High interannual surface p CO 2 variability in the Southern Canadian Arctic Archipelago’s Kitikmeot Sea.

. Warming of the Arctic due to climate change means the Arctic Ocean is now ice-free for longer, as sea ice melts earlier and refreezes later. Yet, iIt remains unclear how the this extended ice-free period will impact carbon dioxide (CO 2 ) fluxes due to scarcity of surface ocean CO 2 measurements. Baseline measurements are urgently needed to understand spatial and temporal air−sea CO 2 flux variability in how air−sea CO 2 fluxes will spatially and temporally vary in a changing Arctic 20 Ocean. It isThere is also uncertainty as to whether the previous basin-wide surveys are representative of the many smaller bays and inlets that make up the Canadian Arctic Archipelago (CAA). By using a research vessel that is based in the remote Inuit community of Ikaluqtuutiak Cambridge Bay (Cambridge Bay Ikaluqtuutiak, Nunavut

determining the circulation in the CAA (Wang et al., 2012). The majority of existing pCO 2 measurements made in the CAA were collected along the southern route through the Northwest Passage on the research icebreaker CGGS Amundsen . This large pCO 2 dataset was used to estimate a −7.7 ± 4 Tg C yr -1 sink for the CAA during the open water season . The CCGS Amundsen pCO 2 dataset provides excellent broad spatial coverage of the CAA, 70 but the vast area surveyed was limited in temporal coverage and fine spatial detail. The CCGS Amundsen typically only transited through the central straits, channels, gulfs, and seas that make upof the southern route through the Northwest Passage once each summer. The numerous bays and inlets that are off the main channel were not sampled, meaning that local-scale pCO 2 variability was potentially unaccounted for during the synoptic scale sampling. This small-scale pCO 2 variability is difficult to predict empirically and may be better observed via regional studies. For example, the model of 75  is was shknown to underestimate pCO 2 by an average of ~26 μatm in Coronation Gulf and Dease Strait regions of the Kitikmeot Sea.  postulated that large river inflow in the region may account for divergences from their model, uAhmed et al. (2019), likely due to river inflow. Understanding what caused this deviation from the modelwhether this is the case warrants further investigation and makes the Kitikmeot Sea a prime location for focused study. 80 Our understanding of the inorganic carbon system in the Kitikmeot Sea region primarily comes from three distinct sources of measurements. Firstly, the 2010−-2016 summertime ship measurements of pCO 2 in the central channel of the Kitikmeot presented by . Their measurements show the region to be slightly undersaturated at the beginning of August, becoming slightly oversaturated supersaturated in the middle of August through to the middle of September, and 85 then becoming undersaturated again in early October. Coronation Gulf is one of the few areas of the CAA that was consistently observed to be supersaturated with CO 2 in summer. Oversaturation Supersaturation of pCO 2 in Coronation Gulf is likely a result of high summer surface seawater temperatures (CO 2 thermodynamics mean that a 1°C temperature increase, increases pCO 2 by 4.23% (Takahashi et al., 1993) (Takahashi et al., 1993)) and high river discharge, particularly to the southwest (Geilfus et al., 2018). The second source of carbonate system measurements in the region are CO 2 flux 90 observations at the Qikirtaarjuk Island observatory oin the Finlayson Islands in Dease Strait (Butterworth and Else, 2018).
Their measurements from the 2017 ice breakup season through to the summer indicate that there is CO 2 drawdown, and thus, undersaturation at breakup and for the first two weeks of open water. Near the end of July, the region transitions into a CO 2 source through to the end of August (Butterworth and Else, 2018). The region reverts to a sink in late August as the sea cools and surface pCO 2 declines; the region remains a sink until almost full ice cover in November (Butterworth et al., 2022). A 95 similar pattern was observed in the summer of 2018, except notably, when pCO 2 began to fall in late August the region did not revert all the way back into a sink (Butterworth et al., 2022). The third source of carbonate system measurements are provided by Duke et al. (2021) who report autonomous pCO 2 measurements at a depth of 7 m from an instrument installed on the Ocean Networks Canada (ONC) underwater sensor mooring in Cambridge Bay between August 2015 and August 2018. The sensor measurements from Cambridge Bay indicate that pCO 2 is oversaturated supersaturated in winter and 100 undersaturated by the start of June at the onset of sea ice melt (Duke et al., 2021). Their measurements show that there is a short period of oversaturation supersaturation in the middle of August coinciding with increased sea water temperature, the ocean then quickly returns to a CO 2 sink and remains undersaturated up until freeze-up (Duke et al., 2021). Duke et al. (2021) confirmed that the biogeochemical measurements at the ONC site were representative of the offshore during most seasons by comparing discrete dissolved inorganic carbon (DIC) and total alkalinity (TA) samples collected at both 2 and 7 105 m at the ONC platform and an offshore station (B1). The surface stratification at ONC breaks down after the 2 week sea ice melt and river runoff period in early July. After the sea ice melt and river runoff period, DIC, TA, salinity, and temperature values recoreded by the ONC mooring are then once again representative of the surface mixed layer.
All three sources of measurements indicate that there is notable interannual variability in surface pCO 2 in the Kitikmeot Sea. 110 The ship-based measurements provide a snapshot of spatial variability across the wider region during the open -water season whereas the time series from Qikirtaarjuk Island observatory and the ONC mooring provide insights into seasonal and interannual variability at specific locations. There are obvious shortcomings to both approaches. Icebreaker-based studies may under-represent small-scale variability that exists in nearshore regions that are inaccessible due to the vessel's large draft. Whereas the fixed observatories may over-represent temporal variability which is location-specific; for example, the 115 ONC mooring is in an enclosed Bay close to the outlet of a large river (Manning et al., 2020) and the flux footprint of the Qikirtaarjuk Island observatory spans a hotspot for mixing and productivity (Dalman et al., 2019). Given the limitations of each of these data sources, there is a need to understand how representative these data sources they are of the wider Kitikmeot Sea region.

