manno et al., 2017;Doney et al., 25 2020; Doo et al., 2020) as the pH drop also leads to a reduction in the CaCO 3 saturation state (Ω) of seawater. Depending on the CO 2 concentration pathway, future projections suggest further reductions of surface ocean pH of 0.1-0.4 until the end of the 21st century from the 1990s (Bopp et al., 2013). While global average acidification rates for surface waters, both from pre-industrial times to present and as projected for the future have been dealt with in several studies (e.g. Caldeira and Wickett, 2003;Raven et al., 2005;Kwiatkowski et al., 2020), less is known about acidification rates on regional scales, especially below 30 the surface.
The Nordic Seas, comprised of the Greenland, Iceland and Norwegian seas ( Fig. 1) and bounded by the Fram Strait in the north, the Barents Sea Opening to the northeast and the Greenland-Scotland Ridge in the south, are of particular interest when it comes to ocean acidification due to its specific dynamic, biogeochemical and ecosystem characteristics. The surface circulation pattern (e.g. Blindheim and Østerhus, 2013;Våge et al., 2013) is characterised by the warm, saline Atlantic waters that flow 35 northward as the Norwegian Atlantic Current in the east, mainly constrained to the Norwegian Sea, and cold and fresh waters of Arctic origin flowing southward as the East Greenland Current in the west. The surface waters are undersaturated in pCO 2 , i.e. their pCO 2 is lower than that of the atmosphere, making them important sinks for atmospheric CO 2 . This undersaturation comes as a result of several processes, including strong primary production, cooling of northward flowing Atlantic waters, and the inflow of pCO 2 undersaturated waters from the Arctic Ocean (Olsen et al., 2008;Ólafsson et al., 2020b). In the Greenland 40 and Iceland seas, deep and intermediate water-masses are formed through open-ocean convection. Some of these water-masses ultimately overflow the Greenland-Scotland Ridge and feed into the North Atlantic Deep Water and consequently help to sustain the lower limb of the Atlantic Meridional Overturning Circulation (AMOC, Dickson and Brown, 1994;Våge et al., 2015;Chafik and Rossby, 2019). The strong connection between surface and deep waters that is created through this deep water formation, would ultimately lead to early and relatively large detection of anthropogenic carbon and acidification in the 45 deep waters of the Nordic Seas and North Atlantic (Tjiputra et al., 2010;Perez et al., 2018), which could have negative impacts on their cold-water coral reefs. Due to the prevailing low temperatures, the Nordic Seas already have naturally low saturation states of CaCO 3 (Ólafsson et al., 2009;Skjelvan et al., 2014), making their cold-water coral reefs particularly exposed to ocean acidification (Kutti et al., 2014).
There has been extensive research on changes in the carbonate system and pH in the Nordic Seas, facilitated by the many 50 research and monitoring cruises in the area (e.g., Bellerby et al., 2005;Olsen et al., 2006;Ólafsson et al., 2009;Skjelvan et al., 2008;Chierici et al., 2012;Skjelvan et al., 2014;Jones et al., 2020;Skjelvan et al., 2021). Acidification rates of -0.0023 to -0.0041 y −1 have been observed in surface waters, which is greater than expected from the increase in atmospheric CO 2 alone (Ólafsson et al., 2009;Skjelvan et al., 2014). This is consistent with the many observations that have indicated that surface ocean pCO 2 , which is closely related to pH, has risen faster than the atmospheric pCO 2 (Olsen et al., 2006;Skjelvan et al., 55 2008; Ólafsson et al., 2009), i.e. a weakening of the pCO 2 undersaturation of the Nordic Seas surface waters might have occurred the past decades. Studies on present and future pH in the Nordic Seas using both a regional and an Earth System Model have also been published (Skogen et al., 2014(Skogen et al., , 2018. There is, however, to our knowledge, no previous work assessing acidification rates from the pre-industrial until the end of the 21st century using both observational and modelling data. In this study, we fill this gap by examining past, present and projected future ocean acidification rates and changes in 60 aragonite saturation in the Nordic Seas, over the full water column, by using the best available information for the various time periods. This includes a combination of in situ observations, gridded climatological data, and Earth System Model (ESM) projections. To get a better understanding of the processes behind the acidification rates, we decompose the pH changes into their thermodynamic and chemical drivers.

pH Drivers -Theoretical Background
The rising atmospheric CO 2 concentration increases the pCO 2 difference between the atmosphere and the ocean; i.e. the oceans become more undersaturated in CO 2 with respect to the atmosphere, which results in a flux of CO 2 from the atmosphere into the ocean. When CO 2 dissolves in seawater, it reacts with water to form carbonic acid (H 2 CO 3 ): which then dissociates into bicarbonate (HCO − 3 ) and hydrogen ions (H + ): Apart from C T , seawater pH is also controlled by temperature, salinity, and A T . The qualitative, instantaneous, effects of an increase in each property are shown in Table 1. Temperature and salinity only affect pH by altering the dissociation constants and thus the partitioning of C T between its different constituents. A T is the sum of the concentration of bases (proton acceptors) in the seawater. The relation between C T and A T influences the pH by affecting the buffer capacity of seawater. Note that the relations in Table 1 are the instantaneous, or thermodynamic, effects from a change in these properties, and does not consider 85 indirect effects on pH, for example from the change in air-sea fluxes that will follow, e.g. from a temperature driven pCO 2 change (e.g. Jiang et al., 2019;Wu et al., 2019).
showing that the dissolution of CO 2 in seawater results in a reduction in CO 2− 3 . This further affects the saturation state of 90 CaCO 3 (Ω), defined as: where K sp is the solubility product. CaCO 3 exists in two different forms in seawater: calcite and aragonite. Aragonite is more soluble than calcite, with a higher K sp . The saturation state of aragonite (Ω Ar ) is therefore lower than that of calcite (Ω Ca ) at a given place and time. When Ω is less than one, the water is corrosive and CaCO 3 dissolves.

