The recent state and variability of the carbonate system of the Canadian Arctic in the context of ocean acidification

Ocean acidification driven by the uptake of anthropogenic CO2 by the surface oceans constitutes a potential threat to the health of marine ecosystems around the globe. The Arctic Ocean is particularly vulnerable to acidification due to its relatively low buffering capacity and, thus, is an ideal region to study the progression and effects of acidification before they become globally 15 widespread. The appearance of undersaturated surface waters with respect to the carbonate mineral aragonite (ΩA < 1), an important threshold beyond which the calcification and growth of some marine organisms might be hindered, has recently been documented in the Canada Basin and adjacent Canadian Arctic Archipelago. Nonetheless, few of these observations were made in the last five years and the spatial coverage in the latter region is poor. Additionally, the strong 20 variability inherent to this dynamic shelf environment renders the temporal imprint of ocean acidification on carbonate system parameters (pH, pCO2, DIC, Ω) virtually indistinguishable on decadal timescales. We use a dataset of carbonate system parameters measured in Canadian Arctic Archipelago (CAA) and its adjacent basins to describe the recent state of these parameters across the Canadian Arctic and investigate the amplitude and sources of the system’s variability. Our 25 findings reveal that, in addition to the surface of the Canada Basin, the entire water column of the Queen Maud Gulf was undersaturated with respect to aragonite in 2015 and 2016. We also estimate that approximately a third of the interannual variability in surface DIC in the CAA results from fluctuations in biological activity.

widespread. The appearance of undersaturated surface waters with respect to the carbonate mineral aragonite (ΩA < 1), an important threshold beyond which the calcification and growth of some marine organisms might be hindered, has recently been documented in the Canada Basin and adjacent Canadian Arctic Archipelago. Nonetheless, few of these observations were made in the last five years and the spatial coverage in the latter region is poor. Additionally, the strong 20 variability inherent to this dynamic shelf environment renders the temporal imprint of ocean acidification on carbonate system parameters (pH, pCO2, DIC, Ω) virtually indistinguishable on decadal timescales. We use a dataset of carbonate system parameters measured in Canadian Arctic Archipelago (CAA) and its adjacent basins to describe the recent state of these parameters across the Canadian Arctic and investigate the amplitude and sources of the system's variability. Our

Introduction
Ocean acidification and its repercussions on marine ecosystems constitute an important consequence of the ongoing rise in atmospheric carbon dioxide (CO2) concentrations. The world's oceans absorbed approximately a third of the anthropogenic CO2 released to the atmosphere over the last two centuries of industrial activity (Sabine et al., 2004) and are currently a sink for ~24% 35 of global annual anthropogenic carbon emissions (Le Quéré et al., 2018). Atmospheric carbon dioxide uptake by the surface oceans has well defined impacts on seawater chemistry, including a decrease of pH resulting from the dissociation of carbonic acid (H2CO3), the product of the reaction between water and dissolved CO2. A large fraction of the hydrogen ions released by this reaction is neutralized by carbonate ions (CO3 2-), leading to a decrease of their concentration and, 40 concomitantly, the saturation state of seawater with respect to the carbonate minerals calcite and aragonite. The saturation state is defined by: (1) ΩC,A= [Ca 2+ ][CO3 2-] / K * SP where square brackets denote concentrations and K * SP is the stoichiometric solubility product of calcite or aragonite, the two most common marine CaCO3 polymorphs, at a given temperature, 45 pressure and salinity.
The combination of these phenomena is most often referred to as ocean acidification (OA).
As it proceeds, the dissolved inorganic carbon concentration (DIC; the sum of [H2CO3*], ] and [CO3 2-]) in the surface ocean is expected to increase relative to the total alkalinity (TA; the 50 capacity of a solution to neutralize protons), as the latter is nearly conservative in the surface ocean (Wolf-Gladrow et al., 2007). The global mean surface ocean pH currently sits ~0.1 units below its preindustrial value (Orr et al., 2005) and, according to Earth System models, under the IPCC's "business as usual" RCP8.5 emission scenario, is predicted to decrease by an additional 0.3 units by the end of this century (Bopp et al., 2013). 55 Marine calcifying organisms, many of which are important primary producers (e.g., coccolithophores), extract the constituents of their calcitic or aragonitic tests (shells) from seawater. In most cases, their ability to do so is directly dependent on the saturation state of the https://doi.org/10.5194/bg-2020-29 Preprint. Discussion started: 20 February 2020 c Author(s) 2020. CC BY 4.0 License.
2 Study area 2.1 Canada Basin 90 The Canada Basin (CB), Canadian Arctic Archipelago and Baffin Bay accommodate the flow of surface waters from the North Pacific to the North Atlantic (Stigebrandt, 1984), as well as circulation of Atlantic waters at greater depths. The water mass structure of the southern Canada Basin is representative of these broad circulation patterns and can be summarized as follows MacDonald et al., 1989;Lansard et al., 2012): a relatively cold and fresh 95 surface layer that contains significant fractions of meteoric water (river discharge and precipitation) and sea-ice melt in the summer and becomes homogeneous in winter; an intermediate layer (~50-200m) of advected Pacific waters, often divided into summer and winter varieties, the latter being distinctively rich in nutrients and metabolic CO2 and recognizable by a temperature minimum in the upper halocline; a layer of warm (~0.5°C) and saline (SP>34) Atlantic 100 water; a cold bottom layer with practical salinities (SP) reaching 34.85. The main surface circulation feature in this area, the clockwise Beaufort Gyre, is the largest freshwater reservoir in the northern oceans, formed through Ekman pumping (Proshutinsky et al., 2009). This feature is reversed at depth. The main source of freshwater to the Beaufort Sea (the southwest portion of the Canada Basin) is the Mackenzie River (Carmack and MacDonald, 2002), although the contribution 105 of sea-ice melt is significantly increasing along with the accelerating reduction in ice cover (Yamamoto-Kawai et al., 2009a). The supply of freshwater at the surface, combined with the advection of pre-acidified waters from the Pacific (100-200 m) and the Atlantic (below 400 m; Luo et al., 2016) Oceans result in the presence of three distinct and expanding undersaturation horizons in the Canada Basin (Wynn et al., 2016).