120
In this paper, we present surface pCO 2 measurements made during annual summertime surveys of the Kitikmeot Sea between 2016 and 2019. We use these new pCO 2 measurements to determine the magnitude of CO 2 uptake in the Kitikmeot Sea shortly after ice breakup. These new pCO 2 measurements also allow us to bridge the gap between previous measurements, which were made at contrasting spatial scales (e.g.,. the low spatial variability point-scale observation from the local carbon observatories and the large-scale CAA-wide pCO 2 measurements),. We use our new measurements to 125 explore whether there are small-scale regional pCO 2 differences in the inlets and bays of the CAA which are not adequately represented by CAA-wide sampling. We also use our new measurements to explore pCO 2 variability in the proximity of these observatories to see determine whether they are representative of the wider region. In attempting to unify existing measurements, we aim to unravel the seasonal and interannual variability of pCO 2 in the region.

Oceanographic setting
The Kitikmeot Sea ( Figure 1) is a shallow shelf sea within the CAA that encompasses Coronation Gulf to the west, linked via Dease Strait to Queen Maud Gulf in the East, Bathurst Inlet to the South, and Chantrey Inlet to the Southeast (Williams et al., 2018). The communities of Cambridge Bay, Kugluktuk, and Gjoa Haven, Nunavut, are the main year-round settlements in the Kitikmeot Sea region. River inputs from mainland Canada and snow and ice melt provide a considerable 135 source of freshwater in the region (Williams et al., 2018), resulting in some of the lowest salinity surface waters in the CAA . The Kitikmeot sea is nutrient-limited (Back et al., 2021), and as a result chlorophyll concentrations are also low in the region (Kim et al., 2020). Modelling results of the physical oceanography of the region suggest demonstrates that the stratification regime in Dease Strait and Queen Maud Gulf is characterised by a ~40 m warm fresh surface layer and a cold salty bottom layer which extends down to around 100 m (Xu et al., 2021). Coronation Gulf has a three layer regime 140 composed of a 40 m warm fresh surface layer, a colder salty layer down to 100 m and a stable deep layer down to 350 m (Xu et al., 2021). Vertical mixing in the Kitikmeot Sea is prohibited by strong stratification throughout most of the year; however after sea ice breakup wind driven mixing gradually deepens the surface mixed layer resulting in an almost fully mixed water column in Dease Strait (Xu et al., 2021).

145
The oceanographic boundary for the Kitikmeot Sea has been designated as where the shelf shoals to <30 m in the west (Dolphin and Union Strait) and northeast (Victoria Strait) (Williams et al., 2018). At the Dolphin and Union Strait, warm fresher surface seawater flows out across the sills while and subsurface flows of more saline nutrient-rich Pacific waters enter the sea. Another feature of the Kitikmeot Sea is that strong tidal currents in narrow channels can keep certain areas icefree in winter (Williams et al., 2018). Strong tidal currents beneath sea ice such as around the Finlayson Islands in Dease 150 Strait act to slow winter sea ice growth and enhance primary production by introducing nutrients (Dalman et al., 2019). Firstyear sea ice dominates the Kitikmeot Sea although some multiyear ice may be blown into Queen Maud Gulf from the northern part of the CAA (Xu et al., 2021). Seawater temperatures across the Kitikmeot Sea vary considerably throughout the year; they are around -2°C in winter and reach upwards of 10°C in summer (Xu et al., 2021). The bounding sills, large freshwater inputs and low nutrient loads make the Kitikmeot Sea unique within the CAA. 155 observatory where the eddy covariance tower is located. Shoreline data is was taken from the World Vector Shoreline database and river data is was taken from the CIA World Data Bank II (WDBII), both of which were accessed via the Global Self-consistent, Hierarchical, High-resolution Geography Database (GSHHG) (Wessel and Smith, 1996). Bathymetry data is was taken from the 2-minute Gridded Global Relief Data (ETOPO2) v2 database (NGDC, 2006). This map was made using tools from the M_Map Matlab plotting package (Pawlowicz, 2020).

Field campaign description
Annual oceanographic surveys of the summertime surface seawater partial pressure of carbon dioxide ( pCO 2 (sw) ) were programs (cruise details in table S1). In each of the four years, an underway pCO 2 system was deployed on cruises conducted under ice-free conditions between early August and mid-September. The Canadian High Arctic Research Station (CHARS) in Cambridge Bay, Nunavut acted as a staging ground for this work since as Cambridge Bay is the home port for the RV Martin Bergmann. 165 Between 2016 and 2019, the cruise track varied from year to year depending on the focus of the workobjectives of the research conducted ( Figure 2). The first week of each summer field season was typically used to complete work for the MEOPAR program, the majority of the ship time for the MEOPAR work was spent in the proximity of Cambridge Bay, the Finlayson Islands, Wellington Bay and the western region of Queen Maud Gulf. Cruises in mid to late August were used to conduct work for the K3S program; for the K3S work the ship typically ventured travelled further from Cambridge Bayafield 170 heading into Bathurst Inlet, the central region of Queen Maud Gulf and ChantryChantrey Inlet. The opportunistic nature of the data collection meant that data density varied between regions, as not every region was surveyed each year.
Sea ice concentrations in the months preceding each annual survey are were taken from the daily gridded 3.125 km AMSR2 satellite radiometer product (Spreen et al., 2008). To determine weeks since open water, the nearest point on the AMSR2 175 grid was determined for each pCO 2 (sw) measurement. The time between the measurement and when sea ice concentration fell constantly below the threshold value for the marginal ice zone (85%) (Cruz-García et al., 2021) was then calculated.

Underway system
The RV Martin Bergmann is a 20 m repurposed commercial fishing trawler from Newfoundland with a draft of 3.4 m ( Figure 3a and 3b). The ship does not have its own dedicated integrated underway system; instead surface seawater was 180 sampled from an inlet at a depth of ~1 m through ~2 m of 1/2" ID PVC tubing securely draped over the bulwark of the vessel through an external hatch (Figure 3c and 3d). A Waterra Tempest WSP-12V-3 submersible pump was used to pump surface seawater through this inlet tubing at a rate of 10 L min -1 . In situ surface seawater temperature (SST (1m) ) was measured by a Campbell Scientific 107 temperature sensor attached to the tubing inlet.