95
Equation 6 shows that lower concentrations of CO 2− 3 , as induced by uptake of anthropogenic CO 2 and increase in C T , result in a reduction in the saturation state. The impact of C T on the saturation state is also seen in the spatial distribution of Ω in the surface ocean, which broadly follows temperature gradients (e.g. Orr, 2011;Jiang et al., 2019). The reason behind this temperature dependency is the higher CO 2 solubility of colder waters that give them the capacity to absorb more CO 2 at a given atmospheric pCO 2 , which decreases the CO 2− 3 concentration. Consequently, cold waters also have a relatively low Ω Ar

100
and Ω Ca and are thus more vulnerable to acidification. Apart from C T , Ω is also influenced by A T , temperature and salinity, as shown in Table 1.
The sensitivity of pH and Ω to uptake of anthropogenic CO 2 is dependent on the buffer capacity of the seawater that is largely determined by the concentration of carbonate ions [CO 2− 3 ] (e.g. Sarmiento and Gruber, 2006;Orr, 2011). Waters with higher concentrations of CO 2− 3 , i.e. a higher buffer capacity, have the capability of converting a larger fraction of the absorbed 105 CO 2 into bicarbonate. A smaller fraction remains as dissolved CO 2 , implying a smaller increase in the seawater pCO 2 . These waters therefore have the capability of absorbing more CO 2 for any given increase in atmospheric pCO 2 (assuming a uniform increase in pCO 2 between water-masses), which also implies a larger decline in CaCO 3 saturation state. The drop in pH, on the other hand, is larger in waters with lower CO 2− 3 concentration as they have less ability to neutralise the carbonic acid since their buffer capacity is lower. Data from 28 research cruises (Brewer et al., 2010;Anderson et al., 2013a, b;Anderson, 2013a, b;Bellerby and Smethie, 2013;Johannessen and Golmen, 2013;Johannessen, 2013a, b;Johannessen and Simonsen, 2013;Johannessen and Olsen, 2013;Johannessen et al., 2013c, a, b;Jones et al., 2013;Omar and Olsen, 2013;Omar and Skogseth, 2013;Pegler et al., 2013;Skjelvan et al., 2013;Wallace and Deming, 2014;Lauvset et al., 120 2016; Tanhua, 2017;Jeansson et al., 2018;Marcussen, 2018;Schauer et al., 2018) in the Nordic Seas were extracted from the GLODAPv2.2019 data product, which provides bias-corrected, cruise based, interior ocean data (Olsen et al., 2019). The GLODAPv2 data product is considered consistent to within 0.005 for salinity, 2% for silicate , 2% for phosphate, 4 µmol kg −1 for C T and 4 µmol kg −1 for A T (Olsen et al., 2019).
The time-series data are from the Norwegian Sea (Ocean Weather Station M) and the Iceland Sea. The data from the Ocean 125 Weather Station M, located at 66 • N and 2 • E, have been described in Skjelvan et al. (2008). At this station, sampling at 12 depth levels between surface and seabed (2100 m) was carried out each month between 2002 and 2009, and 4-6 times each year between 2010 and 2019. Here, the uncertainty related to the sample data is 0.001 for salinity, 0.7 µmol kg −1 for silicate , 0.06 µmol kg −1 for phosphate, 2 µmol kg −1 for C T and 2 µmol kg −1 for A T . The time-series station in the Iceland Sea, covering the period of 1985-2019, is situated at 68 • N and 12.67 • W. It is visited approximately 4 times a year and samples are 130 taken at 10-20 depth levels between surface and seabed (1900 m). The uncertainty related to the sampled data at this station is 0.005 for salinity, 2% for silicate , 2% for phosphate, 4 µmol kg −1 for C T and 4 µmol kg −1 for A T . These data have been described in Ólafsson et al. (2009).
The data from the program" Monitoring ocean acidification in Norwegian waters" covers the period 2011(2011Tilførselsprogrammet and 2013 and are based on water column stations along repeat sections 135 in the Nordic Seas (Chierici et al., 2012(Chierici et al., , 2013(Chierici et al., , 2015(Chierici et al., , 2016Jones et al., 2018Jones et al., , 2019Jones et al., , 2020. Analytical methods for C T and A T follow the Dickson et al. (2007) and accuracy and precision is controlled by Certified Reference Materials (CRM), and by participation in international intercomparison studies (e.g. Bockmon and Dickson, 2015). The uncertainties related to the sampled data is 0.005 for salinity, 0.1 for silicate , 0.06 for phosphate, 2 µmol kg −1 for C T and 2 µmol kg −1 for A T .
Data for the Eastern Fram Strait were collected on cruises with RV Helmer Hansen within the CarbonBridge project, and on 140 cruises with RV Lance (Chierici et al., 2019b) organized by the Norwegian Polar Institute.
For atmospheric CO 2 data, we used the annual mean atmospheric CO 2 mole fraction (xCO 2 ) from the Mauna Loa updated records, downloaded from www.esrl.noaa.gov/gmd/ccgg/trends/ .

Model data
For the estimates of past and future ocean acidification under various climate scenarios, we used output of the fully cou-145 pled Norwegian Earth System Model (NorESM1-ME, Bentsen et al., 2013;Tjiputra et al., 2013Tjiputra et al., , 2016 as well as outputs of an ensemble of ESMs that participated in the Coupled Model Intercomparison Project Phase 5 (CMIP5, Taylor et al., 2012). NorESM1-ME includes the dynamical isopycnic vertical coordinate ocean model MICOM (Bleck and Smith, 1990) and the Hamburg Oceanic Carbon Cycle model (HAMOCC5, Maier-Reimer et al., 2005), adapted to the isopycnic ocean model framework. The HAMOCC5 model simulates lower trophic ecosystem processes up to the zooplankton level, including 150 primary production, remineralization and predation, and full water column inorganic carbon chemistry. While the simulations of NorESM1-ME are used to get a process understanding, the ESM ensemble is used to get an estimate of model uncertainty.
We chose emission-driven historical and future scenarios, rather than concentration driven ones, as only those capture the full impact of carbon cycle feedbacks (Booth et al., 2013). Specifically, we utilise emission-driven historical experiments for the period from 1850 to 2005 and emission-driven future scenarios for the period from 2006 to 2100, with focus on Rep-155 resentative Concentration Pathways 2.6, 4.5 and 8.5 (RCP2.6, RCP4.5, and RCP8.5; Meinshausen et al., 2011;van Vuuren et al., 2011a). RCP2.6 represents a mitigation scenario, RCP4.5 a stabilization scenario and RCP 8.5 a high-emission scenario.
While NorESM1-ME outputs are available for future scenarios with low to high emissions (RCP2.6, RCP4.5 and RCP8.5), the CMIP5 data-portals only contains ESM outputs for the future scenario with high emissions (RCP8.5, referred to as 'esm-rcp85' within the data-portal). We therefore utilised NorESM1-ME to inform about variations in future pH-changes that are 160 dependent on the presumed future emission strength, and our ESM-ensemble to inform about model-dependent uncertainties in those pH-changes, albeit only for the high emission scenario. Our ESM-ensemble contains all ESMs that have participated in experiment 'esmrcp85' and whose output is publicly available in one of the CMIP5 data portals and contains all variables needed for our analysis. This results in an ensemble of 7 ESMs: 1) CESM1(BGC) (The Community Earth System Model, version 1 -Biogeochemistry, Long et al., 2013), 2) CanESM2 (second-generation Canadian earth system model, Arora et al., System Model v1, Yukimoto et al., 2011). For our model ensemble, we only investigate one realisation of each scenario.

