In 2008, surface waters in the Canada Basin of the Arctic Ocean were found to be undersaturated with respect to aragonite. This is associated with recent extensive melting of sea ice in this region, as well as elevated sea surface temperature and atmospheric CO2 concentrations. We have estimated the relative contribution of each of these controlling factors to the calcium carbonate saturation state (Ω) from observations of dissolved inorganic carbon, total alkalinity and oxygen isotope ratio. Results indicate that the increase in atmospheric CO2 has lowered surface Ω by ∼0.3 in the Canada Basin since the preindustrial period. Recent melting of sea ice has further lowered mean Ω by 0.4, and of this, half was due to dilution of surface water and half was due to the change in air-sea disequilibrium state. Surface water warming has generally counteracted the mean decrease in Ω by 0.1.
 The Arctic Ocean has undergone rapid and substantial changes, especially in the last decade, including: warming [Steele et al., 2008] and freshening [Yamamoto-Kawai et al., 2009a] of surface waters, a deepening of the nutricline [McLaughlin and Carmack, 2010] a shift in phytoplankton community size [Li et al., 2009], and an increase in surface carbon dioxide (CO2) concentrations [Cai et al., 2010] observed mostly in the Canada Basin of the Arctic Ocean. These changes are tightly associated with the attendant loss of summer sea ice. A decrease in the saturation state of surface water with respect to calcium carbonate (CaCO3) is another of these changes [Yamamoto-Kawai et al., 2009b]. Surface water in the Arctic Ocean was oversaturated with respect to CaCO3 in 1997 and thus was favourable to the formation of shells and skeletons of calcifying organisms [Jutterström and Anderson, 2005]. In 2008, however, surface waters in the Canada Basin were undersaturated with respect to aragonite-type CaCO3 [Yamamoto-Kawai et al., 2009b]. This is due to a combination of the global phenomenon, “ocean acidification”, and the recent extensive melting of sea ice.
 The CO2 that is taken up by the oceans react with water to form carbonic acid and thus lowers pH. This causes ocean acidification that is accompanied by a reduction in the saturation state of CaCO3 in seawater. The CaCO3 saturation state (Ω) with respect to either aragonite or calcite, the two major types of CaCO3 produced by marine organisms, is expressed by the product of carbonate and calcium ions ([CO32−] and [Ca2+]) in seawater relative to the stoichiometric solubility product (K′SP) at a given temperature, salinity and pressure:
Waters with Ω > 1 are favorable to precipitate CaCO3 but waters with Ω < 1 are corrosive and therefore CaCO3 will begin to dissolve in the absence of protective mechanisms. Since the preindustrial period, the global mean surface pH has already decreased by 0.1 unit and Ωaragonite has lowered by 0.4 [Orr et al., 2005; Steinacher et al., 2009]. Steinacher et al. , using model simulations, predicted that climate feedback will amplify the decrease in Ω of Arctic surface waters through the melting of sea ice. This is because the input of sea ice meltwater dilutes seawater and thus decreases salinity and the concentrations of total alkalinity, dissolved inorganic carbon and Ca2+, resulting in a decreased Ω in surface water. Also, as CO2 in Arctic surface water has been below equilibrium with atmospheric CO2 by photosynthesis in summer (mainly on surrounding shelves) and by the presence of an ice cover which constrains air-sea gas exchange [Bates et al., 2006], the loss of ice cover will increase CO2 in surface water and thereby decrease Ω. Steinacher et al.  predicted that surface aragonite undersaturation in some parts of the Arctic Ocean would occur within a decade and become more widespread with rising atmospheric CO2. However, undersaturation has already been observed [Yamamoto-Kawai et al., 2009b] and calcifiers such as shelled pelagic molluscs Limacina helicina, a key species of polar ecosystems, are already exposed to unfavorable conditions [Comeau et al., 2009]. In order to better understand the relative importance of processes affecting acidification as well as to predict future changes in Ω, we here quantify causes of observed undersaturation of surface water in the Canada Basin in 2008. We use salinity (S), temperature (T), nutrients, dissolved inorganic carbon (DIC), total alkalinity (TA) and the oxygen isotope ratio of water (δ18O). From these observations, we first calculate Ω, fugacity of CO2 (fCO2), pH and fraction of sea ice meltwater (fSIM). Then, by comparing these data and calculations with those in 1997, we estimate preindustrial distribution of Ω (ΩPI) and changes in Ω due to elevated atmospheric CO2 (ΔΩatm), warming (ΔΩtemp), the change in air-sea disequilibrium (ΔΩa-s) and dilution by sea ice melt (ΔΩdil).