185
Upon entering the ship, the flow of seawater passed through a SoMAS MSRC VDB-1 vortex debubbler and was then split between several instruments via Tygon tubing ( Figure 3). The An Idronaut Ocean Seven 315 On-line module thermosalinograph measured seawater temperature (SST (tsg) ) and salinity at a seawater flowrate of 0.5 L min -1 . The A Wetlabs ECO BBFL2B Triplet measured fluorescence at a flowrate of 2.5 L min -1 . The The output of the ECO fluorescence sensor output was post-processed to remove spikes from bubbles and particles but was not calibrated against in situ 190 measurements. A flow of 2 L min -1 was directed to the seawater equilibrator. Instrument flowrates were set with manual flowmeters so that the internal instrument volumes and associated tubing of the Idronaut, ECO and equilibrator were flushed at the same rate, this meant that approximately half of the 10 L min -1 flow from the pump was not analysed and was discarded overboard. A made to ordercommercially available Sunburst Sensors underway SuperCO 2 system measured surface seawater CO 2 ; this an identical system was previously described by Evans et al. (2019), and. The SuperCO 2 system follows the general recommendations of Dickson et al. (2007) SOP5. A Permapure liqui-cel 2.5X8 series membrane contactor served was used as the equilibrator for the pCO 2 system, the waterside seawater flowrate for the equilibrator was approximately 2 L min -1 . 200 Seawater temperature was measured at the equilibrator seawater inlet using a thermistor (T (equ) ). The gas counter flow into the equilibrator was supplied by an air pump at a flowrate of 100 ml min -1 . CO 2 has been shown to fully equilibrate in this model liqui-cel when set up in a single pass setup at these water and gas flowrates (Sims et al., 2017). The system does not utilise a dryer and thus does not requires a water vapour correction in post-processing as the equilibrator is assumed to be at 100 % humidity. For additional accuracy, the inbuilt H 2 O sensor was calibrated with a LI-610 Portable Dew Point Generator 205 on-site before each deployment. The SuperCO 2 system has a standard multi-position valve and alternates between equilibrator air, atmospheric samples, and three gas standards. The timing of the valve switching was set so that each of the three CO 2 standards (CO 2 mixing ratios (χxCO 2 ) of 255.1, 409.9, and 566.4) were flushed through the system at 200 ml min -1 for 5 minutes every 6 hours. Standard gases were certified at the University of Manitoba against standards obtained from Environment and Climate Change Canada, and are thus traceable to World Meteorological Organization standards. The 210 SuperCO 2 system has an integrated air pump configured to make atmospheric measurements; these measurements were not used due to contamination from the ship's exhaust. The SuperCO 2 system also measureds atmospheric pressure P (atm) .
Variables Measurements from the underway system were logged every minute.: χxCO 2 and related variables were logged to the computer of the SuperCO 2 system, the data recorded by the ECO were logged to a separate data file, and the latitude and 215 longitude recorded with a Garmin GPS16X-HVS GPS unit were logged to a Campbell Scientific CR300 data logger. The CO 2 measured by the system was were processed following SOP 5 (Dickson et al., 2007) SOP 5. Partial pressure of CO 2 (χpCO 2 ) is measured the output provided by by the Licor 850 in the SuperCO 2 system. , this is converted to the gas mixing ratio of CO 2 (xCO 2 ) using the pressure in the Licor (P licor ). The χxCO 2 is calibrated using a piecewise linear interpolation in time with the three standards. As there was no dryer the equilibrator is assumed to be at full humidity, tThe partial pressure 220 in the equilibrator (pCO 2 (equ) ) is was therefore then determined in the equilibrator (pCO 2 (equ) ) using thecalculated by multiplying by atmospheric pressure P (atm) and assuming full humidity. pCO 2 (equ) is was converted to pCO 2 (1m) using the T (equ) , SST (1m) , and the fractional temperature change constant of (Takahashi et al., 1993). The depth of the seawater inlet was validated each year by comparing the thermosalinograph salinity and the in situ temperature sensor with surface temperature and salinity from CTD rosette measurements at the surface. As tThere was no in situ temperature sensor during the 2017 and 225 2018 field seasons, the warming was then characterised from T( equ ) and CTD rosette measurements following Ahmed et al.   (Takahashi et al., 1993). This results resulted in an additional 2.09% and 2.74% uncertainty in pCO 2 (1m) , these values are 235 similar to the 2% uncertainty reported by  following the same method. The standard system configuration during the four cruises is detailed above; changes from this configuration during specific cruises are detailed in the supplementary materials (Table S2). There are several logistical aspects associated with deploying, operating, and maintaining an underway pCO 2 system in a remote Arctic location on a small vessel like the RV Martin Bergmann; this is discussed further in supplementary materials. 245

Calculations: Air Air−-sea CO 2 fluxes
In the absence of a reliable ship-based atmospheric CO 2 record, hourly measurements are were taken from the atmospheric observatory in Barrow Alaska (71.32°N,156.61°W) (K.W. Thoning, 2020;Peterson et al., 1987). Despite the long distance between Barrow and the Kitikmeot Sea (around 1800 km), atmospheric CO 2 should beare quite very similar at both locations as the atmosphere is well mixed for a long residence time gas like CO 2 and both locations are remote northern sites away 250 from biogenic and industrial emissions. To validate this assumption a long term (1985−2019) mean difference of 0.246 μatm was calculated between the hourly measurements at Barrow and weekly atmospheric samples from Alert Nunavut (Lan et al., 2022). Wind speed adjusted to a reference height of 10 m (U 10 ) is was taken from the Qikirtaarjuk Island observatory (Butterworth and Else, 2018) for the 2017 and 2018 field seasons whereas a four times daily record of U 10 from the NCEP-DOE v2 reanalysis product (Kalnay et al., 1996) is was used for 2016 and 2019 field seasons. 255 The air--sea fluxes of CO 2 (F, mmol m -2 d -1 ) is was calculated as F (sea-air) = k W k 0 ΔpCO 2 SF The water phase gas transfer velocity (k w , cm hr -1 ) is was calculated using U 10 and the parameterisation of Nightingale et al. (2000), a unitless Schmidt number (Sc) normalised to a Sc of 660 (Wanninkhof, 2014) is was used to scale k w . k W = (0.222 (U 10 ) 2 + 0.333 (U 10 )) (Sc/660) -1/2 260 ΔpCO 2 (μatm) is the partial pressure difference between the seawater interface and air ΔpCO 2 = pCO2 (sw) -pCO 2(air) . The solubility of CO 2 in seawater (k 0, mol L -1 atm -1 ) is was taken from (Weiss, 1974). A unit scaling factor (SF) of 0.24 is used to convert the units of k w to md -1 . The Schmidt number and solubility are were calculated using the in situ temperature and salinity values adjusted for skin effects (Woolf et al., 2019).

