Aragonite saturation states and pH in western Norwegian fjords : seasonal cycles and controlling factors , 2005 – 2009

The uptake of anthropogenic carbon dioxide (CO2) by the ocean leads to a process known as ocean acidification (OA), which lowers the aragonite saturation state (Ar) and pH, and this is poorly documented in coastal environments including fjords due to lack of appropriate observations. Here we use weekly underway data from the Voluntary Observing Ships (VOS) program covering the period 2005– 2009 combined with data from research cruises to estimate Ar and pH values in several adjacent western Norwegian fjords, and to evaluate how seawater CO2 chemistry drives their variations in response to physical and biological factors. The OA parameters in the surface waters of the fjords are subject to strong seasonal and spatially coherent variations. These changes are governed by the seasonal changes in temperature, salinity, formation and decay of organic matter, and vertical mixing with deeper, carbon-rich coastal water. Annual mean pH and Ar values were 8.13 and 2.21, respectively. The former varies from minimum values (≈ 8.05) in late December – early January to maximum values of around 8.2 during early spring (March–April) as a consequence of the phytoplankton spring bloom, which reduces dissolved inorganic carbon (DIC). In the following months, pH decreases in response to warming. This thermodynamic decrease in pH is reinforced by the deepening of the mixed layer, which enables carbon-rich coastal water to reach the surface, and this trend continues until the low winter values of pH are reached again. Ar, on the other hand, reaches its seasonal maximum (> 2.5) in midto late summer (July–September), when the spring bloom is over and pH is decreasing. The lowestAr values (≈ 1.3–1.6) occur during winter (January– March), when both pH and sea surface temperature (SST) are low and DIC is its highest. Consequently, seasonal Ar variations align with those of SST and salinity normalized DIC (nDIC). We demonstrate that underway measurements of fugacity of CO2 in seawater (fCO2) and SST from VOS lines combined with high frequency observations of the complete carbonate system at strategically placed fixed stations provide an approach to interpolate OA parameters over large areas in the fjords of western Norway.


Introduction
The continued emissions of carbon dioxide (CO 2 ) (Le Quéré et al., 2015) are of global concern, not only because they are the main driver of anthropogenic global warming, but also because of the changes in the ocean chemistry they cause (Ciais et al., 2013).The increase in the atmospheric CO 2 concentration drives a net ocean CO 2 uptake, which leads to higher proton (H + ) concentration, i.e. lower pH, lower concentration of carbonate ion (CO 2− For the open ocean, the rate of OA has been relatively well documented and understood during the last decade.Observations from time series stations and voluntary observing ships in different oceanic regions consistently show systematic changes in surface ocean chemistry that result from OA. Specifically, long-term negative trends of pH and saturation state for aragonite ( Ar ) have been observed (e.g.Lauvset et al., 2015;Bates et al., 2014).
For coastal regions observed rates of pH change largely differ from those expected from oceanic CO 2 uptake alone, as variations in other biogeochemical processes, related for example to changes in nutrient loading and eutrophication, are important as well (Clargo et al. 2015;Provoost et al., 2010;Wootton et al., 2008).
The Norwegian west coast (Fig. 1) is dominated by fjords, narrow and deep estuaries, carved by glacial processes, with a sill in the mouth where they connect to the coastal North Sea.Apart from being important recreation areas and marine pathways, these fjords are important ecosystems and their physics, marine life, and associated environmental pressures have been relatively well studied (e.g.Matthews and Sands, 1973;Erga and Heimdal, 1984;Asplin et al., 2013;Brattegard et al., 2011;Stigebrandt, 2012;and references therein).
However, only a few studies on the carbon cycle of Norwegian fjords exist and these are only from the high Arctic at Svalbard (Fransson et al., 2014;Omar et al., 2005).Generally, in the Northern Hemisphere, high latitude coastal regions are thought to be sinks for atmospheric CO 2 , while low-latitude regions are thought to be CO 2 sources (Borges et al., 2005;Cai et al., 2006;Chavez et al., 2007;Chen and Borges, 2009).The few existing studies of Norwegian fjords confirm the above picture; i.e. they act as an annual net sink for atmospheric CO 2 (Fransson et al., 2014;Omar et al., 2005).
The carbon cycle of the northern North Sea, to which the western Norwegian fjords are connected, has been well studied (Thomas et al., 2004(Thomas et al., , 2005(Thomas et al., , 2007(Thomas et al., , 2009;;Bozec et al., 2005Bozec et al., , 2006;;Omar et al., 2010).However, observationbased OA estimates are still scarce.Recently, Clargo et al. (2015) observed a rapid pH decrease in the North Sea, but after accounting for biological processes, they estimated an ocean acidification rate consistent with concurrent atmospheric and open-ocean CO 2 increases over the period they studied, 2001-2011.Filling the knowledge gap on OA (and generally the carbon cycle) in western Norwegian fjords is important because these areas are spawning grounds for different fish species (Salvanes and Noreide, 1993;Johannessen et al., 2014), production sites for pelagic calcifiers (Berge, 1962;Erga and Heimdal, 1984;Frette et al., 2004), the home for some coral reefs (e.g.Fosså et al., 2002), and significant food sources due to the aquaculture industry that operates there.Observations of the carbon cycle dynamics in the fjord system will not only further our understanding and ability for prediction, but also serve as benchmarks against which future changes are compared.
In this study, we present first estimates of OA parameters in surface waters of several adjacent western Norwegian fjords (Fig. 1), based mainly on weekly underway data from the Voluntary Observing Ships (VOS) program covering the period [2005][2006][2007][2008][2009].We combine the underway data with available station data from research cruises to facilitate a complete description of the seawater CO 2 chemistry in accordance with the recommendations of OA core principles by McLaughlin et al. (2015).We focus on analyses of Ar and pH values and evaluate their variations in response to the physical and biological factors: summer warming and stratification, spring phytoplankton bloom, and deep mixing during fall and winter.First we present the mean distribution across the different fjords (Korsfjord-Langenuen-Hardangerfjord) to understand the spatiotemporal patterns, then we collapse all data into a monthly time series to analyse the seasonal controls and resolve any interannual or multiyear temporal patterns.