Canadian Arctic Archipelago
The CAA is a series of islands on the Canadian continental shelf, through which complex circulation patterns unfold in narrow and relatively shallow channels (<500 m Strait, located centrally in the archipelago, inhibits the eastward flow of Atlantic waters, so that only surface and Pacific-origin waters reach Baffin Bay (Bidleman et al., 2007). The properties of these water masses are substantially modified during this transit . East 120 of Barrow Strait, Atlantic waters originating from the Labrador Sea penetrate the archipelago through Baffin Bay. Smaller inflows of water from the deep Arctic Ocean into the archipelago occur through the Queen Elizabeth Islands (North-East) and Nares Strait; minor outflows occur through Jones Sound and into Hudson Bay via Foxe Basin. Notwithstanding the Mackenzie River, whose discharge is limited to the Beaufort Sea and Amundsen Gulf, the southern portion of the 125 CAA receives a considerable amount of freshwater from other large North American rivers (e.g., Coppermine, Burnside, Black).

Baffin Bay
The oceanographic regime of Baffin Bay is distinct from that of the CAA and CB, as it 130 receives multiple inputs from both the Arctic and Atlantic Oceans. Cold and relatively fresh Arctic and Pacific-derived waters enter this 2300-m deep semi-enclosed basin through the Nares Strait as well as Jones and Lancaster Sounds (Muench, 1971;Jones et al., 1998Jones et al., , 2003. Warmer and more saline Atlantic Ocean waters are transported from the Labrador Sea by the West Greenland Current (WGC) into Baffin Bay through the eastern side of Davis Strait, circulate cyclonically, i.e., in an 135 anti-clockwise direction, before joining the southward Baffin Island Current (BIC) which exits Baffin Bay through the western Davis Strait (Bourke et al., 1989, Munchow et al., 2015. Atlantic Ocean waters are modified as they mix with Arctic inflows in Northern Baffin Bay, near the North Water Polynya (Melling et al., 2001). The resulting water mass structure is described by Tang et al. (2004) as: 1) a cold (T<0°C) and relatively fresh (SP<33.7) surface layer, representing the mixed 140 Arctic inputs, 2) a warm (T>0°C) and saline (SP>34) Atlantic Ocean water layer found at depths of ~300 to 800 m, and 3) a deeper layer of nearly constant salinity (SP = 34.5).

Methods
The dataset used in this study comprises data from 420 stations visited during various research cruises carried out aboard the CCGS Amundsen between 2003 and 2016. Table 1 summarizes the timeframe and relevant data acquired during each cruise; figure 2 shows the position of each sampling station. Although ice conditions restricted most observations to the 150 summer months, two winter time-series, acquired in 2003-2004(CASES, Miller et al., 2011(CFL, Shadwick et al., 2011 are included in the dataset.

