Baseline Monitoring of the Western Arctic Ocean Estimates 20% of Canadian Basin Surface Waters Are Undersaturated with Respect to Aragonite

Marine surface waters are being acidified due to uptake of anthropogenic carbon dioxide, resulting in surface ocean areas of undersaturation with respect to carbonate minerals, including aragonite. In the Arctic Ocean, acidification is expected to occur at an accelerated rate with respect to the global oceans, but a paucity of baseline data has limited our understanding of the extent of Arctic undersaturation and of regional variations in rates and causes. The lack of data has also hindered refinement of models aimed at projecting future trends of ocean acidification. Here, based on more than 34,000 data records collected in 2010 and 2011, we establish a baseline of inorganic carbon data (pH, total alkalinity, dissolved inorganic carbon, partial pressure of carbon dioxide, and aragonite saturation index) for the western Arctic Ocean. This data set documents aragonite undersaturation in ∼20% of the surface waters of the combined Canada and Makarov basins, an area characterized by recent acceleration of sea ice loss. Conservative tracer studies using stable oxygen isotopic data from 307 sites show that while the entire surface of this area receives abundant freshwater from meteoric sources, freshwater from sea ice melt is most closely linked to the areas of carbonate mineral undersaturation. These data link the Arctic Ocean’s largest area of aragonite undersaturation to sea ice melt and atmospheric CO2 absorption in areas of low buffering capacity. Some relatively supersaturated areas can be linked to localized biological activity. Collectively, these observations can be used to project trends of ocean acidification in higher latitude marine surface waters where inorganic carbon chemistry is largely influenced by sea ice meltwater.


Introduction
The importance of the Arctic Ocean in the context of global carbon dioxide (CO 2 ) uptake and ocean acidification is widely accepted [1][2][3][4][5][6][7][8][9]. Ocean acidification is projected to occur relatively rapidly in the Arctic due to processes and conditions that are unique to Arctic surface waters [7][8][9]. For instance, a large fraction of the global net CO 2 uptake during recent decades (,2200 Tg C yr 21 ) has occurred over the relatively small surface area of the Arctic Ocean. From the estimated net CO 2 uptake rates of the Arctic Ocean (65-175 Tg C yr 21 ; reviewed in ref. [8]), it is inferred that as much as 7.5% of global oceanic CO 2 uptake may occur in the Arctic Ocean, which comprises only 3.9% of the global ocean's surface. These figures may be even more noteworthy given that the uptake occurs predominantly in seasonally ice-free areas, which are a fraction of the Arctic Ocean surface.
One driving factor behind the disproportionate Arctic Ocean CO 2 uptake and resulting acidification is the relatively cold surface water, which absorbs more CO 2 than warmer seawater [3,5]. Furthermore, over the past decade, summer sea ice extent has rapidly declined [9][10][11][12][13], reaching a record low in 2012 [14]. Multiyear sea ice has experienced the greatest decline [9]. This loss of multiyear ice has exposed the surface mixed layer (typically#50 m thick), which is undersaturated with respect to atmospheric CO 2 ; the result is oceanic CO 2 uptake. Previously, ice cover significantly inhibited CO 2 exchange between this layer and the atmosphere [15][16][17].
This melt-associated exposure of undersaturated waters is unique to the Arctic, where a steep vertical density gradient, attributable primarily to a cold halocline underlying the surface mixed layer, inhibits upward mixing of CO 2 -rich deep waters, even during winter when the mixed layer temperature is lowest and salinity is highest due to sea ice formation [18]. Unlike the Southern Ocean seasonal ice zone, where the under-ice water has a relatively high partial pressure of CO 2 (pCO 2 ,420 matm) due to rapid vertical mixing of Circumpolar Deep Water [19], the mixed layer immediately under Arctic sea ice remains low in salinity and pCO 2 . Thus, when Arctic sea ice melts in summer, the under-ice layer is exposed to the atmosphere and is also diluted by freshwater from sea ice melt. This dilution is an important consequence of multiyear sea ice loss because it reduces surface seawater calcium and carbonate ion concentrations and pH [20,21]. The combination of these processes is expected to drive the expansion of areas of carbonate mineral undersaturation in the Arctic Ocean over the next decades [3][4][5][6].
The in situ melting of sea ice has a number of direct consequences for seawater chemistry. The saturation state index of carbonate minerals (V carb ) is defined by: 2+ ] is the calcium ion concentration, [CO 3 2-] is the  carbonate ion concentration, and K sp is the stoichiometric solubility product of the mineral. This index provides a quantitative measure of the thermodynamic potential for carbonate minerals such as aragonite and calcite to either precipitate (V carb .1) or dissolve (V carb ,1). Importantly, when multiyear sea ice melts, fresh water containing relatively little carbonate and calcium is liberated, thereby lowering V carb of the receiving seawater. The resulting changes in pH, V carb , and pCO 2 affect calcifying organisms at the base of the food chain, creating effects that propagate through higher trophic levels [1,3]. Thus, the expansion of acidification into areas previously isolated from atmospheric contact by multiyear ice cover is likely to have ecological and economic consequences. Ocean acidification models project that under a number of plausible scenarios of increasing atmospheric CO 2 , the Arctic Ocean will become undersaturated with respect to carbonate minerals in the next decade [3][4][5][6]. Recent data indicate that some areas are already undersaturated with respect to the carbonate mineral aragonite [16,[21][22][23][24][25]. These areas, which include regions of the Canadian Archipelago, Canada Basin, and Beaufort Sea, exhibit the lowest aragonite saturation state (V a ) values during late summer, when sea ice is at its minimum annual extent. However, the spatial extent of undersaturated waters in the western Arctic Ocean has been poorly constrained.
Much of the uncertainty of models that project future trends of Arctic Ocean acidification is due to inadequate data coverage, particularly in higher latitudes. As of late 2011, only ,20,000 data points were available in public databases for the entire Arctic Ocean (surface and water column; data from refs. [16,17,[21][22][23][24][25][26] and the data sets of the Japan Agency for Marine Earth Science and Technology, BioChem, and the Carbon Dioxide Information Analysis Center, including CARINA and GLODAP; Figure 1). Most of the data are from the Atlantic/Norwegian Sea ( Figure 1). Establishing carbonate chemistry baselines and improving our understanding of the processes that drive changes in carbonate chemistry are fundamental for constraining the biogeochemical process parameters needed to predict ocean acidification effects on Arctic Ocean flora and fauna. Thus, a systematic approach for measuring the carbonate chemistry of the Arctic Ocean at high spatial resolution from coastal to pelagic waters is needed.
To begin to establish these baselines and fill critical data gaps, we collected carbonate chemistry data at high spatial resolution in the western Arctic Ocean during 2010 and 2011. We supplemented these data with measurements of conservative tracers of water sources as well as proxies of biological processes. This study compares aragonite saturation states in open pelagic waters, shallow shelf waters, and ice-bound high-latitude waters to delineate rates of change and causes of variation in carbonate mineral saturation states.