The study area
The study area covers, from north to south, the interconnected Raunefjord (centred around 60.27 • N, 5.17 • E), the Korsfjord (centred around 60.17 • N, 5.21 • E), Langenuen, and southern parts of the Hardangerfjord, which are all situated along the western coast of mainland Norway (Fig. 1).The area stretches over some 60 km, but the main focus here will be on the area from the Korsfjord to the Hardangerfjord from which the vast majority of the data have been acquired.
The bathymetry and hydrographic conditions of the fjords have been described elsewhere (Helle, 1978;Mathew and Sands, 1973;Bakke and Sands, 1977;Erga and Heimdal, 1984;Asplin et al., 2014).In the following only a brief account, based on the above studies, is given.
The Korsfjord is 690 m deep in its main basin and situated about 25 km south of the Norway's second largest city, Bergen.To the west, Korsfjord is relatively well connected to the coastal North Sea through a 250 m deep sill at Marsteinen.To the north it connects with the Raunefjord through the 100 m deep strait Lerøysundet -between the islands Sotra and Lerøy.At the eastern end the fjord branches into the smaller and shallower fjords Lysefjord and Fanafjord, and to the southwest it connects with the coastal North Sea through the Selbjørnsfjord, which has a sill depth of 180 m at Selbjørn.To the south it connects to the Hardangerfjord through the 25 km long and 300 m deep strait Langenuen.
The Hardangerfjord is a 179 km long fjord ranking as the fourth longest fjord in the world.It stretches from the coastal open ocean in the southwest to the mountainous interior of Norway.Our study includes the southern parts of the fjord.This is bounded by the larger islands Stord and Tysnesøya in the north, the Haugaland peninsula in the south, and the smaller islands Fjellbergøya and Halsnøya on the southeast side.This part of the fjord is over 300 m deep in its basin (around 59.76 • N, 5.55 • E) and connects with the smaller fjords Ålfjord and Bjoafjord in the south.
In the fjord system, run-off from land mixes with salty water originating from the northward flowing Norwegian Coastal Current (NCC) to produce a typically salinity stratified water column with a complex circulation, forced both by external and internal factors.In particular, the coastal winds have a profound influence on the water circulation in western Norwegian fjords producing episodic renewal of the deep water that follows periods of prolonged northerly winds (Svendsen, 1981;Erga and Heimdal, 1984).
Besides wind forcing, the hydrography of the fjords is also influenced by winter cooling, summer warming, and runoff.The fjords also receive freshwater through the NCC, which carries water originating from the Baltic Sea and rivers in the southern North Sea (Skagseth et al., 2011 and references therein).On seasonal timescales, salinity drives stratification during spring-summer and the water column is more homogenous during winter.Additionally, Asplin et al. (2014) reported regular episodes of water exchange between Hardangerfjord and the NCC that homogenized the upper 50 m of the fjord by mixing with coastal water.During these events the water temperature inside the Hardangerfjord regularly becomes identical with that of the adjacent coastal North Sea (Asplin et al., 2014).
Water exchange with the NCC is important for the fjord ecosystems as it supplies nutrients and oxygen to the area (Aure and Stigebrandt, 1989).In response, primary production is enhanced in the fjords, which support rich and diverse marine life (Erga and Heimdal, 1984;Erga, 1989;Salvanes and Noreide, 1993).Erga and Heimdal (1984) studied the dynamics of the spring bloom in the Korsfjord and estimated a total primary production of 74 g C m −2 during the period February-June.Further, they reported light regime and water column stability to be dominant controls of the onset of the bloom.They also pointed out that changes in the alongshore wind component are important for the bloom dynamics, with persistent northerly winds inducing upwelling of nutrient-rich coastal water that promotes blooming, whereas the opposite situation occurs from persistent southerly winds.During calm periods strong stratification develops, which can ultimately lead to nutrient exhaustion in the upper water column.
The study area, with its adjacent waters, is ecologically and economically important because it covers spawning grounds for a number of different fish species (Lie et al., 1978;Johannessen et al., 2014) (Fosså, 2015).The fjord system also contributes to the important aquaculture production that, with its annual fish production of > 700 tonnes, ranks Norway within tenth place worldwide.About one-fifth of this is produced in the Hordaland County where the fjord system studied here is situated (http://www.diercke.com).
2 Data and methods