Gridded climatological data
Climatological distributions of pH and Ω Ar were calculated from the data of C T , A T , temperature, salinity, phosphate and silicate included in the mapped GLODAPv2 data product . The GLODAPv2 climatology of C T is normalized to the year of 2002. It is important to mention that the GLODAPv2 climatology along the northern Greenland coast 175 is mainly based on data from one cruise in 1993, and is therefore likely not representative for the long-term mean. We also determined pre-industrial pH by subtracting the GLODAPv2 estimate of anthropogenic carbon from the mapped climatology of present C T  for comparison with the pre-industrial state estimate from NorESM1-ME. When doing so we assumed that the changes in the temperature, salinity and A T of the Nordic Seas are of minor importance. The GLODAPv2 estimate of anthropogenic carbon hase been calculated with the the transit time distribution (TTD). He et al. (2018) published a 180 thorough analysis of the different sources of uncertainty in this method, and concluded that the overall uncertainty is 7.8-13.6%.
Combining this with the mapping errors Lauvset et al. (2020) estimate that the global ocean anthropogenic carbon inventory calculated from the mapped fields is 167±29 PgC. Note that the GLODAPv2 mapped pre-industrial climatology is referenced to an atmospheric CO 2 level of 280 ppm, and not to a specific time period or year. These data are only used in Fig. 2.

185
To estimate the potential impact of the Nordic Seas acidification on cold-water corals, we used habitat positions in longitude and latitude from EMODnet Seabed Habitats (www.emodnet-seabedhabitats.eu) together with information on depth from ETOPO1 (NOAA National Geophysical Data Center, 2020).

Spatial drivers 190
To elucidate the observed spatial variability of pH and Ω Ar distribution as extracted from the Nordic Seas in the GLODAPv2 climatology, we performed a correlation analysis with the drivers listed in Table 1. When it comes to C T and A T , one has to look at the combined effect, i.e. C T /A T . A potential correlation does not necessarily mean that there is a mechanistic relation, but can be a consequence of the contrasting properties of the Atlantic and polar waters. Therefore, in an attempt to better understand the effect of each driver, we calculated pH and Ω Ar by step by step introducing the spatially varying climatologies 195 of the drivers, while keeping all other drivers constant (set to the spatial mean value of the Nordic Seas surface waters). First, we calculated pH and Ω Ar by using the spatially varying temperature climatology, and keeping all other variables constant (pH(T), Ω Ar (T)). Thereafter, we repeated the same exercise with the spatially varying temperature, C T and A T climatologies to get pH(T, C T , A T ) and Ω Ar (T, C T , A T ). Finally, we added the salinity variability to get pH(T, C T , A T ,S) and Ω Ar (T, C T , A T , S). We started with temperature because it has an initial thermodynamic effect on pH and Ω Ar , and a subsequent, 200 secondary, effect from the resulting air-sea CO 2 exchange and change in C T /A T . Salinity was chosen as the last variable due to the minor effect it has on pH and Ω Ar .

pH changes and its drivers
It is important to keep in mind that changes in pH represents a relative change, and that pH trends are therefore not directly comparable across water-masses with large differences in mean pH (Fassbender et al., 2021). In these cases, it is preferable to 205 evaluate changes in H + concentration that represents an absolute change (Kwiatkowski and Orr, 2018). However, pH variations in the Nordic Seas are relatively small, and we have therefore decided to use pH in this study.

Present
Measurements of C T , A T , temperature, salinity (Figs. S1-S4) phosphate and silicate from the data sets described in Sect. 2.1.1 were used to calculate pH (on total scale) and Ω Ar , at in situ temperature and pressure, using CO2SYS for MATLAB (Lewis 210 and Wallace, 1998;van Heuven et al., 2011). Wherever nutrient data were missing, silicate and phosphate concentrations were set to 5 µmol kg −1 and 1 µmol kg −1 , respectively. For the CO2SYS calculations we used the dissociation constants of Lueker et al. (2000), the bisulfate dissociation constant of Dickson (1990) and the borate-to-salinity ratio of Uppström (1974). This ratio has recently been shown to be suitable for the western Nordic Seas (Ólafsson et al., 2020a). Waters, and we therefore refer to is as the Eastern Fram Strait. Regional trends were computed from annual means for five 225 different depth intervals (0-200, 200-500, 500-1000, 1000-2000, and 2000-4000 m) using linear regression. Although summer mixed layer depths generally is shallower than 200 m, a thickness of 200 m was used for the surface layer since this sets the approximate limit for the influence of seasonal variations associated with, e.g., primary production (e.g. Skjelvan et al., 2008).
The significance of the trends (at 95% confidence level), were determined from the p-value of the t-statistic, (as implemented in MATLAB's fitlm function). For the comparison of trends, we determined 95% confidence intervals of the slopes by the use 230 of the Wald method (as implemented in MATLAB's fitlm and coefCI functions).
The observed long-term changes in pH were decomposed into contributions from changes in temperature (T), salinity (S), C T and A T (Figs. S1-S4), following the proceedure of Lauvset et al. (2015). First, the effect of each of these processes on the CO 2 fugacity (f CO 2 ) is determined following Takahashi et al. (1993) and Metzl et al. (2010): Finally, we converted it to pH by acknowledging that dpH = -([H + ]ln (10) We additionally made a decomposition of the A T and C T drivers into a freshwater component and a biogeochemical compo-240 nent (Supplementary material, Sect. 1). The freshwater drivers of A T and C T are typically of similar magnitude and opposite sign, and consequently cancel each other. We therefore decided to stay with the more simple decomposition as shown in Eq. 7.
The only exception is discussed in Sect. 4.1.
To understand whether the observed pH changes are consistent with the changes in atmospheric CO 2 , the pH change that can be expected in seawater where the pCO 2 perfectly tracks the atmospheric pCO 2 (pH perf ) was determined for each region 245 by adding the observed change in atmospheric xCO 2 to the local mean pCO 2 for the first year with observations, and then calculating the pH with CO2SYS with the local temperature, salinity, A T , phosphate and silicate and their respective changes as inputs. We did not make any corrections for water vapour and atmospheric pressure because the rates of change for xCO 2 and pCO 2 are the same. Any deviation between observed pH change and pH perf is a consequence of changes in seawater pCO 2 that are smaller/larger than in the atmosphere, i.e a change in the air-sea pCO 2 difference.