Materials and Methods
From 3 August to 5 September 2010 and from 16 August to 28 September 2011, we collected seawater chemical and biological measurements as noninterference science on United States (U.S.) Extended Continental Shelf Task Force cruises aboard the U.S. Coast Guard Cutter Healy in the western Arctic Ocean (Figure 1). No specific permissions were required for collecting water samples in Arctic international waters or U.S. or Canadian waters. Our studies did not involve endangered or protected species. Continuous and discrete water samples (Table 1) were collected underway from ,8 m below the waterline, via a port in the ship's hull. Approximately 25,000 data records resulted from the 2010 cruise; approximately 9,000 data records resulted from the 2011 cruise.

Underway Continuous Measurements
Geographic position, salinity (S), temperature (T), and fluorometric data (n = 62,387) were collected using a shipboard Ashtech ADU5 GPS system, a SeaBird SBE45 thermosalinograph, and a Seapoint Chlorophyll Fluorometer [27]. The majority of the carbonate chemical measurements were made using a Multiparameter Inorganic Carbon Analyzer (MICA) [28], a flowthrough system which measured pH, pCO 2 , and total dissolved inorganic carbon (TCO 2 ), along with S and T (n = 34,000). Data were logged approximately every 2 mins on the 2010 cruise and every 7 mins on the 2011 cruise. The MICA was calibrated using Certified Reference Material from Professor A. Dickson of the University of California at San Diego. Precision was 60.002 for pH, 62 matm for pCO 2 , and 62 mmol kg 21 for TCO 2 . A Lamont-Doherty Earth Observatory underway system for surface water measurements was also used to measure pCO 2 during the 2011 cruise (n = 18,500). This system was calibrated every 4 hrs using five calibration gas mixtures certified by the National Oceanic and Atmospheric Administration (NOAA) Climate Monitoring and Diagnostic Laboratory. The precision of the pCO 2 seawater values is about 61.5 matm.

