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

: 8 The uptake of anthropogenic Carbon Dioxide (CO 2 ) by the ocean leads to a process known as 9 ocean acidification (OA) which lowers the aragonite saturation state (Ω Ar ) and pH, and this is 10 poorly documented in coastal environments including fjords due to lack of appropriate 11 observations. 12 Here we use weekly underway data from Voluntary Observing Ships (VOS) covering the 13 period 2005-2009 combined with data from research cruises to estimate Ω Ar and pH values in 14 several adjacent western Norwegian fjords, and to evaluate how seawater CO 2 chemistry 15 drives their variations in response to physical and biological factors. The OA parameters in the surface waters of the fjords are characterized by strong seasonal and spatially coherent variations. These changes are governed by the seasonal changes in 18 temperature, salinity, formation and decay of organic matter, and vertical mixing with deeper, 19 carbon-rich coastal water. Annual mean pH and Ω Ar values were 8.13 and 2.21, respectively. The former varies from minimum values (≈8.05) in late December - early January to 21 maximum values of around 8.2 during early spring (March-April) as a consequence of the 22 phytoplankton spring bloom, which reduces Dissolved Inorganic Carbon (DIC). In the 23 following months, pH decreases in response to warming. This thermodynamic decrease in pH 24 is reinforced by the deepening of the mixed layer, which enables carbon-rich coastal water to 25 reach the surface, and this trend continues until the low winter values are reached again. Ω Ar , 26 on the other hand, reaches its seasonal maximum (>2.5) in mid to late summer (July -Sept),


Weekly underway VOS data
Weekly underway measurements of fugacity of CO 2 in seawater (fCO 2 ) and SST were 161 obtained aboard the containership MS Trans Carrier (operated by Seatrans AS, Norway, 162 www.seatrans.no). During the study period, the ship sailed from Bergen to ports in 163 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 165 roughly at 5°E longitude to Amsterdam, Netherlands, and then back on the same route (Omar   179 We augment the VOS data with station data acquired during scientific cruises in the study  Table 1 summarizes details of these three 182 datasets, which will be referred to as the CS, 2015 and RF datasets, respectively. Environment Agency, and measurements were taken at three stations in the Korsfjord,189 Langenuen and southern Hardangerfjord (Fig.1, red squares).

Cruise and fixed station data
During each of the above cruises water samples were collected for analyses of parameters 191 including DIC, total alkalinity (TA), salinity and temperature at 1-2 stations. The DIC 192 concentrations were determined by the coulometric method (e.g. Johnson et al., 1993) with a 193 precision of ±1 µmol kg -1 . TA was measured by potentiometric titration with strong acid 194 (HCl), and a precision of ±2 µmol kg -1 . Accuracy was checked by using Certified Reference 195 Material supplied by A. Dickson (SIO). Once all samples have been corrected with respect to 196 offsets determined from the CRM measurements, the DIC and TA measurements were 197 accurate to within the respective measurement precision (above). Only surface data 198 (depth<=4m) from within the geographical rectangle 59.74-60. 34°N and 5.17-5.55°E were 199 used in the current study.  were left for 50 hours starting 24.01.2012 10:00 GMT, recording one measurement each hour.

212
A full description of the measurement method for these instruments is found at 213 http://www.sunburstsensors.com/. In addition to pH, these instruments also recorded the 214 seawater temperature and they have measurement precision and accuracy of <0.001 and +/-215 0.003 pH units, respectively. During the test, salinity was also recorded using a Seaguard 216 RCM from Aanderaa Data Instruments. These sensor data were used to assess the uncertainty 217 in our pH values estimated as described in section 3.1. 2.4.1 Complete seawater CO 2 chemistry from SST and fCO 2 A complete description of the seawater CO 2 chemistry from the UW SST and UW fCO 2 data 221 collected onboard MS Trans Carrier has been obtained through a 3-step procedure. This is 222 similar to the procedure described in Nondal et al. (2009) with the main modification being 223 that in the current study, sea surface salinity (SSS) was determined from empirical 224 relationship with SST.

225
First, the RF dataset has been used to determine the regional SSS versus SST relationship.

226
The RF data was chosen for this purpose because it covered all seasons well, both in 2007 and 227 2008. The identified regional SSS-SST relationship allowed us to estimate a SSS value for 228 each UW SST observation from MS Trans Carrier. This step was necessary because the total 229 number of measured SSS values were less than 150 data points, while the available underway 230 SST and fCO 2 data were much more numerous (> 9900 data points), covering most of the 231 study area during the years 2005-2009. The remaining SST and SSS data (CS, and from 232 sensors) were used for evaluation to verify that SST-SSS relationship is valid for the whole 233 study area (section 3.1). Salinity values estimated from SST will be denoted as SSS(sst).