Past and future
As described in Sect. 1.1, the total change in pH and saturation states does not only depend on local changes in C T , A T , temperature, salinity, and nutrients, but also on the initial buffer capacity of the seawater. ESMs are usually biased, i.e., there is an offset between modelled fields and reality. In particular, NorESM1-ME has high A T and low C T relative to observations in deep waters, leading to a biased high pH ( The methodology for calculating pH drivers described in the previous section was also used for calculating the drivers of past (1850-1859 to 1996-2005) and future (1996-2005 to 2090-2099) pH changes, using the changes in temperature, salinity, A T and C T data from NorESM1-ME output together with the climatological values from GLODAPv2.

Uncertainty analysis
There are several sources of uncertainties (σ) involved in our calculations of pH and Ω: measurement uncertainty (σ mes ), mapping uncertainty (σ map ) for the gridded product, and uncertainties related to dissociation constants (σ Kx ) used in the CO2SYS calculations. To estimate the total uncertainties in pH and Ω resulting from these we used the uncertainty propagation routine in CO2SYS (Orr et al., 2018). The uncertainties in the input parameters (A T , C T , temperature, salinity, phosphate 270 and silicate) were set to σ mes for the single measurements, and σ 2 mes + σ 2 map for the mapped product and for the past and future estimates. As σ mes we used the estimated uncertainties from Olsen et al. (2019), and for σ map we used the mapping uncertainty (3D field) from Lauvset et al. (2016). The correlation between uncertainties in A T , C T were set to 0. Including a correlation term would decrease the uncertainty, and possibly overestimating the uncertainty is preferable to including a poorly constrained correlation term . For the dissociation constants we used the default uncertainties in CO2SYS.

275
From here on, the calculated uncertainties will be presented as σ 1 , when σ Kx and σ mes are included, and σ 2 , when σ Kx , σ mes and σ map are included.
For the observations described in Section 2.1.1, the mean, maximum and minimum uncertainties (σ 1 ) in pH, Ω Ar , Ω Ca and pCO 2 obtained from the uncertainty propagation are listed in Table 2. Variations in the uncertainties arise from variations in temperature and salinity, which impact the uncertainty of dissociation constants. As discussed in Orr et al. (2018), random 280 and systematic uncertainties tend to cancel out when calculating trends (i.e. comparing measurements from the same location but from different times), unless there are substantial changes in the local salinity and temperature. Therefore, to estimate to what extent these uncertainties could impact our trend estimates, we investigated whether there is any trend in the uncertainties (Figs. S6-S7). In our observational data, there is also an uncertainty in the annual mean estimates related to seasonal undersampling.

285
Most samples (about 60% in total) from the data sets described in Sect. 2.1.1 were collected during spring and summer (April-September, Figs. S8-S13). The uneven sampling frequency of different seasons introduces uncertainty in the annual means of the uppermost ocean layer, and can bias the trend estimates. Unfortunately, there are not enough data to allow for deseasonalization in order to remove such potential biases. To get an idea of the effect of seasonal undersampling we additionnaly calculated trends by using annual means containing samples from April-September, and June-August, only.

290
Because a large part of this study focuses on process understanding and the driving factors behind pH change, we do not consider model uncertainty in Sect. 3, where the drivers of pH change in the model projections are analysed, here we only use the combined uncertainties of measurements and mapping.
In Sect. 4.2, where the future aragonite saturation horizon is presented, we additionally take into account model and scenario uncertainty. Modelled future projections are uncertain due to incomplete understanding or parameterization of fundamental 295 processes, as well as different and unknown future carbon emission scenarios (Frölicher et al., 2016). We note that internal climate variability is an additional source of uncertainty that we do not take into account in this study. The model dependent

Results
Before going into regional details of pH changes, we will give an overview of surface pH changes from 1850 to 2100 (Fig. 2).
To be consistent with our regional analysis in Sect Seas surface water is reasonably well simulated. The pH trend estimated from the observations for this period, -2.64±0.31 10 −3 yr −1 , is not significantly different (at the 95% confidence level) from the modelled pH trend, -2.21±0.04 10 −3 yr −1 . Because 310 the pH calculated from observational data is based on discrete samples with incomplete spatial and temporal coverage, its representatives for the entire Nordic Seas can be questioned, and we cannot expect an exact agreement with the model.
As expected, the future evolution of surface water pH in the Nordic Seas depends strongly on the CO 2 emission scenario ( Fig. 2). Under the high-CO 2 emission RCP8.5 scenario, NorESM1-ME simulates the pH to decrease by 0.40 between 2020 and 2100, to an average value of 7.66±0.02 by the end of the century (model ensemble range: 7.59-7.79, Fig. S5). For the In 1850, the simulated Nordic Seas average surface pH is 0.11 units higher than the global average, which is related to the undersaturation of CO 2 in the surface waters of the Nordic Seas (Jiang et al., 2019). Note that our global average is lower than the one estimated by, e.g., Jiang et al. (2019) for the surface ocean due to our 200 m thick surface layer. The difference between 325 the global ocean and the Nordic Seas is decreasing with time and by the end of the century the Nordic Seas surface pH is 0.03, 0.07 and 0.08 pH units higher than the global average in the RCP8.5, RCP4.5 and RCP2.6 scenarios, respectively. This is most likely partly due to the colder waters of the Nordic Seas, which gives them a lower buffer capacity, and partly due to a faster warming in the high latitude oceans related to polar amplification (Dai et al., 2019), which would give a faster decrease in the Nordic Seas pH compared to the global mean. Additionally, in RCP8.5, there is an increase in the pCO 2 undersaturation of the 330 global ocean that increases the global average pH (Fig. S14).

Present distribution of pH and Ω saturation states
Due to the contrasting properties of the Atlantic waters (here defined as waters with salinity>34.5) and polar waters (salin-ity<34.5) that meet and mix in the Nordic Seas, there are large spatial gradients in its surface temperature, salinity and chemical properties (Fig. S15). The Atlantic Water, located in the eastern part of the Nordic Seas, is characterized by higher temperature, 335 salinity, and A T , while polar waters are colder and fresher with lower A T . This results in a decrease in temperature, salinity, and A T from south and east to north and west. Within the Atlantic water, C T increases with decreasing temperature, largely as a consequence of the increased CO 2 solubility in colder water. The C T associated with polar waters is lower than that of Atlantic waters. The pH in the Nordic Seas increases from the Atlantic waters to the polar waters (Fig. 3). There is a significant, strong 340 (R<-0.5), negative correlation with all drivers, i.e. pH decreases with increasing temperature, salinity, C T and A T (Table 3 and Fig. S16). Here, only the negative correlation with A T is nonphysical, i.e. we would expect an increasing pH with an increasing A T (Table 1). The correlation with C T /A T is insignificant. Because the drivers are not orthogonal, it is impossible to rule out the contribution of each driver by just looking at these correlations, and we can only conclude that there is a strong water-mass dependency in the spatial distribution of these variables.