Discrete Surface Samples and Analyses
Discrete surface water samples were collected from the underway system every 1-4 hrs following the protocols of ref. [29]. The samples (n>560) were immediately analyzed for pH using purified meta-cresol purple indicator dye and the method and equations of ref. [30].  Additional discrete samples were collected every 1-8 hrs for post-cruise laboratory analysis of TA and TCO 2 (n>305) as well as oxygen isotopic composition (n>307). TA was measured spectrophotometrically and TCO 2 was measured coulometrically, both following the procedures of ref. [29]

Water Mass Mixing Model
To identify sources of water masses, we applied a threecomponent mixing model based on measurements of salinity (S) and oxygen isotopic composition (d 18 O). This mixing model is well suited to quantify fractional contributions (f) from various sourcesin this case, meteoric fresh water and sea ice meltwater mixing with seawater. Meteoric freshwater (MW), derived from terrestrial runoff and direct precipitation, has negligible salinity (S>0) but is relatively 18 O-depleted in the high latitudes of the Arctic (d 18 O>-20%). Sea ice melt (SIM) also has relatively low salinity (S,5), but its oxygen isotopic composition is similar to that of seawater, from which it is formed (d 18 O>0 to -6%). Seawater (SW) has salinity of 32-35 and d 18 O values of -1 to +0.3%, although these values vary between Atlantic and Pacific source regions. This freshwater tracer method, which takes advantage of the large differences in the d 18 O values of the freshwater sources [31], has been used for decades to partition surface waters into constituent water sources [32][33][34][35][36][37][38].
Using discrete measurements of S and d 18 O, each surface water sample can be partitioned into a mixture of the three endmembers, with the fractional contribution of each end-member determined by mass balance:   Table 3. Dunn's Method Q-statistic results for pairwise multiple comparisons of the four Arctic Ocean regions.  [32,38,42]. The advantage of using Atlantic inflow seawater rather than BSI seawater for SW end-member values is that we are then able to make comparisons with other pan-Arctic studies of water sources. The disadvantage is that our calculated f MW values may be overestimates. In other words, our calculated f MW values, attributed to Arctic river runoff and direct precipitation, may also reflect the influence of freshwater that derives from BSI. Adding dissolved silica measurements to the mass balance model could provide a means of independently quantifying the additional BSI source [34,38]. Figure 2 illustrates the sensitivity of estimated f values (mixingmodel output) to uncertainties in end-member compositions. The mixing model is relatively insensitive to analytical errors, which are small (,0.01 and 0.1% for S and d 18 O, respectively) relative to other sources of uncertainty. The model is most sensitive to potential systematic error introduced by incorrect assumptions regarding the SW end-member.

Aragonite Saturation States
In situ aragonite saturation states (V a ) were calculated using the CO2calc carbon calculator application [43]. Input parameters were the TA-pH data pair, carbonic acid dissociation constants (pK 1 and pK 2 ) from Lueker et al. [44], HSO 4 2 dissociation constants from Dickson [45], and the borate dissociation constant of Lee et al. [46] and other measured parameters such as T, S. The solubility product (K sp ) for aragonite was derived from Mucci [47]. Internal consistency was established by comparing V a derived using the TA-pH pair with V a derived using theTCO 2 -pH pair or TCO 2 -pCO 2 pair. On the basis of these consistency test results, we estimate a precision of 60.05 for V a .
Values of V a calculated from coincident MICA and discrete sample analyses (i.e., coincident TA-pH data pairs; n = 68) were statistically compared as an estimation of the equivalence of the two data sets. Spearman rank-order correlation shows the two data sets are highly and positively correlated (R 2 = 0.985; p,0.001). The Mann-Whitney (a = 0.05) test for significant differences (W = 4712.0; p = 0.816) found no significant difference between the two data sets. Based on this equivalence ( Figure 3A), we use V a values calculated from the discrete samples ( Figure 3B) to compare to other discrete sample measurements, and we use V a values calculated from the MICA data ( Figure 3C) to provide more detailed spatial resolution of V a distributions.
The areal extent of aragonite undersaturation (V a ,1) was estimated using ESRI ArcGIS software (version 10.1). V a values calculated from the 2010 and 2011 cruises (.21,000 data points) were projected and then used to interpolate a raster map of aragonite undersaturation in the Makarov and Canada basins, both part of the Canadian Basin. A natural neighbor routine was used for the interpolation. The Makarov Basin (334,224 km 2 ) and Canada Basin (1,524,664 km 2 ) were delineated and measured by tracing the surrounding 1000 m isobath contour on the International Bathymetric Chart of the Arctic Ocean [48]. To determine the percent areal coverage of aragonite undersaturation, the interpolated area of undersaturation (371,525 km 2 ) was divided by the area of the Canadian Basin (1,858,888 km 2 ).