Weekly underway VOS data
Weekly underway measurements of fugacity of CO 2 in seawater (f CO 2 ) and sea surface temperature (SST) were obtained aboard the containership MS Trans Carrier (operated by Seatrans AS, Norway, www.seatrans.no).During the study period, the ship sailed from Bergen to ports in southwestern Norway on a weekly basis.It passed through several fjords including the Korsfjord and the Hardangerfjord (Fig. 1), then crossed the North Sea mostly along a transect roughly at 5 • E longitude to Amsterdam, the Netherlands, and then back on the same route (Omar et al., 2010).The measurement method used aboard MS Trans Carrier was described in Omar et al. 2010).Briefly, the instrument uses a non-dispersive infrared CO 2 / H 2 O gas analyser (LI-COR 6262) to determine the CO 2 concentration in a headspace air in equilibrium with a continuous stream of seawater.One analysis was done every 3 min and the instrument was calibrated roughly every 6 h with three reference gases, which are traceable to reference standards provided by National Oceanic and Atmospheric Administration -Earth System Research Laboratory (NOAA/ESRL).The instrument automatically shut off when the ship approached ports in Bergen (20-30 km from port ≈ 60.2

Cruise and fixed station data
We augment the VOS data with station data acquired during scientific cruises in the study area in the period 2007-2010 and in 2015, and during regular visits (1-4 times per month) to a fixed station in the Raunefjord in 2007 and 2008.Table 1 summarizes details of these three data sets, which will be referred to as the CS, 2015, and RF data sets.Five of the cruises were conducted in the Korsfjord and the Raunefjord (Fig. 1, Table 1) onboard RV Hans Brattstrøm as part of the EU FP7 educational project CarboSchools (CS) in 2007-2010.The CS data set covers mainly the spring and summer seasons reflecting the somewhat opportunistic nature of the sampling campaign.The 2015 cruise took place during fall (24 September) as part of the Ocean Acidification project funded by the Norwegian Environment Agency, and measurements were taken at three stations in Korsfjord, Langenuen, and southern Hardangerfjord (Fig. 1, red squares).
During each of the above cruises water samples were collected for analyses of parameters including dissolved inorganic carbon (DIC), total alkalinity (TA), salinity and temperature at 1-2 stations.The DIC concentrations were determined by the coulometric method (e.g.Johnson et al., 1993) with a precision of ±1.2 µmol kg −1 .TA was measured by potentiometric titration with strong acid (HCl), and a precision of ±4.8 µmol kg −1 .Accuracy was checked by using certified reference material supplied by A. Dickson at Scripps Institution of Oceanography.Once all samples have been corrected with respect to offsets determined from the CRM measurements, the DIC and TA measurements were accurate to within 1.4 times respective measurement precision (above).Only surface data (depth< = 4 m) from within the geographical rectangle 59.74-60.34• N and 5.17-5.55• E were used in the current study.
The Department of Biology, University of Bergen, has acquired data for conductivity, temperature, and depth (CTD) from a fixed station in Raunefjord (RF) during 27 days in 2007 and 35 days in 2008 as part of a monitoring program close to the marine biological field station at Espegrend.These data contained temperature and salinity profiles with 1 m resolution.Averages of the uppermost 5 m have been used in this study and will be referred to as the RF data set.