2. Data and Analytical Method
 The data were collected during the cruise on board the CCGS Louis S. St-Laurent in the Canada Basin from July 20 to Aug. 20 in 2008. S was analyzed using a Guildline salinometer and referenced to IAPSO standard seawater. Samples of DIC were analyzed using a Single-Operator Multi-Metabolic Analyzer coulometer system. The pooled standard deviation (Sp) for duplicate samples was 3.5 μmol/kg (n = 11 pairs). Samples for TA were taken from the DIC bottle and analyzed by an open-cell titration with 0.1 N HCl [Dickson, 1990]. The Sp for duplicate samples was 3.5 μmol/kg (n = 7 pairs). Both DIC and TA measurements were calibrated against a certified reference material provided by Dr. Dickson at Scripps Institute of Oceanography. At four stations where surface stratification was very strong, there seems to be an inconsistency between properties drawn first and the last from the Niskin bottle and these surface data were not included in this study. δ18O samples were measured at Oregon State University on a mass spectrometer connected with CO2-H2O equilibration unit. Sp of δ18O for duplicate analysis was 0.05 ‰ (n = 47). TA and DIC observations from 1997 were taken from Jutterström and Anderson  (available from the CCHDO (http://cchdo.ucsd.edu/cruise/18SN19970924)) and δ18O observations from Macdonald et al. [1999, 2002]. δ18O data from 1997 were corrected by adding +0.35 ‰, determined by a comparison of deep water Canada Basin data (depth > 500 m) collected between 2005–2008. In this study, only data in the surface layer from 0 to15 db were used.
 Carbonate parameters were calculated using the CO2SYS program [Lewis and Wallace, 1998] with coefficients K1 and K2 from Roy et al. , KSP from Mucci  and KSO4 from Dickson . The fraction of sea ice meltwater in each water sample (fSIM) was estimated from mass balance equations for S and δ18O [Yamamoto-Kawai et al., 2009a]. ΩPI, ΔΩatm, ΔΩtemp, ΔΩa-s and ΔΩdil are estimated as follows. The DIC concentration in surface water observed in year t (Cobst) with observed S (Sobst), T (Tobst) and TA (TAobst) is expressed as the sum of DIC of water in equilibrium (CEQ) with atmospheric CO2 (fCO2t) and CO2 difference in air-sea equilibrium expressed in terms of DIC concentration (ΔCdiseqt):
Disequilibrium can be due to cooling, warming, photosynthesis, and mixing with deeper water coupled with insufficient air-sea CO2 exchange. In the preindustrial period, S, T, TA and ΔCdiseq would have been different from our observations that have been influenced by recent extensive melting of sea ice. DIC in this period (CPI), when atmospheric CO2 was 280 μatm, is thus expressed as,
If we know S′, T′, TA′ and ΔCdiseq′, then the preindustrial Ω can be estimated from CPI and TA′. We set T′ to be −1.5°C assuming that sea ice completely covered the study area in the preindustrial period. ΔCdiseq′ was set to be −41 μmol/kg from the mean ΔCdiseq obtained in 1997 at offshore stations covered by multi-year sea ice (n = 8, standard error = 5 μmol/kg) assuming that these stations represent preindustrial conditions. Horizontal distributions of S and TA in surface waters are established by formation/melting of sea ice and by mixing of seawater from the Bering Sea with river water. We assume that only the former process had changed from preindustrial period to the year of observations. S and TA in the absence of sea ice formation/melting (S0 and TA0) should vary regionally and inter-annually reflecting fluctuation of river water pathways [Macdonald et al., 2002; Yamamoto-Kawai et al., 2009a], and can be calculated for each observed water sample by removing the effect of sea ice meltwater/brine from fSIM [Yamamoto-Kawai et al., 2009a]:
SSIM=4 was used following Yamamoto-Kawai et al. [2009a]. TASIM was set to be 441 μmol kg−1 from ice samples observed by Rysgaard et al. . Then, by assuming that distributions of S0 and TA0 for the preindustrial period were the same as observed in 1997 or 2008, preindustrial S′ and TA′ can be estimated from S0, TA0 and preindustrial fSIM (fSIM′):
We set f′SIM to be 0.00 by assuming that the mean fSIM for offshore ice-covered stations in 1997 (fSIM = 0.00; n = 7; standard error = 0.01) represents preindustrial f′SIM. Accordingly, S′ and TA′ at these stations are almost identical to S and TA observed in 1997. Now, Ω in the preindustrial period (ΩPI), as well as Ω with the elevated atmospheric CO2 of 1997 or 2008 (Ω1), Ω with elevated atmospheric CO2 plus increased T (Ω2), Ω with elevated atmospheric CO2 plus change in air-sea disequilibrium (Ω3), Ω with elevated atmospheric CO2 plus dilution by sea ice melt (Ω4), can be calculated from observed or estimated S, T, TA, and DIC as listed in Table 1. ΔΩatm is the difference in Ω due to increase in fCO2 and represented by Ω1 − ΩPI. Similarly, ΔΩtemp is Ω2 − Ω1, ΔΩa-s is Ω3 − Ω1, and ΔΩdil is Ω4 − Ω1.