265
Direct measurements of the air-sea CO 2 fluxes (F (sea-air) ) made using the micrometeorological eddy covariance technique (Butterworth and Else, 2018) can bewere used to infer pCO 2(sw) by rearranging the flux equation. That was achieved as follows using pCO 2(air) from the Licor 7200 at the Qikirtaarjuk Island observatory and SST and SSS from a mooring at a depth of 13 m which was 1 km from the tower (Butterworth et al., 2022). An eddy covariance flux footprint is the area over which the eddy covariance measurements correspond to and varies depending on atmospheric conditions. Using the Kljun et 270 al. (2015) footprint model, Butterworth and Else (2018) showed that the footprint of the Qikirtaarjuk Island observatory during spring and summer can be modelled as an ellipse with an upwind axis that varies between approximately 0.75 -2.0 km and a cross-wind axis that varies between 0.1 -0.2 km. The effective flux footprint is however much smaller as over 90% of the flux signal comes from within 100 m of the tower. Uncertainty in the pCO 2(sw) values derived using eddy covariance arises from uncertainty in the flux measurements (hourly uncertainty of ~20% in the Arctic) (Dong et al., 2021a), 275 uncertainty in the gas transfer parameterisation (~ 5-10%) (Woolf et al., 2019), the small uncertainty in the atmospheric pCO 2 value, uncertainties in k 0 and the schmidt number (including uncertainties in SST and salinity inputs from the 13 m mooring).

Results
To facilitate comparisons between the four summertime cruises made in different years, observations have been partitioned into separate oceanographic zones based on the local geography, observational data density, previous pCO 2 (sw) measurements, and proximity to the local carbon observatories ( Figure 4a). Bathurst Inlet and Chantrey Inlet were designated zones based on their large freshwater inputs. The Finlayson Islands and Cambridge Bay are where the 285 Qikirtaarjuk Island observatory and ONC mooring are located, respectively; these regions were also heavily surveyed because the RV Martin Bergmann often returned to port in Cambridge Bay and passed the islands to access Wellington Bay and Bathurst Inlet. Wellington Bay ( Figure 1) is a shallow, partially enclosed basin for which a relatively large amount of data was collected due to annual fish-tagging surveys associated with the local subsistence char fishery (Harris et al., 2020).
All the measurements in the Dease Strait West zone were made in the central channel and are in the same approximate 290 geographically region to those collected by . Most of the measurements in the Queen Maud Gulf zone were made in the west; the box is large enough to include sparse measurements in the central and Northern regions which do not warrant being considered separately.
Observations of temperature, salinity, pCO 2 (sw) , fluorescence, U 10 , and CO 2 flux during the four field seasons are plotted as 295 time series and coloured by the sub-region of the measurement (Figure 4b-4g). Summary statistics (mean, standard deviation, and range) of each variable in each region for all four cruises are presented in Table 1  There was high interannual pCO 2 (sw) variability (Table 1) Figure 4d). There was also high regional variability in large pCO 2 (sw) interannual variability was larger than the observed regional variability each year, for example in 2018 pCO 2 (sw) ranged from 218 μatm to 387 μatm (Table 1) (Table 1).

Discussion 345
Presented in the results above are the multiyear summertime pCO 2 (sw) observations made on RV Martin Bergmann. These data reveal the spatial and inter-annual variability of pCO 2 (sw) near the beginning ofthroughout the open-water season in the Kitikmeot Sea. To maximise the value of the pCO 2 (sw) observations made on RV Martin Bergmann we will now present and discuss these new measurements alongside previous measurements and in the context of our current understanding of the carbonate system in the region. 350