Bacterial Production Rates
Bacterial production rates (BP; mg C m 23 d 21 ) were calculated using the following relationship: where Chl-a is chlorophyll a (mg L 21 d 21 ) and T is temperature (uC). This equation was derived from data previously collected in the Arctic Ocean during the same season as this study [49].

Statistical Analyses
An initial assessment of the complete data set indicated that waters were not homogenous throughout the sampled area. The data set was therefore sorted into four regions based on differences in salinity and other measured parameters (Table 1): the Beaufort Sea, Makarov Basin, Sever Spur, and Canada Basin regions. The sample sites for the HLY1102 data set extend to higher latitudes than the HLY1002 data set and were therefore used to conservatively delineate these geographic boundaries. Comparisons for significant differences or similarities between each region were performed using the f SIM , f MW , f SW , V a , pCO 2 , S, T, Chl-a, and BP data sets. The data from each region were not normally distributed as determined by the Shapiro-Wilk normality test (a = 0.05). Accordingly, comparisons of variables were performed using the nonparametric Kruskal-Wallis one-way ANOVA (KW ANOVA, a = 0.05), followed by an all-pairwise multiple comparison of median values using Dunn's method (a = 0.05). Dunn's method for multiple comparisons determines which of the comparisons analyzed by KW ANOVA are significantly different. A p-value of ,0.05 indicates a significant difference between the two median values. The Spearman rank-order method was used for all correlation calculations.
Rates of change of V a were obtained by comparing September 1997 data [50] to the Healy 2010 and 2011 data. Surface TA and TCO 2 from similar locations were used to calculate V a as described above. Rate of change for a given location was then calculated as the slope of a linear fit of V a vs. year. The 1997 TCO 2 and TA data were analyzed on board using Dickson Certified Reference Material for every 20 samples to evaluate accuracy. Precision for the 1997 data was ,0.1% for TCO 2 and ,0.2% for TA.

Spatial Coverage
In the course of the late-summer 2010 and 2011 Healy cruises, more than 34,000 water samples (flowthrough and discrete) were analyzed for S, T, pCO 2 , Chl-a, and BP, and calculated V a (Figure 1). The resulting data sets are among the most comprehensive in the region (Previous work in the area shown in Figure 1 had collectively produced ,19,900 data records). Sampling length intervals ranged from approximately 0.24 km to 10 km or more, depending on ship speed and sampling frequency. Table 2 shows median and range values for the HLY1102 data records that contain a complete suite of all of the following parameters: V a , S, T, pCO 2 , f SIM , f MW , f SW , Chl-a, and BP.

Regionally Distinct Seawater Properties
The four delineated regions of the western Arctic (Table 2) are the Beaufort Sea, Canada Basin, Makarov Basin, and Sever Spur regions. Although these names coincide with geographic regions rather than oceanographic regimes, the Beaufort Sea is predominantly shallow and coastal, while the other regions are predominantly deep basins. In pairwise comparisons, the regions exhibit significant differences in four or more of the seawater variables of V a , S, T, pCO 2 , f SIM , f MW , f SW , Chl-a, and BP (Table 3). Figure 4 shows climatological August surface salinity through 1997 [52] and surface salinity encountered on the HLY1002 and HLY1102 cruises. The recent data show a generally lower S than is seen in the estimates of long-term means. The Healy data also show S increasing with increasing latitude (Figure 4) and ice cover ( Figure 5). A concomitant increase in V a with decreasing pCO 2 occurred along this same latitudinal trend ( Figures 5 and 6). An exception to this general pattern is the Beaufort Sea region, which shows higher V a values in the samples near the coast.