Sampling and measurements
Seawater was sampled separately for each measured parameter from Niskin bottles 155 mounted on a Rosette system equipped with a Seabird SBE 911plus Conductivity-Temperature-Depth (CTD) sensor, which collected in-situ practical salinity (SP) and temperature data throughout the water column. The conductivity/salinity probe was calibrated post-cruise against measurements carried out on discrete seawater samples using a Guildline Autosal 8400 salinometer (accuracy of  0.002 or less), itself calibrated with IAPSO standard seawater. Samples 160 used for pH determination were drawn directly from the Niskin bottles into 125 mL low density polyethylene (LDPE) bottles with no headspace to avoid gas exchange with surrounding air and left to thermally equilibrate in a temperature bath set at 25.0 (0.1) °C. pH, on the total proton scale (pHT), was then measured spectrophometrically on a Hewlett-Packard 8453 UV-visible diode array spectrophotometer using m-Cresol purple (Clayton and Byrne, 1993) and Phenol red onboard or at Dalhousie University on a Marianda VINDTA 3C instrument, following the protocol described by Dickson et al. (2007) and calibrated with Certified Reference Materials (CRM) provided by A.G. Dickson (Scripps Institute of Oceanography). The precision of the instrument was found to be  2-3 µmol kg -1 based on repeated CRM analyses. The remaining DIC and TA analyses were performed respectively on a SOMMA instrument (Johnson et al., 1993) (Lewis and Wallace, 1998), using the carbonic acid dissociation constants determined by Mehrbach et al. (1973), refit by Dickson and Millero, (1987), the HSO4dissociation constants of Dickson (1990) and the total boron concentration (BT) from Uppström (1974). pCO2 was normalized, using CO2SYS, to the mean temperature of the top 100 meters of the water column (-0.4 °C). The resulting ∂pCO2 /∂T/pCO2 values range between 0.033 and 0.051 °C -1 , with 190 a mean of 0.046 °C -1 , in good agreement with values obtained by Jiang et al. (2008;0.027-0.042°C -1 ) and Takahashi et al. (1993;0.0423 °C -1 ).

Quality control
In order to assess the robustness of the computed DIC values, we calculated DIC from TA 195 and pHT(25°C) and compared the results with the measured DIC values. The resulting coefficient of determination of the linear fit to the measured and calculated DIC values, R 2 , is 0.989, while the mean difference between calculated and measured DIC values is ~2 µmol kg -1 . We excluded 30 measurements that differed from the calculated values by more than 50 µmol kg -1 (2.5% of the mean DIC). Questionable TA measurements, excluded from the dataset, were identified as those outside a range of 3 standard deviations from the mean salinity-normalized TA for individual regions (CB, CAA, BB) characterized by internally consistent water mass assemblages. TA measurements obtained from the two instrumental methods (VINDTA and Radiometer Titrilab 205 865) used in 2015 and 2016 were also compared to ensure that data originating from both methods could be used interchangeably in the calculation of additional parameters and conjointly in time series. The resulting coefficient of determination between both datasets (R 2 ) is 0.988, the mean of the non-systematic discrepancy between values is 6 µmol kg -1 and its maximum is 36 µmol kg -1 , respectively corresponding to 0.3% and 1.7% of the mean TA. The degree to which the results of 210 this test are representative of the entire dataset is uncertain, but they constitute the best possible estimate of the uncertainty associated with the use of the two analytical methods used to measure TA. When TA measurements obtained from both methods deviated significantly ( >10 µmol kg -1 ), specific alkalinity (TA/SP), which should remain relatively constant in a given water mass of uniform salinity (Millero, 2005;p.268), was used to determine which data to discard. The deviation 215 from TA values calculated from DIC and pHT(25°C) was used to complement the first method, especially at the surface, although the validity of DIC measurements was previously assessed using TA.