345
From the correlation between the pH and pH(T), we note that temperature-induced variations (through the thermodynamic effect) are able to explain 34% of the spatial variability in pH (Fig. 3). The range in pH(T) values (8. 06 -8.29) is very close to the observed one (8.10-8.32), indicating that temperature alone can give rise to the observed pH-range. Adding C T and A T to the picture explains an additional 55% (temperature, C T and A T explain all together 89% ), and are therefore important contributors to spatial variations in pH, in contrast to what is suggested by directly correlating these variables (Table 3). This

350
shows that the influence of C T and A T on pH is masked out by temperature variations in Table 3 and Fig. S16, which can be explained by the two cancelling effects that temperature has on pH described in Jiang et al. (2019). For example, while the instantaneous, thermodynamic, effect of a drop in temperature leads to a pH increase, it also gives rise to a decrease in pCO 2 that leads to a CO 2 uptake from the atmosphere, which subsequently increases the C T /A T ratio and decreases the pH. In the Nordic Seas, the spatial pH variations strongly correlate with surface pCO 2 (R=-0.99, Table 3), which range between 185 and 342 µatm 355 and is lower than the atmospheric pCO 2 of 373 µatm (in year 2002, to which these data are normalized). This undersaturation is partly a result of the large heat release to the atmosphere and cooling of the sea surface, and shows that the sea surface CO 2 did not yet equilibrate with the atmosphere. Because most of the data have been used to produce these climatologies are from the productive season, there is probably also a contribution from primary production to this undersaturation. There is also a negative correlation between C T /A T ratio and temperature, indicating that CO 2 uptake has been taking place. The temperature 360 effect on pH in the Nordic Seas is therefore a combination of the instantaneous thermodynamic effect, and the effect of the subsequent CO 2 exchange and the resulting increase in the C T /A T ratio. The strong relation between pH and pCO 2 , which also has been observed for the global ocean (Jiang et al., 2019), suggests that the processes responsible for the spatial pH variations in the Nordic Seas are heat fluxes and production/remineralization of organic matter. On top of temperature, C T and A T , the addition of salinity explains the last 11% of the spatial variability in pH. The effect of salinity is the largest in the low-saline 365 regions, i.e in the polar waters and in the Norwegian coastal current.
The Ω Ar show an opposite pattern to pH, with low saturation states in polar waters, and high saturation states in Atlantic Water. The Ω Ar distribution is strongly correlated with C T / A T (R=-0.99) (Fig. 3,f) and temperature (R=-0.86). This is related to the strong relation between Ω Ar and C T / A T to the climatological temperature distribution and its impacts on the CO 2− 3 concentration (for which C T / A T is a proxy), as described in Section 1.1. As for pH, temperature has two effects on Ω Ar , but 370 in contrast to pH where these effects are counteracting, they reinforce each other for Ω Ar . From Fig. 3,d, it becomes clear that the temperature effect on the solubility of Ω Ar (Ω Ar (T)) only can explain 11% of the observed Ω Ar range, although it is able to explain 98% of the variability. The strong correlation with temperature is therefore largely a result of the temperature effect on C T / A T (Sect. 1.1 and Orr (2011); Jiang et al. (2019)). When adding C T / A T to the picture, the observed range in Ω Ar is reproduced, and 100% of the variability is explained. Salinity induced variations only have a minor contribution to the spatial variations in Ω Ar . As for pH, the effect of salinity is more prominent in the low salinity-regions.  Table 4. Spatial correlation (r) and explained variance (r 2 , in paranthesis) between pH and pH (

Changes from pre-industrial to present
Maps of surface pH and Ω Ar distributions, and their changes from pre-industrial to present (calculated from the gridded GLODAPv2 data and rates of change from the NorESM1-ME emission-driven historical run as described in Section 2.2.2), are shown in Fig. 4. The spatial pattern of changes in pH and Ω Ar are similar. The strongest decreases, reaching -0.12 and -0.55, 380 respectively, are found in Atlantic waters along the Norwegian coast both for pH and Ω Ar . The smallest change is found in polar waters. The reasons behind these patterns of change will be discussed in Section 4.1.
Due to the longer ventilation time scales of deeper waters, the pH decrease weakens with depth. As shown in the section across 70 • N (Fig. 5) waters below 2500 m are nearly unaffected. While the entire water column remains saturated with respect to calcite, the saturation horizon (Ω=1) of aragonite shoaled from a mean depth of 2200 m (uncertainty range: 2100-2400) 385 during pre-industrial, to a mean depth of 2000 (uncertainty range: 1700-2300) m in present times, across this specific section.
Note that these depths were obtained from the contour interpolation when creating Fig. 5, which has a finer vertical resolution than the GLODAPv2 climatology.

Present day changes (1981-2019)
Regional trends in observed seawater pH between 1981 and 2019 for the five different depth intervals are presented in Fig.   390 4.1 and Table 5. The corresponding trends in H + are shown in Fig. S17 and Table S7. In surface waters (0-200), significant  Table 5. pH trends ± standard error (10 −3 yr −1 ) calculated from the data presented in Fig. 4  Trends of aragonite saturation states are shown in Fig. 7 and Table 6. As for pH, the rates of change is strongest in surface waters. For Ω Ar , the rates of decline are in the order of 10 −2 yr −1 and significant in all regions except for the Greenland Sea.
The weak decline in the Greenland Sea surface layer is a result of a smaller increase in C T in combination with relatively strong increases in A T and temperature, which counteracts the effect of C T on the saturation states (while the temperature amplifies During this period of time we detect trends in the uncertainties of pH and Ω Ar (Figs. S6 and S7). These are, however, about two orders of magnitude smaller than the trends pH and Ω Ar , and they do therefore not significantly impact our estimated trends.