In situ pH sensor data
In January 2012 we carried out an evaluation of two pH sensors of the type Submersible Autonomous Moored Instruments (SAMI_pH, second generation) at the marine biological field station at the eastern shore of the Raunefjord.The sensors were suspended from a wooden frame attached to the floating docks around a raft house in the fjord -some hundred metres from land.The instruments were submersed at about 1 m depth in the fjord and were left for 50 h starting 24 January 2012 10:00 GMT, recording one measurement each hour.A full description of the measurement method for these instruments is found at http://www.sunburstsensors.com/.In addition to pH, these instruments also recorded the seawater temperature and they have measurement precision and accuracy of < 0.001 and ±0.003 pH units, respectively.During the test, salinity was also recorded using a Seaguard Recording Current Meter (RCM) from Aanderaa Data Instru-Ocean Sci., 12, 937-951, 2016 www.ocean-sci.net/12/937/2016/We obtained a complete description of the seawater CO 2 chemistry from the UW SST and UW f CO 2 data collected onboard MS Trans Carrier through a 3-step procedure as described below.This is similar to the procedure described in Nondal et al. (2009) with the main modification being that in the current study, sea surface salinity (SSS) was determined from an empirical relationship with SST.First, the RF data set has been used to determine the regional SSS vs. SST relationship.The RF data were chosen for this purpose because it covered all seasons well, both in 2007 and 2008.The identified regional SSS-SST relationship allowed us to estimate a SSS value for each UW SST observation from MS Trans Carrier.This step was necessary because the total number of measured SSS values were less than 150 data points, while the available underway SST and f CO 2 data were much more numerous (> 9900 data points), covering most of the study area during the years 2005-2009.The remaining SST and SSS data (CS, and from sensors) were used for evaluation to verify that SST-SSS relationship is valid for the whole study area (Sect.3.1).Salinity values estimated from SST will be denoted as SSS(sst).
Second, we determined TA from SSS(sst) and SST using an algorithm we identified for the region using the CS data set.This allows us to estimate a corresponding alkalinity value for each UW f CO 2 observation obtained from MS Trans Carrier.Alkalinity values estimated from measured SSS and SST data will be denoted as TA(sss), whereas TA values estimated from SSS(sst) and SST values will be denoted as TA(sst).
The UW f CO 2 together with TA (sst), UW SST, and SSS(sst) were then used to characterize the full seawater CO 2 chemistry using CO2SYS (Lewis and Wallace, 1998;van Heuven et al., 2011), with K1 and K2 constants from Lueker et al. (2000).The concentration of silicate and phosphate has been put to zero during the calculations, and the errors introduced by this simplification were negligible compared to the uncertainties from other sources (to be described in Sect.3.1).The CO2SYS calculation also gives DIC, pH, Ar , and all other seawater CO 2 chemistry variables.The data estimated using this three stage procedure will be denoted pH(sst) and Ar (sst) and are the main focus of this study.
pH and Ar values based on TA(sss) and f CO 2 will be denoted as pH(sss) and Ar (sss), whereas values that are either measured or computed from measured TA and DIC will be denoted as simply pH and Ar .nDIC denotes the DIC values normalized to constant salinity (the mean value) according to Friis et al. (2003) with freshwater endmember DIC concentration of 1039 µmol kg −1 inferred from the cruise data.An overview of the symbols used for estimated and derived quantities used in this study is given in Table 2.