Error estimation 220
In order to quantify the error associated with the calculated carbonate system parameters reported in this study, we used the CO2SYS program modified by Orr et al. (2018), which applies error propagation to instrumental and constant-related uncertainties. For simplicity, we report the mean uncertainty for each parameter (see Table 2), as the variance is minimal within our dataset.
We found the additional uncertainty associated with the unavailability of nutrient concentrations 225 (P and Si) as input parameters in CO2SYS to be negligible (up to 0.0006 pH units, 1.5 µatm pCO2 and 0.006 Ω units, as determined using nutrient data where available). The uncertainty on Δ DICBio, the biological contribution to temporal variations of the surface DIC pool, was calculated by applying standard error propagation to the procedure described in Sect. 4.3. and Baffin Bay (BB) are presented in Table 3. It is important to note that the mean regional values we report for the Canada Basin may be skewed by the higher density of stations located along the Mackenzie Shelf, and that our sample size for Baffin Bay consists of only 6 stations. Practical 240 salinities considerably below 25 were mostly observed near the mouth of the Mackenzie River, with some in the Queen Maud Gulf (QMG). The discrepancy between SP and TA values observed in the CB/CAA and Baffin Bay clearly illustrates the change in water mass regime west of Lancaster Sound (see Sect. 2), while DIC, which is strongly affected by biological activity, shows a less prominent spatial pattern. In all regions, surface-water pCO2, of which we only consider 245 data acquired over the top 5 meters of the water column in order to render it more indicative of gas exchange potential, was largely undersaturated with respect to the atmosphere, by as much as 150 µatm (Fig. 3). This suggests that the region as a whole acts as a net CO2 sink during the summer,   Kawai et al., 2009b) and increased river discharge (Déry et al., 2016). We estimate, based on a linear regression of surface δ 18 O data against SP, that upwards of 95% of the freshwater at the surface of the QMG in 2015 was of riverine origin (Appendix B). Our dataset does not allow us to directly differentiate the contributions of air-sea gas exchange and biological activity (respiration) to the high pCO2 280 observed at these locations. Nonetheless, the depth of these samples (<5 m) implies some degree of equilibration with the atmosphere. Although the diurnal cycle of biological activity may play a role in the development of peaks in Ω, we rule out this mechanism for the case discussed above, as the 0.73 minimum of 2016 was observed in the early afternoon. Although the ΩA minima represent significant undersaturation, the uncertainty on ΩA computations (0.08 or 0.16 depending 285 on the parameters used in the calculation; see Methods) blurs the saturation threshold in such a way that ΩA values marginally below 1 might in reality represent supersaturated conditions, and vice-versa. It is important to note that, even without the influence of climate change, areas of high riverine discharge naturally harbor lower carbonate mineral saturation states. Thus, undersaturated conditions in the QMG and elsewhere do not solely result from the documented increase of 290 https://doi.org/10.5194/bg-2020-29 Preprint. Discussion started: 20 February 2020 c Author(s) 2020. CC BY 4.0 License. freshwater inputs described above. Nonetheless dilution by freshwater affects the degree of this undersaturation as well as its spatial and temporal extent.
Surface waters throughout the study area are supersaturated with respect to calcite, with ΩC ranges (mean) of 1.34 to 3.25 (2.04), 1.21 to 3.29 (1.97) and 2.38 to 2.70 (2.52) in the CB, CAA 295 and BB, respectively. Uncertainties on ΩC values are on the order of 0.25-0.30, almost twice as large as those of ΩA, due to the larger uncertainty of the calcite stoichiometric solubility product at 25 °C and SP= 35 (Mucci, 1983). The most prominent feature in profiles of carbonate system parameters in the Canada Basin is the Upper Halocline Layer (UHL), a layer of water originating from the Pacific Ocean with a relatively lower pH due to its high metabolic CO2 content (Shadwick et al., 2011 the UHL was characterized by a pHT minimum of 7.82 ± 0.03, a pCO2 maximum of ~652 ± 6 µatm and a ΩA minimum of 0.75 ± 0.16 in the central CB. This pHT minimum migrates upwards from ~180 to ~ 140 meters as the UHL encounters the continental shelf west of M'Clure Strait but maintains its amplitude. The presence of such an acidified layer exacerbates the vulnerability of the planktonic communities in this area, as, in addition to the aragonite undersaturation found at 315 the surface, ΩA drops below one at depths of 100 to 125 meters or even shallower waters in the Canada Basin. As CO2 naturally diffuses or mixes from the UHL to the overlying waters and the combination of gas exchange and freshening continues to generate undersaturated conditions at https://doi.org/10.5194/bg-2020-29 Preprint. Discussion started: 20 February 2020 c Author(s) 2020. CC BY 4.0 License. the surface, the entire photic zone (where ΩA < 1.5) may acidify and become undersaturated with respect to aragonite at a much faster rate than that of other oceans. The shallowest subsurface The Amundsen Gulf and the western portion of the Parry Channel ( Fig. 7; CAA1) essentially exhibit the same water mass structure and carbonate system chemistry as the Canada Basin, as the dominant circulation pattern pushes water eastward from the CB to the CAA. Undersaturation with 335 respect to aragonite does not occur at the surface in these areas, owing to higher salinities.