Aragonite Undersaturation and Water Mass Sources
Our calculations of V a in the combined Canada and Makarov basins indicate that in the summers of 2010 and 2011, approximately 20% of the 1.9 million km 2 surface area was undersaturated with respect to aragonite (i.e., V a ,1; Figure 5). All observed undersaturation was confined to the Canada Basin and Beaufort Sea. Undersaturated water extended poleward from the Beaufort Sea to nearly 80uN between the Canadian Archipelago and the Chukchi Cap (over 3.7610 5 km 2 ). Waters supersaturated with respect to aragonite were seen in the Makarov Basin and Sever Spur areas (1.14#V a #1.61) in association with nearly 100% sea ice cover and relatively low pCO 2 (Figures 5 and 6). Pairwise statistical comparisons show that V a values in these two regions were significantly greater than in the Canada Basin (Table 3). Beaufort Sea coastal waters had V a values significantly lower than Makarov Basin waters.
Sea ice melt showed the greatest contribution to water masses in the Beaufort Sea and Canada Basin (f SIM up to 0.219) and comparatively low contributions in the Makarov Basin and Sever Spur areas (f SIM up to 0.061; Table 1, Figure 7). Values of f MW were more homogenous throughout the study area but were generally lower in the Makarov Basin and Sever Spur areas (0.116-0.177) than in the Beaufort Sea and Canada Basin (0.054-0.307).
Surface water V a was strongly and positively correlated with surface S in all localities (Table 4). Further, S in all localities was strongly but negatively correlated with the fraction of fresh water derived from sea ice meltwater (f SIM ) and/or meteoric water (f MW ; Table 4). Values of f MW showed a consistent negative correlation to V a throughout the entire western Arctic, including waters in icecovered areas such as the Makarov Basin.

Aragonite Undersaturation and Biological Processes
Aragonite saturation state can be influenced by not only air-sea CO 2 exchange but also biological activities such as carbon fixation and respiration. Photoautotrophic organisms may remove CO 2 from the water, thereby tending to increase V a . Respiration, in contrast, adds CO 2 , thereby tending to decrease V a . During our late summer cruises, Chl-a, which is indicative of photoautotrophic activity, showed no significant correlation with V a except in the Canada Basin (Table 4), where the correlation was positive. Chl-a and pCO 2 also positively covaried in the Canada Basin. The pCO 2 : V a relationship was significantly and negatively correlated in the Makarov Basin and showed no significant association in the other regions.

Discussion
Projections of the extent and rate of Arctic Ocean acidification and the associated ecosystem changes require models based on causal relationships and complex feedbacks among rapidly changing physical, chemical, and biological processes. Model formulation and validation both require extensive data sets. Our baseline study significantly expands the body of data available for such work and also provides the first high-resolution data from higher latitude waters (up to 88u489 N into the Makarov Basin; Figure 1). Previous work has focused primarily on coastal areas of the southern Canada Basin and the Bering and Beaufort Seas [20,22,25,[51][52][53][54][55]. The new data provide a basis for detecting future change and analyzing trends in surface water chemistry.
Combined with older data, the 2010-2011 measurements can also provide insight into changes that have already occurred in the Arctic, primarily over the past two decades. (Canadian Basin carbonate chemistry data from before the 2000s are sparse.) For example, a comparison with 1994 cruise data [50] indicates that surface S in the northern Canada Basin and the Makarov Basin This marked surface freshening was accompanied by a large decrease in average V a , from 1.31 to 1.14, which suggests a link between V a and dilution from sea ice melt in these areas. A number of locations that were supersaturated with respect to aragonite in 1997 were undersaturated in 2010 and 2011. Based on the limited data available from sites in the southern Canada Basin, we estimate that summer surface V a decreased at a rate of 2.1% yr 21 between 1997 and 2010-nearly an order of magnitude greater than the average rate observed for the Pacific Ocean (0.36% yr 21 ; [56]).