Local scalecomparisons with the ocean carbon observatories
The two local observatories, the ONC mooring in Cambridge Bay and the Qikirtaarjuk Island observatory (Figure 1),, provide measurements throughout the year that are not readily possible with shipboard observations. pCO 2 (sw) is directly measured on the ONC mooring, whereas pCO 2 (sw) is calculated from the flux derived using measurements from the Qikirtaarjuk Island observatory eddy covariance "EC tower". By takingUsing the pCO 2 (sw) observations from these two 355 observatories alongside the new RV Martin Bergmann measurements allows us to construct enables we can create a multiyear timeline of pCO 2 (sw) in the region to be constructed ( Figure 5). It should be noted that the three measurement sources in Figure 5 are not co-located, the Qikirtaarjuk Island observatory on the Finlayson Islands is 35 km west of the ONC mooring (Figure 1) and the Bergmann measurements span a slightly wider area (Figure 2). (Duke et al., 2021)Despite the spatial disparity in these measurements, it should also be acknowledged that for calculations of global CO 2 flux on a 1° x 360 1° grid, the majority of these measurements would fall within the same grid cell. It might be expected that on these sorts of spatial scales the measurements should agree close to perfectly, but that is not not always the case ( Figure 5). It is possible that the SST measured from the 13 m mooring which is used to calculate pCO 2 d is not representative of the surface interface, which would bias the schmidt number and k 0 used in the calculation of pCO 2 (sw) from the tower; yet, even if this were the case, the magnitude of the impact can not explain the larger pCO 2 (sw) differences between the methods (146 μatm). Even though the RV Martin Bergmann measurements are being made close to the surface (at a depth of 1 m), th; howeverIt is possible that themeasured surface ,which would biasused in our calculations; yet, (XX uatm) as this would 380 require (Xu et al., 2021) an extremely large temperature gradient ~5-10 °C between the RV Martin Bergmann SST at 1 m and SST at the interface. The most likely explanation for the differences in pCO 2 (sw) ibetween the two methods is that even though the RV Martin Bergmann measurements are being made close to the surface (at a depth of 1 m), surface stratification in the surfacethis upper meter is driving the differences being observed. The impact of surface stratification on pCO 2 (sw) has been observed elsewhere in the Arctic (Ahmed et al., 2020;Dong et al., 2021b) including for cases where differences can be 385 up to 200 μatm . Surface stratification in the Kitikmeot Sea is caused by melting of first-year sea ice and the large freshwater input by rivers (, whichrivers alone can alone contribute an estimated 70 cm of freshwater to the surface annually); (Williams et al., 2018). The fact that the EC tower pCO 2 (sw) was higher than the RV Martin Bergmann pCO 2 (sw) would suggest that this is due to river induced stratification, as river Arctic riverine water is often typically higher in pCO 2 (sw) (Cai et al., 2010), indeed this was true between the 30 th June and 2 nd July 2017 for Freshwater Creek (Manning et al., 390 2020). (Kljun et al., 2015).. Interestingly, the predicted pCO 2 (sw) from the EC tower showsshow a peak in early August 2017 and a downwards trend through to the end of August, something that is also seen in the ship ship-based pCO 2 (sw) observations ( Figure 5b). Similarly, the predicted pCO 2 (sw) from the EC tower increases in August 2018 at a similar rate to the increase seen in the shipboard pCO 2 (sw) observations (2.22 μatm d -1 ) (; Figure 5c). The fact that similar trends can be observed in the RV Martin Bergmann and the EC tower pCO 2 (sw) does suggest that seasonal trends in the region are 395 detectable with both methods. However, tThe generalThe disagreement between the RV Martin Bergmann measurements and those from the EC tower highlights the need for year-round pCO 2 (sw) observations in the flux footprint of the EC tower.

Formatted: Superscript
Additionally, interfacial pCO 2 (sw) measurements and vertical profiles may help reconcile the observed disparities seen between the two measurement sources of data.

400
On the other hand, TtThere is good agreement in the pCO 2 (sw) values between the EC tower and the ONC mooring in May, June, and October 2017 (average pCO 2 (sw) EC tower for October 11 th to 14 th is 320 μatm and is 311 for October 24 th to 30 th ) ( Figure 5b) and in May and June 2018 (Figure 5c). The breakdown of stratification at the end of the ice-free summer period and over the winter (Xu et al., 2021) may explain the good agreement between the EC tower and the ONC mooring at these times. In June 2017, the two systems diverge. Specifically,, the pCO 2 (sw) at the ONC mooring decreases due to a spring 405 bloom (Duke et al., 2021), whereas pCO 2 (sw) from the EC tower is not impacted, as does not. As the bloom in Cambridge Bay is caused by wastewater discharge (Back et al., 2021) it might be expected that this signal would not detectable at the EC tower. 365.85 and 370.02406 μatm respectively. pCO 2 (sw) at tThe ONC mooring on 10:00 3 rd August was 326 μatm.11 and on 12:40 415 12 th August was 371 μatm.03. Disagreement between the ONC mooring and the RV Martin Bergmann here may be due to the different intake depths of the two systems. Stratification may mean the ONC mooring is not always representative of pCO 2 (sw) closer to the air−-sea interface, especially during for parts of ice free period of the year; however, , CTD profiles from 2018 do indicate there is stratification in the surface 10 m in the summer (Back et al., 2021). The spring 2016 measurements from the ONC mooring show that pCO 2 (sw) was high in the spring leading into thatthe shipummer field 420 season, and the trend towards increasing pCO 2 (sw) due to warming is was captured in August 2016 by both the ONC mooring and the RV Martin Bergmann observations.
Combining the data sources in this way highlights the value of having these different observatories to look at multiyear changes. The observatories provide context to the variability in the summertime pCO 2 (sw) measurements from local ships. 425 The patchinessintermittence of the measurements from the ONC mooring and the Qikirtaarjuk Island observatory reflects the challenges in making these novel measurements in an extreme environment. Knowledge about how to run themoperate both observatories and prevent instrument outages means that future measurements will build towards much needed continuous and complementary multiyear datasets.