Correlations and validations
In this section we present the regression equations identified in this study in addition to validating the various estimation procedures used by comparing the estimated values with those measured/computed.The results of these comparisons are summarized in Table 3.For each comparison, the coefficient of determination (R 2 ) and the significance level (p-value) are used as metrics for the goodness of the correlation while the associated root mean square error (RMSE) is benchmarked against the maximum target uncertainties developed by the Global Ocean Acidification Observing Network (GOA-ON) and the California Current Acidification Network (C-CAN) of ±0.2 for Ar (McLaughlin et al., 2015), which corresponds to maximum uncertainties of ±0.02, ±1.25, or ±1.8 in pH, SST, or SSS, respectively.
The regional SST-SSS relationship obtained from the RF data set is given by Eq. ( 1) and is depicted in Fig. 2a (filled symbols).Despite a clear covariation between SST and SSS, there is a lot of scatter in the data and the statistics of the regression equation is not particularly strong (Eq.1).The observed correlation most probably arises from the annual cycles; during summer the study area embodies warm water diluted by run-off, whereas during winter the surface water is colder and saltier due to little or no run-off.The magnitude of these annual variations varies with time and space and this is reflected by the high degree of scatter in the relationship.Consequently, the identified regression model is able to explain only 27 % of the salinity variations.Nonetheless, the independent station and sensor data (dots, squares, and stars), which have been acquired from the whole study area in different seasons, falls into a pattern around the relationship described by Eq. ( 1) with a RMSE of 0.81.Thus, we assume that Eq. ( 1) is able to estimate the seasonal SSS variations across the whole study region.To verify this we have compared the monthly averages of RF_SSS data with values obtained using Eq. ( 1) and monthly RF_SST.As shown in the last row of Table 3, the estimated values were significantly correlated with the monthly RF_SSS (R 2 = 0.65 and p = 0.002) and the resulting RMSE of 0.3 was lower than the benchmark values of ±1.8.
As further verification that the RF SST data set is spatially representative, we compared it with the chronologically colocated UW SST that have been acquired onboard MS Trans Carrier across the whole study area.The two data sets were found to be almost identical (Fig. 2b; 3rd row Table 3).The relationship between TA, SSS, and SST obtained from the CS and 2015 data sets is given by Eq. 2 according to TA = 32.09SSS−4.39SST+ 1210 R 2 = 0.90; n = 23; RMSE = 13.0 µmolkg −1 . (2) Alkalinity is a semi-conservative variable and is normally modelled as a linear function of salinity (e.g.Millero et al., 1998;Bellerby et al., 2005;Nondal et al, 2009).However, using a multi-parameter linear regression with SST and SSS as independent parameters improved the regression statistics considerably (R 2 = 0.90; n = 23; RMSE = 13.0 µmolkg −1 ) compared to a linear regression with only SSS (R 2 = 0.67; n = 23; RMSE = 24.0µmolkg −1 ).This is probably because SST acts as an indicator of the effect of nutrient cycling on TA in agreement to what has been reported for the open Atlantic Ocean (Lee et al., 2006).
To estimate a corresponding TA value for each UW f CO 2 observation obtained from MS Trans Carrier, we used salinity values estimated from the UW SST data by using Eq.(1).The results (denoted as SSS(sst)) were then inputted into Eq.( 2) to obtain TA(sst) (see Table 2 for nomenclature).The fact that TA(sst) is based on SSS(sst) rather than measured SSS values introduces an additional error in the estimated pH(sst) and Ar (sst).In order to assess this error we compared pH(sst) and Ar (sst) with values based on the cruise data, i.e. pH(sss) and Ar (sss).First, we computed pH(sss) and Ar (sss) by combining all available measured SSS, estimated TA(sss) from Eq. ( 2), and co-located UW SST and www.ocean-sci.net/12/937/2016/Ocean Sci., 12, 937-951, 2016 TA values determined from estimated SSS(sst) and SST using Eq. ( 2).pH(sst), Ar (sst), DIC(sst) Values of pH, Ar and DIC that have been obtained by combining TA (sst), f CO 2 and ancillary variables.f CO 2@meanSST f CO 2 at the mean temperature nDIC DIC normalized to the mean salinity Table 3. Results of the comparisons between measurement-based and estimated values for pH, Ar , SST ( • C), and SSS.For the first three parameters, the statistics of the linear relationships depicted in Fig. 2b-d are listed.For SSS, monthly averaged data are compared to estimates obtained with Eq. ( 1) using monthly SST.For SST, the comparison is carried out to verify that measurements from Raunefjord are representative for the whole study area (i.e.UW_SST can be estimated by RF_SST), which is implicitly assumed by the use of Eq. ( 1).R 2 is the coefficient of determination, and "RMSE" denotes the root-mean-square error.UW f CO 2 .Then we repeated the calculations, but this time we replaced the measured SSS with estimated SSS(sst) from Eq. (1) to compute pH(sst) and Ar (sst).The very strong linear relationships (R 2 ≈ 1) between the resulting values in Fig. 2c and d (circles) confirm that the estimated pH(sst) and Ar (sst) reproduce closely the measurement-based values of pH(sss) and Ar (sss) for the whole study area.This is also evident from comparison statistics on rows 1 and 2 in Table 3, which show that measured-based values are well correlated (R 2 ≈1, p< 0.001) with those estimated with RMSE values of 0.003 and 0.04 for pH(sss) and Ar (sst), respectively, which are well within the aforementioned maximum target uncertainties developed by the C-CAN (last column in Table 3).
To quantify the total error associated with the pH(sst) and Ar (sst) estimates, we considered two main sources for error.First we computed the residuals (estimated -measurementbased) using the data shown in Fig. 2c and d (including the sensor data).The mean difference for the whole study area was 0.002 ± 0.004 and 0.005 ± 0.08 for pH and Ar , respectively.Thus, the maximum probable error from this source is 0.006 and 0.09 for pH and Ar , respectively.Additionally, we estimated that the computed and/or measured pH values included an error of 0.012 pH units, which under the current conditions (mean TA, f CO 2 , SST, and SSS) would give an error of 0.09 in Ar .These two error estimates were combined (as the square root of sum of squares) to determine the total error in our estimates, which were found to be ±0.013 and ±0.13 for pH and Ar , respectively.It must be noted that the above total error was derived from all available observational data including the in situ sensor data (shown in Fig. 2c and in described Sect.2.3), which are the only wintertime measurements used in this study.This is important because the lack of wintertime data in the CS data set that was used for the identification of TA-SSS/SST relationship (Eq.2), means that wintertime TA(sst) might be overestimated so that corresponding pH(sst) values would be overestimated.In fact, during the aforementioned comparison between pH(sst) and measured pH we noted that for this particular data set pH(sst) overestimated the measurements.However, the estimates were consistent with the observations to within the total error of ±0.01 pH units.Thus, by utilizing the above total errors, we also accounted for the effect of this possible caveat of Eq. ( 2) arising from the lack of wintertime TA measurements.
From the above we conclude that we are able to estimate pH(sst) and Ar (sst) across the whole study area and during all seasons with total errors of ±0.01 and ±0.1 for pH and