234
Second, we determined TA from SSS(sst) and SST using an algorithm we identified for the 235 region using the CS dataset. This allows us to estimate a corresponding alkalinity value for 236 each UW fCO 2 observation obtained from MS Trans Carrier. Alkalinity values estimated from 237 measured SSS and SST data will be denoted as TA(sss), whereas TA values estimated from 238 SSS(sst) and SST values will be denoted as TA(sst).

239
The UW fCO 2 together with TA (sst), UW SST, and SSS(sst) were then used to characterize 240 the full seawater CO 2 chemistry using CO2SYS (Lewis and Wallace, 1998;van Heuven et al., 241 2011), with K1 and K2 constants from Lueker et al. (2000). The CO2SYS calculation also 242 gives DIC, pH, Ω Ar and all other seawater CO 2 chemistry variables. The data estimated using 243 this three stage procedure will be denoted pH(sst) and Ω Ar (sst) and are the main focus of this 244 study. 245 pH and Ω Ar values based on TA(sss) and fCO 2 will be denoted as pH(sss) and Ω Ar (sss), 246 whereas values that are either measured or computed from measured TA and DIC will be 247 denoted as simply pH and Ω Ar . nDIC denotes the DIC values normalized to constant salinity 248 (the mean value) according to Friis et al. (2003) with freshwater end member DIC 249 concentration of 1039 µmol kg -1 inferred from the cruise data. An overview of the symbols 250 used for estimated and derived quantities used in this study is given in Table 2.

253
In this section we present the regression equations identified in this study in addition to 254 validating the various estimation procedures used by comparing the estimated values with 255 those measured/computed. The results of these comparisons are summarized in Table 3. For  Thus, from here on we assume that Eq. 1 is able to estimate the seasonal SSS variations across 274 the whole study region. To verify this we have compared the monthly averages of RF_SSS 275 data with values obtained using Eq.1 and monthly RF_SST. As shown in the last row of Table   276 3, the estimated values were significantly correlated with the monthly RF_SSS (R 2 =0.65 and 277 p=0.002) and the resulting rmse of 0.3 was lower than the benchmark values of ±1.8. (1) 279 As further verification that the RF SST dataset is spatially representative, we compared it with 280 the chronologically co-located UW SST that have been acquired onboard Trans Carrier across the whole study area. The two datasets were found to be almost identical ( Fig. 2b; 3 rd row 282 Table 3).

283
The relationship between TA, SSS and SST is given by Eq. 2 according to:

294
To estimate a corresponding TA value for each UW fCO 2 observation obtained from MS 295 Trans Carrier, we used salinity values estimated from the UW SST data by using Eq.1. The 296 results (denoted as SSS(sst)) were then inputted into Eq.2 to obtain TA(sst) (see Table 2 for 297 nomenclature). The fact that TA(sst) are based on SSS(sst) rather than measured SSS values 298 introduce an additional error in the estimated pH(sst) and Ω Ar (sst). In order to assess this error 299 we compared pH(sst) and Ω Ar (sst) with values based on the cruise data, i.e. pH(sss) and 300 Ω Ar (sss). First, we computed pH(sss) and Ω Ar ( values of 0.003 and 0.04 for pH(sss) and Ω Ar (sst), respectively, which are well within the 309 aforementioned maximum target uncertainties developed by the C-CAN (last column in Table   310 3).

311
To quantify the total error associated with the pH(sst) and Ω Ar (sst) estimates, we considered 312 two main sources for error. First we computed the residuals (estimated -measurement-based) using the data shown in Figs. 2c and 2d (including the sensor data). The mean difference for 314 the whole study area was 0.002 +/-0.004 and 0.005 +/-0.08 for pH and Ω Ar , respectively.

315
Thus, the maximum probable error from this source is 0.006 and 0.09 for pH and Ω Ar,

316
respectively. Additionally, we estimated that the computed and/or measured pH values 317 included an error of 0.007 pH units, which under the current conditions (mean TA, fCO 2 , 318 SST, and SSS) would give an error of 0.038 in Ω Ar . These two error estimates were combined 319 (as the square root of sum of squares) to determine the total error in our estimates, which were 320 found to be ±0.01 and ±0.1 for pH and Ω Ar , respectively. It must be noted that the above total 321 error was derived from all available observational data including the in situ sensor data 322 (shown in Fig. 2c  to the effect of the fall mixing, which enables carbon-rich coastal water to reach the surface 347 layer, as mentioned in section 1, and is reflected by increasing DIC during this period ( Fig.   348 3d).