Regional scalesspatial variability in the underway data 430
Focusing back on the RV Martin Bergmann data, there is clear evidence of spatial regional variability in the underway data.
pCO 2 (sw) was typically lower by ~20−-40 μatm in the small bays (Cambridge Bay, and Wellington Bay) and larger inlets surveyed (Bathurst Inlet, ChantryChantrey Inlet) compared to the central channel (e.g., Dease Strait West, the Finlayson Islands, and Queen Maud Gulf) ( Table 1). The reason for relatively lower pCO 2 (sw) in the Bays and Inlets is not readily apparent. For this trend to be driven by temperature, Using the 4.23 % °C -1 constant from Takahashi et al. (1993) it is 435 possible to test whether the pattern of lower pCO 2 (sw) in the Bays and Inlets was driven by temperature, for a representative 360 μatm value for pCO 2 (sw) to be ~20−40 μatm lower it would need to be between 1.35 and 2.78 °C colderthe bays and inlets would need to be ~2°C colder, which was not observed. Rather than being colder,In fact, many of these regions, such as Bathurst Inlet, were warmer, and which would usuallybased on the Takahashi et al. (1993) constant, would thus have a predictedpredict higher pCO 2 (sw) .
Although the fluorescence sensor was not robustly calibrated against in situ 440 measurements, the fluorescence signal was consistent with previous measurements that showed the region to have widespread low primary production at the surface (Martin et al., 2013). Inspite of the lack of Even though these regions did not have consistently higher surface chlorophyll-a concentrations, biological production at depth can not be ruled out as an explanation for lower pCO 2 (sw) in the bays. For example, wastewater discharge has been shown to cause a deep (20 -30 m) chlorophyll bloom in Cambridge Bay (Back et al., 2021). A large under ice (Arrigo et al., 2012;Mundy et al., 2009) or ice 445 edge (Perrette et al., 2011) phytoplankton bloom earlier could also explain lower values in the season could also explain lower summertime pCO 2 (sw) values in these bays and inlets that persists into summer. . It is also possible that these regional differences are driven by regional freshwater inputs; all four identified regions are fed by rivers and there are sharp salin ity transitions of ~5 that point to the existence of mixing and fronts (Figure 4c). Rivers are typically thought to be highly oversaturated supersaturated in pCO 2 (sw) in the Arctic due to organic matter breakdown (Teodoru et al., 2009), potentially 450 contributing to so it might be expected that there would be higher pCO 2 (sw) in these bays and inlets. However, whilst the freshwater local rivers are high in pCO 2 (sw) (Manning et al., 2020), they are typically unbuffered and thus have much lower DIC relative to seawater. Whilst the average values for riverine TA (565 μmol kg -1 ) and DIC (533 μmol kg -1 ) in the CAA are low, maximum measured values for TA (2,272 μmol kg-1) and DIC (2,252 μmol kg-1) values can be as high or higher than in seawater, depending on the bedrock type underlying the drainage basin (Brown et al., 2020). Dilution by low pCO 2 455 (sw) ice meltwater does lower pCO 2 (sw) (Cai et al., 2010;Meire et al., 2015), so it may be that, a greater impact of sea ice meltwater in these bays and inlets may be contributing to the lower observed pCO 2 (sw) .
The ONC mooring is located in Cambridge Bay in shallow water (sensor depth 79 m), at this depth the mooring is not impacted by the Freshwater Creek plume which is detectable at < 2 m (Duke et al., 2021;Manning et al., 2020). It is still 460 unclear how much of an impact being located in the isolated Bay has on the representativeness of these measurements for the Kitikmeot region. As the RV Martin Bergmann travelled into and out of the Bay multiple times during the four years of observations, differences in pCO 2 (sw) measured in the Bay and outside the Bay may help identify whether the ONC mooring site is representative of the region as a whole. All transects into and out of Cambridge Bay are shown in Figure 6. Two subregions are designated, inside the Bay and outside the Bay, here pCO 2 (sw) from the RV Martin Bergmann was averaged every 465 two days for which there was were data available ( Table 2). As seen in Table 2, pCO 2 (sw) was largely similar (typically < ±15 μatm) inside and outside of the bay with pCO 2 (sw) typically <±12 μatm. On the 17 th August 17 th , 2017, pCO 2 (sw) was much higher (39.6 33.29 μatm) in the Bay. As , as measurements are similar before (8 th /9 th ) and after (19 th /20 th ) the 17 th August, it would appear that this difference is caused by this would point to this being due to a process something only occurring happening in the Bay; possibly related to the river plume. Overall, the agreement between the measurements inside 470 and outside of the Bay is encouraging and suggests that pCO 2 (sw) in Cambridge Bay, at least broadly agrees with that in the main Channel. Without more information, it is difficult to conclude whether the mooring is truly representative of the wider Kitikmeot Sea.