Water column observations
Although the amplitudes of the ΩA, pHT and pCO2 excursions are slightly smaller than those found in the CB, the UHL is considerably shallower in the western CAA. Consequently, ΩA falls below one at depths of 50 to 70 meters, and the upper portion of the water column in those parts of the CAA might become undersaturated with respect to aragonite even more rapidly than in the CB.

340
As reflected by the blue lines in Fig. 7 (CAA1), the UHL becomes progressively less discernable on depth profiles as it undergoes modification and mixing during its transit from the CB to Lancaster Sound. Atlantic waters are found at the bottom of the water column in the Amundsen Gulf and Parry Channel. The saturation maxima at ~400 m (ΩA ~ 1.1 to 1.4) are significantly lower in these areas than at similar depths in the CB.

Time series
Of the 420 stations that make up our dataset, twenty-four were visited at least on two 380 different years and match our comparability criteria for time-series. These criteria are: 1) the stations were sampled within 31 calendar days of each other (this criterion is not ideal since seasonality is highly variable and driven by complex sea-ice processes, including ice break-up) and 2) the stations are located within a 5 km radius. The mean time difference and distance between comparable stations are 12 calendar days and 1.81 km, respectively. Eight were visited three times, To quantify near-surface change, we averaged data from the top 25 meters of the water column. Across this depth interval, between 2007 and 2016, the temperature-normalized pCO2, DIC and DIC/TA at site LS1 rose by 36 ± 9 µatm, 37 ± 10 µmol kg -1 and 0.008 ± 0.005, while 400 pHT and ΩA decreased by 0.042 ± 0.037 and 0.12 ± 0.23, respectively. Given its uncertainty, the change in ΩA is insignificant. The same trend is visible between the same years at station CAA1, at a greater magnitude (+ 78 µatm pCO2, -0.088 pHT unit, -0.37 ΩA unit), except for a DIC decrease  As previously stated, these time series are snapshots in time and cannot be assumed to represent the continuous evolution of the carbonate chemistry in the Canadian Arctic. Nonetheless, even with a small sample size, we can confidently state that the temporal evolution of carbonate system parameters in the region does not display a systematic trend on sub-decadal timescales. 455 Moreover, most of the significant changes our time series exhibit are associated to variations in the physical oceanography of the region (water-mass distribution and circulation) or surface processes (melting of sea-ice). Given the well-documented rapid melting of the sea-ice cover in the region (e.g., Tivy et al., 2011), we did not expect to observe increases in summer surface salinity over time intervals of 5 to 9 years. Our time series, therefore, offer proof of the strong 460 interannual variability of this highly dynamic system. Discerning the ocean acidification signal amid the various physical and biological sources of change would require continuous time series over a longer period of time. We estimated this period to 23 to 35 years for pH, 25 to 37 years for pCO2, 31 to 46 for ΩA and 118 to 177 years for DIC using calculations of Time of Emergence, or the time required for the effects of a process to emerge from the natural variability of a system (see 465 Appendix A). Time of emergence calculations are usually performed with large, continuous datasets from climate models; we therefore do not consider these results to be statistically correct. Nonetheless, this exercise shows the relative variabilities of the carbonate system parameters and highlights the particularly strong variability of DIC, which is the object of the next section. Its results also imply that without accurate measurements of the effects of biological activity and sea-470 ice processes (both major drivers of natural variability), direct detection of the ocean acidification signal will require at least 20-25 years of observations. Gathering data in the CAA in the next few years is therefore critical, as regular ship-based observational campaigns in the region started in the early 2000's (Giesbrecht et al., 2014).