Effects of Sea Ice Melt on Aragonite Saturation State
The observed decrease in aragonite mineral saturation state in the southern Canada Basin since 1997 has been accompanied by a decrease in late summer salinity (Figure 4). Increased freshwater in the Arctic Ocean is known to have diluted surface mixed layer S over the past decade [57][58][59][60][61]. This freshening may have been enhanced in the western Arctic, where recent spin-up of the winddriven convergence of the Beaufort Gyre may have increased freshwater retention [57,60]. The strong vertical density gradient of the Arctic Ocean also acts to maintain low S in the surface mixed layer (,0-50 m depth) throughout the seasons [18]. The cold Arctic halocline at ,50-400 m depth reduces upward mixing of CO 2 -rich deep waters to the surface mixed layer [18], thus playing a significant role in maintaining low surface S and V a .
Comparisons of S and V a show a significant correlation throughout the study area (Figures 4, 5, and 8; Table 4), suggesting that dilution with freshwater provides a direct mechanism for reducing saturation state [8,21]. The water-mass mixing model distinguishes the proportion of freshwater derived from sea ice melt (f SIM ) vs. meteoric water (f MW ; i.e., terrestrial runoff and direct precipitation). Relatively high f SIM was observed in areas south of ,80uN ( Figure 8A), notably in those areas where summer ice extent had recently decreased and particularly where multiyear ice had melted. The strong negative correlation between f SIM and V a in the Beaufort Sea and Canada Basin (Figure 8, Table 4) reflects the dominant influence predicted by dilution from sea ice meltwater on aragonite saturation state. Areas north of 80uN, such as the Makarov Basin and Sever Spur regions, were largely immune to the effects of sea ice melt on aragonite undersaturation, showing weak to no significant correlation between f SIM and V a .
The influence of f MW ( Figure 7B) was more homogenous than f SIM ( Figure 7A). Within the relatively narrow f MW range encountered, highest values occurred in Beaufort Sea coastal waters while lower values occurred in the ice-covered Makarov Basin. Overall, f MW values were negatively correlated to V a in all regions, which reflects the spatially consistent role of meteoric freshwater in reducing V a . Still, those correlations were relatively weak compared to the correlations between f SIM and V a in the Beaufort Sea and Canada Basin (Table 4), where V a is strongly affected by sea ice meltwater.
The variable influence of freshwater sources across the regions is evident in the geographic distribution (clustering) of the data Table 4. Spearman rank-order correlations between the Arctic Ocean regions.