Table 1. Parameters Used to Calculate Ω
CEQ(S′, T′, TA′, fCO2=280) + ΔC′diseq
Ω1 (ΩPI + ΔΩatm)
CEQ(S′, T′,TA′, fCO2t) + ΔC′diseq
Ω2 (ΩPI + ΔΩatm + ΔΩtemp)
CEQ(S′, Tobs,TA′, fCO2t) + ΔC′diseq
Ω3 (ΩPI + ΔΩatm + ΔΩa-s)
CEQ(S′, T′,TA′, fCO2t) + ΔCdiseq
Ω4 (ΩPI + ΔΩatm + ΔΩdil)
CEQ(Sobs, T′,TAobs, fCO2t) + ΔC′diseq
Ωobs (ΩPI + ΔΩatm ΔΩtemp + ΔΩdil + ΔΩa-s)
4. Results and Discussion
 In 2008, most surface waters in the southern Canada Basin were undersaturated with respect to aragonite (Figure 1). The lowest Ωobs values were observed in the southeast part of the basin, where fSIM was high [Yamamoto-Kawai et al., 2009b]. Preindustrial surface waters were oversaturated with respect to aragonite in the whole study area and ΩPI ranged between 1.4 and 1.6. This is similar to the simulated value (Ωaragonite ∼1.5) for the Arctic Ocean in year 1820 from a global coupled carbon cycle-climate model [Steinacher et al., 2009]. As shown in the same model simulation, Ω in the Arctic surface water was the lowest in the global ocean in the preindustrial period (the global mean Ωaragonite was 3.4). This was primarily due to a massive input of freshwater from the North Pacific and surrounding rivers [cf. Carmack et al., 2008] with low TA [Cooper et al., 2008], low S and low Ca2+ concentration and atmospheric cooling. As fSIM, T and ΔCdiseq are assumed to be constant over the study area, small regional variation in ΩPI in Figure 1 reflects distribution of these waters, especially river water [cf. Bates et al., 2009; Chierici and Fransson, 2009].
 The distribution of ΔΩatm shows that the increase in atmospheric CO2 had lowered Ω by ∼0.25 by 1997 and ∼0.3 by 2008 (Figure 2). This is smaller than global mean of −0.4 [Steinacher et al., 2009] because the increase in DIC for a given increase in atmospheric CO2 is smaller in colder waters (high Revelle Factor).
 Although ΔΩatm should be larger (more negative) in warmer water, warming increases Ω itself by decreasing the solubilities of CaCO3 and CO2. The highest sea surface T (SST) was observed in the southern Canada Basin, with the maximum value of 3°C and 8°C in 1997 and 2008, respectively. The increase in T from −1.5°C, assumed for preindustrial period, to current values increases Ωaragonite by 0.25 and 0.5, respectively (Figure 3). ΔΩtemp is ∼0 in the northern Canada Basin even in 2008, where surface water is still mostly covered by sea ice and SST is < −1°C. Continued warming is expected to increase ΔΩtemp in the future.