The biological contribution to interannual DIC variations: ΔDICBio
We define ΔDICBio as the change in the contribution of in-situ biological activity (photosynthesis and respiration) to the DIC of a parcel of water over a given period of time.
ΔDICBio is calculated for each depth at recurrently visited stations according to: where DICReference is computed in CO2SYS using the temperature-normalized seawater pCO2 calculated at a reference time (from measured TA and pHT) and the TA measured at the time of interest (the year for which ΔDICBio is reported). The change in global mean atmospheric CO2 concentrations between the reference year and the year of interest is added to the pCO2 in order to account for gas exchange (data from Dlugokencky and Tans, NOAA/ESRL). This approximation 485 rests on the assumption that the yearly increase in surface water pCO2 follows that of the atmosphere (given stable biological production), as observations from global monitoring stations demonstrate (e.g., González-Dávila et al., 2010), although the validity of this claim is weakened on short spatial and temporal (sub-decadal) scales (Fay et al., 2013, Wanninkhof et al., 2013. This also restricts our calculations to the upper portion of the water column (25 m) that is in direct 490 contact with the atmosphere. Under the additional assumptions that DIC is only affected by gas exchange, biological activity and mixing, and that TA is not significantly affected by biological activity (Zeebe and Wolf-Gladrow, 2001), DICReference represents the DIC of a parcel of water if its in-situ biological component remained unchanged relative to a reference year (i.e., identical contribution, negative or positive, from the balance between photosynthesis and respiration).
affect the results of this analysis, given the salinity range of the data subset used in the calculation of ΔDICBio (25.6 < SP < 33.7). Thus, ΔDICBio can provide insights into the interannual variability of biological activity in the Canadian Arctic, without direct measurements of parameters such as chlorophyll or biomass. 500 Figure 12 shows ΔDICBio, averaged over the top 25 meters of the water column, at the 18 stations where comparable data were available, all located in the Amundsen Gulf and CAA. The magnitude of the calculated ΔDICBio is beyond its uncertainty, which varies from 6.4 to 14.3 µmol kg -1 (mean of ± 8.6 µmol kg -1 ), at 11 locations out of 18. All significant results from the month of balance is driven by many interconnected, often localized processes. For instance, short-lived episodes of upwelling of halocline waters not only directly change the chemical properties at the surface, but also provide nutrients that stimulate biological activity .
Primary production in the Arctic is also closely linked to the seasonal cycle of sea-ice (e.g., Arrigo et al., 2008). Beyond its natural fluctuations, ongoing disruptions to this cycle and other physical 535 properties (stratification, temperature, etc.) forced by climate change are known to affect phytoplankton communities and their productivity (Ardyna et al., 2014;Blais et al., 2017), possibly increasing their imprint on the variability of DIC and other carbonate system parameters.
As previously mentioned, variations in water mass composition cannot directly explain variations in ΔDICBio. Nonetheless, mixing is likely accompanied by changing nutrient concentrations, which 540 influence the balance of photosynthesis and respiration (Tremblay et al., 2015).

Conclusions
Field observations of carbonate system parameters made between 2014 and 2016 in the Canadian Arctic reveal that surface waters of the region serve as a net CO2 sink in the summer and 545 are generally close to saturation with respect to aragonite (1 < ΩA < 1.5). Surface undersaturation

Time of Emergence of ocean acidification signals
The time of emergence (ToE) of a process affecting a natural system is the time required for the measurable effects of this process to emerge from the natural variability of the system. The concept is predominantly applied in global climate change modeling studies, for which the results 585 are either "years of emergence" based on a pre-industrial steady-state (e.g., Friedrich et al., 2012) or time intervals over which observations must be made in order to distinguish an anthropogenic signal from natural variability. Few of these studies have used observations (e.g., Sutton et al. 2016), and, to our knowledge, none of them have focused specifically on the Arctic.
We define the time of emergence according to the following equation:  Although the assumption of relative equilibration with atmospheric pCO2 might be less applicable at these depths, we applied the same technique using data from depths of 100 m and 300 m in order to estimate the time of emergence below the surface, where the interannual variability should be relatively small. Significant acidification at these depths might be due in large part to advection (e.g., Luo et al., 2016) rather than gas exchange.

610
The results of this analysis are presented in Table A2. Despite their similarity, our calculated times of emergence are consistently longer than those reported in modelling studies (Keller et al., 2014;Rodgers et al., 2015) This is consistent 630 with the fact that coastal waters, that comprise a large portion of our dataset, exhibit a much higher variability in pH (and other carbonate system parameters) than open oceans (Duarte et al. 2013).
Furthermore, direct observations are likely to integrate variability on temporal and spatial scales that are too small to be resolved by models. It is also important to note that distinct measurement used by Lansard et al. (2012), the fractions of meteoric (river) water and sea-ice melt would be, 665 respectively, 98% and 2%. A potential source of error affecting this estimate is the use of the δ 18 O of Mackenzie River water as the riverine end-member, which might differ significantly from the oxygen isotope signature of the rivers discharging in the Queen Maud Gulf.

670
The raw data collected as a part of the ArcticNet program, on which most of the observations presented in this paper are based can be accessed through the Polar Data Catalogue (Mucci, 2017).

Competing interests
The authors declare that they have no conflict of interest.