Region
Parameter   shown in Figure 8. For the data set as a whole, V a is negatively correlated with f SIM , but regional distinctions are apparent. The Makarov Basin and Sever Spur areas are generally characterized by relatively supersaturated V a values and generally low f SIM (Tables 2, 4). The Canada Basin and Beaufort Sea data show a strong correlation between V a and f SIM , but for a given f SIM value, Beaufort Sea V a values are higher than those from the Canada Basin. This offset is likely due to the fact that meteoric water from river runoff, which is more abundant in the coastal waters of the Beaufort Sea ( Figure 7B), contributes higher concentrations of dissolved calcium, inorganic carbon, and carbonate alkalinity than does sea ice melt [22]. As a result, Beaufort Sea surface waters are more strongly supersaturated than Canada Basin waters for an equivalent contribution of freshwater from sea ice melt. This effect, observed strongly in the Beaufort Sea, is likely enhanced by the fact that North American rivers carry more calcium, inorganic carbon, and carbonate alkalinity than do Eurasian rivers [39]. Thus, the net effect of meteoric water on lowering V a values is likely less significant in the southern Canada Basin and Beaufort Sea, where the Mackenzie River is the primary meteoric water source, than in the Makarov Basin ( Figure 7B), where Eurasian rivers are the primary source. Over the past decade, the Makarov Basin has received increased input from these rivers [59]. Differences between pelagic and coastal waters may also be partially attributed to a pronounced upwelling event that occurred in the Beaufort Sea during late summer 2011 [52]. This coastal upwelling may account for a reduced MW role, resulting in somewhat higher regional V a values in 2011 than in 2010.
Regarding the relative roles of f MW and f SIM , it is worth noting that our sensitivity analysis ( Figure 2) suggests that f MW values may in some cases be overestimated due to the simplifying assumption that the SW end-member is Atlantic-derived. A portion of the freshwater attributed in our analysis to terrestrial runoff from Arctic rivers may in actuality derive from Bering Strait Inflow. However, the inference of a strong regional relationship between sea ice melt and low V a remains robust, being relatively insensitive to model assumptions.
The observed negative correlation between f SIM and V a (Figure 8) may convolve the two hypothesized effects of reduction of sea ice cover on V a -namely, dilution with meltwater and exposure to the CO 2 -rich atmosphere, both of which may act to reduce V a in areas of recent sea ice melt. Figure 9A illustrates the combined roles of these two effects in a simple model of conservative mixing between end-member seawater and sea ice meltwater. Model lines show variation of the f SIM : V a relationship under conditions in which pCO 2 is varied while f MW is held constant (dashed lines show model curves; gray points show observed data). The model lines show that for a given f SIM contribution, a higher pCO 2 corresponds to a lower saturation state. As predicted by the model, V a is negatively correlated with f SIM , but the slope for the correlation of the observed data is steeper than model predictions. The greater slope associated with the observations is likely due to the combination of the two seaice-related effects on V a . For instance, waters represented by points at the left side of the data cluster derive predominantly from the Makarov Basin, where V a is generally highest. These waters show low f SIM values, due to minimal melting of sea ice, and also low pCO 2 values, reflecting disequilibrium with the atmosphere (labeled ''low f SIM , low pCO 2 '' in Figure 9A). In contrast, waters in areas of recent sea ice melt, such as the Canada Basin or Beaufort Sea, show high f SIM values and generally high pCO 2 (labeled ''high f SIM , high pCO 2 '' in Figure 9A). The combination of the disequilibrium effect with the meltwater effect tends to produce a trend connecting the data from these two regions, with a steeper slope than is predicted by a model that includes seawatermeltwater mixing only. Figure 9B shows a similar model of conservative mixing of seawater and sea ice meltwater, but in this case varying f MW while holding pCO 2 constant. Again, the model lines show that a higher fractional contribution of sea ice melt corresponds to a lower saturation state. However, the addition of meteoric water tends to partially counter the effect of SIM: for a given meltwater contribution (f SIM ), increasing the meteoric contribution (f MW ) corresponds to a higher saturation state. As with Figure 9A, the observed data show a steeper negative slope than that predicted by the simple two-component mixing model. However, f SIM and f MW are observed to be positively correlated throughout the Canada Basin and Beaufort Sea (Table 4). Connecting regions of ''low f SIM , low f MW '' and ''high f SIM , high f MW '' in Figure 9B suggests that the correlation between these variables would tend to counter the ''steeper-than-predicted'' slope of the correlation between f SIM and V a discussed above.
The model shown in Figure 9B also illustrates the proposed explanation for the positive offset of V a values (for a given f SIM ) from the Beaufort Sea as compared to the Canada Basin ( Figure 8).
Model calculations of the f SIM : V a relationship are offset to higher V a values as a result of high modeled f MW . This effect is seen predominantly in the Beaufort Sea. This effect is likely due to terrestrial contributions of dissolved calcium, inorganic carbon, and carbonate alkalinity from the Mackenzie River [39].

Effects of Biological Activity on Aragonite Saturation State
Variation in V a and pCO 2 values show the effects of not only hydrodynamic forcing [24], regional sea ice melt, and terrestrial inputs of freshwater and CO 2 [25], but also local biological processes. Consistent with previous Arctic Ocean surface water observations [62], our Makarov Basin measurements showed low CO 2 partial pressures compared to atmospheric values. Seawater pCO 2 ranged from 242 to 330 matm; atmospheric values at the time were ,380 matm or higher. Seawater pCO 2 values lower than atmospheric values have previously been attributed to (a) incomplete seawater equilibration with the present-day atmosphere and (b) cooling [8,17,20], but also (c) relatively low net community production in areas of offshore basins as compared to coastal waters [8].
Bacterial metabolism is the collective process of bacterial production (BP) ( Table 2) and respiration (BR). Bacterial production pathways produce biomass, while respiration pathways generate energy for cellular processes. The BR reactions release CO 2 when organic and inorganic C are substrates. The efficiency at which bacteria incorporate C into their biomass is a fraction of the total C utilized by the heterotrophic microbial community. This efficiency is described by the bacterial growth efficiency [BGE = BP/(BP+BR)] [63]. In general, BGE values range from 0.03 to 0.27 in open marine waters and from 0.09 to 0.45 in coastal waters. BGE values in the western Arctic Ocean have been shown to be relatively low, averaging only 0.0760.09 [49,64]. The low BGE indicates that a significant proportion of the utilized C is being used in respiration-related processes, which produce and release CO 2 .
Differences in the BGE of heterotrophic microbial communities in the four regions may explain the inconsistent correlations between pCO 2 and BP. These quantities are positively correlated in the Beaufort Sea and Canada Basin regions but are negatively correlated in the Makarov Basin and Sever Spur regions (Table 4). A confounding influence on these inconsistent correlations may be the overwhelming contribution of atmospheric pCO 2 exchange relative to the heterotrophic microbial community production of pCO 2 . The influence of heterotrophic microbial community respiration on pCO 2 and thereby V a would be detected only if the effects (i.e., increasing pCO 2 and decreasing V a ) were greater than the net effect of CO 2 influences from the atmosphere, photoautotrophic communities, and freshwater additions.