Future changes
The future evolution of pH in the Nordic Seas depends strongly on the emission scenario (Fig. 2). Here we will present, in more detail, the future evolution of pH and Ω Ar under low and high emission scenarios, RCP2.6 and RCP8.5 respectively, as simulated by NorESM1-ME.
In RCP2.6, an additional pH decline of 0.06-0.11 in the surface waters is simulated between present (1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005) and future 420 (2090-2099) (Fig. 8c). The largest pH decreases are found in polar waters, leading to a weakening of the zonal gradient in pH that we see in the present and pre-industrial periods. The surface Ω Ar is expected to decrease by about 0.2-0.5. Surface waters are expected to remain supersaturated with respect to both calcite and aragonite under RCP2.6. An interesting feature in this scenario, is that the strongest ocean acidification occurs at depths of 1000-2000 m (Fig. 9c), which leads to a shoaling of the aragonite saturation horizon to a depth of 1100 m (uncertainty range: 800-1200 m). This is discussed in more detail in Sect. Under the RCP8.5 scenario, the pH in surface waters drops by about 0.4-0.5 from present, to a value of 7.6-7.9 in 2100 ( Fig.   10), with the largest decreases taking place in polar waters. The surface Ω Ar drop by around 1.1-1.3. In contrast to RCP2.6, the largest decline take place in the Atlantic Water. The strong ocean acidification in this scenario leads to a reversal of the pH depth dependency so that by the end of the century pH increases from the surface to deep ocean (Fig. 9c), reflecting that the 430 input of anthropogenic carbon at the surface overrides the effect of pressure and organic matter remineralization on the vertical pH gradient. The change in Ω Ar is large enough to bring the entire water column, and consequently also the entire seafloor, to aragonite undersaturation. The only exception is a thin surface layer (above 30 ±10 m) in the Atlantic Water region.

435
To understand what has caused the pH changes presented in Sect. 3, we decompose the trends into their different drivers using Eq. 2 and subsequent transformation of f CO 2 to pH (Fig. 11). In the surface layer (i.e. the upper 200 m) the pH decrease is in agreement (within 95% confidence) with the pH change expected from the increase in atmospheric CO 2 , except in the Norwegian Basin and the Iceland Sea where the trends are larger. This is related to a faster increase in the seawater pCO 2 compared with that of the atmosphere (Fig. S18), meaning that the pCO 2 undersaturation of the Norwegian Basin 440 and the Iceland Sea is diminishing. The significance of these results are, however, sensitive to the choice of months, i.e., trends calculated with data from April-September, or July-August, give different results. The strong positive trend in pCO 2 that we see in our dataset can therefore be a result of seasonal undersampling, and should be verified with a larger dataset.
Notwithstanding, diminishing pCO 2 undersaturation have been observed in earlier studies of the North Atlantic (Lefèvre et al., 2004;Olsen et al., 2006;Ólafsson et al., 2009;Metzl et al., 2010;Skjelvan et al., 2014), and could be a result of a change winter in the North Atlantic subpolar gyre, which they suggested to be a result of a decrease in productivity. One other possible mechanism was suggested in Olsen et al. (2006) and Anderson and Olsen (2002), where they associated the fast increase in seawater pCO 2 with a large advective supply of anthropogenic carbon from the south and corresponding changes in the buffer capacity (see also Terhaar et al. (2020b)).
The main driver of the pH decrease in the surface layer is increasing C T , which is partly offset by A T increases (see also Fig.

455
S4). The effect of increasing A T is strongest in the Barents Sea Opening that, together with an apparent cooling, explain the low, non-significant, pH decline observed there (Fig. 4.1). The overall increase in A T is partly a consequence of increasing salinities in the Nordic Seas in the past decades (Fig. S2, S19), which also have been observed in many studies and has been explained to be a result of changes in the inflowing Atlantic Water related to subpolar gyre strength (Holliday et al., 2008;Lauvset et al., 2018). The increasing salt content does not only affect A T , but also equally C T . This effect is, however, about the same 460 magnitude as the A T driver, but in opposite direction (Fig. S19). The effect of changes in temperature on pH in the surface layer is relatively small. In contrast to several studies pointing towards a warming of the Nordic Seas (e.g. Holliday et al., Figure 11. Contribution of observed changes in temperature, salinity, CT , AT to the observed trend in pH (OBS) over the 1981-2019 period.
Bars showing trends that are significantly different from zero are outlined with a black line. Sum indicates the total trend in pH calculated as the sum of the trends associated with these four driving factors. The dashed line and black asterisks indicate the pH trends expected from the change in atmospheric CO2 during the same period for the whole area and for the separate basins, respectively.
The ambiguous effect of temperature in surface waters is a result of unequal distribution of sampling over the seasons. When 465 calculating trends with all available temperature data, not only the ones that accompanied the C T and A T data, we obtain a clear warming signal (not shown). In an attempt to estimate the effect of seasonal under sampling on our surface pH trends, we also calculated the trends and their drivers by using data from the productive season (April-September) only. The pH trends obtained from these data are not significantly different from the ones in Fig. . However,  for the surface layer, part of the C T and A T increase can be explained by increasing salinites, but there is also a biogeochemical component (Fig. S19). The uncertainties in the freshwater and biogeochemical components are, however, large, making the 480 decomposition uncertain. The warming seen in the deep waters is likely a result of the decreased deep-water formation in the Greenland Sea and the following increased exchange with warmer Arctic deep waters (e.g. Østerhus and Gammelsrød, 1999;Blindheim and Rey, 2004;Karstensen et al., 2005;Somavilla et al., 2013). The relatively strong trends in C T and pH in the upper 2000m of the Greenland and Iceland seas could be a consequence of deep winter mixing (Våge et al., 2015;Brakstad et al., 2019). However, the convection in the Iceland Sea has only been documented to reach depths of 200-400 m (Ólafsson, 485 2003; Våge et al., 2015). The signal in the deep Iceland Sea is therefore likely a result of spreading of intermediate waters from the Greenland Sea (Messias et al., 2008;Jeansson et al., 2017). Also in the Norwegian Basin there is a significant trend down to 2000 m, although weaker than in the other basins. This is likely also a result of advection from the Greenland Sea (Blindheim, 1990;Blindheim and Rey, 2004;Jeansson et al., 2017). The water-masses in the 2000-4000 m range are increasingly dominated by old Arctic deep waters (e.g. Somavilla et al., 2013). With ages of around or more than 200 years (Jutterström and Jeansson, 2008;Stöven et al., 2016), they have been isolated from the increasing anthropogenic CO 2 , which explains the weak trends at these depths.
The exceptionally strong trends in A T in the surface and the 200-500 m layer in Barents Sea Opening are intriguing.  ering that the strong A T trend also exists in the 200-500 m layer, it is likely not a result of seasonal undersampling. Further, the salinity decomposition in Fig. S19 in the Supplementary material suggest that it is not a result of changing salinity, but rather of biogeochemical processes. While this decomposition gives clear results in the 200-500 m layer, the uncertainty of the freshwater component is as large as the biogeochemical component in the surface layer, making the decomposition, and therefore the role of changes in freshwater content and biogeochemical processes, uncertain. This is a result of the uncertainty in the salinity 500 trend (Fig. S2), which could be caused by the presence of the relatively fresh, Norwegian Coastal Current that has been shown to occasionally, under specific wind conditions, spread into the Barents Sea Opening Olsen et al. (2003). One biogeochemical process that could have a potential impact the Barents Sea A T trend is the recurrent blooms of calcifying coccolithophorids (Giraudeau et al., 2016), which consumes A T during growth, and releases A T when their shells are decomposed. There is an indication of an increase in their presence in the Barents Sea (Giraudeau et al., 2016;Oziel et al., 2020). In which direction this 505 would impact the A T depends on horizontal advection, remineralization and burial, and deserves separate dedicated process studies.
For past and future changes, the drivers of surface pH change show similar spatial patterns, except for temperature (Fig.   12). As for the present day changes, the main driver of pH change is an increase in C T , which is larger in Atlantic Water than in polar waters. The larger increase in C T in the Atlantic Water, which is in agreement with what has been observed 510 over the last 2-3 decades (Olsen et al., 2006), can partly be related to their higher buffer capacity (Sect. 1.1). In polar waters, C T is additionally diluted by the increased freshwater export from the Arctic Ocean (Shu et al., 2018) that to varying degree counteracts the effect of atmospheric CO 2 uptake. The increasing freshwater export also results in a dilution of A T and salinity in polar waters that have, respectively, a negative and positive contribution to the pH trend. While the effect of A T dilution is on the same order of magnitude as the effect of C T dilution, the effect of the reduction in salinity is minor. The Atlantic