349
The mean distribution of Ω Ar (sst) also shows a significant seasonal variation. There are three values. During winter, the surface water is cold with low Ω Ar (sst) and pH(sst) values.  Fig. 4 (a-e) where it can be seen that DIC is the most important 377 driver followed by SST and TA, whereas SSS had a negligible effect (not shown) on the 378 seasonal pH variations. We also note that the effects of SST and TA combined are nearly 379 equal to, but opposite to that of DIC (Fig. 4c,d,e). As a result, the sum of all effects is <0.06 380 pH units, and compares well to the observed amplitudes (Fig. 4a), meaning that the 381 decomposition model is able to account for the observed seasonal changes. Note also the TA 382 control is identical to that of SST (Fig. 4c,e). The reason for this is that TA values used here 383 are obtained from SSS(sst) and SST using Eq. 2, which in effect means that they are based on observed in Ω Ar (sst) (Fig. 4a). We therefore conclude that seasonal changes in DIC and TA 395 are the most important driver for changes in Ω Ar (sst).

396
From the above we conclude that the main drivers of Ω Ar (sst) are DIC and TA, whereas for 397 pH(sst), SST also has a significant impact. This means that the formation and destruction of 398 organic matter together with upwelling of carbon-rich coastal water, seasonal warming and 399 cooling, and runoff inputs, are the processes that govern most of the seasonal variability of 400 OA parameters within the study area. It then follows that interannual variability in the above 401 processes would lead to corresponding variations in pH(sst) and Ω Ar (sst). Such interannual 402 changes are evident from the monthly time series (Fig. S1), where the rate of seasonal change 403 differs between the years, both for SST and DIC normalized to the mean salinity (nDIC)  Table 4. For pH, IAV was normally much lower than the seasonal changes and 408 ranged between 0.01 and 0.02 although higher changes were observed during the months which is much lower than the seasonal changes (section 3.2). Higher IAVs were observed for performed showed that year-to-year differences in pH were related to those in fCO 2 rather 419 than SST changes, whereas year-to-year differences in Ω Ar (sst) were more related to those in 420 SST than fCO 2 . In any case, the observed year-to-year differences were not systematic, and no 421 multiyear temporal trend was apparent from the 4-year time series analyzed in this study.

Inference of OA parameters from VOS underway data 423
Changes in the oceanic CO 2 -system variables are related through ratios called Buffer Factors.

424
Specifically, changes in Ω Ar and pH in response to CO 2 variations can be quantified by partial while the CO 2 system is fully determined only occasionally, an easy way of interpolating the seasonality in pH and Ω Ar, is to predict them from fCO 2 . We have implemented this 439 alternative way of estimating pH and Ω Ar using the CS cruise data. For the estimation of Ω Ar 440 we used fCO 2@meanSST , which is fCO 2 adjusted to constant temperature (i.e. at mean SST), 441 because these normalization improved the regression significantly. Since we were interested 442 in pH and Ω Ar we plotted these parameters directly against ln(fCO 2 ) or ln(fCO 2t@meanSST ). The 443 results are shown in Fig. 5 and conform to tight relationships between computed pH and 444 ln(fCO 2 ) values (Fig. 5a), and between computed Ω Ar and ln(fCO 2t@meanSST ) (Fig. 5b) which is comparable to the residuals associated with pH(sst) and Ω Ar (sst) ( Table 3). An 451 advantage of this procedure, however, is that it utilizes much tighter empirical relationships, 452 involves fewer computational steps, and is based on UW data, which are much more available. However, both the slopes and the intercepts of these correlations vary slightly with 504 DIC and TA. Therefore, the most accurate interpolations will be achieved if the relationships 505 are calibrated with high frequency observations of the complete carbonate system, measured 506 at few strategically placed fixed stations.

507
The Ω Ar -fCO 2@meanSST relationship, and the rate of change of its slope and intercept with 508 DIC, have been used to project the time when under-saturation of calcium carbonate could be 509 expected to occur in the study area. This is expected to occur in the year 2070, if we assume 510 business as usual emission scenario (RCP 8.5), and that oceanic CO 2 concentrations follow 511 that of the atmosphere (i.e. constant disequilibrium between ocean and atmosphere).