Seasonal Trends
Seasonal trends of V a in the Arctic Ocean are generally poorly documented (for an exception see [65]). Winter data collected in 2011 and 2012 between ,83uN and 87uN [62] provide, in combination with our Makarov Basin data, a rare opportunity for a seasonal comparison. In general, the winter and summer values of surface water V a and pCO 2 were comparable. In addition, Makarov Basin under-ice mixed layer water (15-40 m below the ice) was sampled in April 2012 along a transect of the Switchyard Program (data archived by the Advanced Cooperative Arctic Data Information Service, aoncadis.org). The under-ice winter water was characterized by T = 21.6960.04uC, S = 31.060.7, pCO 2 = 312613 matm, and V a = 1.3160.06. These data are consistent with our Makarov Basin summer data (Table 2), indicating little seasonal variation in the surface water chemistry of the Makarov Basin.

Implications for Modeling of Future Chemical Changes
Models have projected that large areas of the Arctic Ocean will become undersaturated with respect to aragonite in the next decade [3][4][5][6]. The data presented here indicate that the area of undersaturation presently extends to approximately 20% of the Canadian Basin in the late summer months, when sea ice is near its minimum extent. A number of other studies have shown areas of aragonite undersaturation in the Canadian Archipelago and on the Beaufort Sea shelf [21,23,24].
Projection of future freshening and undersaturation in areas currently under perennial ice cover, such as the Makarov Basin, may be facilitated by water source identification based on the use of salinity and oxygen isotopic composition as conservative tracers (e.g., Figures 7-9). Our analysis provides insight into how sea ice meltwater and meteoric water influence aragonite saturation states. The negative correlation between V a and f SIM (Figure 8, Table 4) characterizes the dependence of saturation state on the melting of multiyear sea ice and suggests a pattern along a transect from the Beaufort Sea to the Makarov Basin. The Makarov Basin represents an end-member whose chemical composition is essentially free of the effects of reduced sea ice cover and input of additional freshwater from melting of multiyear ice, while the southern Canada Basin represents conditions resulting from reduced sea ice. Given that summer ice-free conditions are projected to expand in the near term [10][11][12], these spatial trends allow prediction of the conditions under which the carbonate chemistry of the Makarov Basin will resemble that of the southern Canada Basin. Predictive models of ocean acidification must account for not only the effects of equilibration with atmospheric CO 2 but also freshwater dilution and biogeochemical processes.
Data from this study also suggest that an assumption of metabolic coupling or co-dependence between heterotrophic and photoautotrophic organisms in the Arctic Ocean is not appropriate for inclusion in predictive models of ocean acidification processes. Recent studies of the relation between bacterial production and primary production in coastal and pelagic Arctic waters have shown that these carbon cycling processes can proceed uncoupled [49,66,67]. These studies found that the degree of correlation depends on time of year, percentage of ice cover, hydrodynamic regime, bacterial abundance, and bacterial growth efficiency. The absence of predictable correlations between biological variables and V a in this study indicate that complex interactions are driving the biogeochemical processes. Uncoupling of these processes and variability in their rates and efficiencies have important implications for not only the retention of C at shallower productive depths [68] but also for the formulation of models that predict the influence of heterotrophic and photoautotrophic processes on pCO 2 .
In conclusion, the data presented here collectively suggest that recent decreases in western Arctic Ocean V a can be predominantly attributed to recent melting of multiyear sea ice and the associated seawater freshening and uptake of atmospheric CO 2 ; biogeochemical processes exert an additional influence. Despite increasing awareness of rapid global change in the Arctic, much remains to be done with respect to predicting changes in marine surface water chemistry. For example, few data are available for the polar winter, and it is not known whether aragoniteundersaturated areas decrease in size with the seasonal freezing of sea ice. Also, while the effects of sea ice melt may diminish over the long term, the effects of increasing terrestrial runoff and direct precipitation are likely to persist [69]. Quantitative documentation of these processes in the Arctic Ocean is needed for refinement of the next generation of global ocean acidification models.