Interannual variability and large scale seasonal trends
We have identified local scale differences between the pCO 2 (sw) values from the RV Martin Bergmann and , the ONC, and the Qikirtaarjuk Island observatories and regional scales differences between the bays and inlets and the main channel. The very low pCO 2 (sw) values (261 289 μatm) observed in 2018 (Table 1) could be caused by a combination of low SST (1m) , springtime CO 2 depletion by primary production and recent dilution by sea ice melt (Else et al., 2012;Ahmed et al., 2021;Geilfus et al., 2015) or river runoff at where salinities are >20 (Cai et al., 2010), yet .. Without identifying clear chemical signatures that can be attributed to each process it is difficult to we cannot say with certainty which of these 490 processes was most important in producing these low pCO 2 (sw) values. As the ice breakup was late in 2018 (resulting in samples collected shortly after breakup), it can be assumed that surface ocean CO 2 exchange with the atmosphere was limited by the ice cover until just before these measurements were made, as sea ice is essentially impermeable to gases (Loose et al., 2011;Butterworth and Else, 2018). Additionally, the presence of sea ice through to the end of Julycover in 2018 prevented warming ofmeant there was far less warming of the surface seawater as(average SST (1m) was = 4.32 °C low 495 in 2018), this explaination rules out surface cooling lowering SST (1m) and thus pCO 2 (sw) . Light penetrating through sea ice between March and June could have driven primary production below and within the ice . Indeed, an increase in under-ice chlorophyll a concentration together with a draw-down of surface nutrients between April to June 2018 indicate supported under-ice phytoplankton production during this period (Dalman et al., 2019). However, chlorophyll a concentrations did not exceed 0.6 g L -1 , as production is limited by surface nutrient availability in the region (Back et al., 500 2021) . It is likely that the melting sea ice stratified the surface and diluted surface pCO 2 (sw) as has been observed in other parts of the Arctic Ahmed et al., 2020); low surface ocean salinity values in the first weeks of the survey support this. Measurements several weeks into the 2018 cruise show that pCO 2 (sw) increased quickly in the following weeks (to XXX ~300 μuatm), likely due to a combination of air-sea exchange and the observed surface warming. Interestingly,  did not observe pCO 2 (sw) values below 300 μatm at any point during the five years of passing through 505 the Kitikmeot Sea. Therefore, 2018 could be an anomalously low year for pCO 2 (sw) , or the discrepancy could highlight the fact that  did not make any measurements immediately after sea ice breakup in this region in the region.
Furthermore, the discrepancy could be influenced by the difference in sampling depth of the two pCO 2 systems between the CCGS Amundsen (7 m) and RV Martin Bergmann (1 m). The best way to assess the impact of the sampling depth would be to take simultaneous measurements via the ships intake and at the interface as in Ho and Schanze (2020). 510 The processes driving the changes in pCO 2 (sw) that have been discussed above can be partially quantified using back of the envelope calculations with several assumptions. The individual impact on pCO 2 (sw) of dilution by melting sea ice, air-sea gas exchange, net community production (NCP) and warming of seawater are explored across the region for the month of August in 2018. 515 Firstly, the impact of dilution by sea ice melt can be tested by assuming conservative mixing of TA, DIC, and salinity as in (Meire et al., 2015). For the seawater mixing endmember, surface TA (2034.43 μmol kg -1 ) and DIC (1958.82 μmol kg -1 ), SST (-1.38°C) and salinity (28.64) are taken from seawater bottle data on the 18 th June 2018 (Duke et al., 2021) alongside surface silicate (4 μmol L -1 ) and phosphate (0.5 μmol L -1 ) from 2018 (Back et al., 2021). Average values from spring 2019 520 for TA (356.60 μmol kg-1), DIC (340.24 μmol kg-1) and salinity (4.56) in first year sea ice are used for the sea ice mixing end member . Taking a sea ice thickness of 1.8 m and assuming water expands 10% when it freezes to form sea ice, would suggest melting all the sea ice would add 1.64 m of water, to reach the final salinity of 24.82 (the average recorded value from the RV Martin Bergmann measurements) with conservation of salinity would require this freshwater to mix with 8.68 m of seawater. The ratio of these two depths can then be used to provide the predicted TA 525 (1768.26 μmol kg -1 ), and DIC (1702.05 μmol kg -1 ), for the seawater at a salinity of 24.82. Using CO2SYS (Lewis et al., 1998;Van Heuven et al., 2011) the calculated pCO 2 (sw) value for the initial seawater conditions is 369 μatm and after the melting of sea ice pCO 2 (sw) is 302 μatm,under constant temperature.. The dissociation constants of carbonic acid used in the CO2SYS calculations were those by Mehrbach et al. (1973) refit by Dickson and Millero (1987) and the HSO 4 dissociation constants from (Dickson, 1990). For these calculations temperature was kept constant. As the average measured pCO 2 was 530 289 μatm in 2018, sea ice melt and conservative mixing of seawater can account for the majority (66.75 μatm) of the total change in pCO 2 (80 μatm) from the initial seawater conditions in 2018.
Secondly, using the same approach as DeGrandpre et al. (2020) an estimate of the individual and combined impact of air-sea exchange and NCP on pCO 2 (sw) can be made using a simple model with the following assumptions: taking the average flux 535 from the 2018 cruise of -16.79 mmol m -2 d -1 , a 40 m mixed layer depth for Dease Strait (Xu et al., 2021), with a density of ( 996.49 kg m -3 ) from SST (-1.38°C) and salinity (28.64), an upper estimate of NCP (6.63 g C m −2 ) which is the average integrated rate for Cambridge Bay during the open water season of 2018 (Back et al., 2021). With this configuration a change in DIC (+0.0176 μmol kg -1 hr -1 ) due to air-sea exchange and NCP (-0.003 μmol kg -1 hr -1 ) can be calculated. Taking the combined change in DIC (+0.0142 μmol kg -1 hr -1 ) and substituting it into CO2SYS (Van Heuven et al., 2011;Lewis et al., 540 1998) with the same initial TA, DIC, silicate and phosphate concentrations as on the 18 th June 2018, produces a pCO 2 (sw) change of 0.0459 μatm hr -1 for one time step. Scaling this DIC change for the month of August, with no other changes in the system, would increase pCO 2 (sw) by 36.31 μatm (with NCP component reducing pCO 2 (sw) by 9.4 μatm and air-sea exchange component increasing pCO 2 (sw) by 47.34 μatm).
, we can derive. Here we make for the 2018 cruise:air-sea CO2 ;determined ;and. From this we can calculate theS due to air-545 sea exchange and NCP (Van Heuven et al., 2011;Lewis et al., 1998), Thirdly, using the 4.23 % °C -1 Takahashi et al. (1993)   can be used to model pCO 2 (sw) , where X is weeks since ice breakup. Both models predict very similar pCO 2 (sw) in the first seven weeks after sea ice breakup, the average difference between the models for this period is 8.01 μatm. The models differ more after 7 weeks after sea ice breakup,. At 14 weeks after sea ice breakup, the model of  predicts a pCO 2 (sw) that is 81.2 μatm higher than the model fit to the RV Martin Bergmann pCO 2 (sw) observations. Fundamentally, understanding the drivers of the large interannual variability in pCO 2 (sw) seen in the Kitikmeot Sea requires an understanding 605 of the interconnected processes involved and their timing. The interannual variability SST (1m) and salinity are comparable to the modelling results of Xu et al. (2021), by e. Expandingexpanding on that modelling work with a complex biogeochemical model that can incorporate all the known processes impacting pCO 2 (sw) , it may make it be possible to accurately reproduce the pCO 2 (sw) observations in this region.

The Kitikmeot Sea as a sink for atmospheric pCO 2
The RV Martin Bergmann pCO 2 (sw) measurements indicate that the region is a CO 2 sink in early August, most years (Table   1). At sea ice breakup, low SST (1m) values are low impacts increases solubility resulting inand there are large ΔpCO 2 gradients between the surface ocean and the atmosphere, these conditions persist for several weeks after sea ice breakup.
Warming of the surface ocean when pCO 2 (sw) is slightly undersaturated is the likely cause of pCO 2 (sw) oversaturation 615 supersaturation in some years, resulting in the region becoming a net source once the saturation threshold is met later in the season.. Decreasing SST (1m) at the end of the ice-free season lowers pCO 2 (sw) producing a second period when there are larger ΔpCO 2 gradients between the ocean and the atmosphere, this is partially identifiable in the RV Martin Bergmann measurements from late in 2017. Whilst not demonstrable with the RV Martin Bergmann measurements, cooling decreased SST (1m) at the end of the ice-free season should lower pCO 2 (sw),) thereby providing a second period when there are large 620 ΔpCO 2 gradients between the ocean and the atmosphere.. The magnitude of the ΔpCO 2 , and thus the size of the CO 2 sink throughout the summer, appears to not only to be driven by time since ice breakup, but also by the absolute surface ocean pCO 2 (sw) value at the time of ice breakup.  used remote sensing products to identify this region as a net sink when the flux is integrated over the full ice-free period,. oOur measurements corroborate these findings.