Spatiotemporal variations
In order to present the mean distributions across the different fjords throughout the annual cycle, we condensed the data into one virtual year by projecting it onto non-equidistant rectangular grids using the "weighted-average gridding" method of the Ocean Data View software (Schlitzer, 2015).As evident from Fig. 3, there is a clear seasonality (for the interannual changes see Sect.3.3) in both pH(sst) and Ar (sst).The former varies between minimum values (8.05) around the 1 January to typical maximum values of around 8.25, which occur during the late winter and/or spring (March-April).This increase is due to the reduction of DIC (Fig. 3d), induced by the phytoplankton spring bloom.This clearly counteracts and outweighs the negative effect on pH of warming the water column during this period.However, during April/May, the effect of warming begins to dominate and pH(sst) starts decreasing.By September the SST starts decreasing, while pH continues to drop.This is due to the effect of the fall mixing, which enables carbon-rich coastal water to reach the surface layer, as mentioned in Sect. 1, and is reflected by increasing DIC during this period (Fig. 3d).
The mean distribution of Ar (sst) also shows a significant seasonal variation.There are three factors that drive this: (i) reduced concentrations of DIC by the spring bloom increases the concentration of carbonate ions, (ii) Ar (sst) increases with rising temperature so that warming during the summer actually reinforces the increase of Ar initiated by biological carbon uptake, and (iii) reduced TA due to freshwater input from run-off and mixing of deeper carbon-rich water into surface layer reduce Ar (sst) during fall.Thus, Ar (sst) reaches its maximum (> 2.5) in July-September, when the spring bloom is over and pH has already started decreasing (Fig. 3a, c).The lowest Ar (sst) values (≈ 1.3-1.6)occur during winter (January-March) when both pH and SST are low, despite TA being high due to high SSS values.The decoupling in the seasonal cycles of pH and Ar clearly supports the case that pH alone is not an adequate measure of ocean acidification, in accordance with the C-CAN recommendation that "measurements should facilitate determination of Ar and a complete description of the carbonate system, including pH and pCO 2 " (McLaughlin et al., 2015).
The above-described seasonal variations in pH(sst) and Ar (sst) are spatially more or less coherent within the whole study area, except for the slight south-north gradient during May-September, with the highest values south of 60 • N (see Fig. 3a, c).All in all, during summertime the study area em- bodied warm surface water with high Ar (sst) and intermediate pH(sst) values.During winter, the surface water is cold with low Ar (sst) and pH(sst) values.

Controls of seasonal variability and trends
To investigate the seasonal variability more thoroughly, we computed monthly averages of pH(sst), SST, Ar (sst), and nDIC(sst) for one composite year.Then we quantified the effect of DIC, TA, SST, and SSS on the monthly changes of pH(sst) and Ar (sst) in order to gain more insight into the processes governing the seasonal variations and their relative importance (Fig. 4).
For pH(sst) we used the decomposition method described in Lauvset et al. (2015) to quantify the importance of different parameters.This method estimates the monthly pH changes expected from corresponding changes observed in SST, SSS, DIC, and TA as well as their sum.The results are shown in Fig. 4a-e where it can be seen that DIC is the most important driver followed by SST and TA, whereas SSS had a negligible effect (not shown) on the seasonal pH variations.We also note that the effects of SST and TA combined are nearly equal to, but opposite of, that of DIC (Fig. 4c, d, e).
As a result, the sum of all effects is < 0.06 pH units, and compares well to the observed amplitudes (Fig. 4a), meaning that the decomposition model is able to account for the observed seasonal changes.Note also the TA control is identical to that of SST (Fig. 4c, e).The reason for this is that TA values used here are obtained from SSS(sst) and SST using Eq. ( 2), which in effect means that they are based on SST.This emphasizes the need for measured SSS and TA values when the objective is to analyse the controls of pH and Ar (sst) variations.
For Ar (sst) we investigated the importance of different controls (DIC, TA, SST, SSS) by varying them independently over their observed range, while holding all other drivers constant, and re-computing Ar (sst).The magnitude of the standard deviation of the results is indicative of the importance of the varying drivers.The result of this exercise is shown in Fig. 4f-i.Evidently, the variations of SST and SSS are the least important drivers for Ar (sst) seasonal changes, since varying these parameters induces changes that are about an order of magnitude less than the observed seasonal amplitude in Ar (sst).On the other hand, changing DIC and TA (Fig. 4h, i) induces changes that are comparable to the seasonal amplitude observed in Ar (sst) (Fig. 4a).We therefore conclude that seasonal changes in DIC and TA are the most important drivers for changes in Ar (sst).
From the above we conclude that the main drivers of Ar (sst) are DIC and TA, whereas for pH(sst), SST also has a significant impact.This means that the formation and destruction of organic matter together with upwelling of carbon-rich coastal water, seasonal warming and cooling, and run-off inputs, are the processes that govern most of the seasonal variability of OA parameters within the study area.It then follows that interannual variability in the above processes would lead to corresponding variations in pH(sst) and Ar (sst).Such interannual changes are evident from the monthly time series (Fig. S1 in Supplement), where the rate of seasonal change differs between the years, both for SST and DIC normalized to the mean salinity (nDIC) according to Friis et al. (2003).Additionally, for SST, the extreme values also change between the years.These changes are in turn reflected in the pH(sst) and Ar (sst) for which the amplitude of the interannual variability (IAV), calculated as the temporal standard deviation, is presented in Table 4.For pH, IAV was normally much lower than the seasonal changes and ranged between 0.01 and 0.02 although higher changes were observed during the months April (0.04), and July and October (0.03).Similarly, for Ar (sst), the IAV was typically 0.1, which is much lower than the seasonal changes (Sect.3.2).Higher IAVs were observed for June (0.2) whereas November and December showed the lowest IAVs (< 0.05).Quan-titatively, the above IAVs are probably lower limits due to the use of constant empirical relationships for the estimation of SSS and TA (Eqs. 1 and 2).That is, there may be interannual changes in the relationship between SST and SSS (Eq. 1) and/or between SSS/SST and TA (Eq.2).Thus, the use of a constant relationship over the years may have led to underestimation of the resulting IAV.Consequently, a comprehensive analysis of the drivers of the IAV was not carried out in this study.However, the sensitivity computations we performed showed that year-to-year differences in pH were related to those in f CO 2 rather than SST changes, whereas year-to-year differences in Ar (sst) were more related to those in SST than f CO 2 .In any case, the observed year-to-year differences were not systematic, and no multiyear temporal trend was apparent from the 4-year time series analysed in this study.