 The change in Ω due to air-sea disequilibrium (ΔΩa-s) ranged from −0.5 near the coast to 0.2 offshore in 1997. Close to the coastal, high S and high fCO2 (not shown) indicate upwelling of subsurface water in 1997. Subsurface water with high fCO2 should also have low Ω and thus upwelling of this water lowers surface Ω. ΔΩa-s of the offshore stations are ∼0 as mean ΔCdiseq for these ice-covered stations were used as the preindustrial ΔCdiseq. In 2008, ΔΩa-s was negative all over the study area with higher negative values in the southern region. Negative values are due to higher fCO2 closer to equilibrium with atmospheric CO2 than in the preindustrial period (as exemplified by the ice covered area in 1997). Cai et al.  observed that fCO2 in the western part of the Canada Basin in 2008 was higher and closer to equilibrium than in the 1990s. They attributed this to the increased area of open water and longer ice-free period. Based on our estimates, the mean ΔCdiseq changed from −35 μmol/kg of 1997 to −12 μmol/kg in 2008 (corresponding to an air-sea fCO2 difference of −88 and −41 μatm, respectively). This change would be achieved if surface water had been exposed to the atmosphere for 47 more days in 2008 than in 1997. Considering the increase in initial seawater fCO2 due to warming before air-sea gas exchange started, it takes 38 more days. This calculation was done following the similar procedure as Cai et al. , assuming the mixed layer depth of 20 m and a wind speed of 7.3 m/s. Cai et al.  also estimated from satellite data that the ice-free period was 1.5–2 months in 2008 and 1 month in 1999. Therefore, 38 more days of gas-exchange is not unrealistic. The Arctic Ocean is predicted to be almost ice-free in summer by as early as 2030 [Stroeve et al., 2008]. Although increasing atmospheric CO2 concentration will continue to lower Ω (more negative ΔΩatm), the continued disappearance of sea ice will not decrease Ω further because the disequilibria observed in 2008 is already close to 0 and surface water fCO2 will equilibrate with atmosphere within a short time in the future [Cai et al., 2010]. Some studies indicate that primary production in the Arctic basin will increase under ice-free conditions due to the longer growing season [Arrigo et al., 2008]. This will lower surface fCO2 and increases Ω. However, because of low nutrient concentrations in surface waters, photosynthesis in summer in the Canada Basin interior occurs mostly in the subsurface layer at ∼50 m where nutrients are available during summer, and the depths of this nutricline has deepened in recent years [McLaughlin and Carmack, 2010]. Therefore low primary productivity of the surface Canada Basin will be maintained in the future and will not significantly change Ω.
 The change in Ω by dilution (ΔΩdil) was positive at shelf/slope stations in 1997 (Figure 2), corresponding to observations of negative fSIM values [Macdonald et al., 1999, 2002]. Negative fSIM indicates salinization by brine rejection from sea ice formation which increases Ω (opposite to the dilution by sea ice melt). The presence of brine-enriched water may also be associated with coastal upwelling of subsurface water with negative fSIM values. In 2008, ΔΩdil ranged from 0 to −0.3. Naturally, large (more negative) values were found in the area where fSIM was high in 2008 [Yamamoto-Kawai et al., 2009b]. The high fSIM is due to the increased melting of sea ice in recent years and also the accumulation of fresh surface water within the Beaufort Gyre in the Canada Basin, which is forced by Ekman convergence associated with the anticylonic wind field, and this has enhanced recently due to a more mobile sea-ice cover [Shimada et al., 2006; Proshutinsky et al., 2009]. We expect that both the mobility of ice/ocean and fSIM will remain higher than the preindustrial conditions, but will not continue to increase in the future because much of the multi-year ice has already disappeared from the Arctic Ocean [Kwok et al., 2009].
 In winter, cooling of surface water to the freezing T sets ΔΩtemp to be 0, and ΔΩdil will be less negative during winter because of sea ice formation and convective mixing with deeper waters increase surface layer S, TA, and Ca2+. However, winter convection is now limited to shallow layer (20–40 m), and in recent years the surface S is lower throughout the year due to increased melting of sea-ice [Jackson et al., 2010]. Thus, winter ΔΩdil is likely low now. Rapid cooling and ice formation may increase the air-sea disequilibrium during winter. Therefore, overall, the undersaturation of surface water is likely a year-round phenomenon, although there are insufficient observations to assess the annual conditions of Ω.
 In summary, the recent melting of sea ice has contributed to the rapid decrease of surface Ω in the 2000s and made surface waters in the Canada Basin undersaturated with respect to aragonite (Figure 3). These waters are still marginally oversaturated with respect to calcite (Ωcalcite is 1.1–2.0 in 2008). We expect that surface waters will continue to be undersaturated with respect to aragonite and will become undersaturated with respect to calcite in the near future. However, the rate of decline in Ω will slow down with the disappearance of multi-year sea ice. Atmospheric CO2 concentration and temperature will then be factors that will control future surface Ω in the Canada Basin. Changes in ocean dynamics due to climate change and consequent biological responses may also alter future Ω.
 We thank A. Proshutinsky as a principal investigator of the JOIS project, funded by NSF. We are grateful to the captain, crew, and all scientific participants on the cruises of the CCGS Louis S. St-Laurent. M. Davelaar is thanked for the analysis of DIC and TA.