625
The large variability in pCO 2 (sw) measured in the four years of observations highlights the fact that, in the Arctic, single cruises in only part of the ice-free season are likely not capturing the full seasonal variability in these regions. Many pCO 2 (sw) observations in the Arctic are temporally biased towards the middle of the ice-free season, when moving vessles through the Arctic Oocean is easiest.. As these single cruises are the only measurements in many of these regions in databases like SOCAT (Bakker et al., 2016), they could result in a biased regional flux estimates in these regions. In particular, iIt should 630 be acknowledged that the majority of the CAA is not included in the state of the art observational based products (Landschützer et al., 2020).

Conclusions
The ONC mooring and EC tower both provide similar pCO 2 (sw) values in spring and autumn showing good agreement between the two platforms. Measured pCO 2 (sw) from the EC tower was sometimes similar to much higher than what was 635 measured from the RV Martin Bergmann whereas at other times it was much higher. , but Ssimilar seasonal trends which are likely related to temperature were seen in pCO 2 (sw) from the EC tower and the RV Martin Bergmann both data sources which may be attributable to surface stratification caused by sea ice melt and riverine flowsinputs. Comparing measurements collected by the RV Martin Bergmann in and out of Cambridge Bay indicates that Cambridge Bay surface ocean pCO 2 (sw) is not drastically different fromsimilar to that in the main channelDease Strait in August. This may indicate that pCO 2 (sw) at the 640 ONC mooring may be broadly representative of Dease Strait.
The Kitikmeot Sea was a CO 2 sink from the atmosphere or a very week weak CO 2 source over the summers of 2016 -2019, consistent with previous measurements from . The CO 2 sink was highly variable from year to year at the beginning of August (average observed fluxes of +3.58, -2.96, -16.79 and -0.57 0.41, -7.70, -21.26 and -2.08 mmol m -2 645 d -1 during the 2016, 2017, 2018, and 2019 cruises respectively) with average pCO 2 (sw) as low as 288.55 261.19 ± 19.70 μatm and as high as 445.08 403.65 ± 44.44 μatm. pCO 2 (sw) was much lower in 2018 due to the much lower SST (1m) that year. The magnitude of the air−-water ΔpCO 2 throughout the summer appears to be controlled by the absolute pCO 2 (sw) value at the time of ice breakup. Low pCO 2 (sw) values increase in August due to exchange with the atmosphere and warming broadly following the predicted trends using the model developed by . In years where pCO 2 (sw) is high when ice 650 breakup occurs, warming can cause a period of slight pCO 2 (sw) oversaturation supersaturation in summer, in these situations the magnitude of this oversaturation supersaturation is likely moderated by the air air-sea flux reducing pCO 2 (sw) . pCO 2 (sw) was found to be ~20−-40 μatm lower in the Bays and Inlets that were surveyed; this could be driven by increased freshwater inputs into these isolated regions. Lower pCO 2 in the bays and inlets would represent an observational bias in the CAA-wide surveys . Local fFreshwater fluxes into the southern CAA are much greater than elsewhere in the CAA, 655 meaning that this bias might be more prominent in the Kitikmeot Sea. Further observations in these regions may complement the basin-level pCO 2 mapping.
These findings provide a more nuanced picture of the considerable inter-annual variability in pCO 2 (sw) observed during repeat cruises in the same region, underscoring how much may be missed by relying on data collected during one-off cruises 660 along the dynamic Arctic coasts. In particular, tThe pCO 2 (sw) at the time of ice melt is very important as it dictates the magnitude and direction of the flux for much of the ice-free period; however,. Aa. A better understanding of pCO 2 (sw) through the ice covered period is needed to help unravel the seasonal and interannual variability.

Acknowledgements
Parts of this research were completed on or adjacent to Inuit Owned Lands under the authority of the Nunavut Land Claim 665 Agreement, and the work was licensed by the Nunavut Research Institute. We thank the Ekaluktutiak Hunters and Trappers Organization and the community of Cambridge Bay for their hospitality and support of this project. Richard Sims was Emingak for all their assistance in the field, and Sophia Ahmed for her work on data interpretation.

Author Contributions
This manuscript was written by RPS, all co-authors made contributions to the final paper. BTGE installed the underway 680 system at the start of each field season. The performance of the underway system was monitored by BTGE with help from SFJ, SFG, KAB, PJD and RPS. RPS organised and processed the data. RPS made the figures and interpreted the results and with support from MA and BTGE. BJB analysed the data from the EC tower and provided that data for this paper. PJD provided the data from the ONC mooring. BTGE, KAB, CJM and WJW secured have been were central in planning the cruise programme. BTGE oversaw completion of the work. 685

Data and code availability
The processed underway data from the RV Martin Bergmann which is the new data described in this paper is available in the supplement as .mat files. The raw and processed underway data from the RV Martin Bergmann data will also be available via Zenodo. The final processed data will also be submitted to the Surface Ocean Carbon Atlas (SOCAT). The wind data and inferred seawater pCO 2 data from the EC tower are included in the supplement as .mat files. The ONC mooring data is freely 690 available at https://data.oceannetworks.ca/home. The AMSR2 sea ice data https://seaice.unibremen.de/data/amsr2/asi_daygrid_swath/n3125/ the NCEP winds https://psl.noaa.gov/data/gridded/data.ncep.reanalysis2.html and the atmospheric pCO 2 data from Barrow ftp://aftp.cmdl.noaa.gov/data/greenhouse_gases/co2/in-situ/surface/ which were used in this paper are all freely available from their online repositories. Processing code and the code needed to reproduce the figures was written in Matlab 2016a. 695 The code is provided in the supplement and is also available at https://github.com/Richard-Sims/Sims_2022_Bergmann_pCO2.