Inference of OA parameters from VOS underway data
Changes in the oceanic CO   1), and the slope of these relationships can be expressed mathematically by where expressions for the carbonate alkalinity Alk C and the parameters P and S are defined in Egleston et al. (2010).
We have evaluated the right-hand sides of Eqs. ( 3) and ( 4), using the CS cruise data, and the results showed that these quantities change only a few per cent (1.3 and 3.4 %, respectively) due to seasonal changes in the various variables.The ratio γ DIC /ω DIC changed by 1-6 % and ranged from −1.08 to −0.980, while γ DIC /β DIC changed by 0.5-3 % and ranged from 0.84 to 0.88.This, together with the fact that equations 3 and 4 can be defined in terms of ln(f CO 2 ) instead of ln(CO 2 ) (Egleston et al., 2010;Takahashi et al., 1993), suggests that in situations where underway surface f CO 2 and SST are frequently measured, whereas the CO 2 system is fully determined only occasionally, an easy way of interpolating the seasonality in pH and Ar , is to predict them from f CO 2 .We have implemented this alternative way of estimating pH and Ar using the CS cruise data.For the estimation of Ar we used f CO 2@meanSST , which is f CO 2 adjusted to constant temperature (i.e. at mean SST), because this normalization improved the regression significantly.Since we were interested in pH and Ar we plotted these parameters directly against ln(f CO 2 ) or ln(f CO 2t@meanSST ).The results are shown in Fig. 5 and conform to tight relationships between computed pH and ln(f CO 2 ) values (Fig. 5a), and between computed Ar and ln(f CO 2t@meanSST ) (Fig. 5b).
The magnitude of the residuals (computed -estimated) associated with pH and Ar values obtained from the above re-  lationships were 0.000 ± 0.005 and 0.01 ± 0.06, respectively, which is comparable to the residuals associated with pH(sst) and Ar (sst) (Table 3).An advantage of this procedure, however, is that it utilizes much tighter empirical relationships, involves fewer computational steps, and is based on UW data, which are much more numerous than station data from oceanography cruises.Thus, it minimizes errors introduced by intermediate results such as the TA-SSS/SST regression in Eq. ( 2) and/or seasonal data coverage.Furthermore, a direct comparison revealed that values obtained from Eqs. ( 5) and ( 6) were almost identical to those of pH(sst) and Ar (sst) (Fig. S2) with values for R 2 , p value, and RMSE of 1, 0, and 0.003 for pH; and 1, 0, and 0.02 for Ar .However, it is important to realize that for the above procedure too, a representative full description of the carbonate system is necessary for up-to-date determinations of Eqs. 5 and 6.Further, this calibration data ideally should include high-frequency time series observations, since the slopes (i.e.Eqs. 3 and 4) change slightly with the carbonate system variables (e.g.DIC and TA; see Eqs. 3 and 4), which vary on multiple timescales (hours-days-years).Furthermore, the procedure is based on measurements of only one of the four master parameters constituting the carbonate system (i.e.f CO 2 ).Therefore, it only provides a way to interpolate pH and Ar values, but cannot support the analyses of controls that have been provided in the proceeding section.From Fig. 5b we note that lowest Ar values are associated with the highest f CO 2@meanSST values, which occur during late fall and winter.Monitoring of these extreme val-ues are of special interest because (i) during late fall and early winter the upwelling of carbon-rich water occurs and surface water also reflects the properties of the deeper water, and (ii) the rate of change at this point (lowest Ar , highest f CO 2@meanSST ) indicates the time when under-saturation of calcium carbonate can be expected in these waters.To estimate this for the current data we used Eq. 5 and the observation that the slope (i.e.Eq. 3) and intercept decreased by about 0.0008 and 0.004 for every 1 µatm increase in mean f CO 2@meanSST.We also took into account an uncertainty of ±0.2 in the Ar estimates and found that Ar becomes undersaturated (< 1) when mean annual f CO 2@meanSST is about 310 ± 70 µatm higher than its present value (310 µatm).For business as usual emission scenario (RCP 8.5), this is equivalent to about year 2070 ± 10 if we assume that the development in the ocean follows that of the atmosphere (i.e.constant disequilibrium between ocean and atmosphere).