515
Waters show a tendency towards increasing A T and salinity that partly reduces/amplify the decrease in pH. The temperature has an overall negative effect on the pH trend as a result of an overall warming. From past to present, present to future RCP2.6, the temperature increase is almost non-existing in polar waters, indicating that it has been shielded from warming through the presence of sea ice. In some smaller regions there is even a sign of a cooling, which could be a result of an increased presence of polar waters due to the increasing freshwater export.

520
The combined effect of these drivers explain the zonal gradients in the pH decrease that we saw in Sect. 3.2 and 3.4. From past to present, the largest pH decrease take place in the Atlantic Water due to a stronger uptake of anthropogenic carbon and a stronger warming in these waters. The increasing freshwater export from the Arctic, and the dilution of A T , plays an important role in the eastern Nordic Seas, but it does not override the acidification rate in the Atlantic Water. From present-future, the freshwater export and dilution of A T plays a bigger role, and the acidification becomes larger in polar waters compared to 525 Atlantic Water. For the changes from past-present, and present to future RCP2.6, the zonal gradient in Ω Ar drops follows that of pH, showing the importance of the competing effect of A T dilution in polar waters, and C T uptake in Atlantic Water, respectively. In RCP8.5, there is, in contrast to pH, a larger drop in the easter part. This can be explained by the larger changes in temperature, which affects Ω Ar in the opposite direction.
In the historical run and all three future projections of NorESM1-ME, the change in surface ocean pCO 2 differs from the 530 change in the atmosphere (Fig. S14). From past to present, there is an increase in the undersaturation, i.e. the positive trend in the oceanic pCO 2 lags the trend in the atmosphere. This means that the pH decrease is less than that expected from the increase in atmospheric CO 2 . The lag continues into all the future scenarios, but from around 2040 and onward the oceanic pCO 2 increases faster than that of the atmosphere, resulting in a decreasing undersaturation. In RCP2.6 and 4.5 this gives rise to, on average, stronger decreases in pH (from 1996-2005 to 2090-2099) than expected from the rise in atmospheric CO 2 . In RCP8.5,however,535 the difference between the end-of-the century ocean and atmospheric pCO 2 is still larger than the present day, meaning that the decrease in pH is less than expected. As detailed above there are several mechanisms underlying undersaturation of surface ocean pCO 2 in the Nordic Seas, but further analyses of these, including their potential future changes, is beyond the scope of this paper.
Below the surface layer, C T is also the main driver of past and future pH changes (Fig. 13). The change from pre-industrial to present indicates a gradually weaker impact of C T with depth, except for a tongue at about 1000 m depth that connects to the surface in the Iceland sea. This is most likely related to the deep water formation in this region that spreads at depth.
The end-of-the-century C T increase under the RCP2.6 scenario is larger in the deep than in the surface layer, resulting in the stronger pH reduction at mid-depths as seen in Fig. 9. This mid-depth layer with a strong acidification is partly a result of the higher atmospheric CO 2 concentrations in the middle of the 21st century , in combination with the rapid ventilation of the 545 water column in this area, i.e. when these waters were at surface they were exposed to peak atmospheric CO 2 . However, the large C T increase in deep waters is also partly explained by increased remineralization, as indicated by a ∼1 ml O 2 l −1 increase in the apparent oxygen utilization (AOU) at depths of 1800-2100 m in both RCP2.6 and RCP8.5 (not shown) throughout the Nordic Seas. Assuming Redfield (O 2 :C=132:106) this corresponds to a change in C T of ∼30 µmol kg −1 , which results to a pH decrease of ∼0.1 at the alkalinity in question. Impacts of changes in A T , salinity and temperature, are relatively modest at 550 depth.
The residual between the sum of the four drivers and the actual pH change (Figs. 12 and 13) can be attributed to approximations involved in the decomposition, including the assumption of a linear trend and the use of temporal means (Takahashi et al., 1993;Lenton et al., 2012;Lauvset et al., 2015). These assumptions are least appropriate for the RCP8.5 scenario, where the changes are largest, and therefore the residual is especially large for this scenario. Although the absolute numbers related to 555 the drivers should be taken with care, this decomposition still gives a good estimate of the relative importance of temperature, salinity, C T , and A T on pH changes.

Implications for cold-water corals
Cold-water corals build their structures out of aragonite, which is the more soluble form of calcium carbonate. To some degree, living corals can compensate for aragonite undersaturation in seawater and increase their internal pH by 0.3-0.6 (McCulloch 560 et al., 2012; Allison et al., 2014). For some time these corals can therefore continue to calcify in waters with Ω Ar <1, however, the calcification rates and breaking strength of the structures of the most abundant coral organism, Lophelia pertusa, is reduced under such conditions (Hennige et al., 2015). Furthermore, dead coral structures, which compose a major part of the reefs, cannot resist corrosive waters and experience increased dissolution rates in a situation with Ω Ar <1. Cold-water coral reefs, along with their ecosystems, are therefore likely to collapse if the water they live in becomes undersaturated in aragonite.