Summary and conclusions
We have been able to determine, for the first time, the seasonal changes and drivers of pH and Ar across western Norwegian fjords (59.74-60.23 • N and 5.16-5.6 • E), based on underway f CO 2 and SST data combined with sporadic data from research cruises and empirical relationships.
During summertime the study area embodied warm surface water with high Ar (< 2.5) and intermediate pH values (8.12-8.17).During winter, the surface water was cold with low Ar and pH values, 1.3-1.6 and 8.05-8.07,respectively.Maximum pH of 8.25 was encountered during the phytoplankton spring bloom (March-April).Seasonal changes in DIC, TA, and SST were the most important drivers of pH and Ar changes, although the SST influence on Ar was only weak.
Measurement errors together with seasonal and spatial gaps in the data used to identify the empirical relationships are considered as two main sources for uncertainties, and the computed total errors associated with the estimated values (±0.01 and ±0.1 for pH and Ar) are about 50 % of the maximum target uncertainties developed by the Global Ocean Acidification Network.
We have shown that the strong correlations of pH and Ar with f CO 2 and f CO 2@meanSST (f CO 2 adjusted to the mean temperature), respectively, provide an approach to interpolate pH and Ar values both seasonally and spatially.Furthermore, the Ar-f CO 2@meanSST relationship, as well as the rate of change of its slope and intercept with DIC, has been used to estimate that under-saturation of the aragonite mineral can occur in the study area by the year 2070 ± 10, if we assume a business as usual emission scenario (RCP 8.5) and constant CO 2 disequilibrium between ocean and atmosphere.
The Supplement related to this article is available online at doi:10.5194/os-12-937-2016-supplement.

Figure 1 .
Figure 1.An overview map of western Norway with a detailed map of the study area showing the positions from where cruise and underway data have been acquired.The thick arrow on the right panel indicates the approximate position of the Norwegian Coastal Current (NCC).

Figure 2 .
Figure 2. (a) RF SSS as a function of SST (filled symbols) with the regression line described by Eq. (1).Sampling month is indicated by the colour of the data points.The CS (dots), 2015 (squares), and sensor (stars) data sets are also shown for comparison with the regression line.(b) Compares RF SST with chronologically co-located UW SST acquired from the whole study area during 2008 (blue) and 2007 (red).(c) Compares pH(sst) with pH values that have been measured or computed from TA and DIC.Symbols are as in Fig. 1.(d) Compares Ar (sst) with Ar values that have been computed from measured TA and DIC or from measured pH and UW f CO 2 .Symbols are as in Fig. 1.In the legend, "reg."means regression line.

Figure 3 .
Figure 3. (a) Estimated pH(sst), (b) UW SST, (c) estimated Ar (sst), and (d) estimated DIC, which have been normalized to the mean salinity of 30.5 as a function of latitude and time of the year.All data from 2005 to 2009 have been condensed into one virtual year to resolve the spatial and seasonal variations.

Figure 4 .
Figure 4. Left panel: monthly pH changes ( pH) as observed (a) and expected due to sum of all derivers (b), SST changes (c), DIC changes (d), and by TA changes (e).Right panel: standard deviations in monthly mean Ar as a result of variations in all parameters (f) or only in SST (g) in DIC (h) in TA (i).

Figure 5 .
Figure 5. (a) and (b) pH and Ar from CS (dots) and 2015 (red squares) cruises plotted as a function of ln(f CO 2 ) and ln(f CO 2 meanSST ), respectively.
. Additionally, the largest concentration of coral reefs in western Norway is found in

Table 1 .
Details of the CarboSchools (CS) and Raunefjord (RF) cruise data sets.The plus signs denote the parameters for which sampling/measurement were carried out.For the RF data set, each data point represents the average of five measurements acquired in the upper 5 m.

Table 2 .
Overview of the symbols used for quantities estimated and/or derived from the measurement-based variables SSS, SST, TA, pH, DIC, f CO 2 , and Ar .
The latter is compared against benchmarks derived from maximum target uncertainties (max.uncertainty) developed by Global Ocean Acidification Observing Network (Sect.3.1).The p value is the probability of no linear relation between the estimated and measurement-based values.

Table 4 .
Monthly mean values for pH(sst) and Ar (sst) and associated interannual variability (IAV), computed as standard deviations, in the study area for the period2005-2009.