565
It has been estimated that globally about 70% of the deep sea corals will be below the aragonite saturation horizon by the end-of-the-century under high-emission-scenarios (Guinotte et al., 2006;Zheng and Cao, 2014).
Most of the reef sites that have been identified in the Nordic Seas (321 out of the 324 within the region defined in Fig.   1) are at depths of 0-500 m (Fig. 14,  saturation horizon will shoal to 1100 m (uncertainty: 900-1300 m) by the end of the century. In the emission-driven RCP4.5 scenario, the saturation horizon is projected to shoal to 600 m depth (uncertainty: 400-800 m) by the end of this century. This 575 implies that the deepest observed reefs will be exposed to corrosive waters, and thus experience elevated costs of calcification and dissolution of dead structures. The majority (315 out of 324) of the coral sites in the Nordic Seas are, however, found at shallower depths than the projected saturation horizon with its uncertainty, although the margins are small. Also García-Ibáñez et al. (2021) suggested that cold-water corals in the subpolar North Atlantic will be exposed to corrosive waters if the 2-degree goal (which is the aim of RCP2.6) is not met. In the RCP8.5 scenario, NorESM1-ME projects the whole water column to be 580 undersaturated in aragonite at the end of this century, such that all cold-water coral reefs in the Nordic Seas will be exposed to corrosive waters. Because of the low Ω Ar in surface waters, the uncertainty of Ω Ar related to mapping, measurements and dissociation constants does not result in any uncertainty in the saturation horizon in this scenario (i.e. Ω Ar < 1 in the surface waters also when taking into account the uncertainties). For RCP8.5 the NorESM1-ME results are consistent with our CMIP5 model ensemble that suggests that the future saturation horizon lies in the range of 0 and 100 m. Comparison with the CMIP5 585 ensemble is not possible for RCP2.6 and RCP4.5 because few of the models have performed emission-driven runs under these scenarios. However, NorESM1-ME simulates among the stronger drops in pH in all depth layers considered in Fig. S5, and have also been shown be in the upper end of absorption of anthropogenic carbon in the Arctic Ocean (Terhaar et al., 2020a), suggesting that out estimates of the future saturation horizon lies in the upper bound of possible future states.

590
We have provided a detailed analysis of spatial and temporal variations of past, present and future acidification, and its drivers, in the Nordic Seas. We have further assessed the potential impacts of this acidification on aragonite saturation and cold-water coral reefs.
pH changes and its potential ecosystem impacts From 1850 to 1980 both the model simulation of NorESM1-ME and observational data, together with the GLODAPv2 pre-595 industrial estimate, suggest that the pH of Nordic Seas surface waters has dropped by 0.06, which is similar to the pH decrease of the global surface ocean. During this period, the aragonite saturation horizon has slightly shallowed, but has remained well below the depths of known cold-water coral habitats. During the last 39 years covered by this study, when regular sampling of carbon system variables have been made in the region, the pH of the Nordic Seas surface waters has decreased at a rate of -2.79±0.3 10 −3 yr −1 on average, resulting in a pH decline of 0.11 between 1981 and 2019. This decrease is stronger than 600 the decrease observed for the global ocean of -1.80±0.4 10 −3 yr −1 for the period 1991-2011 . The pH reductions are significant all over the Nordic Seas surface waters, except in the Barents Sea Opening. In some regions the acidification is detectable down to 2000 m, which we attribute to the deep water formation an how these water-masses spreads at depth, and the waters at 1000-2000 m throughout the Nordic Seas have approached aragonite undersaturation. An additional pH drop of 0.1-0.4 in the surface waters is projected until the end of the century, depending on the emission scenario. In the 605 high-emission scenario, RCP8.5, all cold-water coral reefs will be exposed to corrosive waters by the end of the 21st century, threatening not only their existence, but also that of their associated ecosystems. This is confirmed by an CMIP5-ensemble of 7 models, whose members all agree on these consequences. The NorESM1-ME simulations suggest that some cold-water corals will be exposed to undersaturation also under the RCP4.5 scenario, and that this only can be avoided by keeping the emissions within the limits prescribed in the RCP2.6 scenario. In comparison to our ESM-ensemble, NorESM1-ME tends to simulate 610 shallow saturation horizons. These results can therefore be considered as careful estimates.

pH drivers
The acidification during the last 39 years is mainly driven by increasing C T coming from the uptake of anthropogenic carbon.
The effects of increasing C T is slightly opposed by increasing A T , which partly comes as a result of the increasing salinities, i.e. "the salinification", of the Nordic Seas. While in the deep waters there is a clear warming signal, which has contributed 615 to the decreasing pH, the impact of temperature in the surface is ambiguous, and even shows a cooling in some places. We find this apparent cooling to be a result seasonal undersampling, which further complicates a comparison of the changes in sea surface pCO 2 to the atmospheric one. In the Barents Sea Opening, there is an exceptionally strong increase in A T , which we cannot relate to increasing salinity. The reasons behind this strong increase is then either a result of biogeochemical processes, or can also be a result sampling issues. Unfortunately, we cannot pin this down with the dataset we have, and this remains as 620 an open question for future investigations.
Also for past and future changes, increasing C T is the main driver of pH change in the Nordic Seas, but here we can distinguish some regional differences related to the different water-masses. In the Atlantic Water, the pH change is mainly driven by increasing C T and temperatures, and slightly opposed by increasing A T related to a salinification, as we also see in our observational dataset for the period of 1981-2019. In polar waters, however, there is a clear signal of increasing freshwater 625 export from the Arctic which has an important impact on the acidification through dilution of C T , A T , and salinity. The dilution of C T slightly opposes the effect of uptake of anthropogenic carbon, which increases the relative impact of decreasing A T on the pH drop. The absence of this freshwater signal in our observational dataset might be a result of the relatively short time scale, but it is also possible that our regions are located to far to the East. The data from the time-series station in the Iceland Sea can be obtained from the NCEI database (Ólafsson, 2012; Ólafsdóttir et al., 2020) 635 The data from the Norwegian ocean acidification monitoring program (Chierici et al., 2019a), and from the Eastern Fram Strait (Chierici and Fransson, 2019) is available at the Norwegian Marine Data Centre (NMDC).
The ESM simulations can be downloaded at https://esgf-node.llnl.gov/search/cmip5/ The cold-water coral positions have been derived from data that is made available under the European Marine Observation Data Network (EMODnet) Seabed Habitats initiative (www.emodnet-seabedhabitats.eu), financed by the European Union under Regulation (EU) No Author contributions. AO, FF and FF designed the research. FF, FF, and AO performed the data-analysis with inputs from NG, IS, MC and EJ. FF lead the writing of the manuscript with inputs from all co-authors. JT designed, tested, and performed the NorESM1-ME model simulations.
Competing interests. The authors declare that they have no conflict of interest. Kwiatkowski, L. and Orr, J. C.: Diverging seasonal extremes for ocean acidification during the twenty-first century, Nature Climate Change, 8, Ólafsson, J., Ólafsdóttir, S. R., Takahashi, T., Danielsen, M., and Arnarson, T. S.: Enhancement of the North Atlantic CO2 sink by Arctic Waters,